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 |
|---|---|---|---|---|---|
Stop codon readthrough ( SCR ) occurs when the ribosome miscodes at a stop codon . Such readthrough events can be therapeutically desirable when a premature termination codon ( PTC ) is found in a critical gene . To study SCR in vivo in a genome-wide manner , we treated mammalian cells with aminoglycosides and performed ribosome profiling . We find that in addition to stimulating readthrough of PTCs , aminoglycosides stimulate readthrough of normal termination codons ( NTCs ) genome-wide . Stop codon identity , the nucleotide following the stop codon , and the surrounding mRNA sequence context all influence the likelihood of SCR . In comparison to NTCs , downstream stop codons in 3′UTRs are recognized less efficiently by ribosomes , suggesting that targeting of critical stop codons for readthrough may be achievable without general disruption of translation termination . Finally , we find that G418-induced miscoding alters gene expression with substantial effects on translation of histone genes , selenoprotein genes , and S-adenosylmethionine decarboxylase ( AMD1 ) .
To complete synthesis of a mature protein , ribosomes must terminate translation accurately at the end of each coding sequence . Translation termination requires recognition of the normal termination codon ( NTC ) in the A site of the ribosome by release factors eRF1 and eRF3 in eukaryotes , a process distinct from the RNA-RNA mediated decoding of mRNAs by tRNAs during translation elongation ( Schuller and Green , 2018 ) . Following stop codon recognition by release factors , the nascent peptide is hydrolyzed by eRF1 releasing the mature protein product ( Frolova et al . , 1999; Zhouravleva et al . , 1995 ) and ribosomes are subsequently removed from the mRNA by recycling ( Barthelme et al . , 2011; Pisarev et al . , 2010; Shoemaker and Green , 2011 ) . Normally , termination is a highly efficient process , with an estimated accuracy of greater than 99% ( Floquet et al . , 2012; Harrell et al . , 2002; Namy et al . , 2001 ) ensuring maintenance of proteome fidelity . Nonsense mutations perturb the process of translation by insertion of a premature termination codon ( PTC ) within the coding sequence ( CDS ) of a gene . Upon encountering a PTC , a ribosome will terminate translation prematurely resulting in the production of a truncated peptide , which typically lacks proper functionality , and may even exert dominant-negative effects ( Miller and Pearce , 2014 ) . To guard against the negative consequences of mRNAs harboring PTCs , the conserved nonsense-mediated decay ( NMD ) machinery , working together with the ribosome , identifies premature stop codons and targets the message for decay ( Celik et al . , 2015; Kim and Maquat , 2019 ) . Discrimination of normal and problematic termination contexts must occur independently of the nucleotide sequence of the stop codon since identical stop codons ( UAA , UAG , and UGA ) signal translation termination at both NTCs and PTCs . In mammals , a strong signal for NMD derives from the position of the stop codon relative to that of a protein complex known as the Exon-Junction-Complex ( EJC ) deposited upstream of each splice junction during splicing of the mRNA ( Le Hir , 2000; Singh et al . , 2012 ) . While NTCs are typically found in the terminal exon of protein coding genes , PTCs often are found in upstream exons , and in these cases are recognized as aberrant when the ribosome encounters a termination codon upstream of a deposited EJC . As nonsense mutations account for approximately 11% of inherited genetic disorders in humans ( Mort et al . , 2008 ) , targeted treatments for these particular mutations could substantially alleviate human disease . One class of compound proposed as a treatment for nonsense mutations are collectively known as nonsense-suppression therapeutics ( Keeling et al . , 2014; Lee and Dougherty , 2012 ) . Acting at the level of translation , compounds in this class force the ribosome to ‘read through’ a PTC and continue translation thereby restoring synthesis of full-length protein . Typically , such stop codon readthrough ( SCR ) involves a process in which a near-cognate tRNA ( nc-tRNA ) base pairs with a termination codon , forcing the ribosome to continue elongation instead of terminating translation ( Brody and Yanofsky , 1963; Smith et al . , 1966 ) . Achieving such specificity for PTC readthrough without globally disrupting termination at NTCs remains a critical challenge for nonsense suppression therapies and will require a more complete understanding of translation termination and stop codon readthrough in different sequence contexts . While translation termination is generally the predominant reaction at stop codons , termination efficiencies do vary considerably between different stop codon contexts . Many factors have been reported to influence the probability of termination , readthrough , or frameshifting ( where the ribosome slides on an mRNA , changing the frame of translation ) including the identity of the stop codon and surrounding sequence contexts ( Anzalone et al . , 2019; Bonetti et al . , 1995; Floquet et al . , 2012; Harrell et al . , 2002; McCaughan et al . , 1995; Namy et al . , 2001 ) , proximal RNA structures ( Firth et al . , 2011; Steneberg and Samakovlis , 2001 ) , RNA modifications ( Karijolich and Yu , 2011 ) , presence of RNA binding proteins ( Amrani et al . , 2004 ) , and availability of aminoacylated nc-tRNA ( Beznosková et al . , 2019; Blanchet et al . , 2015; Roy et al . , 2015 ) . Intriguingly , high rates of SCR have been documented in diverse organisms including viruses ( Li and Rice , 1993; Wills et al . , 1991 ) , yeast ( Namy et al . , 2003; Williams et al . , 2004 ) , flies ( Dunn et al . , 2013 ) , and humans ( Loughran et al . , 2014; Loughran et al . , 2018 ) . Many sequences promoting readthrough show evolutionary conservation suggesting functional importance ( Jungreis et al . , 2011 ) . Some organisms even differentially decode all three stop-codons as sense or stop in a manner thought to depend on proximity to the polyA tail ( Heaphy et al . , 2016; Swart et al . , 2016; Záhonová et al . , 2016 ) . Numerous documented examples of substantial SCR ( on the order of 15% ) demonstrate the potential for restoring significant levels of full-length protein in the context of nonsense mutations . Despite extensive study , comprehensive understanding of the rules dictating readthrough efficiencies of stop codons has remained elusive . To explore the processes of translation termination and stop codon readthrough , we utilized a class of compounds known as aminoglycosides ( AG ) . AGs have been abundantly characterized in bacterial systems where they bind the decoding center of the ribosome ( Carter et al . , 2000; Fourmy et al . , 1996; Moazed and Noller , 1987 ) and promote miscoding ( Davies and Davis , 1968; Pape et al . , 2000 ) . Similarly , in eukaryotic systems , a subset of the AGs have been shown to promote miscoding ( Palmer et al . , 1979 ) by binding in the decoding center of the ribosome ( Garreau de Loubresse et al . , 2014; Prokhorova et al . , 2017 ) . Again , in these systems , the manner in which AGs increase the likelihood of SCR is by promoting the accommodation of nc-tRNAs ( Burke and Mogg , 1985; Howard et al . , 1996; Manuvakhova et al . , 2000 ) , and simultaneously blocking the action of termination factors ( Eyler and Green , 2011 ) . While a correlation between basal and AG-stimulated SCR frequencies has been reported for a collection of stop codon contexts ( Floquet et al . , 2012 ) , the extent of global perturbations on translation termination by AGs has not been documented . Here , we performed ribosome profiling ( Ingolia et al . , 2011 ) to systematically investigate the activities of AGs in promoting readthrough of stop codons genome-wide . This approach allows for an unbiased examination of translation termination for all stop codons in their native sequence contexts . We observe broad stimulation of SCR following treatment with AGs that is especially robust for G418 . Using these compounds , we uncover a general role for termination codon identity , as well as surrounding sequence contexts , in determining genome-wide rates of SCR . G418 stimulates readthrough at multiple classes of stop codons including PTCs , NTCs , and 3′UTR termination codons ( 3′TCs ) . Importantly , we find that G418 more potently induces readthrough of 3′TCs relative to NTCs . Finally , we define several biological processes that are disrupted by high levels of G418-induced stop codon readthrough including translation of histone mRNAs , selenoproteins , and S-adenosylmethionine decarboxylase 1 ( AMD1 ) .
We initially investigated the ability of AGs to promote readthrough of a PTC using a dual-luciferase reporter assay . To directly compare activity of various AGs , and additionally optimize AG concentrations for downstream ribosome profiling experiments , we developed a reporter expressing Firefly ( FLuc ) and Nano Luciferase ( NLuc ) from a bidirectional CMV promoter ( Figure 1A ) . Normal protein synthesis was measured by the expression of full-length FLuc , and SCR was measured by the expression of NLuc when a PTC was inserted at position R154X ( UGA ) . While expressing each luciferase from a separate mRNA can introduce variability in RNA levels between messages , this strategy avoids many of the pitfalls that can confound interpretation of bicistronic reporters for studying stop codon readthrough ( Loughran et al . , 2017; Terenin et al . , 2017 ) . Stimulation of ribosome readthrough was tested for six different AGs ( depicted in Figure 1—figure supplement 1 ) : G418 , gentamicin , paromomycin , neomycin , tobramycin , and amikacin . To compare induction of ribosome readthrough by these AGs , HEK293T cells were transiently transfected with the reporter , treated with AGs , and measured 24 hr later for luciferase activity . Consistent with the role of AGs as general protein synthesis inhibitors ( Blanchard et al . , 2010 ) , production of FLuc decreased with increasing concentrations of AGs ( Figure 1B ) . Despite this reduction in FLuc synthesis , AGs promoted NLuc synthesis on the R154X reporter , revealing strong stimulation of PTC readthrough ( Figure 1C ) . When we calculated the level of normalized readthrough by normalizing the ratio of NLuc to FLuc for AG-treated cells relative to untreated cells ( Figure 1D ) , all AGs tested , with the exception of tobramycin , were able to stimulate readthrough of the R154X PTC . Of the six AGs examined , G418 and gentamicin potently stimulated PTC readthrough in this assay , while paromomycin , neomycin , and amikacin showed lower but detectable stimulation of PTC readthrough . While reporter assays provide a powerful tool for studying readthrough of a given stop codon , the throughput of these assays is inherently limited , their output can be biased by the identify of the amino acid incorporated at a given stop codon ( Xue et al . , 2017 ) , and most importantly , synthetic constructs do not adequately capture termination codons in their endogenous sequence contexts . To address these limitations , we performed ribosome profiling to globally examine readthrough of NTCs in live mammalian cells . By monitoring the presence of ribosomes translating downstream of stop codons , we can identify individual readthrough events independent of protein output . Based on the results from our luciferase reporter , we performed ribosome profiling in HEK293T cells treated with AGs at concentrations that maximized NLuc signal while inhibiting no more than 50% of FLuc signal ( Figure 1B , dashed-line ) . Following a 24 hr treatment , cells were lysed and sequencing libraries were prepared for two biological replicates for each treatment condition . Densities of ribosome-protected fragments ( RPFs ) in coding sequences ( CDSs ) showed strong correlations between replicates ( R > 0 . 99 for all samples , Figure 2—figure supplement 1 ) . Biological replicates were pooled for further analysis to increase read depth for these initial samples . To measure readthrough of stop codons genome-wide , we performed an average gene ( or metagene ) analysis , aligning all transcripts at their annotated stop codons and calculating normalized ribosome densities in this window . Examining translation in untreated cells ( Figure 2A , black-line ) reveals strong three-nucleotide periodicity in coding regions upstream of stop codons as expected for elongating ribosomes in the CDS . At the stop codon itself , RPFs are enriched ( Figure 2A , black arrow ) as observed in previous studies in multiple organisms , reflecting that termination is slower than elongation ( Ingolia et al . , 2011; Schuller et al . , 2017 ) . Finally , ribosome density drops precipitously following the stop codon due to the high efficiency of termination , resulting in few ribosomes present in 3′UTRs . Two immediate differences emerge when comparing global termination in untreated cells to AG-treated cells . First , the peak of terminating ribosomes decreases in cells treated with G418 and paromomycin ( Figure 2A , orange and green arrows ) . Second , there is an increase in ribosome density in the 3′UTRs of cells treated with AGs , with the largest increases observed for G418 and paromomycin ( Figure 2B ) . The reduction in the peak of ribosomes at stop codons is consistent with increased rates of decoding of NTCs by nc-tRNAs . And , as a corollary , increased rates of readthrough relative to termination at NTCs predict increased density of ribosomes in 3′UTRs . Importantly , the magnitude of the reduction in the peak of terminating ribosomes correlates with the increased density of 3′UTR ribosomes for paromomycin and G418 . A 10-minute G418 treatment resulted in equivalent levels of 3′UTR ribosomes as the 24 hr treatment , and treatment with a higher concentration of G418 led to a dose-dependent increase in 3′UTR ribosome density ( Figure 2—figure supplement 2A ) . Next , we compared the overall density of ribosomes in a window immediately upstream ( −147 to −16 nts , CDS ) and downstream ( +5 to +100 nts , 3′UTR ) of the stop codon to quantify the extent of 3′UTR ribosome enrichment upon AG treatment ( Figure 2C ) . At the concentrations tested here , G418 promotes the highest levels of 3′UTR ribosome density ( preserving 20% of CDS density ) , followed by paromomycin , and then gentamicin and neomycin while amikacin and tobramycin only minimally promote readthrough above basal levels . Relative levels of readthrough as measured by luciferase activity ( Figure 1D ) generally agreed with 3′UTR ribosome densities for the AGs . We next evaluated SCR on a per transcript basis . To accomplish this task , we defined a metric referred to here as the Ribosome ReadThrough Score ( RRTS , Figure 2—figure supplement 2B ) where we calculated the density of ribosomes in the region of the 3′UTR ( with the assumption that 3′UTR RPFs were generated by SCR ) between the NTC and the first in-frame 3′TC , and divided this value by the density of ribosomes in the CDS for every annotated transcript . For each protein-coding gene , only a single transcript isoform possessing the 3′-most termination codon was analyzed to ensure that RPFs aligning downstream of the stop codon reflect translation of the 3′UTR rather than translation of the CDS of an alternatively spliced transcript ( see Methods regarding transcript selection ) . RRTS values were well correlated between biological replicates ( R > 0 . 7 , Figure 2—figure supplement 2C ) . As NTC readthrough is normally a rare event , the vast majority of transcripts showed zero RPFs in 3′UTRs of untreated cells in these initial libraries . Treatment with AGs broadly increased RRTS values ( Figure 2D ) indicating that stimulation of NTC readthrough by AGs is a global phenomenon . In addition to these observed effects of aminoglycoside treatment on translation termination , we also examined effects on translation initiation and elongation . For both of these phases of translation , G418 again proved the most disruptive of the AGs tested here . G418 treatment led to increases in ribosome occupancy at initiation codons ( Figure 2—figure supplement 3A ) while other AGs showed only modest effects on this process . The G418-dependent enrichment of initiating ribosomes was comparable between the 10 min and 24 hr time points , and was increased with a higher concentration of G418 . To measure perturbation of translation elongation , we calculated average codon occupancies for all 61 sense codons by comparing the ribosome density at a specified codon relative to the ribosome density of the CDS ( Figure 2—figure supplement 3B ) , and averaged these measurements across all occurrences of the given codon . AG treatment globally disrupted codon occupancies to varying extents resulting in a substantial loss-of-correlation between AG-treated and untreated cells . Of note , G418 treatment revealed both amino acid-specific changes wherein codon occupancies on all glycine ( orange ) and aspartic acid ( cyan ) codons were increased , as well as codon-specific changes wherein codon occupancy of a single isoleucine ( green ) codon ( AUA ) was decreased while occupancy of the other two codons ( AUC and AUU ) was unaffected . These data reveal that aminoglycosides interfere with every phase of translation and provide additional evidence for the general inhibition of translation observed using luciferase assays ( Figure 1B ) . As termination is normally a very efficient process , ribosomes are rarely found in 3′UTRs . While NTC readthrough events result in translation of 3′UTRs , several additional reactions could also be responsible for the increased 3′UTR ribosome density observed following AG treatment . Previous reports have demonstrated that ribosomes enter 3′UTRs due to readthrough ( Dunn et al . , 2013 ) , frameshifting ( Michel et al . , 2012 ) , failures in ribosome recycling or rescue ( Guydosh and Green , 2014; Mills et al . , 2016 ) , or reinitiation of translation ( Young et al . , 2015; Young et al . , 2018 ) . To establish the primary mechanism responsible for the increased abundance of the ribosomes found in 3′UTRs upon AG treatment , we generated additional deeper ribosome profiling datasets in G418-treated cells as this AG showed the strongest induction of 3′UTR ribosomes . We utilized the increased sequencing depth of these libraries ( Figure 3—figure supplement 1A and B ) to determine whether the ribosomes found in 3′UTRs were derived from readthrough events or other competing pathways . To verify that sequencing reads aligning in 3′UTRs derive from ribosomes , we compared read length distributions in 3′UTRs and CDSs . Ribosomes in mammalian cells protect , on average , 29–31 nucleotide fragments ( Ingolia et al . , 2011; Wolin and Walter , 1988 ) of RNA while other species , such as RNA binding proteins ( Ji et al . , 2016 ) , protect more variable fragment sizes . When we compare the sizes of 3′UTR and coding sequence reads , we see strong agreement in the distribution of fragment lengths consistent with the argument that 3′UTR reads derive from ribosomes ( Figure 3—figure supplement 1B ) . A useful diagnostic for validating bona fide translation in ribosome profiling data is the presence of three-nucleotide periodicity , a signature of elongating ribosomes . Examination of a metagene plot of 3′UTRs in the deep ribosome profiling libraries reveals strong three-nucleotide periodicity in G418-treated cells , but not in untreated cells ( Figure 3A ) . By mapping ribosomal A sites of RPFs to single-nucleotide positions , we next calculated the proportions of ribosomes translating in each of the three possible reading frames . As anticipated , both untreated and G418-treated cells show strong enrichment of ribosomes translating in the frame ( Frame 0 ) of the CDS . Strikingly , while the reading frame of the CDS is completely lost in untreated cells ( Figure 3B - equal representation of RPFs in all three frames ) , we see strong conservation of frame in the 3′UTRs of G418-treated cells ( Figure 3C ) . As the alternative processes that might be responsible for generating 3′UTR ribosomes ( including frameshifting , recycling failure , and reinitiation ) should not result in reading frame maintenance , these data strongly indicate that G418 increases 3′UTR ribosome density by stimulating readthrough of NTCs . Ribosomes that read through the NTC and continue translation are predicted to encounter subsequent in-frame stop codons in 3′UTRs at some frequency . Indeed , over 90% of mRNAs have an in-frame stop codon found in the 3′UTR ( Figure 3—figure supplement 2A ) . Given the presence of these in-frame stop codons , ribosomes that read through the NTC will once again face a decision of whether to terminate translation , frameshift , or read through the downstream stop codon . As termination remains the predominant reaction at stop codons , even when cells are treated with G418 , ribosome density is predicted to significantly decrease downstream of any in-frame stop codon . For example , examination of HSPA1B ( Figure 3D ) in cells treated with G418 reveals that 3′UTR ribosome density for this transcript remains relatively stable between the NTC and the first in-frame 3′TC , but ribosome density substantially drops after every in-frame 3′TC ( Figure 3D , red boxes ) and not at out-of-frame stop codons ( Figure 3D , gray boxes ) . A gene containing no in-frame 3′TCs , such as APRT ( Figure 3D , right ) , shows ribosome density throughout the entire 3′UTR as well as ribosomes that reach the poly A tail ( data not shown ) . We next analyzed termination globally at the first stop codon encountered in 3′UTRs as a function of reading frame ( Figure 3E ) . In untreated cells , small peaks of ribosomes are present at the stop codons in all three reading frames suggesting that low levels of translation normally occur in all frames of 3′UTRs ( black traces ) . In contrast , when cells are treated with G418 , translation proceeds primarily in the same reading frame as the coding sequence . Upon reaching the in-frame ( Frame-0 ) 3′TC , ribosomes are enriched at the stop codon , followed by a major depletion of downstream ribosome density , similar to what we observe at NTCs ( Figure 2A ) . For out-of-frame 3′TCs ( −1 and +1 frames ) ribosome density is maintained downstream of stop codons indicating that most ribosomes are not translating in these frames . Given that a majority of ribosomes reaching in-frame stop codons terminate translation , we next asked whether the gradual decline of ribosome density observed in Figure 3A can be explained by the fraction of transcripts that have encountered in-frame stop codons . As distance from the NTC increases , so does the probability of encountering a termination codon ( Figure 3—figure supplement 2B ) . Satisfyingly , this trend mirrors that of the global decline in ribosome density ( Figure 3A ) as distance from the stop codon increases . Taken together , we conclude that G418-induced 3′UTR ribosomes derive primarily from SCR events . Numerous reports have demonstrated that the identity of the stop codon influences the likelihood of stop codon readthrough using reporter assays across diverse systems ( Schueren and Thoms , 2016 ) . To systematically compare the influence of stop codon identity on SCR , we inserted each possible PTC ( UAA , UAG and UGA ) at six additional positions in our NLuc reporter ( W12X , V40X , E51X , H88X , G113X , W134X , and R154X ) . Importantly , these inserted stop codons were distributed relatively equally along the length of the mRNA and only inserted at positions not predicted to disrupt protein secondary structures ( Lovell et al . , 2016 ) and thus luciferase reporter activity . Cells were transfected with these reporters and treated or not treated with G418 to evaluate readthrough at each stop codon . Readthrough efficiencies varied considerably between stop codon positions in a manner that did not appear to be a function of PTC proximity to the poly ( A ) tail ( Figure 4A with G418 , Figure 4—figure supplement 1A without G418 ) . Consistent with previous findings , a general trend emerged that for a particular position , UAA stop codons were least likely to be read through while UGA stop codons were most likely to be read through ( readthrough likelihood: UGA > UAG > UAA ) . It is possible that the variability in SCR between the different PTCs is a consequence of different sequence contexts surrounding each stop codon . To query readthrough as a function of stop codon identity genome-wide , we again utilized the deeper ribosome profiling data sets . We sorted transcripts by stop codon identity and measured RRTS values for all transcripts in the absence and presence of G418 ( Figure 4B ) . While in general , these data provide support that readthrough of NTCs is a rare event under normal conditions regardless of the identity of the stop codon , we did find that UAA and UGA stop codons were significantly more likely to be readthrough than UAG stop codons ( p=1 . 62×10−3 and p=4 . 84×10−7 , Mann-Whitney U ) in untreated cells . More importantly , in G418-treated cells , we identify the same trend as in the reporter assays ( UGA > UAG > UAA ) but with strong significance as determined by our genome-wide analysis . Numerous reports have identified an influence of the nucleotide immediately following the stop codon ( +4 position ) on SCR using reporter assays . Based on structural studies revealing compaction of the mRNA in the A site of the ribosome during termination ( Brown et al . , 2015 ) , stop codons may be more appropriately considered as four-nucleotide ( 4 nt ) rather than three-nucleotide ( 3 nt ) signals as simply derived from the codon table for tRNA decoding . Indeed , as previously reported ( Ingolia et al . , 2011 ) , measurement of read-length distributions of ribosome protected fragments ( RPFs ) at stop codons reveal that ribosome profiling accurately captures the protection of an additional nucleotide of mRNA by the ribosome at stop codons ( Figure 4C ) . Next , we compared the effects of the twelve possible 4 nt stop codon signals on NTC readthrough ( Figure 4D ) for cells treated with G418 . Generally , a purine at the +4 position decreases likelihood of NTC readthrough , likely due to stabilization of the eRF1-mRNA interaction through base stacking between the mRNA and rRNA ( Brown et al . , 2015 ) . For each individual 3 nt stop codon , the presence of a C in the fourth position significantly increases the likelihood of readthrough relative to all other nucleotides ( p<0 . 01 for all stop codons , Mann-Whitney U ) . And , combining the highest-readthrough three-nucleotide codon ( UGA ) with the highest-readthrough +4 position ( C ) results in the highest observed readthrough for UGAC compared to all other 4 nt termination codons in both untreated ( Figure 4—figure supplement 1B ) and G418-treated cells ( Figure 4D ) . In light of the influence of the +4 nucleotide on stop codon readthrough , we next asked whether additional positions surrounding the stop codon might also influence stop codon readthrough . We analyzed the sequence context covered by the footprint of the ribosome , fifteen nucleotides upstream and twelve downstream ( −15 to +15 ) of the termination codon using two different approaches . As a first approach , we weighted each stop codon context by the RRTS for that transcript and performed a one-sided two-sample Student’s t-test using kpLogo ( Wu and Bartel , 2017 ) to test whether a given nucleotide at each position increased or decreased the likelihood of stimulating SCR . P values were adjusted using the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) and plotted using Logomaker ( Tareen and Kinney , 2019 ) for untreated ( top ) and G418-treated ( bottom ) cells ( Figure 5A and magnification of G418 data in Figure 5—figure supplement 1A ) . As an alternative approach , we used a linear regression model to calculate regression coefficients for every nucleotide in the sequence window ( Figure 5—figure supplement 1B ) , a strategy that has been previously applied to readthrough of luciferase reporters ( Schueren et al . , 2014 ) . These distinct approaches yielded striking agreement on the identity of sequence features that yield increased SCR . Looking broadly across this defined sequence window , several features emerge that influence SCR . Generally , the presence of A’s or U’s increase SCR probability while C’s and G’s decreases SCR probability , especially in 3′UTRs , for both untreated ( Figure 5A , top ) and G418-treated ( Figure 5A , bottom ) cells . In untreated cells , along with the stop codon , the most influential positions are the +4 and +5 positions; in G418-treated cells , the influence of the +4 nt and +5 nt are evident , but are now diminished relative to the stop codon . In agreement with previous observations , we also identified the tobacco mosaic virus readthrough signal CAAUUA as the strongest readthrough promoting hexanucleotide signal in the 3′UTR in G418 treated cells ( Anzalone et al . , 2019; Namy et al . , 2001; Skuzeski et al . , 1991 ) . Considering the influence of C’s and G’s in promoting efficient translation termination relative to A’s and U’s , we next examined the nucleotide usage of protein coding transcripts 40 nts upstream and 60 nts downstream of the stop codon ( Figure 5B ) . As nucleotide usage in the CDS is constrained by the genetic code , we see that many patterns in this region repeat with three-nucleotide periodicity . Intriguingly , despite the inherent A/U richness of 3′UTRs , C’s and G’s are enriched in the immediate vicinity of the NTC . As distance from the NTC increases , C’s and G’s occur less frequently until U’s and A’s eventually become the predominant nucleotides in 3′UTRs . A notable exception to these trends occurs at the +4 position where G’s and A’s are enriched while C’s and U’s are depleted , in agreement with the large influence of these nucleotides on SCR at this position ( Figure 5A ) . Taken together , it seems that evolution has finely tuned 3′UTRs to promote efficient translation termination . We next asked what the likelihood of SCR would be at a different class of termination codon , the collection of first in-frame stop codons in the 3′UTR ( 3′TCs ) . While 3′TCs are rarely translated under normal conditions , the pervasive translation of 3′UTRs in G418-treated cells allowed for direct comparison between SCR efficiency at NTCs and 3′TCs . For this comparison , we calculated relative downstream ribosome density in a window 30 nts upstream and downstream of the stop codon for NTCs and the first in-frame 3′TC , excluding transcripts that overlapped additional stop codons ( Figure 5D , Figure 5—figure supplement 2 ) . In comparison to NTCs , which preserved 33% downstream ribosome density in this window , stop codons in 3′UTRs were more likely to be read through with 46% of the ribosome density maintained downstream of the 3′TC . These observations are consistent with the fact that A’s and U’s are generally enriched in the stop codon contexts of 3′TCs ( Figure 5C ) . It seems likely that purifying selection tightly maintains NTCs but not 3′TCs in mammals ( Belinky et al . , 2018 ) . In addition to investigating ribosome readthrough at NTCs and 3′UTR stop codons , we applied ribosome profiling to study readthrough of a natural PTC . For this set of experiments , we used Calu-6 cells that harbor a nonsense mutation ( R196X – UGAG ) in the tumor suppressor gene TP53 . To initially verify that G418 was able to at least partially restore levels of full-length TP53 protein , we analyzed TP53 protein levels by western blotting . In untreated cells , no detectable band was observed for TP53 , whereas G418 treatment restored full-length TP53 synthesis in a dose-dependent manner ( Figure 6A ) . We also observed production of a second , more prominent band corresponding to truncated TP53 . The higher level of truncated TP53 relative to full-length TP53 suggests that even with high levels of G418 , termination occurs more frequently than readthrough at the PTC of this mRNA . We next treated Calu-6 cells with G418 for 24 hr , generated lysates , and sequenced samples using ribosome profiling and RNA-seq . As with HEK293T cells , G418 stimulated global SCR in Calu-6 cells ( Figure 6—figure supplement 1 ) . Canonical models of EJC-dependent NMD maintain that displacement of EJC’s by translating ribosomes protects mRNAs from decay by NMD . Given that TP53 has only a single PTC , we predicted that readthrough of the PTC , and subsequent translation to the NTC , would displace EJCs and protect this message from decay . We measured mRNA levels of TP53 using RNA-seq ( Figure 6B ) and observed a dramatic 16-fold stabilization of TP53 mRNA levels following G418 treatment . To verify that ribosomes were in fact reading through the PTC , we analyzed ribosome profiling reads on this message for both untreated ( Figure 6C ) and G418-treated ( Figure 6D ) cells . As predicted by low mRNA levels , few RPFs mapped to the TP53 mRNA in untreated cells . Further , all RPFs , save one , were found upstream of the PTC demonstrating little readthrough of the PTC under basal conditions . In contrast , treatment with G418 stimulated readthrough of the PTC , yielding many downstream RPFs , while also enriching reads on the mRNA overall . RPF density upstream of the PTC was greater than that downstream of the PTC in concordance with the western blot analysis ( Figure 6A ) . These observations lend support to a model wherein readthrough of a stop codon displaces RNA binding proteins ( such as the EJC ) , protecting the mRNA from NMD ( Keeling et al . , 2004 ) . To investigate dysregulation of gene expression by G418 , we compared changes in RNA levels ( RNA-seq ) and translation ( ribosome profiling , RPFs ) allowing calculation of ribosome occupancies ( ROs ) by taking the ratio of RPF densities to RNA-seq densities for each transcript ( Figure 7—figure supplement 1; Ingolia et al . , 2009; Xiao et al . , 2016 ) . For the purposes of our discussion , we will treat the RO metric as a reflection of the translational efficiency of a given mRNA . Untreated cells were compared to cells treated with G418 at both the 10 min and 24 hr time points ( Figure 7A and B ) . We reasoned that a 10 min treatment would capture differences in translation before activation of major transcriptional reprogramming . Indeed , we observe induction of the unfolded protein response ( chaperones and foldases ) in the levels of both RNA-seq and RPF data at the 24 hr time point ( purple dots , upper right quadrant Figure 7B ) but not at the 10 min time point . These observations are consistent with the fact that the cell must contend with high quantities of misfolded proteins produced by miscoding events as well as C-terminal protein extensions arising from readthrough ( Oishi et al . , 2015 ) . Additionally , we see translational upregulation of ATF4 at the later time point ( Figure 7B , orange ) , a critical transcription factor known to be translationally activated during stress ( Vattem and Wek , 2004 ) . While the majority of mRNAs are neither dramatically changed in abundance nor differentially translated in response to G418 , there are several interesting outliers that we highlight here . First , the histone mRNAs revealed consistent changes in gene expression . For these mRNAs we see that translation is overall decreased and mRNA levels are stabilized , even after only 10 min of G418 treatment ( Figure 7A/B , red dots ) . As for the majority of genes , we observe robust readthrough of NTCs on these mRNAs on treatment with G418 ( Figure 7C ) . Uniquely , however , we see subsequent translation into a well-characterized conserved hairpin at the terminus of histone 3′UTRs ( Figure 7C , hairpin highlighted in red ) . A second class of genes impacted by G418 treatment is the selenocysteine-containing proteins . Often referred to as the 21st amino acid , selenocysteine tRNA decodes a consensus UGA stop codon as a sense codon , usually in the presence of a downstream SECIS element ( Low and Berry , 1996 ) . In normal conditions , insertion of selenocysteine appears to be the rate-limiting step for translating many selenoprotein mRNAs , as this position represents the largest peak of ribosomes on these messages in untreated cells as demonstrated for two well translated selenocysteine genes ( Figure 7D , black lines marked by green arrow ) . Since G418 potently stimulates readthrough of UGA codons , we wondered whether this AG would alter translation of selenocysteine containing mRNAs . Analyzing expression of selenocysteine genes ( Figure 7A and B , green dots ) revealed that G418 enriches RPFs on several selenocysteine genes relative to mRNA levels . Also , G418 treatment led to a reduction in the peak of ribosomes at UGA selenocysteine codons ( Figure 7D , orange lines ) and increased downstream ribosome density on these mRNAs . Finally , we noticed a substantial increase in RPFs at both time points along with a decrease in RNA levels at the 24 hr time point for AMD1 , a gene whose expression levels are proposed to be regulated by ribosome stalling in both the 5′UTR and 3′UTR . It has been proposed that an upstream open reading frame ( uORF ) which encodes a specific peptide ( MAGDIS* ) stalls ribosomes at its termination codon and thus prevents scanning 40S subunits from reaching the authentic start codon of the downstream CDS ( Law et al . , 2001 ) . We find that G418 treatment reduces the peak of ribosomes stalled at the end of the uORF ( Figure 7E , left ) and dramatically increases the RPF density in the CDS at both the early and late time points . A second level of translational regulation was proposed for AMD1 wherein ribosomes that naturally read through the stop codon are trapped on another peptide-regulated stop codon in the 3′UTR ( Yordanova et al . , 2018 ) . It was further proposed that these accumulated ribosomes might form queues that eventually block initiation of the CDS . As previously observed , we identify the peak of ribosomes stalled at the first in-frame 3′TC of AMD1 in untreated cells ( Figure 7E , right , black arrow ) . Upon G418 treatment , we see that this peak of ribosomes is strongly enriched ( Figure 7E , orange arrow ) , supporting the claim that these stalled ribosomes originate from readthrough of the NTC . We do not , however , observe evidence of queuing ribosomes upstream of this stall site in the 3′UTR , even with the very high level of G418-stimulated readthrough .
Here we used ribosome profiling to examine readthrough of stop codons genome-wide in human cell lines in untreated and aminoglycoside-treated cells . We find that aminoglycosides , and especially G418 , generally disrupted the normally accurate process of translation termination leading to high levels of SCR and , in turn , broad perturbation of several cellular processes . Genome-wide levels of NTC readthrough varied considerably between different stop codons allowing investigation into sequence features driving the differences in SCR on the various stop codons . Examination of both 3 nt and 4 nt termination codons revealed a clear relationship between stop codon identity and SCR in response to G418 treatment; we define genome-wide an increased likelihood of G418-induced SCR with the order UGA > UAG > UAA for the 3 nt stop codons and C > U > A > G for the +4 nt position ( Figure 4B , D ) . In addition to the identity of the termination codon , nucleotides more broadly surrounding the NTC also influence likelihood of SCR , especially at the +5 nt position in untreated cells . Generally , the presence of A’s or U’s near the termination codon increases the probability of SCR while C’s and G’s favor translation termination ( Figure 5A ) . The general observed enrichment of C’s and G’s following NTCs within the first 20–30 nts of 3′UTRs that we document likely results from evolutionary pressure to terminate translation efficiently ( Figure 5B ) . Critically , the first in-frame 3′TCs show higher average levels of SCR in G418-treated cells in comparison to NTCs ( Figure 5D ) . This difference likely arises from the enrichment of A’s and U’s in the stop codon contexts of 3′TCs relative to NTCs , providing further evidence for the important role of stop codon context in regulating translation termination . We also noted that treatment of cells with G418 changed the relative importance of features that impact SCR . In untreated cells , SCR is strongly driven by the identity of the +4 and +5 nucleotide of the stop codon motif ( Figure 5A , top ) . By contrast , in G418-treated cells , SCR is predominantly driven simply by the identity of the stop codon itself ( and in particular by UGA ) ( Figure 5A , bottom ) . We suggest that these striking changes in prediction of readthrough probability between the 3 nt stop codon and the +4 nt contribution could be explained by differences in A site binding mechanisms employed by eRF1 and tRNAs . In untreated cells , where termination by eRF1 is the dominant outcome relative to stop codon readthrough , the sequence context that best stabilizes the eRF1-mRNA interaction effectively predicts the readthrough probability; in particular , the strong contribution of the +4 nucleotide to eRF1 stabilization makes purines ( especially G ) a strong negative correlate with readthrough and pyrimidines ( especially C ) a strong positive correlate , as clearly predicted by the structure ( Brown et al . , 2015 ) . On the other hand , in the presence of G418 when miscoding by nc-tRNAs becomes more favorable , the influence of the 3 nt stop codon that is directly recognized by the nc-tRNA now better predicts SCR probability; the cellular tRNA levels may well contribute to the specificity that we observe . Together , these observations suggest that SCR will be impacted by multiple constraints – the strength of interactions between eRF1 and the stop codon context and the likelihood of decoding a particular stop codon in that particular cellular environment . The high levels of readthrough stimulated by G418 disrupted several biological processes with one of the most substantial consequences being perturbation of histone gene expression resulting in mRNA stabilization and decreased ribosome density ( Figure 7A and B ) . Unique among metazoan mRNAs , histones are not poly-adenylated and instead possess a conserved hairpin structure at the 3′ end of the mRNAs . These hairpins are necessary for all stages of the histone mRNA lifecycle , including translation and mRNA decay ( Marzluff and Koreski , 2017 ) . Due to the massive requirement for histone proteins needed during cell division , exquisite control of protein synthesis and subsequent histone mRNA decay is required as dividing cells transition through the cell cycle . The substantial SCR resulting from G418 treatment results in ribosomes translating into these conserved hairpin structures ( Figure 7C ) . This SCR in turn likely disrupted mRNP complexes that normally stimulate translation initiation ( von Moeller et al . , 2013 ) , consistent with the reduced ribosome density that we observe . Furthermore , displacement of 3′hExo/ERI1 ( the 3′ to 5′ exonuclease required to initiate histone mRNA decay [Slevin et al . , 2014] ) by translating ribosomes may similarly explain the observed mRNA stabilization of these transcripts . Resolution of ribosomes stalled at the ends of histone mRNAs may require the eRF1 homolog PELO , which has been shown to rescue ribosomes from 3′UTRs ( Guydosh and Green , 2014; Mills et al . , 2016 ) , and specifically , promote decay of histone mRNAs ( Slevin et al . , 2014 ) . Interestingly , deletion of PELO in mouse epidermal stem cells increases ribosome occupancy on these messages ( Liakath-Ali et al . , 2018 ) . More generally , the role of ribosome rescue factors in dealing with genome-wide readthrough events remains an outstanding question . In contrast to the histones , ribosome density was substantially increased on the selenoprotein mRNAs following G418 treatment . The large peaks of ribosomes normally stalled at the UGA codon sites for selenocysteine insertion were decreased in G418-treated cells; this likely reflects the fact that UGA now becomes recoded by an nc-tRNA instead of the cognate selenocysteine tRNA . Simple replacement of selenocysteine with other amino acids has been shown to decrease function of several of these proteins ( Axley et al . , 1991; Berry et al . , 1991 ) . Thus , while increased readthrough of UGA codons may generally increase the amount of full-length protein product for these mRNAs , functional levels of selenoproteins may decline with G418 treatment ( Renko et al . , 2017 ) . While selenocysteine insertion is normally specified by an additional RNA sequence element ( SECIS ) at particular UGA codons , the extent to which the general availability of selenocysteine tRNA ( which normally reads UGA in a cognate fashion ) contributes to the relatively higher levels of readthrough of UGA ( C ) stop codons genome-wide remains unclear . One of the mRNAs exhibiting the largest changes in mRNA level and RPF density on G418 treatment was the enzyme AMD1 which is critical in determining the balance between polyamines and S-adenosylmethionine levels in the cell . At the earliest time point , we see AMD1 is the most translationally upregulated mRNA in the transcriptome while at the later time point , translation still remains very high , but mRNA levels decrease substantially ( Figure 7A and B ) . Previous studies have highlighted two distinct translational regulatory features in the AMD1 mRNA . First , there is conserved uORF encoding a specific stalling peptide that prevents termination at the uORF and blocks 40S subunits from reaching the downstream ORF ( Law et al . , 2001 ) . We believe that the increased RPF levels that we observe on AMD1 result from increased readthrough of the uORF stop codon that in turn allows scanning ribosomes to find the downstream start site . Second , previous studies showed striking accumulation of 3′UTR ribosomes in AMD1 which the authors suggested resulted from readthrough of the NTC and led to decreased synthesis of AMD1 and decay of the mRNA through a ribosome queuing mechanism ( Yordanova et al . , 2018 ) . The G418-stimulated SCR on AMD1 similarly led to increased levels of 3′UTR ribosomes at the stall site and reduction in mRNA levels . Importantly , however , we see no evidence of ribosome queuing even under conditions of dramatically increased SCR that might support models of translation initiation inhibition . We wonder whether the observed reduction of AMD1 RNA levels that we observe may instead be mediated by the no-go decay pathway . Recent insights into the processes of cellular mRNA surveillance have converged on collided ribosomes as a minimal signal ( Ikeuchi et al . , 2019; Juszkiewicz et al . , 2018; Simms et al . , 2017 ) for triggering no-go decay and ribosome quality control events . In such a model , possibly as few as two readthrough events could trigger quality control thus providing immediate feedback to downregulate AMD1 expression . Further experiments will be required to explore a potential role of ribosome quality control in regulating expression of AMD1 . Despite the potential for interference with specific processes that we have highlighted here , nonsense-suppression therapeutics provide a promising option to target key disease-causing stop codons for selective readthrough . As such , an understanding of the factors that dictate readthrough probability could help guide efforts in achieving specificity for stimulating readthrough of specific termination codons . While G418 successfully induced readthrough of a PTC in TP53 ( the most frequently mutated gene in cancers [Kandoth et al . , 2013] ) , it simultaneously caused readthrough of the majority of NTCs . Given the generally random nature of nonsense mutations introduced into coding sequences ( for example , when associated with genetic disease ) , stop codon contexts of PTCs are highly variable . Considering that normal stop codons face evolutionary pressure to terminate efficiently , and that we have found that stop codons in 3′UTRs are more likely to be read through than normal stop codons , we predict that most PTCs will exhibit higher levels of SCR than NTCs and thus there may exist a therapeutic ‘Goldilocks’ window in which restoration of the targeted gene could be achieved without global disruption of translation termination at NTCs . This is especially true given that only minimal levels of protein restoration may be required to treat certain disorders ( Keeling et al . , 2014 ) . Our data provide strong support for this possibility . Targeting a central process such as translation without causing deleterious effects remains challenging . General toxicity may result from AGs for multiple reasons including dominant effects from miscoding during elongation or from SCR leading to gain-of-function effects arising from C-terminal extensions ( Freitag et al . , 2012; Schueren et al . , 2014 ) . Moreover , a therapeutic drawback of AGs , and more generally for all compounds that stimulate miscoding , are the general inhibitory loss of function effects on both translation elongation and termination . Targeting translation termination codons through the use of suppressor tRNAs ( Lueck et al . , 2019 ) presents an alternative approach to aminoglycosides with potential for fewer off-target effects as suppressor tRNAs should only impact the fidelity of stop codon recognition for roughly one third of stop codons , but not the overall fidelity of translation . Fortunately , cells also possess the capacity to manage some level of miscoding and SCR by degrading readthrough products ( Arribere et al . , 2016 ) . The increased resolution of the role of stop codon context in regulating translation termination presented here will hopefully aid future development of nonsense suppression therapeutics .
HEK293T cells and Calu-6 cells , purchased from ATCC , were cultured in Dulbecco’s Modified Eagle Medium with high glucose , L-glutamine , and sodium pyruvate , supplemented with 10% fetal bovine serum , certified endotoxin-free ( Thermo Fisher Scientific ) . Cells were grown in a 37 °C cell culture incubator in the presence of 5% CO2 . Antibiotics were not added to the media at any point , with the exception of the AGs tested in the study . Cell lines were purchased directly from ATCC and passaged fewer than 15 times . HEK293T cells tested negative for mycoplasma contamination . AGs ( Sigma – Gentamicin sulfate , Paromomycin sulfate , Neomycin sulfate , Tobramycin , Amikacin sulfate; Thermo Fisher – G418 sulfate ) were prepared by dissolving compounds in either water or 150 mM Tris pH 8 . 0 to prepare 50 mg/mL stock solutions of each compound . Liquid preparations were also purchased for G418 and Gentamicin ( Thermo Fisher Scientific ) . Activity of compounds was not influenced by preparation methods ( data not shown ) , so compounds prepared in 150 mM Tris pH 8 . 0 were used for the majority of experiments . Prior to addition of AGs for ribosome profiling and RNA-seq , AGs were added to media and equilibrated in a cell culture incubator to allow stabilization of pH and temperature before addition to cells . All plasmids in this study were generated using the pcDNA5/FRT/TO vector ( Thermo Fisher Scientific ) . The dual expression cassette containing a shared enhancer and minCMV promoters was subcloned from the pBI-CMV1 ( Clontech ) into the pcDNA5 vector using restriction enzyme digest and Gibson assembly . Nano luciferase ( Promega ) and Firefly luciferase ( a gift from David Bedwell ) were inserted into this construct also using Gibson assembly to generate a template expressing full-length versions of both luciferase constructs . PTCs were then inserted via site-directed mutagenesis using the QuikChange Lightning Multi Site-Directed Mutagenesis Kit ( Agilent ) to generate the constructs tested here . To measure luciferase activity , HEK293T cells were transfected using Lipofectamine 3000 transfection reagent ( Thermo Fisher Scientific ) in 96 well plate format . Lipofectamine 3000 ( 0 . 225 µL/well ) and P3000 reagent ( 0 . 2 µL/well ) was diluted in serum free media ( Opti-MeM with GlutaMAX , Thermo Fisher Scientific ) , and mixed with plasmid DNA ( 100 ng/well ) . Four hours after transfection , AGs were added to each well testing all concentrations in triplicate . After 24 hr of AG treatment , cells were removed from the incubator and equilibrated to room temperature . Luciferase activity was measured using the Nano-Glo Dual-Luciferase Reporter Assay System ( Promega ) . Cells were lysed through addition of Firefly luciferase reagent by direct injection , followed by measurement of Nano luciferase activity using a Synergy H1 microplate reader ( BioTek ) . Cells were lysed in RIPA lysis buffer ( 25 mM Tris pH 7 . 6 , 150 mM sodium chloride , 1% NP-40 , 0 . 1% SDS , 1% Sodium deoxycholate ) supplemented with Roche cOmplete , mini , EDTA-free ( 1 tablet per 3 mL of lysis buffer ) , 100 uM PMSF , 1x mammalian protease inhibitor cocktail ( Sigma ) , after brief wash with 1x phosphate buffered saline ( Thermo Fisher Scientific ) . Cells were scraped in the presence of ice-cold lysis buffer and lysed on ice for 10 min . Lysates were then clarified by centrifugation at 21 , 000 g , 4°C , for 10 min and soluble fractions were flash frozen in liquid nitrogen and stored at −80°C . Total protein levels were quantified using a standard curve by BCA Protein Assay ( Thermo Fisher Scientific ) and absorbance was measured on a Synergy H1 microplate reader ( BioTek ) . 6x SDS loading buffer was added to lysates , samples were denatured at 98°C for 10 min , and 20 µg of protein were loaded to 4–12% Criterion XT Bis-Tris protein gels in 1x XT MES buffer ( BioRad ) . Proteins were then separated using SDS-PAGE at 150V for 1 hr , and transferred to PVDF membrane via Trans-Blot Turbo Transfer System ( BioRad ) . Following blocking with 2 . 5% milk ( Santa Cruz Biotechnology ) dissolved in 1x TBS-T for 1 hr at room temperature , the primary antibody to TP53 ( Santa Cruz Biotechnology , p53 ( DO-1 ) ) diluted 1:100 in 2 . 5% milk , 1x TBS-T was added and membranes were incubated overnight at 4°C with gentle agitation . Membranes were then washed 3x with 1x TBS-T , incubated with HRP-conjugated secondary antibody ( goat-anti-mouse IgG #32430 , Thermo Fisher Scientific ) diluted 1:1000 for 1 hr at room temperature , and washed again 3x with 1x TBS-T . HRP activity was probed by adding SuperSignal West Femto Maximum Sensitivity Substrate ( Thermo Fisher Scientific ) , and chemiluminescence was detected using a G:BOX ( Syngene ) . Membranes were stripped using OneMinute Advance Western Blot Stripping Buffer ( GM Biosciences ) , reprobed with β-Actin-HRP ( Cell Signaling Technologies ) diluted 1:5000 for 1 hr at room temperature , and imaged again for detection of luminescent signal . Ribosome profiling samples were prepared as previously described Ingolia et al . ( 2012 ) with several modifications . AGs were added to cells cultured in 10 cm tissue culture dishes for either 24 hr , or 10 min prior to harvest . Following removal from the incubator , cells were briefly dislodged by scrapping , transferred to 15 mL conical tubes , and pelleted via centrifugation at room temperature for 5 min at 400 g . Cycloheximide was not added to cells at any point prior to lysis . Importantly , we omitted the 1x PBS-wash commonly performed during sample harvesting , as this dramatically decreased the stimulation of SCR otherwise observed for AGs in this assay ( data not shown ) . Media was aspirated and cells were lysed in 200 µL of ice-cold lysis buffer ( 20 mM Tris-Cl pH 7 . 4 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT , 100 µg/mL cycloheximide , 1% Triton X-100 , 2 . 5 U/mL Turbo DNase ( Thermo Fisher Scientific ) ) on ice for 10 min . Lysates were clarified via centrifugation at 21 , 000 g , 4°C , for 10 min . The soluble ribosome-containing supernatant was removed , and separate aliquots were flash-frozen with liquid nitrogen for later quantification , ribosome profiling , and RNA-seq . Sample lysates were quantified using the Quant-iT RiboGreen RNA Assay ( Thermo Fisher Scientific ) and fluorescence ( excitation 485 nm , emission 530 nm ) was measured using a Synergy H1 microplate reader ( BioTek ) . 20 µg of total RNA was diluted to a final volume of 300 uL using polysome buffer ( 20 mM Tris-Cl pH 7 . 4 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT , 100 µg/mL cycloheximide ) and digested with 750 U of RNase I ( Ambion ) for generation of ribosome protected fragments . After 1 hr of digestion at 25°C while shaking at 400 RPM on a thermomixer , digestions were stopped with 10 uL of SUPERase*In ( Thermo Fisher Scientific ) . Nuclease-treated lysates were underlaid with 0 . 9 mL of sucrose dissolved in polysome buffer in 13 × 51 mm polycarbonate ultracentrifuge tubes , centrifuged at 100 , 000 RPM , 4°C , for 1 hr in a TLA 100 . 3 rotor using a Beckmann-Coulter Optima MAX ultracentrifuge , and RPFs were extracted using the miRNeasy mini kit ( Qiagen ) . The extracted RNA was size-selected by denaturing PAGE using a 15% TBE-Urea gel , and fragments corresponding to 15–35 nucleotides ( or 26–35 nucleotides for select libraries ) were excised . RNA fragments were dephosphorylated using T4 PNK ( New England Biolabs ) , and ligated to a 3′ oligonucleotide adapter containing a unique molecular identifier ( UMI ) hexanucleotide degenerate sequence using T4 RNA ligase 2 – truncated ( New England Biolabs ) for 3 hr at 37°C . Ribosomal RNA was depleted using RiboZero ( Illumina ) omitting the final 50°C incubation step in an effort to reduce contamination of short fragments of rRNA . RNA was then reverse-transcribed to cDNA using Superscript III ( Thermo Fisher Scientific ) using RT primer with a second four-nucleotide UMI sequence , RNNN . cDNA was then circularized using circLigase ( Lucigen ) and amplified by PCR using Phusion high-fidelity polymerase ( New England Biolabs ) . PCR-amplified libraries were quantified with a Bioanalyzer 2100 ( Agilent ) using the High Sensitivity DNA kit and pooled in equimolar ratios . Sequencing was performed using a HiSeq2500 ( Illumina ) at the Johns Hopkins Institute of Genetic Medicine . Total RNA was extracted from 50 or 100 µL of lysate using 1 mL of TRIzol ( Thermo Fisher Scientific ) . Following precipitation , the RNeasy kit ( Qiagen ) was used to clean-up RNA prior to library preparation . Starting with 1 . 0 µg of total RNA , RNA-seq libraries were prepared using the low-throughput TruSeq Stranded Total RNA Library Prep Gold kit ( Illumina ) . As for ribosome profiling , cDNA libraries were quantified using a Bioanalyzer 2100 ( Agilent ) with the High Sensitivity DNA kit and sequenced with a HiSeq2500 ( Illumina ) . Following alignment to the hg38 genome , reads were assigned to a custom transcriptome annotation based on the GENCODE ( version 30 ) ( Frankish et al . , 2019 ) annotations for protein coding genes . Protein coding genes were initially filtered , and all transcripts without completed coding sequences or UTRs were discarded . All protein coding transcripts where then compared , and a single transcript was selected for each gene based on the following procedure for genes with multiple annotated coding transcripts . All transcripts with the 3′ most stop codon were initially selected , ensuring that all 3′UTRs did not overlap with the CDS of alternatively spliced transcripts when quantifying SCR . Genes with multiple transcripts sharing this stop codon were then filtered , first selecting genes with primary APPRIS transcripts ( Rodriguez et al . , 2013 ) , and then alternative APPRIS transcripts , followed by inclusion in the consensus CDS gene set ( CCDS ) ( Pujar et al . , 2018 ) , and lastly selecting transcripts with the longest coding sequence for genes without APPRIS or CCDS transcripts for this 3′ most stop codon . For genes still containing multiple isoforms , transcripts were finally filtered by choosing transcripts with the shortest 3′UTRs , and lastly selecting those with the shortest 5′UTRs . After selecting a specific transcript , each transcript was checked for overlap with all transcripts of the nearest gene both upstream and downstream on the same chromosome strand as the transcript being queried and transcripts overlapping any nearby transcripts were discarded . Transcripts with UTRs overlapping annotated pseudogenes were also discarded , as these were observed to bias RRTS calculation for some genes ( data not shown ) . In total , 17 , 937 transcripts remained after filtering procedures and were used for downstream analyses . The 5′ ends of all aligned reads , between 15 and 40 nucleotides in length , were then assigned to protein coding transcripts without data normalization . These uniquely assigned reads were summed to calculate the total quantity of reads mapped to valid transcripts . Reads were assigned a second time to calculate normalized read counts in reads per million ( RPM ) by dividing raw counts by millions of mapped reads for every read length . Mapping to ribosomal A or P sites was next performed by shifting RPF reads from the 5′ end to the position in the center of the A site or P site codon . Density was calculated for all full-length ( 28–35 nt reads ) and empty A site ( 20–23 nt reads ) ( Wu et al . , 2019 ) ribosomes , using reads mapping to the start codon to calibrate the correct shift of each read length for every sequencing library . These measurements generally agreed with previous reports ( Wu et al . , 2019 ) corresponding to offsets for full-length reads of {28:[16] , 29:[16] , 30:[16] , 31:[17] , 32:[17] , 33:[17] , 34:[17] , 35:[17]} and {20:[16] , 21:[16] , 22:[17] , 23:[17]} for empty A site ribosomes . All analysis was performed using full-length ribosome footprints ( 28–35 nt ) . Average gene plots were calculated for regions surrounding the start or stop codons as previously reported ( Schuller et al . , 2017; Wu et al . , 2019 ) . For each transcript , ribosome density at every position was normalized by the overall density of reads mapping to the coding sequence for that transcript . Genes with features of insufficient length ( start codon: 100 nt upstream to 150 nt downstream , stop codon: 150 nt upstream to 100 nt downstream ) were discarded . For analysis in Figure 2C , 3′UTR density was calculated by dividing normalized 3′UTR ( defined as the region from +5 to +100 ) density by normalized CDS ( positions −147 to −16 ) density and this ratio is displayed as a percent . Regions surrounding the stop codon were excluded to avoid biases due to the large density of ribosomes at this position . For every transcript RRTS was calculated by dividing the density of ribosomes in the 3′UTR between the NTC and first in-frame 3′TC by the density of ribosomes in the coding sequence . First , transcripts with fewer than 128 mapped reads were discarded to limit noise from lowly expressed genes ( Ingolia et al . , 2009 ) . To calculate CDS densities , the first 18 nucleotides and last 15 nucleotides of the CDS were excluded from analysis to avoid bias from the large peaks at start and stop codons , and mRNA lengths were adjusted accordingly . For calculation for 3′UTR RPF densities between the NTC and 3′TC , transcripts with fewer than 5 codons between the NTC and 3′TC were discarded , and the first 6 nucleotides following the NTC were not included in these calculations . To test for significant differences between RRTS values as a function of stop codon identity , we used the Mann-Whitney U test ( Mann and Whitney , 1947 ) . To perform log-transformation of RRTSs , scores of zero were assigned to the arbitrarily small value of 2−15 to facilitate plotting , but discarded from statistical analysis when calculating Pearson correlations of log-transformed data . First , the position of all 61 sense codons was determined across all annotated transcripts . Then , to calculate average codon occupancies ( or pause scores ) , the ribosome density within the 3 nt window of a queried codon ( provided that the codon was not within the first or last two codons of the mRNA ) was divided by the average ribosome density for the CDS of each transcript ( excluding the first 15 and last 15 nts ) . This ratio was then calculated for all instances in the transcriptome of a queried codon , and the average of these ratios was determined to be the codon occupancy for that codon . The fraction of full length RPFs ( 28–35 nts ) were assigned to the frame of translation in the CDS ( frame 0 ) , or the two other two possible frames ( frame −1 , frame +1 ) . For calculation of reading frame in Figure 3B–C , normalized ribosome densities presented in metagene analysis ( Figure 3A ) were used to compute reading frames for the CDS and 3′UTR . For calculation of reading frame relative to the first 3′TC found in 3′UTRs presented in Figure 3E , transcripts were sorted by the frame of the first 3′TC encountered in the 3′UTR . Total reads , normalized by sequencing depth , were summed in the window from 12 nt on either side of the stop codon . Transcripts with additional stop codons in this window were discarded . For comparison between NTCs and first in-frame 3′TCs presented in Figure 5D , normalized ribosome densities were again calculated ( as in Figure 2A and B ) in a window spanning 30 nts on either side of each stop codon , for all transcripts except those that ( 1 ) lacked any in-frame 3′TCs , ( 2 ) the first in-frame 3′TC occurred within the first 30 nts of the 3′UTR , ( 3 ) had fewer than 30 nts downstream of the first in-frame 3′TC , or ( 4 ) had additional in-frame 3′TCs in the 30 nt window downstream of the 3′TC . Normalized ribosome densities were calculated upstream and downstream of the stop codons , excluding the codon immediately before and after the stop codon ( plotted in Figure 5—figure supplement 2 ) . The ratio of downstream density divided by upstream density was plotted ( Figure 5D ) and a paired t-test was performed testing the difference between two biological replicates . Stop codon contexts in the region spanning the footprint of the ribosome ( defined as 15 nts upstream and 12 nts downstream of the stop codon ) was retrieved for all transcripts along with the RRTS for that transcript . Any transcripts with 3′UTRs shorter than 12 nts were discarded . For all positions in this sequence window , we used kpLogo ( Wu and Bartel , 2017 ) , to perform one-sided two-sample Student’s t-tests to ask whether a given nucleotide increased or decreased RRTS values relative to all other nucleotides at this position . We used the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) to control the false discovery rate and plotted the P values of the nucleotides in the stop codon context using Logomaker ( Tareen and Kinney , 2019 ) . Regression coefficients for each position within the defined window 15 nt upstream and 12 nt downstream of the stop codon were estimated using a regularized least-squares ( ‘ridge’ ) regression ( Hoerl and Kennard , 1970 ) of RRTS values with sequence contexts of all annotated transcripts , similar to a previous study using luciferase reporter data ( Schueren et al . , 2014 ) . The sequences surrounding the stop codon were formalized as binary vectors with 27 × 4 positions ( A , C , G , or U ) and three positions for the stop codon ( UAA , UAG , UGA ) . The absence ( 0 ) or presence ( 1 ) of each nt was indicated for each sequence . Regression coefficients were calculated using the sklearn . linear_model . Ridge function from the sklearn python package ( Pedregosa et al . , 2011 ) where X is 111 x d ( and d is the number of transcripts ) and y is the vector for RRTS values for each transcript . A value of 103 was used for alpha , as this minimized residual error . Regression coefficients were plotted using Logomaker ( Tareen and Kinney , 2019 ) . Nucleotide frequencies were calculated in a window surrounding the NTC ( 40 nt upstream to 60 nt downstream ) discarding all transcripts with 3′UTRs shorter than 60 nt , and also for the first in-frame 3′TC ( 12 nt upstream to 12 nt downstream ) discarding transcripts with 3′UTRs shorter than 12 nt downstream of the 3′TC in addition to transcripts with other 3′TCs in this window . The fraction of each nucleotide at every position ( stop codons were collapsed to a single position ) was then calculated and plotted using Logomaker ( Tareen and Kinney , 2019 ) . Differential expression of RNA-seq and ribosome profiling datasets were analyzed using DESeq2 ( Love et al . , 2014 ) ( data not shown ) . Logarithmic fold changes were estimated using the ‘apeglm’ package ( Zhu et al . , 2019 ) . For estimation of differences in translation efficiency , we used Xtail ( Xiao et al . , 2016 ) with default parameters . All statistical analysis was performed using the SciPy package in python ( Jones et al . , 2001 ) . Statistical details are described in figure legends and available in Supplementary file 3: Statistical Test Results . Asterisks in figures denote significance levels of p>0 . 05 = ns , p<0 . 05 = * , p<0 . 01 = ** , p<0 . 001 = *** , and p<0 . 0001 = **** . Raw sequencing data and count tables for each sample were deposited in the GEO database ( GSE138643 ) . Data analysis was performed using custom software written in Python 2 . 7 and R 3 . 4 . 4 available at https://github . com/jrw24/G418_readthrough ( Wangen , 2020; copy archived at https://github . com/elifesciences-publications/G418_readthrough ) . | Many genes provide a set of instructions needed to build a protein , which are read by structures called ribosomes through a process called translation . The genetic information contains a short , coded instruction called a stop codon which marks the end of the protein . When a ribosome finds a stop codon it should stop building and release the protein it has made . Ribosomes do not always stop at stop codons . Certain chemicals can actually prevent ribosomes from detecting stop codons correctly , and aminoglycosides are drugs that have exactly this effect . Aminoglycosides can be used as antibiotics at low doses because they interfere with ribosomes in bacteria , but at higher doses they can also prevent ribosomes from detecting stop codons in human cells . When ribosomes do not stop at a stop codon this is called readthrough . There are different types of stop codons and some are naturally more effective at stopping ribosomes than others . Wangen and Green have now examined the effect of an aminoglycoside called G418 on ribosomes in human cells grown in the laboratory . The results showed how ribosomes interacted with genetic information and revealed that certain stop codons are more affected by G418 than others . The stop codon and other genetic sequences around it affect the likelihood of readthrough . Wangen and Green also showed that sequences that encourage translation to stop are more common in the area around stop codons . These findings highlight an evolutionary pressure driving more genes to develop strong stop codons that resist readthrough . Despite this , some are still more affected by drugs like G418 than others . Some genetic conditions , like cystic fibrosis , result from incorrect stop codons in genes . Drugs that promote readthrough specifically in these genes could be useful new treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2020 | Stop codon context influences genome-wide stimulation of termination codon readthrough by aminoglycosides |
The hantavirus envelope glycoproteins Gn and Gc mediate virion assembly and cell entry , with Gc driving fusion of viral and endosomal membranes . Although the X-ray structures and overall arrangement of Gn and Gc on the hantavirus spikes are known , their detailed interactions are not . Here we show that the lateral contacts between spikes are mediated by the same 2-fold contacts observed in Gc crystals at neutral pH , allowing the engineering of disulfide bonds to cross-link spikes . Disrupting the observed dimer interface affects particle assembly and overall spike stability . We further show that the spikes display a temperature-dependent dynamic behavior at neutral pH , alternating between ‘open’ and ‘closed’ forms . We show that the open form exposes the Gc fusion loops but is off-pathway for productive Gc-induced membrane fusion and cell entry . These data also provide crucial new insights for the design of optimized Gn/Gc immunogens to elicit protective immune responses .
Hantaviruses ( order Bunyavirales , family Hantaviridae , genus Orthohantavirus ) persistently infect rodents throughout the world . When transmitted to humans they can cause serious disease such as hemorrhagic fever with renal syndrome and hantavirus pulmonary syndrome ( HPS ) with case fatalities up to 10% and 40% , respectively ( Krüger et al . , 2015 ) . Yet there are no efficient preventive nor therapeutic measures approved against these diseases . As other members of the Bunyavirales order , their RNA genome is single stranded with negative polarity , composed of three segments . The medium ( M ) segment encodes a membrane-anchored polyprotein precursor GPC , which is processed by host cell signal peptidases to generate glyoproteins Gn and Gc ( Löber et al . , 2001 ) . Virion-like particles ( VLPs ) are formed when GPC is expressed in the absence of other viral proteins ( Acuña et al . , 2014 ) , indicating an important role of the glycoproteins in virion budding and in cell exit of the progeny . Authentic virions and VLPs have been shown to project spikes organized in a square lattice ( Acuña et al . , 2014; Martin et al . , 1985 ) . The work of Hepojoki et al . ( 2010 ) revealed that Gn/Gc species can be covalently crosslinked on the surface of virions and suggested oligomeric models for spike assembly based on the characterization of detergent-solubilized spikes . Electron cryo-tomography ( cryo-ET ) data revealed spikes with volumes that can accommodate the molecular mass of ( Gn/Gc ) 4 hetero-octamers , related by 2-fold symmetry axes oriented radially in the particle ( Battisti et al . , 2011; Huiskonen et al . , 2010 ) . A higher resolution 15 . 6 Å cryo-ET map allowed the docking of the Gn ectodomain into the central lobes on the tetrameric spike , at the membrane distal side , and masking the Gc fusion loops , suggesting that the 2-fold related spike-spike interactions are made by the Gc moiety ( Li et al . , 2016 ) . Moreover , a central role of Gn in self-association to form spikes has recently been confirmed by number and brightness analysis in single live cells showing that separate Gn expression allows detection of Gn oligomers while separate Gc expression predominantly leads to Gc monomers and some Gc dimers ( Sperber et al . , 2019 ) . The Gn/Gc spikes on the viral surface are key in directing entry into new cells ( Cifuentes-Muñoz et al . , 2014 ) . Hantavirus cell entry occurs by the interaction of the envelope glycoproteins with host cell receptors , which leads to viral uptake into endosomes . Cell entry is completed when Gc induces the fusion of the viral envelope with the endosomal membrane at acidic pH ( Acuña et al . , 2015 ) . Structure-function studies have also confirmed that Gc is a class II fusion protein , and have provided insight into its fusogenic conformational change triggered by low pH ( Barriga et al . , 2016; Guardado-Calvo et al . , 2016; Tischler et al . , 2005; Willensky et al . , 2016 ) . This irreversible structural rearrangement of Gc into a stable post-fusion trimer involves several steps , including the initial exposure of the Gc fusion loops , which then insert into the target cell membrane via an extended trimeric intermediate . The individual Gc trimer subunits then adopt a ‘hairpin’ conformation that forces apposition of viral and cellular membranes , to allow the bilayers to fuse . On the virion , the Gn and Gc residues involved in intra- and inter-spike interactions have not been identified . These interactions control the fusion activity of Gc ( Guardado-Calvo and Rey , 2017 ) , by maintaining it at neutral pH in a functional metastable conformation ( Harrison , 2015 ) . Here , we show that the 2-fold Gc:Gc contacts between adjacent ( Gn/Gc ) 4 spikes at the surface of the hantavirus particles are mediated by the same interface observed in a crystallographic dimer revealed by the available X-ray structure of Gc in a pre-fusion form . These contacts regulate viral assembly together with spike stability and subsequent disassembly for entry . We further demonstrate that at physiological temperature , the spikes exhibit a dynamic temperature-dependent equilibrium between a ‘closed’ form in which the fusion peptides are masked by Gn , and an ‘open’ form that allows the particle to bind to liposomes at neutral pH .
The X-ray structure of the Hantaan virus Gc ectodomain at neutral pH ( PDB: 5LJY ) ( Guardado-Calvo et al . , 2016 ) revealed a pre-fusion monomer which , among other crystal contacts , presented two Gc molecules packing about a crystallographic 2-fold axis . The two Gc monomers in this dimer cross at an angle of roughly 50 degrees ( Figure 1a ) similar to the 2-fold related spike interactions in the cryo-ET map ( Li et al . , 2016 ) . The ‘dimer’ interaction observed in the hantavirus Gc crystals is reminiscent of the crystallographic dimer contacts presented by class II alphavirus fusion protein E1 ( Roussel et al . , 2006 ) , which is recapitulated by the 2-fold related contacts between hetero-hexameric ( E2/E1 ) 3 spikes at the surface of alphavirus particles ( Sun et al . , 2013; Voss et al . , 2010 ) ( Figure 1a ) . To test whether the Gc:Gc interface observed in the crystallographic dimer is involved in contacts between adjacent spikes at the surface of hantavirus particles , we introduced cysteine substitutions of candidate Gc residues at the putative 2-fold interface , such that they could form inter-spike disulfide bonds . We surveyed residues facing each other across the crystallographic Gc dimer interface with Cα-Cα distances ranging between 4 and 10 Å and selected the highly conserved His303 and Gly187 for single cysteine substitutions . These residues face their counterpart on the 2-fold axis of the crystallographic dimer with Cα-Cα distances of 8 . 4 Å ( His303-His303 ) and 5 . 6 Å ( Gly187-Gly187 ) ( Figure 1b ) . Because the Gc:Gc interface residues are highly conserved ( Figure 1—figure supplement 1 ) , the observed contacts made by Hantaan virus Gc should be maintained in all hantaviruses . We therefore turned to an Andes virus ( ANDV ) glycoprotein expression and VLP producer system for the functional experiments ( Acuña et al . , 2014 ) , which has the advantage of corroboration of ANDV VLP data with authentic ANDV , which we can manipulate . We tested ANDV mutant Gn/Gc constructs having either the Gc H303C or the G187C substitutions ( the amino acid sequence numbering is the same between Andes and Hantaan virus Gc ) for protein production in 293FT cells , surveyed their transport to the plasma membrane as a measure for proper protein folding , and monitored their assembly into VLPs released in the cells’ supernatant . We found that the cysteine mutants were properly expressed and trafficked to the plasma membrane ( Figure 1c ) . We next analyzed the presence of VLPs in the concentrated cells’ supernatant by reducing and non-reducing SDS polyacrylamide gel electrophoresis ( PAGE ) and western blot . The wild type Gc migrated as a monomer under both conditions , while the Gc cysteine mutants migrated predominantly with a molecular mass of ~100 kDa , corresponding to Gc dimers , under non-reducing conditions . In the presence of a reducing agent , these Gc dimers were readily dissociated to monomers ( Figure 1d ) . Together , these results confirm , in a biological context , that the residues forming the Gc dimer contacts in the X-ray structure of a pre-fusion form of Gc are proximal enough to each other on viral particles to allow for disulfide formation while still forming VLPs , thereby supporting the biological relevance of the crystallographic Gc dimer . Furthermore , when comparing the yields of VLP production , the G187C mutant resulted in higher production levels than H303C ( Figure 1d ) . This result correlates with the better geometry and distances between the Cα atoms in the crystallographic Gc dimer for G187C compared to H303C ( Figure 1b ) . We compared the spikes of the G187C mutant VLPs to the wild type VLPs in terms of their migration profile in blue native PAGE ( BN-PAGE ) combined with western blotting upon detergent-solubilization of the spikes . When we incubated Andes VLPs bearing wild type ( Gn/Gc ) 4 spikes at 20˚C and neutral pH , the detergent-solubilized spike was identified as a single band recognized by both , anti-Gn and anti-Gc MAbs ( Figure 2a ) . This band , corresponding to a molecular weight of ~500 kDa matching a ( Gn/Gc ) 4 spike , migrated roughly as expected in BN-PAGE , given the migration of the individual Gn and Gc monomers ( see migration at 50˚C ) , of the Gc post-fusion homotrimer ( see migration at acidic pH ) , and of the standard reference bands at 480 and 720 kDa . In contrast , in the detergent-solubilized G187C VLPs the band containing both , Gn and Gc , barely entered the gel at temperatures up to 40˚C , indicating a high molecular mass , as expected if a disulfide bond interconnects multiple adjacent spikes ( Figure 2b ) . The above data on the disulfide bonds suggested that the dimer interface observed in the Hantaan virus Gc crystals is indeed involved in lateral interactions between spikes on hantavirus particles . The crystal contacts at this interface bury a surface area of ~545 Å2 per subunit , which is a relatively small contact patch ( Figure 1b ) , consistent with the requirement for particle dissociation for entry into cells . We noted that this interface contains several strictly conserved polar residues , three charged ( Glu26 , Asp28 , Arg301 ) , and one ionizable ( His303 ) ( Figure 1—figure supplement 1 ) , which make a network of hydrogen bonds , including inter-chain salt bridges , as well as π-stacking of the His303 imidazole rings ( Figure 1b ) . To further assess the potential functional relevance of this interface , we introduced the following individual site-directed mutations in Gc: E26A , D28A , R301A and H303A . As with the cysteine mutants described above , we tested the new mutant ANDV Gn/Gc constructs for glycoprotein production in 293FT cells , transport to the plasma membrane and assembly into VLPs with concomitant release from cells ( Figure 3a and Figure 3—figure supplement 1a and 1d ) . Of the single Ala substitutions , only mutant D28A passed all the above tests , implying that this mutation was well tolerated , albeit yielding significantly reduced amounts of VLP release ( Figure 3a ) . The three other mutants were either not detected by western blot ( E26A ) or led to the expression of truncated versions of the protein ( 30 kDa ) in 293FT and Vero E6 producer cells ( R301A and H303A ) ( Figure 3a and Figure 3—figure supplement 1a and 1e ) . Yet , substitution of these residues by more chemically similar amino acids , such as the E26Q , R301Q and H303F mutants , still allowing interactions across the interface , resulted in their detectable expression , transport to the plasma membrane and equivalent amounts of assembly into VLPs ( Figure 3a and Figure 3—figure supplement 1b ) . To introduce repulsion at the Gc:Gc interface , we also replaced the residues with opposite charges: Gc E26K , D28K , R301E and H303E . Although these mutants could be detected in the intracellular and plasma membrane fractions , their release into the supernatant was significantly decreased ( Figure 3a and Figure 3—figure supplement 1c ) , suggesting that mutations interfering with interface interactions , such as alanine and opposed charges , strongly impair VLP formation . Taken together , these results indicate that the observed 2-fold related Gc:Gc inter-chain contacts are crucial for viral particle assembly , as would be expected if this interface were indeed the site of lateral packing between adjacent hantaviral spikes . We used BN-PAGE to compare the stability at increasing temperatures of the detergent-solubilized hantavirus wild type and mutant spike complexes . When we treated the wild type spike complex at different temperatures up to 50°C , we were able to visualize on the gel that the band corresponding to the ( Gn/Gc ) 4 spikes at 20°C and 30°C gradually disappeared at higher temperatures with a concomitant appearance of smaller migration bands , which corresponded to several oligomeric Gn forms and to a monomeric Gc species ( Figure 2a ) . The absence of intermediate Gn/Gc dissociation products suggested a two-state behavior , in a highly cooperative spike dissociation process . Quantification of the temperature-induced dissociation of the detergent-solubilized wild type ANDV Gn/Gc spike revealed a melting temperature ( Tm ) of 37 . 7 ± 0 . 4°C . In comparison , the Gn/Gc band of the G187C VLPs did not dissociate into homooligomeric Gn or Gc species up to temperatures of 45°C , revealing a 10˚C increase of its Tm to 48°C ( Figure 2b ) . The dissociated Gc species migrated to a further distance than the band corresponding to a wild type Gc homotrimer ( see Figure 2a , right panel ) , in line with the expected migration of a disulfide-linked Gc dimer ( Figure 2b ) . With the other Gc dimer interface mutants , we found a Tm decreased by 2–5˚C for the mutants E26Q , D28A , R301Q - each affecting ionic interactions at the interface ( Figure 3d and Figure 3—figure supplement 2 ) . In the case of the H303F mutant , the Tm dropped by 7˚C . This more important effect in the thermostability of H303F may be explained by the fact that the phenylalanine side chain is bulkier than that of histidine , hence forcing the Gc protomers to re-accommodate to the change and affect the overall Gc:Gc interaction . Together , these data suggest that the intra-spike Gn/Gc interactions are affected by the lateral inter-spike Gc:Gc contact by allostery; when the Gc:Gc contacts are strengthened by a disulfide bond the Tm of the Gn/Gc spike raise , and lower the spike Tm when the Gc:Gc contacts are weakened . We also monitored the effect of mutations at the Gc:Gc inter-spike interface on the pH required for spike activation for membrane fusion . Among all Gc:Gc interface mutants , only Gc E26Q induced syncytia formation of cells expressing the mutant Gn/Gc construct upon incubation at pH 5 . 5 , retaining ~50% of the fusion activity of the wild type ( Figure 4a ) . All other mutants were fusion inactive , even at a pH as low as 4 . 5 . When we assayed the pH for triggering cell-cell fusion by the E26Q mutant we observed an activation pH of 6 . 2 , that is 0 . 2 units higher than wild type ( Figure 4b ) . Such a raise in the activation pH may be explained by the loss of a salt bridge between Glu26 and Arg301 in the Gc:Gc inter-spike dimer ( Figure 1b ) , facilitating spike dissociation and therefore leading to fusion at a less acidic pH compared to wild type . In order to monitor the activation step allowing interaction with membranes of wild type and Gc:Gc interface mutant VLPs , we carried out liposome co-flotation studies at pH values ranging from 5 . 0 to 6 . 4 . For this purpose , we mixed fluorescently labeled liposomes with the VLPs at each pH , and loaded the mixture to the bottom of a sucrose step gradient . After centrifugation , we monitored each fraction for the presence of liposomes ( by fluorescence ) and VLPs ( by western blot against Gc ) . At pH 6 . 2 , the liposomes migrated to the top of the gradient while the wild type VLPs remained in the bottom fractions ( Figure 4c and Figure 4—figure supplement 1a ) , but increasing amounts of the VLPs were observed in the top fractions at more acidic pHs , depending on the mutant . The D28A mutant began to float at pH 6 . 2 , 0 . 2 units higher compared to wild type VLPs ( Figure 4c and Figure 4—figure supplement 1b ) . Contrary to E26Q , mutant D28A is inactive in syncytia formation . The X-ray structures show that Asp28 not only contributes to the Gc dimer interface ( Figure 1b ) , but its side chain is also involved in a network of inter-subunit polar interactions stabilizing the post-fusion Gc trimer ( Guardado-Calvo et al . , 2016 ) . Again , in this mutant , like in E26Q , the destabilization of the Gc dimer interaction caused fusion activation at a less acidic pH . And the destabilization of the post-fusion D28A Gc trimer likely renders it incompetent for inducing fusion , unlike E26Q . When we analyzed the H303F mutant for pH-induced liposome coflotation , we found that it was more resistant to activation and required a 0 . 2 units lower pH for fusion activation compared to wild type ( Figure 4c and Figure 4—figure supplement 1c ) , opposite to the effect of the E26Q and D28A mutants ( Figure 4b–c ) . Hence , although this mutation led to considerable destabilization of the spike in terms of its thermal resistance ( Figure 3d ) , it turned out to be more resistant to acidification ( Figure 4c ) . Taking into account that the His303 imidazole rings face each other across the interface with a distance of 4 . 1 Å ( Figure 1b ) , they very likely undergo a strong electrostatic repulsion upon protonation at acidic pH . Thus , when His303 is substituted by phenylalanine , this effect does not occur , and additional residues elsewhere must become protonated in order to trigger spike dissociation and fusion . Given that the H303F mutant was fusion inactive at any tested pH , the His303 role in fusion remains to be understood . Together , from these data we conclude that the hantavirus lateral inter-spike Gc:Gc interactions indirectly control spike stability , influencing at the same time Gn/Gc dissociation to induce membrane fusion . Previous data on the hantavirus spike organization ( Li et al . , 2016 ) – and that of other class II enveloped viruses such as alphaviruses and phleboviruses ( Guardado-Calvo and Rey , 2017; Halldorsson et al . , 2018; Voss et al . , 2010 ) – suggest that Gn conceals the Gc fusion loops at the top of the spikes , keeping them from premature membrane insertion until exposure to low pH . We hypothesized that the temperature-induced Gn/Gc dissociation observed by BN-PAGE could reflect a conformational change within the spike , which would lead to a looser interaction between Gn and Gc at the VLP surface . The lateral inter-spike interaction - absent in the detergent-solubilized spikes - may restrain full dissociation of Gn and Gc on VLPs , and the Gn/Gc dissociation observed by BN-PAGE may reflect a temperature-induced transition of the spike into a state in which the fusion loops become exposed at the top of the hantavirus surface at neutral pH . To test this notion , we incubated wild type Andes VLPs with liposomes at pH 7 . 4 at different temperatures and assayed them in the VLP/liposome coflotation assay . When the incubation was performed at temperatures in the range from 20°C to 30°C , we found the fluorescence-labeled liposomes at the top of the gradient while the VLPs remained in the bottom fractions ( Figure 5a ) , confirming previous data that VLPs and liposomes do not interact under these conditions ( Acuña et al . , 2015 ) . However , when increasing the temperature to 37°C and above , we observed that at neutral pH ANDV particles floated with liposomes to the upper fractions , increasing gradually with temperature ( Figure 5a ) . To assess whether membrane interaction at temperatures > 37°C was specifically induced by the Gc fusion loops , and not by non-specific interactions , we tested liposome coflotation of Andes VLPs bearing the W115A/F250A mutations in Gc . These two substitutions of aromatic residues to alanine , respectively at the tip of the cd and ij loops ( which are two of the three fusion loops of Gc ) , do not interfere with VLP formation but were previously shown to abolish insertion into target membranes at low pH ( Guardado-Calvo et al . , 2016 ) . In contrast to wild type , high temperature treatment of the fusion loop mutant VLPs up to 50°C at pH 7 . 4 did not lead to flotation with liposomes to the upper fractions ( Figure 5b ) , indicating that both , the acid-pH-induced and the temperature-induced interaction of VLPs with liposomes , were specifically driven by the Gc fusion loops . In contrast , treatment at 56˚C resulted in flotation of the fusion loop mutant with the liposomes , indicative of non-specific interactions with membranes likely due to partial protein denaturation and concomitant exposure of hydrophobic regions ( Figure 5b ) . Comparison of the temperature-induced Andes VLP-liposome interaction at neutral pH ( Figure 5a ) with the temperature-induced dissociation of detergent-solubilized ANDV spikes ( Figure 2a ) , showed the same profile ( Figure 5c ) . It revealed a Tm for the conformational transition - defined as the temperature at which 50% of the VLPs floated with the liposomes - of 37 . 8 ± 1 . 1˚C , matching the Tm of 37 . 7 ± 0 . 4˚C for the detergent-solubilized spike dissociation . Both ANDV Gn/Gc dissociation and fusion loop exposure as a function of temperature followed a sigmoidal curve , indicative of a two-states system . To test the reversibility of the observed transition , we incubated the VLPs for 15 min at 50˚C to induce fusion loop exposure at neutral pH , and then back-treated them for 1 h at 4˚C to see if the spikes would return to their initial , ‘closed’ conformation . When we then assayed these VLPs in the liposome flotation assay , we found that after sucrose step centrifugation they did not float with liposomes and remained in the bottom fraction ( Figure 5d ) . When we performed the same experiment without allowing the VLPs to recover , or when we only back-treated them for 5 or 15 min at 4°C , the 50°C-treated VLPs still floated with liposomes and were found in the upper fractions . Similarly , the sample still floated with liposomes when we back-treated for 1 h at 37°C instead of 4°C ( Figure 5d ) . Together , these data revealed that the ANDV surface is at a thermodynamic equilibrium , which at the physiological temperature of 37°C dynamically fluctuates between closed and open forms of spikes . We then examined the G187C Gc mutant VLPs in the same way . We found that 50% of the G187C mutant VLPs floated with the liposomes at 37°C , similar to wild type ( Figure 5e and Figure 5—figure supplement 1 ) . The inter-spike disulfide bond at the Gc:Gc interface therefore does not prevent the conformational equilibrium between closed and open forms of the spike , despite the higher Gn/Gc dissociation Tm observed by BN-PAGE ( Figure 2b ) . This result is in line with the Gc fusion loops being away from the Gc:Gc dimer contacts in the spikes ( Figure 1a ) , and corroborates that the observed conformational transition is mainly an intra-spike effect . To assess whether the temperature-induced fusion loop exposure had an effect on viral infectivity , we incubated ANDV for 15 min at different temperatures ranging from 20°C to 56°C . Subsequently we infected cells through adsorption for 1 h at 37°C and quantified viral infection 16 h later . We found that the infection of cells by ANDV was strongly reduced depending on the temperature of the pre-treatment when the virus was adsorbed at 37°C ( Figure 6a ) . But when after high temperature treatment the ANDV particles were allowed to recover the closed conformation of the spikes for 1 h at 4°C during adsorption to cells , their infectivity was completely restored , indicating that the observed inactivation is reversible ( Figure 6b ) . Importantly , ANDV particles treated at 56°C did not recover their infectivity , in agreement with the reported temperature of 56°C required for hantavirus inactivation ( Kallio et al . , 2006 ) . To understand how the temperature-induced fusion loop exposure affects viral infectivity in mechanistic terms , we next analyzed the fusion activity of viral particles with liposomes . For this purpose , we labeled Andes VLPs with octadecyl rhodamine B ( R18 ) and incubated them with liposomes . We observed R18 dequenching upon dropping the pH to 5 . 5 , which indicated lipid mixing between the labeled VLPs and the unlabeled liposomes , as described previously ( Guardado-Calvo et al . , 2016 ) . In turn , when we first incubated the Andes VLPs at 50˚C and then mixed them with liposomes at acidic pH , we detected no lipid mixing , confirming the infectivity results obtained with authentic ANDV . We further tested whether the fusion activity could be restored when the 50°C-treated VLPs were incubated for 1 h at 4˚C and then subjected to liposome fusion assays at acidic pH . Under these conditions , we found that the VLP-liposome fusion activity was restored ( Figure 6c ) , again in line with the reversibility of the closed-to-open states transition and the virus infection results . To test the validity of our observation across hantaviruses , we produced Sin Nombre VLPs as well as Hantaan VLPs and labeled them with R18 . The labeled VLPs gave a clear lipid mixing signal upon acidification ( Figure 6d–e ) , which was lost upon treatment at 50˚C , as observed with Andes VLPs . When the 50°C-treated Sin Nombre or Hantaan VLPs were allowed to recover for 1 h at 4˚C , they fully restored lipid mixing activity when incubated with liposomes at low pH ( Figure 6d–e ) . Hence , these data suggest that the temperature-induced fusion loop exposure and reversible reduction of fusion activity is not an ANDV specificity but a property shared by hantaviruses in general . To understand the molecular basis of the drop in viral infectivity and membrane fusion after treatment of hantavirus particles at high temperature , we tested whether the temperature treatment would still allow for Gc homo-trimerization towards the post-fusion form , as a measure of an early step in the virus-cell fusion mechanism . We thus incubated Andes VLPs at different temperatures and assayed them for low-pH induced Gc trimerization by sedimentation on a sucrose gradient . As expected , this method allowed the detection of Gc running as monomer at pH 7 . 4 and homotrimer at pH 5 . 5 when using untreated VLPs . But when we pre-treated the VLPs at 50°C and then incubated them at pH 5 . 5 , we found Gc migrating predominantly as monomer . The temperature-treatment of the VLPs thus resulted in a form of Gc unable to undergo trimerization at acidic pH to induce membrane fusion ( Figure 6f ) . When VLPs incubated at 50˚C were back-treated for 1 h at 4˚C and then incubated at pH 5 . 5 , we found Gc again sedimenting as homotrimer , confirming the reversibility of the effect . This result indicates that in the open spike Gc is maintained in a form that cannot react to low pH by undergoing the fusogenic conformational change . Only when the spikes were allowed to adopt the closed conformation , they re-acquired the capacity to respond to low pH by allowing Gc homotrimerization to induce membrane fusion .
Here we have addressed the surface organization of pleomorphic hantavirus particles . By combining structural , biochemical and functional analyses , we revealed the molecular interface by which individual ( Gn/Gc ) 4 hetero-octameric spikes associate laterally via 2-fold related Gc:Gc contacts ( Figure 1 ) akin to the contacts observed in alphavirus particles , and that an inter-spike disulfide bond across a 2-fold related Gc:Gc dimer interface improved the overall spike stability ( Figure 2b–c ) . An assembly model in which 2-fold Gc contacts relate individual ( Gn/Gc ) 4 spikes is consistent with electron microscopy observations showing a continuous surface lattice of spikes that interact sidewise to form a grid-like pattern ( Battisti et al . , 2011; Huiskonen et al . , 2010; Li et al . , 2016; Martin et al . , 1985 ) and with molecular assembly models proposed earlier ( Hepojoki et al . , 2010 ) . Contrary to the study of Hepojoki et al . ( 2010 ) suggesting that Gc is mostly dimeric when solubilized from spikes , this and previous work show that Gc is monomeric when solubilized from viral particles ( Acuña et al . , 2015; Barriga et al . , 2016 ) or when recombinantly expressed in the absence of the transmembrane segment ( Guardado-Calvo et al . , 2016; Willensky et al . , 2016 ) . These observations are consistent with the small Gc:Gc contact patch , which appears too weak to maintain Gc dimers in solution . Similarly , membrane-anchored Gc expression in the absence of Gn was predominantly monomeric in live cells , although some Gc dimers were detected ( Sperber et al . , 2019 ) . Therefore , it is likely that Gc anchoring in the context of lateral inter-spike constrains may be required for efficient Gc:Gc association . The residues involved in Gc:Gc contacts are highly conserved across hantaviruses in general , suggesting that our results can be extended to viruses across the mammal-infecting branch of the Hantaviridae family ( Figure 1—figure supplement 1 ) . Our results also show that mutation to residues that interfere with interface contacts significantly decrease virus particle production ( Figure 3a ) , implying a role for these contact residues in viral assembly by connecting spikes laterally , building the viral surface lattices . Because hantaviruses do not have a matrix protein to induce membrane curvature , as do most of the other enveloped RNA viruses , the Gc:Gc contacts driving lateral interactions between spikes on the membrane are likely to play an important role in virion budding . In spite of the unique four-fold symmetry of the hantavirus spikes , their overall arrangement on the particles has clear similarities to that of other class II enveloped viruses , such as alphaviruses , which display 3-fold symmetric spikes ( Lescar et al . , 2001 ) . The inter-spike contact area of ~500 Å2 of the 2-fold related hantavirus Gc:Gc ( Guardado-Calvo et al . , 2016 ) and alphavirus E1:E1 dimers ( Roussel et al . , 2006 ) include a highly conserved His residue at the center the dimer interface; Gc His303 and E1 His125 . Their substitution decreases the pH threshold for acid-induced activation in hantaviruses ( Figure 4c and Figure 4—figure supplement 1c ) and in alphaviruses ( Qin et al . , 2009 ) , suggesting in both cases that inter-spike dissociation is driven by repulsion upon protonation . The results reported here have also revealed that the hantavirus spikes exhibit a dynamic equilibrium between closed and open forms , with the latter exposing the Gc fusion loops at physiological temperatures even at neutral pH ( Figure 5a ) . We found that at 37°C - the physiological temperature of its rodent hosts – about 50% of the Andes VLPs bound to liposomes via Gc fusion loop exposure . The steep sigmoidal curve of liposome binding and detergent-solubilized spike dissociation as a function of temperature ( Figure 5c ) suggests a strongly cooperative effect . A possible explanation for these observations can be provided by assuming that the spikes change conformation intermittently , in a stochastic fashion , as represented in Figure 7 . Contrary to alphaviruses , hantavirus particle maturation does not involve proteolytic processing of the spikes during exocytosis . It is therefore possible that the observed reversibility of fusion loop exposure is related to the absence of irreversible proteolytic priming for fusion . Some mechanism must ensure , however , that Gc escapes from undergoing a premature irreversible conformational change triggered by the low pH environment of the exocytosis pathway of the cell , a mechanism that awaits to be discovered . Our observation that hantavirus particles with spikes in the open form are not capable of inducing low-pH triggered membrane fusion ( Figure 6c–e ) correlates with the inability of Gc in these particles to form homotrimers ( Figure 6f ) , suggesting that low pH treatment in the presence of multiple spikes in the open form engages Gc in non-functional interactions with itself from adjacent spikes , such that it cannot reach its trimeric post-fusion form ( Figure 7g ) . There appears , however , to be a thin divide between the open conformation of the particles ( around 50°C , Figure 7c ) which allows recovery of their fusogenicity by returning to the ground state ( i . e . , a closed particle , Figure 7b ) , and an irreversible state in which they cannot recover it ( i . e . , treatment to 56°C ) likely involving partial protein denaturation and aggregation ( Figures 5b and 7e ) . Figures 2a and 5c show that at room temperature the VLPs appear to be essentially in the closed form , where no flotation with liposomes is observed . This suggests that the stability of the infectious particle is much higher in the external environment , contributing to their propagation in nature . Within a mammalian host , there would be a much more rapid particle turnover . This notion is in line with previous data showing that hantaviruses are highly labile at 37°C , while displaying prolonged infectivity outside a host ( Kallio et al . , 2006 ) . It is possible that the dynamic spike behavior at 37°C results from adaptation to their rodent hosts , providing advantages for establishing chronical infections; on the one hand the major lability of hantaviruses at 37°C may help to restrict their dissemination within the host by decreasing the time window for viral spread , while on the other hand the conformational diversity may represent an important decoy for escape from the hosts´ immune response . Furthermore , the observed dynamic behavior of the spikes is likely important for virus infectivity . In contrast to liposomes , the plasma membrane is a crowded environment with multiple proteins and glycosaminoglycans , and whether fusion loop exposure can already allow for particle binding to cells is an open question . It could take place , for instance , after binding to a protein receptor ( such as β3αV integrins ) ( Choi et al . , 2008; Gavrilovskaya et al . , 1998; Jangra et al . , 2018 ) that could bring the hantavirus surface into proximity of a patch of naked membrane before its uptake by endocytosis . At any rate , at 37°C there should be enough spikes in the closed conformation on the virion to allow for productive fusion once in the acidic environment of the endosomes . Similar conformational dynamics have been observed for a number of unrelated viruses ( Heyrana et al . , 2016; Kuhn et al . , 2015; Lewis et al . , 1998; Munro et al . , 2014 ) . It has been shown that a similar dynamic behavior of dengue virus particles elicits highly cross-reactive but poorly neutralizing antibodies targeting the conserved but cryptic fusion loop ( Beltramello et al . , 2010; Oliphant et al . , 2006 ) . These antibodies are believed to be responsible for antibody-dependent enhancement of the infection , which is the main obstacle to developing an efficient vaccine against dengue virus ( Rey et al . , 2018 ) . The observed dynamic behavior of the hantavirus particles described here can have therefore an important impact in the development of suitable immunogens capable to confer protection against these pathogens ( Graham , 2017; Kotecha et al . , 2015; McLellan et al . , 2013; Rey and Lok , 2018 ) . Our data thus suggest the possibility of designing a subunit vaccine that exposes an inert ‘closed spike’ conformation only would elicit the strongest antibody response , similar to the closed form of the HIV Env trimers ( Torrents de la Peña et al . , 2017 ) . Our results now pave the way for testing this type of approaches against hantavirus infections .
ANDV isolate CHI-7913 ( Galeno et al . , 2002 ) ( kindly provided by Héctor Galeno , Instituto de Salud Pública , Chile ) was propagated in Vero E6 cells ( ATCC ) as described before ( Barriga et al . , 2013 ) . All work involving the infectious ANDV was performed under strict biosafety level three conditions ( Centro de Investigaciones Médicas , Pontificia Universidad Católica de Chile , Chile ) . 293FT cells ( Thermo Fisher Scientific ) were propagated in DMEM supplemented with 10% fetal calf serum ( FCS ) . Vero E6 cells ( ATCC ) were grown in MEM containing 10% FCS , non-essential amino acids and 1 mM sodium pyruvate ( Thermo Fisher Scientific ) . STR profiling was performed for human cell line authentication ( ATCC ) and mycoplasma testing was negative for all used cell lines . For ANDV Gn/Gc expression we used the plasmid pI . 18/GPC that codes for the full length GPC of ANDV strain CHI-7913 under the control of the cytomegalovirus promotor ( Cifuentes-Muñoz et al . , 2010 ) . Site-directed mutations were generated by DNA synthesis and sub-cloning into pI . 18/GPC using intrinsic restriction sites ( GenScript ) . To express the full length GPC from Hantaan virus or Sin Nombre virus , the plasmids pWRG/HTN-M ( x ) ( Hooper et al . , 2006 ) ( kindly donated by Dr . Jay Hooper , USMARIID ) and pcDNA/Sin Nombre virus-GP plasmid ( Kleinfelter et al . , 2015 ) were used ( kindly provided by Drs . Kartik Chandran and Rohit Jangra from Albert Einstein College of Medicine ) . For Gn/Gc expression , 293FT cells ( Thermo Fisher Scientific ) were grown in 100 mm plates and calcium-transfected with the corresponding GPC encoding plasmid . 48 h later , cell surface proteins were labeled with biotin in order to separate the biotinylated ( surface proteins ) from non-biotinylated ( intracellular proteins ) fractions using a cell surface protein isolation kit ( Pierce ) . For protein detection by western blot , primary anti-Gc monoclonal antibody ( MAb ) 2H4/F6 ( Godoy et al . , 2009 ) , anti-Gn monoclonal antibody 6B9/F5 ( Cifuentes-Muñoz et al . , 2010 ) or anti-β-actin MAb ( Sigma ) were used at 1:2500 and subsequently detected with an anti-mouse immunoglobulin horseradish peroxidase conjugate ( Thermo Fisher Scientific ) 1:5000 and a chemiluminescent substrate ( Pierce ) . All these antibodies were previously characterized concerning their reactivity with negative controls ( Cifuentes-Muñoz et al . , 2011; Cifuentes-Muñoz et al . , 2010 ) . VLPs were harvested from supernatants of 293FT cells transfected with the pI . 18/GPC wild type or the different mutant constructs at 48 h post-transfection and concentrated as previously established ( Acuña et al . , 2014 ) . The amount of GPC encoding plasmid used for each mutant was adjusted in order to reach similar amounts of cell surface accumulation to wild type . The discontinuous native protein gel electrophoresis was performed similar to as previously described ( Niepmann and Zheng , 2006 ) . Briefly , VLP samples harvested from pI . 18/GPC wild type or the different mutant constructs were incubated with Coomassie G-250 0 . 25% and Triton X-100 0 . 5% for 15 min at different temperatures just before loading onto a 3–16% gradient polyacrylamide gel . The native gel electrophoresis was run at 130 mV for 15 h at 4°C . The buffering system was 200 mM Tris for anode buffer and 50 mM Tris , 100 mM glycine cathode buffer . The size of Gn and Gc species was estimated using the migration rate of a molecular standard ( Native Mark Unstained Protein Standard , Invitrogen ) which was independently stained . The rest of the gel was incubated in transfer buffer at room temperature . After the transfer , the nitrocellulose was blocked in PBS including 5% skim milk and next Gn and Gc glycoproteins were stained separately by using anti-Gn MAb 6B9/F5 ( Cifuentes-Muñoz et al . , 2010 ) and anti-Gc MAb 2H4/F6 or 5D11/G7 ( Godoy et al . , 2009 ) at a 1:2500 dilution each . Primary antibody staining was detected as described above . For the quantification of the Gn/Gc spike dissociation , the densitometry values of dissociated Gc ( monomeric or dimeric Gc ) were divided by the densitometry values of the total signal for Gc , using the ImageJ software ( Schneider et al . , 2012 ) . The average value and standard derivation ( s . d . ) of biological replicates was calculated for each temperature condition and a sigmoidal curve fitted using Equation 1 . ( 1 ) Gn/Gc dissociation ( % ) =Gn/Gc disociation ( 20°C ) +Gn/Gc dissociation MAX/ ( 1+e- ( T-Tm ) /b ) were Gn/Gc dissociation ( 20°C ) is the basal dissociation at 20°C , Gn/Gc dissociation MAX is the maximal dissociation value , Tm is the temperature at 50% Gn/Gc dissociation and b is the Hill’s slope of the curve , indicating its steepness . The curve was fitted using a sigmoidal four parameters equation in SigmaPlot 12 . 0 , Systat Software . Liposomes were prepared fresh by the freeze-thaw and extrusion method ( Castile and Taylor , 1999 ) . PC ( phosphatidylcholine , from chicken egg ) and PE ( phosphatidylethanolamine , from chicken egg ) , sphingomyelin ( from porcine brain ) , cholesterol ( from ovine wool ) , were purchased from Avanti Polar Lipids and large multilaminar vesicles ( liposomes ) were prepared using PC/PE/sphingomyelin/cholesterol in a 1/1/1/1 . 5 ratio respectively . The coflotation of viral particles with liposomes was performed as previously established ( Acuña et al . , 2015 ) . First , liposomes were labeled with 200 mM 1 , 6-diphenyl-1 , 3 , 5-hexatriene ( DPH ) and VLPs prepared from wild type or mutant pI . 18/GPC constructs were incubated at pH 5 . 5 for 15 min at 37°C as positive control or at pH 7 . 4 using different temperatures . The VLP-liposome mixture was then added to the bottom and adjusted to 25% ( w/v ) sucrose . Additional sucrose steps of 15% and 5% were then over-layered . After centrifugation for 2 h at 300 , 000 g , liposomes were detected by the fluorescence emission of DPH ( λex = 230 nm; λem = 320 nm ) and VLPs by western blot using anti-Gc MAb 2H4/F6 . The western blot signal was quantified using the ImageJ software ( Schneider et al . , 2012 ) and VLP-liposome coflotation calculated by dividing the densitometry value of liposome associated Gc by the densitometry value of the total signal for Gc . The average value and s . d . of biological replicates was calculated and a sigmoidal VLP-liposome coflotation curve fitted using Equation 2 . ( 2 ) VLP coflotation %=VLP coflotation 20°C+VLP coflotation MAX* ( 1+e- ( T-Tm ) /b ) were VLP coflotation ( 20°C ) is the VLP coflotation at 20°C , VLP coflotation MAX is the maximal VLP coflotation percentage , Tm is the temperature at 50% VLP coflotation and b is the Hill’s slope of the curve . The curve was fitted using a sigmoidal four parameters equation in SigmaPlot 12 . 0 , Systat Software . ANDV was incubated at different temperatures for 15 min and subsequently added to Vero E6 cells ( MOI = 1 ) . After 1 h of adsorption at 4°C or 37°C , the cells were washed in excess and next infection was allowed to proceed for 16 h by incubation in MEM 10% FBS at 37°C . Quantification of viral infection was performed as previously described ( Barriga et al . , 2013 ) by detecting viral nucleoprotein expressing cells by using flow cytometry . Briefly , infected cells were detached and fixed with 2% paraformaldehyde for virus inactivation . Subsequently , the fixed cells were permeabilized using 0 . 1% Triton X-100 and then stained with primary MAb 7B3/F6 anti-ANDV nucleoprotein ( Tischler et al . , 2008 ) by incubation for 2 h at RT , which in turn was detected by goat anti-mouse immunoglobulin conjugated to Alexa Fluor 488 ( Thermo Fisher Scientific ) . Flow cytometry was performed in a cytometer ( FACS CAN II , Becton Dickinson ) counting at least 5 , 000 cells . The gate for ANDV nucleoprotein positive cells was established using as negative control non-infected cells labeled with the same primary and secondary antibodies . For the lipid mixing assay , VLPs were labeled with 1 μg/ml of R18 ( Invitrogen ) . Labeled VLPs were then mixed with liposomes in a continuously stirred fluorimeter cuvette at 37°C and lipid mixing was monitored by the decrease in R18 fluorescence generated by the dilution of the R18 probe with the unlabeled phospholipids in the liposome membrane . Fluorescence was recorded continuously at 580 nm using a fluorescence spectrophotometer ( Varian Eclypse , Agilent Technologies ) at an excitation wavelength of 560 nm using a 10 nm slit width for excitation and emission . After a stable base line was established at pH 7 , it was subtracted from recording and established as base line value corresponding to 0% lipid mixing . The reaction initiation time ( t = 0 ) corresponds to the lowering of the pH to 5 . 5 . The maximal extent ( 100% ) of excimer dilution was defined by the addition of Triton X-100 0 . 1% ( v/v ) after lipid mixing of each condition had concluded . Acid-induced Gc homotrimerization was tested as established before ( Acuña et al . , 2015 ) . VLPs constituted of ANDV Gn/Gc were incubated for 30 min at 37°C at the indicated pH to induce multimerization changes . Next , Triton X-100 1% ( v/v ) was added to allow the extraction of the membrane glycoproteins from the viral particle . The extracted glycoproteins were then added to the top of a sucrose gradient ( 7–15%; w/v ) and centrifuged at 150 , 000 g for 16 h . Next , fractions were collected and the presence of Gc was analyzed by western blot using MAbs anti-Gc 2H4/F6 . The molecular mass of fractions was determined experimentally by a molecular marker ( Biorad ) . This three-color fluorescence assay was performed as previously described ( Cifuentes-Muñoz et al . , 2011 ) . Vero E6 cells ( ATCC ) seeded into 16 well chamber slides were transfected with the pI . 18/GPC wild type or the different mutant constructs using lipofectamine 2000 ( Thermo Fisher Scientific ) . The DNA amounts were adjusted to obtain similar levels of Gc at the cell surface . 48 h later , the cells were incubated in E-MEM ( pH 5 . 5 ) at 37°C for 5 min , subsequently washed with PBS , and the incubation continued for 3 h at 37°C in E-MEM ( pH 7 . 2 ) . The cell cytoplasm was then stained for one hour with 1 μM of 5-chloromethylfluorescein diacetate ( Cell Tracker green CMFDA , Thermo Fisher Scientific ) and cells then fixed for 20 min with 4% paraformaldehyde . For immunelabelling , the cells were then permeabilized with PBS 0 . 1% Triton X-100 and Gc stained using the monoclonal antibody 2H4/F6 1:500 and secondary antibody goat anti-mouse immunoglobulin conjugated to Alexa Fluor555 1:500 ( Thermo Fisher Scientific ) . Finally , nuclei were stained for 5 min with DAPI 1 ng/μL and samples examined under a fluorescence microscope ( BMAX51 , Olympus ) . The fusion index of Gc expressing cells was calculated using the formula: 1- [number of cells/number of nuclei] . Approximately 200 nuclei per field were counted ( 10X magnification ) and five fields used to calculate the fusion index for each sample of at least three biological replicates . All statistical analyses were carried in GraphPad Prism , version 6 , and SPSS software ( SPSS , Inc ) . For protein structure analyses and graphics PyMOL Molecular Graphics System Version 2 . 0 ( Schrödinger , LLC ) was used . | Hantaviruses infect rodents and other small mammals , but do not harm them . When transmitted to humans , often through rodent urine , feces or saliva , they can cause serious and even fatal diseases . Currently , there are no known methods that effectively prevent hantavirus infections or treat the diseases that they cause . During an infection , viruses invade the cells of their host . A hantavirus interacts with target cells through proteins on its surface called Gn and Gc glycoproteins . Previous work has shown that these glycoproteins are organized in bundles of four Gn and four Gc proteins , termed spikes , which project from the membrane that surrounds the virus . The Gc protein changes shape when it is activated and exposes a hidden region that can insert into the membrane of the target cell . The Gc proteins then change shape again to force the cell to fuse with the viral membrane . This process allows the virus to be taken up into the cell , where it can replicate . While the structures of each viral glycoprotein have been determined in isolation , it was not known how they interact within the Gn/Gc spike . Such information is crucial to understand how the viruses infect cells and which areas are exposed to the immune system of the host – and so could be targeted by antiviral treatments . Bignon et al . have now identified the molecular contacts that occur between spikes and interconnect them into a grid-like lattice on the surface of the virus . Genetically altering specific sections of the Gc glycoprotein strengthened or weakened these contacts , which correspondingly increased or decreased how stable the spike was . Preventing the contacts from forming resulted in cells releasing fewer virus-like particles . Bignon et al . also show that at the body temperature of mammals , the shape of the spike fluctuates between an ‘open’ form that exposes the region of Gc that inserts into the cell membrane , and a ‘closed’ form that hides this region . However , when Gc is activated , the open form becomes unable to cause the viral and cell membranes to fuse together . Together , the results presented by Bignon et al . help us to understand how changes to the hantavirus surface enable the virus to infect cells . This knowledge will help researchers to design vaccines that protect against hantavirus infections . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Molecular organization and dynamics of the fusion protein Gc at the hantavirus surface |
How oocytes are transferred into an oviduct with a receptive environment remains poorly known . We found that glands of the Drosophila female reproductive tract , spermathecae and/or parovaria , are required for ovulation and to promote sperm storage . Reducing total secretory cell number by interferring with Notch signaling during development blocked ovulation . Knocking down expression after adult eclosion of the nuclear hormone receptor Hr39 , a master regulator of gland development , slowed ovulation and blocked sperm storage . However , ovulation ( but not sperm storage ) continued when only canonical protein secretion was compromised in adult glands . Our results imply that proteins secreted during adulthood by the canonical secretory pathway from female reproductive glands are needed to store sperm , while a non-canonical glandular secretion stimulates ovulation . Our results suggest that the reproductive tract signals to the ovary using glandular secretions , and that this pathway has been conserved during evolution .
The oviduct must interact extensively with the ovary to receive ovulated eggs in a manner that maximizes successful reproduction and minimizes egg loss and ectopic pregnancy . In humans , oviduct-ovary signaling also likely influence serous ovarian carcinoma development , a disease now thought to originate from the secretory epithelia of the distal oviduct ( Fallopian tube ) following an extended number of ovulatory cycles ( Levanon et al . , 2008; Bowtell , 2010; Kurman and Shih Ie , 2010; King et al . , 2011; Berns and Bowtell , 2012 ) . How the oviduct normally influences mammalian ovulation remains unclear , although a great deal has been learned about the hormonal control of ovulation within the ovary itself ( Kim et al . , 2009; Conti et al . , 2012 ) , and genes such as the nuclear hormone receptor Lrh-1 are known to be essential ( Duggavathi et al . , 2008 ) . The growing realization that important aspects of gamete biology have been conserved during evolution suggests that insights into oviduct-ovary signaling may come from studies of model systems . The Drosophila oviduct plays important roles during egg production that may involve communication with the ovary . The oviduct must be prepared to transport each oocyte released from the ovary to the uterus , to mediate its water uptake and eggshell crosslinking , and to position it for efficient fertilization ( reviewed in Spradling , 1993 ) . During each cycle of ovulation , just one of the many mature oocytes present in the two ovaries is released into an oviduct . Octopaminergic neurons innervating oviduct muscle and epithelia are needed for ovulation , probably to activate oviduct muscle contraction and to stimulate epithelial secretion by activating the Oamb octopamine receptor ( Lee et al . , 2003 , 2009; Monastirioti , 2003 ) . The steroid hormone ecdysone is produced in the adult ovary and is required to maintain egg production ( Buszczak et al . , 1999 ) , although a specific role in ovulation has not been tested . Glandular secretions from male reproductive tracts in both invertebrates and vertebrates facilitate reproduction at multiple steps ( Bloch Qazi et al . , 2003; Suarez , 2008; Heifetz and Rivlin , 2010; Holt and Fazeli , 2010; Ikawa et al . , 2010; Jeong et al . , 2010; Manier et al . , 2010; Dunlap et al . , 2011 ) . Multiple proteins produced in the Drosophila male accessory glands are mixed with sperm upon ejaculation and transferred to the female reproductive tract where they mediate sperm storage , capacitation , and maternal reproductive behavior ( Avila et al . , 2011 ) . For example , sex peptide ( SP ) increases egg laying and reduces female receptivity ( Chen et al . , 1988; Chapman et al . , 2003; Liu and Kubli , 2003 ) by binding to a specific receptor , SPR , in three sets of fru+ppk+ sensory neurons in the female reproductive tract ( Yapici et al . , 2008; Hasemeyer et al . , 2009; Yang et al . , 2009 ) . Ovulin , a protein transferred in male seminal fluid induces ovulation shortly after copulation ( Herndon and Wolfner , 1995; Heifetz et al . , 2005 ) . Another transferred peptide , Acp36DE , facilitates sperm storage ( Neubaum and Wolfner , 1999a; Avila and Wolfner , 2009 ) . Ejaculate components produced in the mammalian testis , prostate , and epididymis also play important roles in reproduction ( Suarez , 2008 ) . For example , mammalian spermadhesins secreted from seminal vesicles mediate sperm attachment to the oviduct epithelia ( Talevi and Gualtieri , 2010 ) . Female reproductive tract secretions also boost reproduction by interacting with transferred sperm and seminal proteins in many species ( Holt and Fazeli , 2010; Jeong et al . , 2010; Dunlap et al . , 2011; Franco et al . , 2011; Schnakenberg et al . , 2011; Wolfner , 2011 ) . Drosophila spermathecae and parovaria , the major exocrine glands of the female reproductive tract , are required for fertility and sperm storage ( Anderson , 1945; Allen and Spradling , 2008; Schnakenberg et al . , 2011 ) . Whether Drosophila female secretory products regulate other aspects of reproduction remains poorly understood , however . Recently , reproductive secretory cell development in the spermathecae and parovaria was shown to follow a stereotyped cell lineage and to depend on the transcription factor Lozenge ( Anderson , 1945 ) and Hr39 ( Allen and Spradling , 2008; Sun and Spradling , 2012 ) , a nuclear hormone receptor homologous to Lrh-1 . Here we used our new understanding of reproductive gland development to manipulate the number and activity of secretory cells in adult females . In addition to documenting a role for protein secretion in sperm storage , we show that adult Hr39 expression and a non-canonical secretion from the adult female reproductive glands are required for normal ovulation . Thus , ovulation in both Drosophila and mice depends on the homologous nuclear hormone receptors Hr39 and Lrh-1 . Our results suggest that a conserved program of reproductive tract secretion mediates oviduct-ovary signaling and may be relevant to the origin of ovarian cancer .
The overall function of female reproductive glands can be assessed by studying adult females bearing mutations in lz or Hr39 which disrupt their development ( Anderson , 1945; Allen and Spradling , 2008 ) . Mutants retain only rudimentary glands or lack them entirely and show defects in sperm storage and egg laying ( Anderson , 1945; Allen and Spradling , 2008 ) . All lz−/− females completely lack reproductive glands , while Hr39−/− females either lack glands ( >90% ) or retain a single defective spermathecae with very few secretory cells ( Figure 1A–C ) . The ovaries in Hr39 and lz mutant females contain a full complement of mature oocytes , however , both lay significantly fewer eggs than controls ( Figure 1D–E ) , indicating that secretory products are required for one or more steps downstream from oocyte completion , such as ovulation , mating , sperm storage , fertilization , or egg laying . 10 . 7554/eLife . 00415 . 003Figure 1 . Female reproductive glands are essential for ovulation . ( A ) – ( C ) DIC images of Oregon-R ( A ) , Hr397154/Ly92 ( B ) and lz3/34 ( C ) mutant female lower reproductive tracts . Both spermathecae ( yellow arrowheads ) and parovaria ( magenta arrows ) are absent in the mutant animals . Bar graphs display the rate of egg laying ( D and E ) , ovulation frequency ( F and G ) , and copulation frequency ( H and I ) for the two mutant genotypes , and heterozygous controls . In all figures , the number of egg laying groups or mating pairs is shown in brackets . Error bars are SEM , or 95% confidence intervals . *p<0 . 01 ( Fisher's exact test , or Student t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 003 It would be particularly interesting if reproductive glands were required for ovulation , since this might indicate that secretory products coordinate activities of the ovary and reproductive tract . Consequently , to distinguish whether ovulation was specifically affected , individual female flies were examined to determine if oocytes had left the ovary within 6 hr after mating to wild type males . In controls , about 50% of females initiated ovulation within this time interval , as indicated by the presence of an egg in either the oviduct , the uterus , or the food vial ( Figure 1F , G ) . In contrast , none of the Hr39 mutant and only 2% of lz mutant females initiated ovulation . The failure of mutant animals to ovulate was not due to defects in mating , as both mutant and control females showed similar rates of copulation success ( indicated by the presence of sperm in the female reproductive tract ) ( Figure 1H , I ) . Thus both Hr39 and lz are required for ovulation , at least initially . Before determining which cells within the glands were needed for ovulation , we investigated whether the failure of lz and Hr39 mutant females to ovulate was due to a secondary requirement of these genes outside of the reproductive glands . For example , lz and Hr39 might function in the octopaminergic neurons that innervate oviduct muscle and stimulate oviduct epithelial cells prior to ovulation ( Lee et al . , 2003 ) , or in the fru+ppk+ sensory neurons of the female reproductive tract . However , lz expression could not be detected in the octopaminergic neurons innervating the oviduct nor in the oviduct muscle or epithelial cells ( Figure 2A ) . Lineage tracing also showed that oviduct cells are not derived from lz+ cells ( Sun and Spradling , 2012 ) . Yet blocking the proliferation of lz-expressing cells during pupal development was sufficient to disturb ovulation ( Table 1 ) . 10 . 7554/eLife . 00415 . 004Figure 2 . lz and Hr39 are not required in reproductive tract neurons . ( A ) lz expression ( lzGal4 driving UAS-mCD8::GFP ) in control female reproductive tract . Spermathecae ( yellow arrowheads ) ; parovaria ( magenta arrowheads ) . Ov: Ovuduct; SR: Seminal receptacle; Ut: Uterus . Two sets of lz+ sensory neurons are illustrated at higher magnification in ( A1 and A2 ) . ( B ) lz expression in female reproductive tract expressing lzGal4>UAScycA ( lz>cycARNAi ) . lz+ sensory neurons are not affected ( B1 and B2 ) . ( C ) Egg production is not affected by expressing Hr39-RNAi in ppk+ neurons of the reproductive tract . ( D ) Ectopic expression of mSP in fru+ reproductive tract neurons reduces virgin female copulation rate , even when neurons are mutant for Hr39 . ( E ) Ectopic mSP in fru+ neurons is sufficient to induce egg laying in control virgin females but not in Hr39−/− females even in the presence of males . * indicates p<0 . 01 and NS indicates p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 00410 . 7554/eLife . 00415 . 005Table 1 . The effect of altering secretory cell ( SC ) number in female reproductive glands on egg laying , ovulation , copulation , and sperm storage in spermathecaeDOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 005GenotypeFemale glandsEgg laying in 2 daysOvulation in 6 hrCopulation in 6 hrSperm storage in 6 hrNSC/Female ( Avg . ± SD ) NEggs/female/Day ( Avg . ± SEM ) NOvulation ( % ) NCopulation ( % ) NSpermathecae with sperm ( % ) lzGal410197 . 0 ± 18 . 04538 . 9 ± 3 . 93076 . 71889 . 0lz>cycARNAi362 . 0 ± 2 . 6*251 . 0 ± 1 . 0*303 . 3*2356 . 5lz>hr39RNAi1510 . 4 ± 7 . 4*255 . 2 ± 0 . 8*3020*1369 . 2lz>hntRNAi12311 . 2 ± 8 . 4*458 . 0 ± 1 . 9*306 . 7*27100 . 0lz>hntRNAi22239 . 4 ± 12 . 1*2516 . 5 ± 1 . 5†lz>PsnDN251 . 8 ± 1 . 9*252 . 0 ± 1 . 8*254*25100 . 0lz>Su ( H ) DN160 . 9 ± 1 . 0*152 . 5 ± 2 . 0*dpr5Gal45192 . 0 ± 15 . 44043 . 0 ± 4 . 53164 . 525100 . 05098 . 0dpr5>cycARNAi259 . 3 ± 3 . 7*253 . 3 ± 1 . 5*358 . 6*3591 . 46515 . 4*dpr5>hr39RNAi10191 . 3 ± 15 . 31542 . 1 ± 2 . 62445 . 82190 . 53783 . 8dpr5>hntRNAi11817 . 1 ± 6 . 3*508 . 4 ± 1 . 8*342 . 9*3494 . 1649 . 4*dpr5>hntRNAi21399 . 5 ± 20 . 6*2550 . 4 ± 2 . 82479 . 22495 . 84981 . 6dpr5>NRNAi259 . 0 ± 3 . 2*250 . 9 ± 0 . 6*251*2391 . 3424 . 8*dpr5>PsnDN1683 . 9 ± 11 . 5*2537 . 4 ± 7 . 52646 . 226100 . 05288 . 5dpr5>Su ( H ) DN2016 . 1 ± 4 . 5*2526 . 2 ± 1 . 7†147 . 1*1492 . 92615 . 4**p<0 . 001 . T-test was used for secretory cell number and egg laying . Fisher's exact test was used for ovulation , copulation , and sperm storage . †p<0 . 01 . Further experiments argued against a requirement of lz and Hr39 in the fru+ppk+ sensory neurons for the female postmating behaviors elicited by the SP/SPR signaling pathway ( Hasemeyer et al . , 2009; Yang et al . , 2009 ) . lz was found to be expressed in a subset of these neurons near the oviduct-uterus junction , but fru+ppk+ neuron number was not affected by knocking down cycA or Hr39 using the lzGal4 driver ( Figure 2A , B ) . Likewise , Hr39 does not function in these neurons because no defects were observed in egg laying when Hr39 levels were reduced by driving Hr39RNAi expression using ppkGal4 ( Figure 2C ) . Furthermore , ectopically expressing a membrane-attached form of sex peptide ( mSP ) in fru+ neurons blocked virgin female receptivity even when carried out in an Hr39 mutant background ( Figure 2D ) ( Hasemeyer et al . , 2009; Yang et al . , 2009 ) , indicating that Hr39 mutant females have intact neural circuitry for post-mating behavior . Yet these same females still did not lay eggs ( Figure 2E ) . Thus , the disruption in ovulation observed in lz and Hr39 mutant females is unlikely to be caused by altered SP/SPR signaling or to other neural defects within the reproductive tract . In order to analyze which cells within the reproductive glands are required for ovulation , we developed methods for perturbing gland development more precisely than is possible using lz and Hr39 mutations . Several previous observations during studies of pupal spermathecal and parovarial development ( Sun and Spradling , 2012 ) revealed a likely role for Notch signaling in their developing secretory cells ( SCs ) . The stereotyped divisions of secretory unity precursor cells ( SUPs ) resemble the Notch-requiring divisions during peripheral nervous system development ( Lai and Orgogozo , 2004; Sun and Spradling , 2012 ) . Consistent with this idea , we discovered that a Notch signaling reporter is dynamically expressed in this lineage ( Figure 3A–D , green ) . Hindsight ( Hnt ) , a transcription factor that acts downstream from Notch during ovarian follicle cell development ( Sun and Deng , 2007 ) , was also expressed in developing and adult SCs but not epithelial cells ( Figure 3B–D ) . Within the SUP lineage , secretory cells displayed the highest level of Notch activity and Hnt expression ( Figure 3D , D' ) . 10 . 7554/eLife . 00415 . 006Figure 3 . Notch signaling and Hindsight are required to form reproductive gland secretory cells . ( A ) The cell lineage underlying secretory cell development ( Sun and Spradling , 2012 ) . Notch signaling activity ( green ) ; Hnt expression ( red ) . Ac: Apical cell; Bc: Basal cell; LEP: Lumen epithelial precursor; Sc: Secretory cell; SUP: Secretory unit precursor . ( B ) – ( D ) Notch activity ( green ) and Hnt ( red ) in spermathecae of adults ( B ) , 26 hr pupae ( APF ) ( C ) , and 48 hr APF ( D ) . ( D' ) shows the boxed region from ( D ) . Yellow arrowhead: Epithelial cell . ( E ) – ( G ) Adult spermathecae from females expressing lz>Psn[DN] ( E ) or lz>hntRNAi ( F–G ) during gland development . lz ( green ) marks epithelial cells; Hnt ( red ) marks secretory cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 006 We extensively documented that Notch signaling and Hnt function during gland development using knockdown experiments ( Table 1 ) . Expression of NotchRNAi driven by lzGal4 causes pupal lethality , however , flies in which Notch signaling is disrupted using dominant negative forms of the pathway components Psn or Su ( H ) survive to adulthood . When we examined the reproductive glands in females of these genotypes , no secretory cells were observed , the gland lumen was collapsed and the duct was malformed ( Figure 3E and Table 1 ) . Depletion of Hnt with two different hntRNAi lines driven by lz-Gal4 almost completely blocked secretory cell formation , while the gland lumen and duct developed normally ( Figure 3F , G and Table 1 ) . It would be worthwhile to further investigate the roles Notch signaling plays during specific steps in the secretory cell lineage . Because , these differential cell divisions ( Figure 3A ) probably resemble those extensively characterized during peripheral nervous system development , we initially focused on using this new information to generate glands containing reduced numbers of secretory cells , without disturbing the epithelial portion of the gland . Adult females whose reproductive glands are deficient in secretory cells were generated by knocking down hnt expression during pupal development using a lzGal4 driver ( Figure 3F , G ) , and tested for their ability to ovulate and lay eggs . SC-deficient females showed strong ovulation defects and laid significantly fewer eggs than controls ( Table 1 ) , indicating that secretory cells per se are required for ovulation . Females whose reproductive glands lack secretory cells were independently generated by expressing dominant negative ( DN ) forms of Psn ( Figure 3E ) or Su ( H ) , and these females also had greatly reduced ovulation and laid few eggs ( Table 1 ) . To further limit possible secondary defects present in animals that develop with reduced numbers of secretory cells , we searched for Gal4 drivers expressed specifically in female reproductive gland precursors among the Janelia Gal4 collection ( Pfeiffer et al . , 2008 ) . From approximately 1000 lines screened , one Gal4 driver , 51B02 ( termed dpr5Gal4 ) is specifically expressed in developing but not in mature secretory cells , nor in other reproductive tract or ovarian tissue ( Figure 4—figure supplement 1 ) . Using dpr5Gal4 to drive a lineage marker confirmed its specificity for the secretory lineage of the reproductive tract and its absence in sensory neurons ( Figure 4—figure supplement 2 ) . Females with different numbers of secretory cells were generated by knocking down cycA , Hr39 , hnt , N , Psn , or Su ( H ) expression with dpr5Gal4 ( Table 1 ) . Regardless of which gene was targeted , the ability of these females to ovulate and to lay eggs depended on the number of secretory cells in their reproductive glands ( Figure 4A , B ) . Copulation was not affected ( Figure 4C ) . These results demonstrate that one or more products produced in the secretory cells of the reproductive tract are required for adult Drosophila females to ovulate and lay eggs . 10 . 7554/eLife . 00415 . 007Figure 4 . Female reproductive tract secretory cells mediate ovulation and sperm storage . ( A ) – ( C ) Relationship between secretory cell ( SC ) number and egg laying rate ( A ) ; percent ovulation ( B ) ; or percent copulation ( C ) . Pooled data from genotypes in Table 1 . Female reproductive tracts ( D and E ) and spermathecae ( yellow circles in D and E; shown at higher magnification: D' and E' ) from normal females ( dpr5Gal4 alone ) ( D ) or females lacking SCs ( dpr5Gal4>hntRNAi ) ( E ) 6 hr after mating to males whose sperm nuclei are marked with protB-GFP ( green ) . Seminal receptacle ( white arrow ) . ( F ) Relationship between secretory cell number and the percentage of spermathecae with >5 sperm . Pooled data from genotypes in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 00710 . 7554/eLife . 00415 . 008Figure 4—figure supplement 1 . The expression pattern of the dpr5Gal4 line in spermathecae at 26 hr ( using UAS-GFPnls ) , 39 hr APF ( using UAS-GFP ) and in the adult female lower reproductive tract ( using UAS-GFP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 00810 . 7554/eLife . 00415 . 009Figure 4—figure supplement 2 . Lineage-marked progeny of dpr5+ cells ( green ) in the female reproductive tract , showing labeling of SC cells . Spermathecae ( yellow arrowheads ) ; parovaria ( magenta arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 009 Studies of animals lacking reproductive secretory organs ( Anderson , 1945; Allen and Spradling , 2008 ) and of adults whose spermathecal secretory cells were partially ablated ( Schnakenberg et al . , 2011 ) have strongly argued that reproductive secretory cells produce products involved in sperm storage . We examined spermathecae for the presence of stored sperm in females generated as described above using dpr5Gal4 that differ in secretory cell number . Within 6 hr after copulation , most wild type females had finished transferring sperm from the uterus to the storage organs ( seminal receptacle and spermathecae ) ( Figure 4D and Table 1 ) ( Manier et al . , 2010 ) . In contrast , less than 20% of females with a severe deficit of secretory cells ( e . g . , dpr5Gal4>hntRNAi1 ) had sperm inside the spermathaecal lumen at this time ( Figure 4E and Table 1 ) . Even among rare spermathecae that contained sperm inside the lumen , the number stored was much less than in controls . The absence of stored sperm is unlikely to be due to a physical block to the spermathecae , since sperm were found in the spermathecal duct and were stored in the seminal receptacle ( Figure 4E ) . Our experiments showed that a minimum of about 80 secretory cells are needed for females to store a normal number of sperm ( Figure 4F ) , indicating that a quantitative requirement exists for the products of these cells . To further investigate the role of reproductive gland cells in adult female fertility , we sought to disrupt the activity of these cells during adulthood in glands that had developed normally . We shifted conditional mutations to the restrictive temperature only after eclosion ( Figure 5A ) and also used the Gal4 line syt12Gal4 , which is expressed in mature secretory cells ( Figure 5B ) , but not in the rest of the reproductive tract and ovary , to limit manipulations to adult secretory cells . The process of canonical protein secretion via the ER/Golgi/plasma membrane pathway was disrupted by expressing a dominant negative temperature sensitive allele of dynamin ( shits ) ( Yang et al . , 2009 ) , or by knocking down the betaCOP or sec23 genes using RNAi controlled by the temperature sensitive Gal80 repressor ( Lee et al . , 2004; Bard et al . , 2006; Aikin et al . , 2012 ) . When syt12Gal4 UAS-shits adults were raised to the non-permissive temperature at eclosion , membrane trafficking in secretory cells was rapidly disrupted as expected ( Figure 5—figure supplement 1 ) . After 6 days at the non-permissive temperature , these females showed severe defects in sperm storage within the spermathecal lumen ( Figure 5B–D ) , and those sperm that were stored exhibited an abnormal morphology characterized by a twisted sperm head ( Figure 5E–F ) . Despite this , the females contained many sperm within the seminal receptacle ( Figure 5C ) and laid a near normal number of eggs ( Figure 5G ) . Even stronger reductions in the number of sperm within the spermathecae were observed when betaCOP or sec23 were knocked down in adults following the temperature shift to inactivate Gal80 ( Figure 5D ) . 10 . 7554/eLife . 00415 . 010Figure 5 . Canonical protein secretion from glandular secretory cells is required for sperm storage but not for ovulation . ( A ) Experimental scheme for testing adult secretory cell function using temperature sensitive shits or GAL80ts . ( B ) and ( C ) Dynamin ( Shi ) is required for sperm storage . Female reproductive tract of syt12Gal4 control ( B ) or syt12Gal4 driving shits expression ( C ) 6 hr after mating to protB-GFP males at 29°C . syt12Gal4 expression is restricted to secretory cells as showed by UAS-RFP ( red ) . ( B' ) and ( C' ) : Higher magnification of boxed spermathecae; seminal receptacle contain sperm ( white arrows ) . ( D ) Sperm content of spermathecae ( three classes ) is reduced in flies with indicated genotype ( x axis ) at 29°C . Bracket: Number analyzed . *p<0 . 01 ( chi-square test ) . ( E ) and ( F ) Abnormal morphology of spermathecal sperm in shits females at 29°C ( F ) compared to control ( E ) . Egg laying rate ( G ) and ovulation time ( from Table 2 ) ( H ) in flies with the indicated genotypes ( x axes ) . *p<0 . 05 ( Students t-test or Fisher's exact test ) . ( I ) Secretory cells use distinct secretory pathways to control sperm storage and ovulation . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 01010 . 7554/eLife . 00415 . 011Figure 5—figure supplement 1 . Membrane trafficking defects are observed in SCs when protein secretion is disrupted . Control secretory cells ( left ) : syt12Gal4 > UAS-RFP; shits-expressing secretory cells ( right ) : syt12Gal4 > UAS-RFP UAS-shits . RFP foci are visible in shits-expressing SCs but not in controls . Single confocal sections are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 011 This experimental paradigm also revealed an ongoing requirement for Hr39 in adults . When Hr39 function was knocked down in adult secretory cells under the control of Gal80ts ( Figure 5A ) , severe reductions in the number of sperm stored in the spermathecae were also observed ( Figure 5D ) . Those sperm that were present showed the same morphological defects seen in animals where canonical secretion had been reduced . The effects on ovulation of knocking down Hr39 expression or disrupting canonical protein secretion were particularly interesting . We modified our ovulation assay so that it could be applied not only to the initial oocytes , but to ongoing ovulation throughout several days of mature adulthood ( see ‘Materials and methods' ) . By determining the total number of eggs laid as well as the steady-state fraction of females that contained an egg in the uterus , we could calculate the average time oocytes spend during ovulation and within the uterus ( Table 2 ) . 10 . 7554/eLife . 00415 . 012Table 2 . The effect of disrupting protein secretion or Hr39 expression during adulthood on the rate of egg laying and uterine egg contentDOI: http://dx . doi . org/10 . 7554/eLife . 00415 . 012GenotypeEgg laying in 2 days*Egg distribution in 6 hrEgg laying time ( min ) NEggs/female/dayNUterus with egg ( % ) Total timeOvulation timeUterus timesyt12Gal42577 . 3 ± 2 . 32842 . 9 ± 18 . 317 . 1 ± 0 . 59 . 8 ± 3 . 17 . 3 ± 3 . 1syt12>shits2572 . 2 ± 2 . 43268 . 8 ± 16 . 118 . 3 ± 0 . 65 . 7 ± 2 . 912 . 6 ± 3 . 0syt12Gal42565 . 4 ± 2 . 32962 . 1 ± 17 . 720 . 2 ± 0 . 77 . 6 ± 3 . 612 . 5 ± 3 . 6syt12>βCOPRNAi2553 . 8 ± 6 . 32885 . 7 ± 13 . 024 . 5 ± 2 . 9†3 . 5 ± 3 . 221 . 0 ± 4 . 0†syt12>sec23RNAi2551 . 5 ± 6 . 22986 . 2 ± 12 . 625 . 7 ± 3 . 1†3 . 5 ± 3 . 322 . 1 ± 4 . 2†syt12>Hr39RNAi2535 . 8 ± 2 . 2†3026 . 7 ± 15 . 8†36 . 9 ± 2 . 3†27 ± 6 . 1†9 . 8 ± 5 . 9*1 day = 22 hr at 29°C . †p<0 . 05 . All data are mean ± 95% confidence interval . T-test was used for egg laying , while Fisher's exact test was used for egg distribution . Interestingly , females with ectopic adult expression of shits in secretory cells were not defective in egg laying or ovulation ( Figure 5G , H ) despite their compromised ability to store sperm in the spermathecae . In particular , the time required per ovulation event was not increased compared to control ( Figure 5H ) . Females in which canonical secretion was disrupted by knocking down betaCOP or sec23 also laid eggs and ovulated similarly to controls ( Figure 5G , H ) . Nonetheless , the secretory cell requirement that we had previously documented for the early ovulation was confirmed when we knocked down Hr39 in adult secretory cells . After six days at the non-permissive temperature , these animals showed a significantly lower rate of egg laying , about half of normal ( Figure 5G ) and ovulated much more slowly than controls ( Figure 5H ) , requiring an average of 27 ± 6 . 1 min per egg compared to 7 . 6 ± 3 . 6 min in controls ( Table 2 ) . Our results show that at least two types of secretory cell products are released from reproductive gland secretory cells using different mechanisms . Products produced by the canonical protein secretory pathway are required to attract and store sperm in the spermathecae and to maintain their normal morphology . In addition , products needed to achieve a normal rate of ovulation require the function of Hr39 in secretory cells , but do not utilize the canonical secretory pathway .
Our experiments extend previous knowledge about the role reproductive tract secretions play in storing sperm . Sperm storage in the female reproductive tract is a general phenomenon in the animal kingdom including humans and insects ( Neubaum and Wolfner , 1999b ) . In mammals , carbohydrate-dependent binding of sperm to the oviduct epithelia is important in order to form a sperm reservoir ( Talevi and Gualtieri , 2010 ) . In the absence of glands and hence of secretions , Drosophila sperm are still stored in the seminal receptacle , but they are poorly motile and fertility is low ( Anderson , 1945; Allen and Spradling , 2008 ) . The same outcome is observed when spermathecal secretory cells are partially ablated in adults prior to mating ( Schnakenberg et al . , 2011 ) . The work reported here allows several additional conclusions . First , the initial attraction of sperm to the spermathecae within 6 hr of mating requires a minimum amount of secretion from the secretory cells ( SCs ) . Females with fewer than 25 SCs rarely contain sperm . In contrast , females with 80 or more SCs show the same high frequency of sperm in their spermathecae as wild type . The secreted attractive factor might interact with the accessory gland protein Acp36DE to facilitate uterine contraction ( Avila and Wolfner , 2009 ) , or act directly on sperm to regulate flagellar function ( Kottgen et al . , 2011; Yang et al . , 2011 ) . Second , the fact that sperm still move to the seminal receptacle in the absence of SCs shows that different mechanisms are involved in transporting sperm to the two different storage organs . Third , we found that female reproductive tract secretions are required to maintain sperm structurally . In the absence of secretory cells , sperm are not attracted to the spermathecae and those in the seminal receptacle aggregate and are difficult to individually assess ( Anderson , 1945; Allen and Spradling , 2008 ) . However , when protein secretion in SCs is disrupted using shits , sperm that do make it to the spermathecae exhibit distinctive morphological abnormalities . Finally , we documented that secretory cells competent to carry out canonical protein secretion and expressing Hr39 are required to store sperm during adulthood . When protein secretion was disrupted after eclosion , sperm storage in the spermathecae was drastically compromised . Since we did not mate these females until the day they were tested , our experiments show that any initial accumulation of secretory products during pupal development turns over or is insufficient to store new sperm . The requirement for new secretion from the reproductive glands is consistent with previous studies showing that some proteins in these glands are induced by mating ( Mack et al . 2006 ) . The fact that Hr39 is required confirms that this gene , which is known to play a prominent role during reproductive gland development and to be expressed in adult secretory cells ( Allen and Spradling , 2008 ) , does play a key functional role in the adult gland . Many but not all spermathecal protein mRNAs , including many that are likely to be involved in sperm maintenance , are greatly reduced in an Hr39 mutant that retains spermathecae ( Allen and Spradling , 2008 ) . A major finding of this study is that the secretory cells of Drosophila female reproductive glands are required for efficient ovulation . When secretory cell number is deficient initially , ovulation is drastically reduced , despite normal copulation and the presence of sperm in the reproductive tract . When secretory cell function is reduced during adulthood by knocking down expression of Hr39 , the time required for ovulation is greatly increased . Unlike the secretions that attract and stabilize sperm , ovulation is not disrupted by knocking down the protein secretory pathway . However , there are many possible reproductive tract secretions that might be released from the gland secretory cells by other mechanisms . A previous study by Schnakenberg et al . ( Schnakenberg et al . , 2011 ) examined the role of spermathecal secretory cells by partially ablating them during adulthood using secretory protein regulatory elements to drive the apoptosis inducer Hid . Like these authors , we found that reproductive tract secretions are required to attract sperm to the spermathecae and to maintain their normal structure . We extended these observations by showing that a minimum number of about 80 secretory cells is required for normal sperm attraction and storage , and that these functions require the canonical protein secretory pathway and Hr39 expression . Schnakenberg et al . reported that egg release from the uterus frequently but sporadically is reduced in females with a deficit of secretory cells , and suggested that secretory cells produce an initial , long-lasting lubricant that coats the uterus . In contrast , when secretion was compromised , we saw that egg laying was strongly and consistently reduced due to defects in ovulation rather than in egg release . Schnakenberg et al . did not study ovulation independently from egg laying . In contrast , we used assays that separate these processes , allowing the role of secretion in ovulation to emerge . In mammals , mature Graffian follicles compete for ovulation based on complex hormonal and biochemical signals that are closely tied to products locally produced by the follicle's granulosa and nascent luteal cells ( Mihm and Evans , 2008 ) . Much less is known about how individual follicles in the Drosophila ovary are selected for oviduct entry from a large pool . While nervous control of ovulation is clearly required to coordinate egg release with environmental and circadian factors ( Yang et al . , 2008 ) , the underlying mechanism of egg selection is likely to be more complex and involve local interactions as well as communication between the ovaries . Identification of the secretory cell product ( s ) that are required for ovulation would provide an important clue to uncovering these mechanisms . A particularly attractive possibility is that communication signals between the reproductive tract and the ovary have been partially conserved between mammals and Drosophila . This prospect is now strengthened by the finding that Hr39 is required for ovulation , like its mammalian counterpart , Lrh-1 . Prostaglandin-like molecules are known to regulate ovulation in mammals ( Dinchuk et al . , 1995; Lim et al . , 1997 ) , and Lrh-1 is thought to function by regulating expression of the prostaglandin-generating enzyme COX-II in mouse granulosa cells ( Duggavathi et al . , 2008 ) . A prostaglandin-like signal already is known to function during egg maturation in Drosophila ( Tootle and Spradling , 2008 ) , and a COX-II-like enzyme , CG10211 , is expressed in spermathecae under the control of Hr39 ( Allen and Spradling , 2008 ) . It will be interesting to determine if Lrh-1 functions in reproductive tract secretory cells . Finally , identifying additional glandular products acting in the Drosophila reproductive tract may elucidate additional pathways of communication between oviduct and ovary that are relevant to the induction of ovarian cancer .
Flies were reared on standard cornmeal-molasses food at 25°C unless otherwise indicated . Trans-heterozygous combinations of Hr397154/Ly92 ( Allen and Spradling , 2008 ) and lz3/34 ( from Bloomington Drosophila Stock Center , BDSC , Bloomington , IN ) were used for loss-of-function analysis . For the rescue experiment with fruGal4 driving UAS-mSP ( Yang et al . , 2009 ) , Hr397154/Cyo; fruGal4 females were crossed to Hr397154/Cyo; UAS-mSP or Hr397154/Cyo . Knockdown with RNAi or dominant negative constructs were carried out at 29°C and the following lines were used: UAS-cycARNAi ( V32421; Vienna Drosophila RNAi Center , Vienna , Austria ) , UAS-cdc2RNAi ( V41838 ) , UAS-Hr39RNAi ( V37694 ) , UAS-hntRNAi1 ( V101325 ) , UAS-hntRNAi2 ( V3788 ) , UAS-NRNAi ( gift from Sarah Bray ) , UAS-PsnDN ( UAS-Psn . 527 . D447A , BDSC ) , UAS-Su ( H ) DN ( Mukherjee et al . , 2011 ) , UAS-mCD8:GFP , lzGal4 ( BDSC ) , dpr5Gal ( 51B02; Pfeiffer et al . , 2008 ) , ppkGal4 ( Yang et al . , 2009 ) . In order to inhibit membrane recycling , the canonical exocytosis pathway , or Hr39 function in adult secretory cells , syt12Gal4 ( 47E02; Pfeiffer et al . , 2008 ) was crossed to UAS-shits ( Yang et al . , 2009 ) , while UAS-dcr2; syt12Gal4 , tubGal80ts was crossed to the RNAi line against bCOP ( BDSC 33741 ) , sec23 ( BDSC 32365 ) , or Hr39 ( V37694 ) at 18°C . Virgin females were selected 4 hr after eclosion and immediately shifted to 29°C for 6 days . UAS-RFP was used to monitor syt12Gal4 expression . For lineage tracing experiments , specific Gal4 driver was crossed to G-Trace lines to monitor real-time expression and lineage expression ( Evans et al . , 2009 ) . Clonal labeling and pupae preparation were as previously described ( Sun and Spradling , 2012 ) . The Notch activity reporter Su ( H ) GBE-Gal4 , UAS-mCD8:GFP was used to monitor Notch activation ( Zeng et al . , 2010 ) , and ProtB-GFP was used to visualize sperm DNA ( Manier et al . , 2010 ) . Control flies were derived from specific Gal4 driver crossed to wild-type Oregon-R . 4- to 6-day-old virgin females were fed with wet yeast 1–2 days before egg laying experiments . Five females were mated to 10 Oregon-R males in each bottle covered with the molasses plate at 25°C and the number of eggs was counted every day for 2 days except for experiments that perturb the exocytosis pathway or Hr39 function in adult secretory cells , which were carried out at 29°C for 2 days . For ovulation and copulation tests , single-pair matings between a 4- to 6-day-old virgin female and a ProtB-GFP male were carried out in the morning at 25°C , except for experiments that perturb the exocytosis pathway or Hr39 function in adult secretory cells , which were carried out at 29°C . 6 hr after mating , females were dissected to examine eggs inside the reproductive tract , and the corresponding vials were also examined for laid eggs . Female reproductive tracts were then fixed with paraformadehyde and sperm inside them were examined to determine copulation success . The number of sperm inside spermathecae was manually counted . In Table 2 , egg laying time ( in minutes ) = 22 × 60/number of eggs; the ovulation time = the egg laying time × ( 1 − egg distribution in uterus ) ; and uterus time = the egg laying time × egg distribution in uterus . The 95% confidence intervals were calculated correspondingly . Pupal and adult reproductive tract staining was carried out as previously described ( Sun and Spradling , 2012 ) . Briefly , tissues were dissected in Grace's media , fixed in 4% EM Grade Paraformadehyde for 15–20 min , and blocked in PBTG ( PBS + 0 . 3% Triton + 0 . 5% BSA + 2% normal goat serum ) . Incubation with primary antibody overnight was followed by a 2-hr incubation with secondary antibody and DAPI staining . Tissues were then mounted in Vectashield mounting media . The following primary antibodies were used: mouse anti-Hnt ( 1:75; Developmental Study Hybridoma Bank ) , rabbit anti-GFP ( 1:4000; Invitrogen ) , and chicken anti-β–Gal ( 1:1000; Abcam ) . Secondary antibodies were Alexa 488 and 546 goat anti-mouse , anti-rabbit , and anti-chicken ( 1:1000; Invitrogen ) . Images were acquired using the Leica TCS SP5 confocal microscope or the Zeiss Axioimager ZI microscope , and assembled using photoshop software . | Mammalian oviducts , or Fallopian tubes , convey egg cells from the ovaries to the uterus . Signalling between the ovary and oviduct , and secretory products produced throughout the reproductive tract , help to increase the likelihood of conception , minimise the loss of egg cells , and reduce the risk of ectopic pregnancy ( in which an embryo implants outside the uterus ) . These processes may also influence the development of ovarian cancer , since Fallopian tube secretory cells were recently identified as the source of the most common and lethal subtype of epithelial ovarian cancer , high grade serous ovarian cancer . Oviduct to ovary signalling is poorly understood in mammals . However , experiments using model organisms such as the fruit fly ( Drosophila melanogaster ) provide a potentially powerful approach to the problem , since many mechanisms in gametogenesis are conserved between species . In particular , secretions within the Drosophila female reproductive tract appear to boost reproductive success by interacting with sperm cells and seminal proteins , as in mammals . But whether these secretions reach the ovary and influence ovulation , or simply act on other aspects of reproduction such as mating , sperm storage , fertilisation or egg laying , remained unknown . In this study , Sun and Spradling identified new genes controlling reproductive gland development and used this knowledge to elucidate secretory cell function . By mutating these genes , or the nuclear hormone receptor Hr39 , they were able to reduce the total number of secretory cells that developed in the female reproductive tract , or to alter their function in adults . The ovaries of flies with abnormal secretory cell function contained as many egg cells as those of normal flies , but the mutant females laid fewer eggs . This indicates that secretory cells are required for at least one stage of reproduction . By comparing ovulation rates in mutant and normal flies , Sun and Spradling showed that the secretory cells generate a product that is specifically required for ovulation , and that production depends on Hr39 activity . This Hr39-dependent secretion is a good candidate for a conserved signal between the reproductive tract and ovary because mouse Lrh-1 , a mammalian gene closely related to Hr39 , is expressed in oviduct secretory cells and is itself required for ovulation . The secretory cells were also found to produce protein secretions that are necessary for female flies to store sperm in the reproductive tract after mating . By elucidating the roles played by female reproductive tract secretions , and demonstrating that they include a signal to the ovary that stimulates ovulation , the work of Sun and Spradling may lead to an increased understanding of ovarian cancer in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2013 | Ovulation in Drosophila is controlled by secretory cells of the female reproductive tract |
The cereblon modulating agents ( CMs ) including lenalidomide , pomalidomide and CC-220 repurpose the Cul4-RBX1-DDB1-CRBN ( CRL4CRBN ) E3 ubiquitin ligase complex to induce the degradation of specific neomorphic substrates via polyubiquitination in conjunction with E2 ubiquitin-conjugating enzymes , which have until now remained elusive . Here we show that the ubiquitin-conjugating enzymes UBE2G1 and UBE2D3 cooperatively promote the K48-linked polyubiquitination of CRL4CRBN neomorphic substrates via a sequential ubiquitination mechanism . Blockade of UBE2G1 diminishes the ubiquitination and degradation of neomorphic substrates , and consequent antitumor activities elicited by all tested CMs . For example , UBE2G1 inactivation significantly attenuated the degradation of myeloma survival factors IKZF1 and IKZF3 induced by lenalidomide and pomalidomide , hence conferring drug resistance . UBE2G1-deficient myeloma cells , however , remained sensitive to a more potent IKZF1/3 degrader CC-220 . Collectively , it will be of fundamental interest to explore if loss of UBE2G1 activity is linked to clinical resistance to drugs that hijack the CRL4CRBN to eliminate disease-driving proteins .
The ubiquitin-proteasome system ( UPS ) is a highly regulated component of the protein homeostasis network that dictates multiple cellular processes in eukaryotes ( Hershko and Ciechanover , 1998 ) . Through the orchestrated actions of ubiquitin-activating enzymes ( E1 ) , ubiquitin-conjugating enzymes ( E2 ) and ubiquitin-ligating enzymes ( E3 ) , the ε-amine of a lysine residue in a target protein is covalently conjugated with K48- or K11-linked poly-ubiquitin chains , thereby marking the target protein for proteasomal degradation ( Jin et al . , 2008; Komander and Rape , 2012; Pickart , 2001 ) . Recently , repurposing the Cullin-Ring E3 ligase complexes CRL4CRBN ( Cul4-RBX1-DDB1-CRBN ) and CRL2VHL ( Cul2-RBX1-EloB/C-VHL ) with small-molecule degraders to remove disease-driving proteins otherwise considered ‘undruggable’ has emerged as a novel therapeutic modality that has the potential to transform drug discovery and development ( Bondeson and Crews , 2017; Huang and Dixit , 2016; Lebraud and Heightman , 2017 ) . There are two types of small-molecule degraders that have been used to date . The first is represented by the immunomodulatory drugs ( IMiDs ) thalidomide ( THAL ) , lenalidomide ( LEN ) and pomalidome ( POM ) , as well as other cereblon modulating agents CC-122 , CC-220 , and CC-885 . This class of molecule docks into a tri-tryptophan pocket in the thalidomide-binding domain of cereblon , the substrate receptor of CRL4CRBN , to create a hotspot for protein-protein interactions thereby enhancing the binding of unique neomorphic substrate to cereblon , resulting in substrate ubiquitination and degradation ( Fischer et al . , 2014 ) ( Chamberlain et al . , 2014 ) ( Matyskiela et al . , 2016 ) ( Petzold et al . , 2016 ) . All three IMiD drugs promote the degradation of two hematopoietic transcription factors IKZF1 and IKZF3 to achieve anti-myeloma activity ( Krönke et al . , 2014 ) ( Lu et al . , 2014a ) ( Gandhi et al . , 2014 ) , whereas only LEN targets CK1α for effective degradation ( Krönke et al . , 2015 ) , which is presumably linked to its efficacy in myelodysplastic syndrome with chromosome 5q deletion . CC-220 is a significantly more potent IKZF1 and IKZF3 degrader than IMiD drugs ( Matyskiela et al . , 2018 ) ( Nakayama et al . , 2017 ) ( Schafer et al . , 2018 ) , and it is currently in clinical trials for relapsed/refractory multiple myeloma and systemic lupus erythematosus . By contrast , CC-885 is the only aforementioned cereblon modulating agent that allows cereblon to recognize translation termination factor GSPT1 for ubiquitination and degradation ( Matyskiela et al . , 2016 ) . The second type of small-molecule degraders is generally referred to as a proteolysis-targeting chimera ( PROTAC ) ( Sakamoto et al . , 2001 ) , which is composed of two linked protein binding ligands , with one engaging a target protein and the other interacting with an E3 ubiquitin ligase such as CRL4CRBN or CRL2VHL to trigger proximity-induced substrate ubiquitination and degradation ( Deshaies , 2015 ) ( Neklesa et al . , 2017 ) . Many PROTACs have been described recently , but the clinical value of this approach has not yet been established . CRL4CRBN belongs to the multi-subunit cullin-RING E3 ubiquitin ligase ( CRL ) family containing over 200 members ( Petroski and Deshaies , 2005 ) ( Lehti et al . , 2013 ) . The mammalian cullin scaffold proteins ( including Cul1 , Cul2 , Cul3 , Cul4A , Cul4B , Cul5 , and Cul7 ) bring their substrates into close proximity with E2 ubiquitin-conjugating enzymes , thereby enabling effective substrate ubiquitination ( Petroski and Deshaies , 2005 ) ( Lehti et al . , 2013 ) . SCF ( Skp1-Cul1-F-box ) Cdc4 , the founding member of the cullin-RING E3 ligase family , was first discovered in the budding yeast Saccharomyces cerevisiae , in which SCFCdc4 works in conjunction with a single E2 ubiquitin-conjugating enzyme Cdc34 to promote the polyubiquitination of a variety of SCF substrates ( Feldman et al . , 1997 ) ( Skowyra et al . , 1997 ) ( Blondel et al . , 2000 ) ( Jang et al . , 2001; Perkins et al . , 2001 ) . Human Cdc34/UBE2R1 can substitute for yeast Cdc34 in Saccharomyces cerevisiae underscoring their functional conservation ( Plon et al . , 1993 ) . However , in contrast to its dominant role in catalyzing the ubiquitination of SCF substrates in yeast , Cdc34 coordinates ubiquitination with UBE2D3/UbcH5c via a sequential ubiquitination mechanism to improve reaction rate and efficiency in human cells . In brief , Cdc34 acts as an ubiquitin chain elongation enzyme that assembles the K48-linked ubiquitin chains on mono-ubiquitins pre-conjugated to SCF substrates by UBE2D3 ( Pan et al . , 2004 ) . Such sequential ubiquitination by two E2 enzymes was first reported for the anaphase-promoting complex ubiquitin ligase ( Rodrigo-Brenni and Morgan , 2007 ) . More recently , the RING1-IBR-RING2 ( RBR ) E3 ligase ARIH1 was shown to tag client substrates of CRL1 , CRL2 and CRL3 with monoubiquitin , thereby enabling CDC34-dependent K48-linked ubiquitin chain elongation ( Scott et al . , 2016 ) . This finding points to a potentially more prevailing mechanism of ubiquitin chain priming and extending carried out by two distinct E2s . Several ubiquitin conjugation E2 enzymes have been reported to regulate CRL4 substrates as well . For instance , in response to UV irradiation , the CRL4Cdt2 ligase complex mediates the proteolysis of Cdt1 with the help of E2 enzymes UBE2G1 and its paralog UBE2G2 , while working together with a different E2 enzyme UbcH8/UBEL6 to trigger the degradation of p21 and Set8 in human cells ( Shibata et al . , 2011 ) . Despite the proven cellular efficacy and clinical success of many cereblon modulating agents , it remain unknown whether unique ubiquitin E2 enzymes control the ubiquitination of each specific cereblon neomorphic substrate , and whether loss of E2 enzymes contributes to resistance to these agents .
The clinical course of multiple myeloma typically follows a recurring pattern of remission and relapse with resistance to IMiD drugs based combination regimens ( Harousseau and Attal , 2017 ) . Such relapse is not frequently associated with cereblon downregulation and/or mutation ( Kortüm et al . , 2016; Qian et al . , 2018 ) ( Zhu et al . , 2011 ) . Hence , we reasoned that resistance to IMiD drugs in myeloma could be ascribed to reduced degradation of IKZF1 and IKZF3 as a result of inactivation of other essential components of the CRL4CRBN ligase complex , for instance the E2 ubiquitin conjugation enzyme . To look for such proteins , we devised a high-throughput CRISPR-Cas9 screen approach to monitor the effect of individual knockout of a gene of interest on POM-induced degradation of IKZF1 protein tagged with enhanced ProLabel ( ePL ) , a small β-galactosidase N-terminal fragment ( Figure 1A ) , and created a single guide RNA ( sgRNA ) library containing three sgRNAs for each of the 41 annotated E2 enzymes in the human genome , as well as three non-targeting sgRNAs in arrayed format ( Supplementary file 1 ) . The ePL tag complements with the large β-galactosidase C-terminal fragment to form an active enzyme that hydrolyzes substrate to emit a chemiluminescent signal , allowing the measurement of ePL-IKZF1 fusion protein level in a high-throughput fashion . To determine the robustness of this screening approach , we transduced U937 cells stably expressing Cas9 and ePL-IKZF1 ( U937_Cas9_ePL-IKZF1 ) with lentiviral vector expressing a non-targeting or CRBN-specific sgRNA . We chose U937 cells for assessing the effect of individual gene knockout on POM-induced IKZF1 degradation because we were able to express the ePL-tagged IKZF1 fusion protein below the level of endogenous IKZF1 protein and POM can induce the efficient degradation of both endogenous and ePL-tagged IKZF1 ( Figure 1—figure supplement 1 ) . Four days post transduction , cells were seeded into 384-well plates pre-dispensed with either DMSO vehicle control or POM at varying concentrations . Sixteen hours after incubation , IKZF1 degradation was assessed using the ePL luminescent assay . As expected , cereblon knockout completely abrogated the degradation of ePL-tagged IKZF1 fusion protein ( Figure 1—figure supplement 1 ) . Using this approach we then evaluated the effect of individual knockout of each E2 enzyme on ePL-IKZF1 degradation induced by POM . Out of 41 E2 enzymes , UBE2G1 and to a lesser extent UBE2M , UBE2D3 , and UBE2D2/UbcH5b , when depleted , imposed statistically significant inhibition on the ePL-IKZF1 degradation ( Figure 1B , and Figure 1—figure supplement 2D , F , I ) . UBE2M , also called UBC12 , is a NEDD8-conjuating enzyme , which regulates , via neddylation , the activity of all Cullin Ring E3 ligases including CRL4CRBN ( Petroski and Deshaies , 2005 ) ( Gong and Yeh , 1999 ) ( Pan et al . , 2004 ) . Indeed , MLN4924 , an inhibitor of NEDD8-activating enzyme ( Soucy et al . , 2009 ) , prevented the degradation of ePL-IKZF1 induced by POM ( Figure 1—figure supplement 2V ) . The effect of UBE2M knockout , however , was much less pronounced ( Figure 1—figure supplement 2I ) . Given the well-established role of UBE2M in cell proliferation and survival , we reasoned that this difference could be explained by that U937 cells only with partial UBE2M loss survived four days after CRISPR gene editing . Consistent with this notion , cellular fitness was markedly reduced by 48 hr treatment of MLN2924 at concentrations that elicited near-complete blockage of POM-induced IKZF1 degradation in the U937_Cas9_ePL-IKZF1 cell line used in the screen ( Figure 1—figure supplement 2V , W ) . CRISPR knockout of UBE2G1 also attenuated the destabilization of endogenous IKZF1 by POM in U937 cells , and this defect could be rescued by wild-type UBE2G1 , but not a UBE2G1 enzymatically-dead mutant ( C90S , Figure 1C ) . Knockout of UBE2D3 , on the other hand , showed very little effect on the degradation of endogenous IKZF1 ( Figure 2C ) . Thus , we reasoned that there are additional E2 enzyme ( s ) modulating the degradation of IKZF1 cooperatively with UBE2G1 . To identify such E2 ( s ) , we evaluated the effect of double knockout of UBE2G1 and one of the 41 E2 enzymes on IKZF1 degradation using a dual gRNA-directed gene knockout approach ( Figure 2A and Figure 2—figure supplement 1 ) . Notably , double knockout of UBE2G1 and UBE2D3 produced more inhibition of POM-induced degradation of ePL-tagged and endogenous IKZF1 than either single knockout alone ( Figure 2B and C ) . Combinatorial ablation of UBE2G1 with UBE2E1 or UBE2M also demonstrated subtle but noticeable further inhibition on IKZF1 degradation ( Figure 2—figure supplement 1E , I ) . Although UBE2D2 knockout slightly attenuated the POM-induced ePL-IKZF1 degradation ( Figure 1—figure supplement 2D ) , double knockout of UBE2D2 and UBE2G1 did not significantly augment the inhibition of ePL-IKZF1 degradation imposed by UBE2G1 single knockout ( Figure 2—figure supplement 1D ) . To assess the substrate selectivity of UBE2G1 , we determined the effect of UBE2G1 knockout on the degradation of IKZF1 , its paralog IKZF3 and other well-characterized cereblon neomorphic substrates , triggered by their respective cereblon modulating agents including cereblon-based PROTACs . Ablation of UBE2G1 significantly diminished the degradation of IKFZ1 ( Figure 3A , and Figure 3—figure supplement 1A , B ) , IKZF3 ( Figure 3A , and Figure 3—figure supplement 1A , B ) and ZFP91 ( Figure 7B , and Figure 7—figure supplement 1B , D ) by LEN , POM , and CC-220 , as well as CK1α degradation ( Figure 7B , and Figure 7—figure supplement 1B , D ) by LEN in OPM2 , DF15 and MM1S myeloma cells . UBE2G1 loss also reduced the degradation of GSPT1 induced by CC-885 in myeloma cell lines OPM2 , DF15 and MM1S ( Figure 3B , and Figure 3—figure supplement 1C , D ) , AML cell lines OCI-AML2 , U937 , MOLM-13 and MV4-11 ( Figure 3—figure supplement 1E–H ) , as well as 293T human embryonic kidney cells ( Figure 3—figure supplement 1I ) . The GSPT1 degradation defect conferred by UBE2G1 depletion could also be rescued by UBE2G1 wild-type but not C90S mutant in 293 T cells ( Figure 3—figure supplement 1I ) . UBE2G1 loss also blocked the degradation of BRD4 induced by the cereblon-based BET PROTAC dBET1 ( Winter et al . , 2015 ) , but not the VHL-based BET PROTAC MZ1 ( Zengerle et al . , 2015 ) in 293 T cells ( Figure 3C ) . Consistently , UBE2G1 depletion prolonged the protein half-lives of IKZF1 and IKZF3 in OPM2 cells treated with POM ( Figure 3D ) . Lastly , UBE2G1 loss did not affect stability of SCF substrates p27 and c-Myc , indicating the specific regulation of CRL4CRBN by UBE2G1 ( Figure 3D ) . UBE2G1 and its paralog UBE2G2 share similar domain structures with CDC34 . A common structural feature of these three E2 enzymes is an acidic loop ( Figure 4—figure supplement 1A , highlighted with red ) in the vicinity of their respective active site cysteines ( Figure 4—figure supplement 1A , highlighted with blue ) , which facilitates the direct binding with ubiquitin and enables K48-linked ubiquitin chain assembly in the absence of associated E3 ligases ( Choi et al . , 2015 ) . In association with gp78 , an ER membrane bound RING finger E3 ubiquitin ligase , UBE2G2 directly tags misfolded proteins with K48-linked ubiquitin chains preassembled on the catalytic cysteine of UBE2G2 , resulting in ER associated protein degradation ( ERAD ) ( Li et al . , 2007 ) . Although it has been shown that UBE2G1 and UBE2G2 redundantly mediate the destabilization of the CRL4Cdt2 substrate Cdt1 in response to UV irradiation , the direct transfer of ubiquitin to Cdt1 by UBE2G1 or UBE2G2 has not been demonstrated ( Shibata et al . , 2011 ) . Next , we employed a reconstituted in vitro ubiquitination assay to address the role of UBE2G1 and UBE2D3 in ubiquitination of IKZF1 and GSPT1 induced by POM and CC-885 , respectively . We monitored the production of ubiquitin conjugates of IKZF1 and GSPT1 catalyzed by UBE2D3 alone , UBE2G1 alone , or in combination , in the presence of Ube1 ( E1 ) , Cul4-Rbx1 , cereblon-DDB1 , ubiquitin and ATP with or without POM or CC-885 . UBE2D3 alone produced ubiquitin conjugates of IKZF1 or GSPT1 mainly with a single or di-ubiquitin moiety in a POM-or CC-885- dependent manner ( Figure 4A and B , lanes 1 and 2 ) . By contrast , UBE2G1 alone failed to display any ubiquitin conjugating activity for both IKZF1 and GSPT1 ( Figure 4A and B , lanes 3 and 4 ) . However , when combined with UBE2D3 , UBE2G1 significantly promoted the extent of ubiquitination of both substrates ( Figure 4A and B , lanes 2 and 6 ) . Moreover , the ubiquitin conjugates of IKZF1 or GSPT1 formed with both UBE2G1 and UBE2D3 , but not UBE2D3 alone , were exclusively K48-linked , because the wild-type ubiquitin used in the reconstituted ubiquitination reaction could be functionally replaced by ubiquitin mutant K48-only ( with six remaining lysine residues mutated to arginine ) ( Figure 4C and D , lanes 2 and 6 ) , but not by K48R ( remaining lysine residues were intact ) ( Figure 4C and D , lanes 8 and 12 ) , and the ubiquitination pattern of IKZF1 or GSPT1 catalyzed by UBE2D3 exhibited no obvious difference between wild-type ubiquitin and K48-only or K48R mutant ( Figure 4A and B , lane 2; Figure 4C and D , lanes 2 and 8 ) . To further explore the mechanism underlying the cooperativity between UBE2G1 and UBE2D3 , we separated the ubiquitination reaction of GSPT1 into two steps . First , following GSPT1 ubiquitination by UBE2D3 alone , we isolated GSPT1 ubiquitin conjugates from the rest of reaction components using a size-exclusion column ( Figure 4E , lanes 1 and 2 ) . We then incubated the purified GSPT1 ubiquitin conjugates with UBE2G1 , Ube1 ( E1 ) , cereblon-DDB1 , ubiquitin and ATP with or without CC-885 and Cul4-Rbx1 . We found that UBE2G1 was capable of catalyzing the further ubiquitination of GSPT1 only with prior-conjugated ubiquitin ( Figure 4E , lanes 1–4 , note the conversion of the mono-ubiquitinated GSPT1 into di- or tri-ubiquitinated forms ) , and this action required the presence of CC-885 and Cul4A-Rbx1 ( Figure 4E , lanes 3–6 ) . To rule out the possibility that the UBE2G1 function observed above is simply an artifact of bacterial recombinant protein , we reconstituted the ubiquitination reaction using FLAG-tagged UBE2G1 and FLAG-tagged UBE2D3 proteins purified from 293T UBE2G1-/- cells , in which ectopic overexpression of FLAG-tagged UBE2G1 and UBE2D3 , but not their respective enzymatically-dead mutant , could partially rescue the defect in CC-885-induced GSPT1 degradation elicited by UBE2G1 loss ( Figure 4—figure supplement 1B ) . In agreement with our previous findings , GSPT1 ubiquitination catalyzed in vitro by FLAG-UBE2G1 and FLAG-UBE2D3 , alone or in combination , was similar to what was observed with bacterial recombinant UBE2G1 and UBE2D3 ( Figure 4—figure supplement 1C ) . In addition , we examined the in vivo function of UBE2G1 and UBE2D3 in the regulation of POM-induced ubiquitination of IKZF1 ectopically expressed in 293 T cells . Ablation of both UBE2G1 and UBE2D3 significantly reduced , but did not completely block , the mono- and polyubiquitination of IKZF1 induced by POM , indicating the existence of additional E2 enzymes with redundant function ( Figure 5A and Figure 5—figure supplement 1A ) . Reintroduction of UBE2G1 alone via transient transfection in 293T UBE2G1-/-; UBE2D3-/- cells dramatically enhanced the extent of POM-induced IKZF1 polyubiquitination ( Figure 5B and Figure 5—figure supplement 1B , lanes 5–8 vs 1–4 ) . Transient overexpression of UBE2D3 alone promoted IKZF1 monoubiquitination as well as polyubiquitination but to a lesser extent ( Figure 5B and Figure 5—figure supplement 1B , lanes 9–12 vs 1–4 ) . Overexpression of both achieved an additive or possibly synergistic effect on IKZF1 ubiquitination ( Figure 5B and Figure 5—figure supplement 1B , lanes 13–16 vs 5–12 , note the conversion of monoubiquinated IKZF1 into its polyubiquitinated forms ) . In keeping with the in vitro finding ( Figure 4A ) , IKZF1 was mainly mono-ubiquitinated in 293T UBE2G1-/- cells following POM treatment , and UBE2G1 wild-type but not C90S mutant could restore IKZF1 polyubiquitination ( Figure 5C and Figure 5—figure supplement 1C ) . Since UBE2G1 depletion significantly attenuated the degradation of all cereblon neomorphic substrates , we reasoned that UBE2G1 protein downregulation , gene deletion or mutation might lead to reduced CRL4CRBN activity , thereby leading to resistance to cereblon modulating agents . To test this hypothesis , we surveyed the protein expression level of UBE2G1 in myeloma cell lines with variable sensitivity to LEN and POM ( Figure 6A ) . LEN sensitive cell lines MM1S , OPM2 , DF15 and NCI-H929 displayed higher expression level of UBE2G1 than LEN medium-sensitive cell line ANABL-6 and LEN resistant cell lines EJM , L363 , SKMM-2 , CAG and ARH-77 , and more strikingly UBE2G1 expression in SKMM-2 was undetectable ( Figure 6B ) . As expected , reintroduction of UBE2G1 wild-type but not C90S mutant significantly augmented the antiproliferative effect of both LEN and POM , which was linked to the enhanced degradation of IKZF1 and IKZF3 ( Figure 6C and D ) . UBE2G1 loss also conferred resistance to CC-885 in OCI-AML2 , U937 , MOLM-13 , and MV4-11 AML cells ( Figure 6—figure supplement 1A–D ) , as well as 293 T cells , and this defect could be rescued by UBE2G1 wild-type but not C90S mutant ( Figure 6—figure supplement 1E ) . Moreover , UBE2G1 deletion conferred resistance to BET PROTAC dBET1 but not MZ1 in 293 T cells ( Figure 6—figure supplement 1F ) . Lastly , UBE2G1 knockout in DF15 , MM1S and OPM2 myeloma cells conferred significant resistance to LEN and POM ( Figure 7A , and Figure 7—figure supplement 1A , C- ) . Importantly , these cells were only partially resistant to CC-220 , in keeping with the increased efficiency in triggering IKZF1 and IKZF3 degradation ( Figure 7A and B and Figure 7—figure supplement 1A–D ) . These results indicate that UBE2G1 deficiency may be a key differentiator in the clinical success of cereblon modulating agents .
The sequential recruitment of two functionally distinct E2s by a single E3 is a general mechanism for substrate ubiquitination conserved from yeast to human ( Rodrigo-Brenni and Morgan , 2007 ) ( Wu et al . , 2010 ) ( Kleiger and Deshaies , 2016 ) . In this work , we provide evidence that CRL4CRBN deploys the same mechanism to mark its neomorphic substrates with K48-linked poly-ubiquitin chains for proteasomal degradation . Mechanistically , UBE2D3 transfers the first ubiquitin onto the lysine residue ( s ) of the cereblon neomophic substrate , thereby enabling UBE2G1 to assemble the K48-linked ubiquitin chains onto the initial anchor ubiquitin ( Figure 4F ) . This orchestrated action between UBE2D3 and UBE2G1 closely resembles the cooperativity of UBE2D3 and Cdc34 in promoting IκBα polyubiquitination mediated by SCFβTRCP2 ( Wu et al . , 2010 ) , except that unlike Cdc34 , UBE2G1 does not possess the ability to transfer ubiquitin onto substrates without prior ubiquitin conjugation . UBE2G1 was known to produce K48-linked poly-ubiquitin chains in the absence of an E3 ubiquitin ligase ( Choi et al . , 2015 ) , but UBE2G1 cannot promote GSPT1 ubiquitination in the absence of CC-885 or Cul4A-Rbx1 ( Figure 4E ) , indicating that close proximity of UBE2G1 to cereblon neomorphic substrates bridged by CRL4CRBN is required to increase the processivity of UBE2G1 at physiological concentrations . This provides an explanation to how cereblon modulating agents can induce effective substrate degradation via K48-linked polyubiquitination , as well as speaks to the potential role of UBE2G1 in mediating the ubiquitination and degradation of other CRL4 cognate and/or neomorphic substrates . This is supported by the impaired degradation of p21 and RBM39 induced by UV irradiation and E7070 treatment , respectively , in 293T UBE2G1-/- cells ( Figure 3—figure supplement 2A , B ) . Although UBE2D3 and UBE2G1 cooperatively regulate the ubiquitination of cereblon neomorphic substrates , ablation of UBE2D3 exhibited very little impact on substrate degradation as compared to loss of UBE2G1 , suggesting that additional E2s might fulfill the role of UBE2D3 , for example , other UBE2D family proteins with a high degree of sequence homology including UBE2D1/UbcH5a , UBE2D3 , and UBE2D4 ( Figure 4—figure supplement 2A ) . Indeed , UBE2D1 and UBE2D2 acted synergistically with UBE2G1 in catalyzing the in vitro ubiquitination of GSPT1 in the presence of CC-885 , whereas cooperativity among UBE2D1 , UBE2D2 and UBE2D3 cannot be detected ( Figure 4—figure supplement 2B ) . Knockout of UBE2D1 , UBED2 or UBE2D4 , however , did not further enhance the inhibition of POM-induced IKZF1 degradation imposed by UBE2G1 knockout in U937 cells ( Figure 2—figure supplement 1C , D , E ) . Below we consider various possible explanations for the surprising finding . First , among all 4 UBE2D family proteins , UBE2D3 is likely the major but not the only E2 utilized by CRL4CRBN to promote neomorphic substrate ubiquitination , possibly owing to their differences in interaction with CRL4CRBN , binding affinity with CRL4CRBN or simply relative protein abundance . Knockout of UBE2D3 could potentially remove the gridlock on CRL4CRBN , enabling its binding with other UBE2D family proteins which otherwise are much less engaged by CRL4CRBN under physiological conditions . If so , it is not surprising to observe the lack of cooperativity between UBE2G1 and UBE2D1 , UBE2D2 or UBE2D4 in promoting IKZF1 degradation in UBE2D3 wild-type cells , as well as the lesser effect of UBE2D3 knockout on IKZF1 degradation than that of UBE2G1 knockout . Second , it is possible that similar to UBE2M/UBC12 , UBE2D1 , UBE2D2 or UBE2D4 is essential in U937 cells , and therefore only cells with partial inactivation of these three proteins survived after CRISPR knockout , resulting in the underestimation of their involvement in IKZF1 ubiquitination . Third , redundancy among UBE2D1 , UBE2D2 and UBE2D4 in replacing the function of UBE2D3 could also prevent the identification of their cooperativity with UBE2G1 by a single gene inactivation approach as used in our screen . Further studies exploiting the combinatorial inactivation of UBE2D3 with one , two or all three remaining UBE2D family proteins will best address this issue . Lastly , we could not rule out the following less satisfying explanations: 1 ) ineffective gene editing efficiency of the sgRNAs we used to knockout UBE2D1 , UBE2D2 and UBE2D4; 2 ) insufficient assay sensitivity leading to failure in detecting subtle changes of IKZF1 degradation conferred by double knockout of UBE2G1 with less critical E2 ( s ) versus UBE2G1 knockout alone; 3 ) indirect impact of UBE2D3 inactivation or off-target effect of UBE2D3 sgRNAs on IKZF1 ubiquitination . It is also interesting to point out that UBE2D3 knockout further blocked the degradation of IKZF1 in UBE2G1 deficient cells ( Figure 2 ) , whereas the in vitro reconstituted ubiquitination assay clearly suggests that these two enzymes work in sequence ( Figure 4 ) . Although double knockout of UBE2G1 and UBE2D3 significantly blocked the ubiquitination and degradation of IKZF1 in U937 and 293 T cells ( Figures 2 and 5 ) , this effect is still much less pronounced than that of CRBN knockout ( Figure 1—figure supplement 1 ) , indicating the existence of additional mechanism ( s ) modulating IKZF1 degradation . Based on the following observations , we speculate that UBE2D family proteins or other yet-to-be-identified E2 ( s ) could promote the assembly of mixed polyubiquitin chains on cereblon neosubstrates in the absence of UBE2G1 , resulting in their ineffective yet significant degradation . First , UBE2D1 and to a lesser extent UBE2D2 are capable of catalyzing CC-885-dependent in vitro polyubiquitination of GSPT1 without the help of UBE2G1 ( Figure 4—figure supplement 2B , lanes 2 , 4 and 6 ) . Second , UBE2D2 knockout impaired the POM-induced degradation of ePL-tagged IKZF1 ( Figure 1—figure supplement 2D ) . Third , UBE2D3 overexpression promoted in vivo polyubiquitination of IKZF1 in 293T UBE2G1-/-; UBE2D3-/- cells ( Figure 5B ) , as well as the degradation of GSPT1 in 293T UBE2G1-/- cells ( Figure 4—figure supplement 1B ) . Moreover , we might also underappreciate the role of additional ubiquitin chain extending E2 ( s ) that could functionally substitute for UBE2G1 . For instance , a UBE2G1 paralog UBE2G2 , which was shown to cooperate with UBE2G1 to mediate the polyubiquitination of Cdt1 by CRL4Cdt2 ( Shibata et al . , 2011 ) , could play a similar redundant role in coordinating the ubiquitin chain assembly on cereblon neomorphic substrates . However , significant loss of cellular fitness following UBE2G2 knockout in the presence or absence of UBE2G1 in U937 cells could prevent the identification of UBE2G2 in our screen . Further experimentation is again required to explore these hypotheses . Taken together , we postulate that the CM-induced proteolysis of cereblon neomorphic substrates is both redundantly and cooperatively regulated by UBE2D family proteins and UBE2G1/2 , with UBE2G1/2 playing a key role in enhancing the rate and extent of K48-linked substrate ubiquitination , resulting in rapid and efficient substrate degradation . Although loss of UBE2G1 conferred resistance to LEN and POM in human myeloma cell lines , it remains to be seen whether UBE2G1 deficiency occurs in human myeloma patients with inherent or acquired resistance to IMiD drug treatment , especially those with normal cereblon expression . The myeloma cell line SKMM-2 , which lost both copies of the UBE2G1 gene ( based on gene copy number characterization in Cancer Cell Line Encyclopedia ) , and has undetectable UBE2G1 protein expression ( Figure 6B ) , was derived from a human myeloma patient who never received any prior treatment with IMiD drugs ( Eton et al . , 1989 ) , warranting the further clinical evaluation of UBE2G1 activity in myeloma patients . Given that UBE2G1 inactivation conferred resistance to all CMs tested and also to cereblon-based PROTACs , patient stratification approaches based on UBE2G1 status might be applicable to the development of IMiD drugs and other novel cereblon modulating agents for a variety of human diseases . Lastly , CC-220 , a novel CM that targets IKZF1 and IKZF3 for degradation much more effectively than does LEN or POM , retained strong antitumor activity at clinically achievable concentrations ( Schafer et al . , 2018 ) in UBE2G1-deficient myeloma cells , suggesting that human patients with resistance to CM drugs owing to diminished UBE2G1 function may be responsive to next-generation CMs that possess higher efficiency and/or potency for degrading the same target protein .
Purified E1 , E2 , ubiquitin , Cul4A-Rbx1 , and cereblon-DDB1 proteins were used to reconstitute the ubiquitination of MBP-fused GSPT1 or IKZF1 substrates in vitro . Purified recombinant human Ube1 E1 ( E-305 ) , UbcH5a/UBE2D1 ( E2-616-100 ) , UbcH5b/UBE2D2 ( E2-622-100 ) , UbcH5c/UBE2D3 ( E2-627-100 ) , wild-type ubiquitin ( U-100H ) , K48R ubiquitin ( UM-K48R-01M ) , and K48-only ubiquitin ( UM-K480-01M ) were purchased from R and D systems . For the ubiquitination of IKZF1 or GSPT1 shown in Figure 4A and B , and Figure 4—figure supplements 1C and 2B , reaction components were mixed to final concentrations of 80 mM ATP , 1 . 5 µM Ube1 , 275 µM Ub , 2 µM Cul4-Rbx1 , 2 µM cereblon-DDB1 , 5 uM IKZF1 ( a . a . 140–168 ) or 5 µM MBP-GSPT1 ( a . a . 437–633 ) as indicated , and then 5 µM UBE2D1 , 5 µM UBE2D2 , 5 µM UBE2D3 ( purified from either from E . coli or human cells ) , or 7 . 5 µM UBE2G1 ( purified from either from E . coli or human cells ) , was added alone or in combination , as indicated . Reactions were incubated in the presence of either DMSO or 80 µM compound ( pomalidomide or CC-885 ) in ubiquitination assay buffer ( 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 10 mM MgCl2 ) . To start the reactions , E1 , E2 , ATP and ubiquitin were pre-incubated for 30 min , and separately MBP-substrate , CRBN-DDB1 , Cul4-Rbx1 , and compound were pre-incubated for 5 min at room temperature , before ubiquitination reactions were started by mixing the two pre-incubations . Reactions were incubated at 30°C for 2 hr before separation by SDS-PAGE followed by immunoblot analysis using anti-MBP antibody ( MBP-probe R29 . 6 , Santa Cruz ) . For the ubiquitin mutant reactions shown in Figures 4C , D , 275 µM K48-only or K48R ubiquitin was substituted for wild-type ubiquitin as indicated , and the E . coli-purified UBE2G1 was used . Reactions were incubated at 30°C for 2 hr before separation by SDS-PAGE followed by immunoblot analysis using anti-MBP antibody ( MBP-probe R3 . 2 , Santa Cruz ) . For the ubiquitination of a pre-ubiquitinated substrate shown in Figure 4E , MBP-GSPT1 ( a . a . 437–633 ) was incubated with 80 mM ATP , 3 µM Ube1 , 600 µM Ub , 4 µM Cul4-Rbx1 , 4 µM cereblon-DDB1 , 5 µM UBE2D3 ( purified from E . coli ) , and 80 uM CC-885 for 4 hr before separation of the reaction over a 10/300 S200 GL ( GE 17-5175-01 ) size exclusion chromatography column to separate the substrate from the rest of the ubiquitination reaction components . 1 . 25 µM purified MBP-GSPT1 was then used as the substate in ubiquitnation reactions including 80 mM ATP , 1 . 5 µM Ube1 , 600 µM Ub , 2 µM Cul4-Rbx1 , 2 µM cereblon-DDB1 , 80 µM CC-885 , and 7 µM UBE2G1 ( purified from E . coli ) . Reactions were incubated at 30°C for 2 hr before separation by SDS-PAGE followed by immunoblot analysis using anti-MBP antibody ( MBP-probe R29 . 6 , Santa Cruz ) . Human embryonic kidney cell line 293T ( Clontech ) was maintained in Dulbecco’s Modified Eagle’s medium ( DMEM; Invitrogen ) supplemented with 10% fetal bovine serum ( FBS; Invitrogen ) , 1x sodium pyruvate ( Invitrogen ) , 1x non-essential amino acids ( Invitrogen ) , 100 U/mL penicillin ( Invitrogen ) , and 100 µg/mL streptomycin ( Invitrogen ) . Acute myeloid leukemia cell lines U937 , MOLM-13 , and MV4-11 and myeloma cell line MM1S were purchased from American Tissue Culture Collection ( ATCC ) . Acute myeloid leukemia cell line OCI-AML2 cell line and myeloma cell lines OPM2 was purchased from Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH ( DSMZ ) . Myeloma cell line DF15 was obtained from Dr John Shaughnessy ( University of Arkansas , Little Rock , AR , USA ) . U937 , MOLM-13 , OPM2 , MM1S and DF15 cell lines were maintained in Roswell Park Memorial Institute ( RPMI ) 1640 tissue culture medium ( Invitrogen ) supplemented with 10% FBS , 1x sodium pyruvate , 1x non-essential amino acids , 100 U/mL penicillin , and 100 µg/mL streptomycin . MV4-11 cell line was maintained in Iscove’s Modified Dulbecco’s medium ( IMDM; ( Invitrogen ) supplemented with 10% FBS , 1x sodium pyruvate , 1x non-essential amino acids , 100 U/mL penicillin , and 100 µg/mL streptomycin . OCI-AML2 cell line was maintained in minimal essential medium ( MEM; Invitrogen ) supplemented with 10% FBS , 1x sodium pyruvate , 1x non-essential amino acid , 100 U/mL penicillin , and 100 µg/mL streptomycin . All cell lines were cultured at 37°C with 5% CO2 in the relevant media mentioned above . All cell lines were authenticated by autosomal STR profiling . All cell lines were tested routinely using the MycoAlert Mycoplasma Detection Kit ( Lonza ) , and confirmed mycoplasma-negative . UBE2G1 and UBE2D3 complimentary deoxyribonucleic acid ( cDNA ) clones were purchased from Dharmacon . The coding regions of UBE2G1 and UBE2D3 were polymerase chain reaction ( PCR ) -amplified and shuttled into pDONR223 via BP ( attB and attP ) recombination to generate pDONR223-UBE2G1 , pDONR223-FLAG-UBE2G1 , and pDONR223-FLAG-UBE2D3 . Site-directed mutagenesis using overlapping PCR was then carried out to generate pDONR223-UBE2G1-CR ( CRISPR resistant ) , pDONR223-UBE2G1-C90S-CR , pDONR223-FLAG-UBE2G1-CR , and pDONR223-FLAG-UBE2G1-C90S-CR , and pDONR223-FLAG-UBE2D3-C85S . Next , gateway donor vectors pDONR223-UBE2G1-CR and pDONR223-UBE2G1-C90S-CR were shuttled into plenti-Ubcp-gateway-IRES-Pur or plenti-PGK-gateway-IRES-Pur via LR ( attL and attR ) recombination to generate plenti-Ubcp-UBE2G1-CR-IRES-Pur , plenti-Ubcp-UBE2G1-C90S-CR-IRES-Pur , plenti-PGK-UBE2G1-CR-IRES-Pur , and plenti-PGK-UBE2G1-C90S-CR-IRES-Pur . Gateway donor vectors pDONR223-FLAG-UBE2G1-CR , pDONR223-FLAG-UBE2G1-C90S-CR , pDONR223-FLAG-UBE2D3 , and pDONR223-FLAG-UBE2D3-C85S were shuttled into plenti-EF1α-gateway-IRES-Pur via LR recombination to generate plenti-EF1α-FLAG-UBE2G1-CR-IRES-Pur , plenti-EF1α-FLAG-UBE2G1-C90S-CR-IRES-Pur , plenti-EF1α-FLAG-UBE2D3-IRES-Pur , and plenti-EF1α-FLAG-UBE2D3-C85S-IRES-Pur . Constructs pDONR221-U6-sgRNA-EF1a-Cas9-P2A-GFP , plenti-EF1α-Cas9-IRES-Bla , pDONR223-IKZF1 , pcDNA3-IKZF1-V5 , pcDNA3-CRBN , and pcDNA3-8xHis-Ub were described previously ( Matyskiela et al . , 2016; Hagner et al . , 2015 ) . The Cas9-P2A-GFP coding region of pDONR221-U6-sgRNA-EF1a-Cas9-P2A-GFP was subcloned into pcDNA3 . 1 ( Invitrogen ) to generate pcDNA3 . 1-Cas9-P2A-GFP . The IKZF1 coding region of pDONR223-IKZF1 was shuttle into plenti-EF1α-ePL-gateway-IRES-Bla via LR recombination to generate plenti-EF1α-ePL-IKZF1-IRES-Bla . Complementary oligonucleotides containing three non-targeting sgRNAs or three gene-specific sgRNAs targeting CRBN or each of the 41 annotated E2 enzymes were annealed and cloned into pRSG16-U6-sgEV-UbiC-TagRFP-2A-Puro or pRSG16-U6-sgEV-UbiC-Hyg , both of which were modified from pRSG16-U6-sg-HTS6C-UbiC-TagRFP-2A-Puro ( Cellecta ) . All sgRNA sequences used in this report ( see Supplementary file 1 ) were selected from the human genome-wide CRISPR sgRNA library ( Cellecta ) . Lentiviral plasmid was cotransfected with the 2nd Generation packaging system ( ABM ) into 293 T cells ( Clontech ) using Lipofectamine 2000 . After 16 hr of incubation , media was changed to fresh DMEM media supplemented with 20% FBS . At 48 hr post transfection , viral supernatant was collected and cleared via centrifugation at 2000 rpm for 5 min , and then filtered through a 0 . 45 micron cellulose acetate or nylon filter unit . Acute myeloid leukemia and myeloma cell lines were spin-inoculated with lentivirus at 2500 rpm for 120 min . After twelve hours , viral supernatant was removed and complete culture media was added to the cells . Forty-eight hours later , cells were incubated with 1 ~ 2 μg/mL puromycin ( Thermofisher ) , 10 ~ 20 μg/mL blasticidin ( Thermofisher ) , or 250 ~ 500 μg/mL hygromycin B ( Thermofisher ) for an additional 2 ~ 7 days to select cells stably integrated with lentiviral vectors . AML and MM cell lines were transduced with plenti-EF1a-Cas9-IRES-Bla , followed by limiting dilution and blasticidin selection in 96-well plates ( Corining ) to generate single clones stably expressing Cas9 . The expression of Cas9 in stable clones were validated by immunoblot analysis . Next , Cas9-expressing cells were transduced with pRSG16-U6-sgNT-1-UbiC-TagRFP-2A-Puro , pRSG16-U6-sgNT-3-UbiC-TagRFP-2A-Puro , pRSG16-U6-sgUBE2G1-1-UbiC-TagRFP-2A-Puro , pRSG16-U6-sgUBE2G1-5-UbiC-TagRFP-2A-Puro , pRSG16-U6-sgUBE2D3-4-UbiC-TagRFP-2A-Puro , or pRSG16-U6-sgCRBN-8-UbiC-TagRFP-2A-Puro . One days after transduction , cells were selection with puromycin for 2 days . Then , gene editing efficiency was verified by immunoblot analysis with antibodies recognizing the targeted proteins . For gene editing of both UBE2G1 and UBE2D3 , U937-Cas9 cells were first transduced with pRSG16-U6-sgUBE2G1-5-UbiC-Hyg . After 24 hr , cells were selected with hygromycin B selection for additional 5 days , and then transduced with pRSG16-U6-sgUBE2D3-4-UbiC-TagRFP-2A-Puro , followed by puromycin selection for 2 days and immunoblot analysis . 293 T cells were transiently transfected with pcDNA3 . 1-Cas9-P2A-GFP , pRSG16-U6-sgUBE2G1-5-UbiC-TagRFP-2A-Puro with or without pRSG16-U6-sgUBE2D3-4-UbiC-TagRFP-2A-Puro . Three days after transfection , cells were subjected to limiting dilution into 96-well plates . After two weeks , stable clones were cherry-picked , expanded and subjected to immunoblot analysis . 293T UBE2G1-/- clone 13 was validated to be UBE2G1 deficient , and UBE2G1-/-;UBE2D3-/- clone 4 was proven deficient for both UBE2G1 and UBE2D3 . For the single-guide RNA directed E2 CRISPR screen , U937 Cas9 cells were first transduced with plenti-EF1a-ePL-IKZF1-IRES-Bla to generate U937_Cas9_ePL-IKZF1 cells , which were then transduced with the focused lentiviral CRISPR library containing three non-targeting control sgRNAs and three gene-specific sgRNAs targeting each of the 41 E2 enzymes . For the dual-guide RNA directed E2 CRISPR screen , U937_Cas9_ePL-IKZF1 were transduced with pRSG16-U6-sgUBE2G1-5-UbiC-Hyg , followed by hygromycin B selection for additional 5 days . Then , cells were transduced with the focused lentiviral CRISPR library targeting all annotated E2s as described above . Four days after transduction with lentiviral vectors expressing non-targeting , CRBN-specific or UBE2-specific sgRNA , U937_Cas9_ePL-IKZF1 cells were dispensed into a 384-well plate ( Corning ) pre-spotted with pomalidomide at varying concentrations . Twenty-five microliters of RPMI-1640 growth media containing 5000 cells was dispensed into each well . After incubation at 37°C with 5% CO2 for 16 hr , 25 µL of the InCELL Hunter Detection Reagent Working Solution ( DiscoverX ) was added to each well and incubated at RT for 30 min protected from light . After 30 min , luminescence was read on an EnVision Multimode Plate Reader ( Perkin Elmer ) . For each cell line , The IKZF1 degradation induced by pomalidomide at each indicated concentration was normalized with the DMSO control . Then , a four parameter logistic model ( sigmoidal dose-response model ) was used to plot the IKZF1 destruction curves . All percentage of control IKZF1 destruction curves were processed and graphed using GraphPad Prism Version 7 . To assess the effect of single knockout of a given UBE2 gene on pomalidomide-induced IKZF1 degradation as shown Figure 1B and Figure 1—figure supplement 2 , an unpaired two-sided t-test was used to compare the difference between sgUBE2 and parental control . To assess the effect of dual knockout of UBE2G1 and a given UBE2 gene as shown in Figure 2B and Figure 2—figure supplement 1 , an unpaired two-sided t-test was used to compare the difference between sgUBE2G1 alone versus sgUBE2G1 plus sgUBE2 . p<0 . 05 is considered as significant . AML cell lines ( 5000 cells per well ) , MM cell lines ( 5000 cells per well ) or 293 T cells ( 2000 cells per well ) in 50 μL complete culture media were seeded into black 384-well plates containing with DMSO or test compounds . After 3 or 5 days , cell proliferation was assessed using CTG according to manufacturer’s instructions . Relative cell proliferation was normalized against the DMSO control . The growth inhibitory curve of each test compound was processed and graphed using GraphPad Prism Version 7 . p<0 . 05 , unpaired two-sided t-test , is considered as significant . Following treatment with test compounds at 37°C for the indicated time , cells were washed in ice-cold 1X PBS twice before harvest in Buffer A [50 mM Tris . Cl ( pH 7 . 6 ) , 150 mM NaCl , 1% Triton X-100 , 1 mM EDTA , 1 mM EGTA , 1 mM β-glycerophosphate , 2 . 5 mM sodium pyrophosphate , 1 mM Na3VO4 , 1 µg/mL leupeptin , one tablet of Complete ULTRA protease inhibitor cocktail ( Roche ) , and one tablet of PhosSTOP phosphatase inhibitor cocktail ( Roche ) ] . Whole cell extracts were collected after centrifugation at top speed for 10 min , resolved by SDS-PAGE gel electrophoresis , transferred onto a nitrocellulose membrane using the Turboblot system ( Bio-Rad ) , and probed with the indicated primary antibodies . Bound antibodies were detected with IRDye-680 or −800 conjugated secondary antibodies using a LI-COR scanner . 293T parental and UBE2G1-/- cells were pretreated with DMSO or pomalidomide ( 1 μM ) for 30 min , followed by the addition of 100 μg/ml cycloheximide ( EMD ) into the culture medium . At various time points as indicated in Figure 3D , cells were collected and subjected to immunoblot analysis . The ubiquitination assays were carried out as described previously ( Lu et al . , 2014b ) . In brief , 293T parental , UBE2G1-/- , and UBE2G1-/-;UBE2D3-/- cells seeded in six well plates were transiently transfected with pcDNA3-IKZF1-V5 , pcDNA3- CRBN , pcDNA3-8 x His-Ub , or pcDNA3 empty vector . In Figure 5B and C and Figure 5—figure supplements 1B , C , 293T UBE2G1-/- and UBE2G1-/-;UBE2D3-/- cells were also transfected with plenti-EF1a-UBE2G1-IRES-Pur , plenti-EF1a-UBE2D3-IRES-Pur or both . Forty-eight hours post transfection , cells were treated with 10 μM MG-132 ( R and D systems ) with or without an increased concentrations of pomalidomide as indicated . Eight hours later , cells were washed twice with ice cold PBS and resuspended in 1 mL PBS . Twenty uL of the cell suspension was boiled in LDS loading buffer , and the remaining cells were collected via centrifugation and lysed in Buffer C ( 6M guanidine-HCL , 0 . 1M Na2HPO4/NaH2PO4 , 20 mM imidazole , pH 8 . 0 ) . Next , whole cell extracts were sonicated for 12 pulses , and mixed with 20 μL of HisPur Ni-NTA Magnetic Beads ( Thermofisher ) at 37°C for 4 hr . Ni-NTA beads were then washed three times with Buffer C , three times with Buffer D ( 1 vol of Buffer C: 3 volumes of Buffer E ) , and three times with Buffer E ( 25 mM Tris . CL , 20 mM imidazole , pH 6 . 8 ) . Bound proteins were eluted by boiling in 2x LDS loading buffer and subjected to immunoblot analysis . Rabbit anti-human CRBN65 monoclonal antibody ( mAb ) ( Celgene , San Diego , CA ) ; rabbit anti-human GSPT1 polyclonal antibody ( pAb; Abcam , #ab49878 ) , rabbit anti-human IKZF1 mAb ( Cell Signaling , #14859 ) , rabbit anti-human IKZF3 mAb ( Cell Signaling , #15103 ) , rabbit anti-human CK1α pAb ( Abcam , #ab108296 ) , rabbit anti-human ZFP91 pAb ( LifeSpan Biosciences , #LS-B14788 ) , mouse anti-human UBE2G1 mAb ( Santa Cruz , #SC-100619 ) , rabbit anti-human UBE2G1 pAb ( Abcam , #SC-101371 ) , mouse anti-human UBE2D3 mAb ( Abcam , #ab58251 ) , rabbit anti-human UBE2D3 pAb ( Sigma , #SAB2102622 ) , rabbit anti-human Cul4A pAb ( Cell Signaling , #2699 ) , rabbit anti-human DDB1 pAb ( Cell Signaling , #5428 ) , rabbit anti-human Rbx1 pAb ( Cell Signaling , #4397 ) , rabbit anti-human Cdt1 mAb ( Cell Signaling , #8064 ) , rabbit anti-human Cdt2 pAb ( Cell Signaling , #A300-948A ) , rabbit anti-human Set8 pAb ( Cell Signaling , #2996 ) , rabbit anti-human RBM39 pAb ( Sigma , #HPA001591 ) , rabbit anti-human p21 mAb ( Cell Signaling , #2947 ) , rabbit anti-human p27 mAb ( Cell Signaling , #3686 ) , rabbit anti-human c-Myc mAb ( Cell Signaling , #5605 ) , rabbit anti-human BRD4 pAb ( Abcam , #ab128874 ) , mouse anti-penta-HIS mAb ( Qiagen , #34660 ) , mouse anti-human Actin mAb ( Sigma , #A5316 ) and mouse anti-human Tubulin mAb ( Sigma , #T9026 ) ( Sigma ) were used as primary antibodies . Goat anti-mouse 800 antibody ( LICOR Biosciences ) , goat anti-rabbit 680 antibody ( LICOR Biosciences ) , goat anti-mouse 800 antibody ( LICOR Biosciences ) and goat anti-rabbit 680 antibody ( LI-COR Biosciences ) were used as secondary antibodies . | Cells routinely breakdown damaged or unwanted proteins to recycle their building blocks . In humans , most of these unwanted proteins are first tagged with a chain of smaller proteins called ubiquitin , in a process known as ubiquitination . Three kinds of enzymes – named E1 , E2 and E3 – act one after the other to recruit and transfer ubiquitin onto the protein . Any problem with this protein-disposal system may cause diseases including cancers . Several drugs such as thalidomide are known to hijack the ubiquitination process by binding to the E3 enzyme . Instead of targeting unwanted proteins , the E3 enzyme-drug complex targets proteins that are driving a disease . These drugs are particularly useful for treating blood cancers . The problem is patients often become resistant to these drugs , and not always because the activity of the E3 enzyme is impaired . An alternative suspect would be an E2 enzyme , but the role of these enzymes remains unclear . Lu et al . have now asked whether a faulty E2 enzyme can lead to drug resistance in a form of blood cancer called multiple myeloma . The experiments tested how proteins relevant for the growth of cancerous myeloma cells were degraded in the presence of different drugs . Genes for the E2 enzymes were inactivated one at a time using a gene editing approach to see which ones would affect the degradation of the proteins and result in drug resistance . Two E2 enzymes , UBE2G1 and UBE2D3 , were found to be critical . UBE2D3 first links the disease-driving proteins with one ubiquitin before UBE2G1 can subsequently assemble a chain of ubiquitin proteins . If either of these E2 enzymes was missing from myeloma cells treated with drugs , the disease-driving proteins could not be properly tagged with ubiquitin . This interfered with the degradation of the proteins and allowed the myeloma cells to continue to grow . Yet , myeloma cells that did not have UBE2G1 and were resistant to certain drugs could still respond to other more potent drugs . This suggests that the success of the drugs depends on UBE2G1 . Therefore , a better understanding of the activity of this E2 enzyme may be useful for the development of future anticancer drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cancer",
"biology"
] | 2018 | UBE2G1 governs the destruction of cereblon neomorphic substrates |
The HIV capsid is semipermeable and covered in electropositive pores that are essential for viral DNA synthesis and infection . Here , we show that these pores bind the abundant cellular polyanion IP6 , transforming viral stability from minutes to hours and allowing newly synthesised DNA to accumulate inside the capsid . An arginine ring within the pore coordinates IP6 , which strengthens capsid hexamers by almost 10°C . Single molecule measurements demonstrate that this renders native HIV capsids highly stable and protected from spontaneous collapse . Moreover , encapsidated reverse transcription assays reveal that , once stabilised by IP6 , the accumulation of new viral DNA inside the capsid increases >100 fold . Remarkably , isotopic labelling of inositol in virus-producing cells reveals that HIV selectively packages over 300 IP6 molecules per infectious virion . We propose that HIV recruits IP6 to regulate capsid stability and uncoating , analogous to picornavirus pocket factors . HIV-1/IP6/capsid/co-factor/reverse transcription .
Despite a wealth of assays and data , HIV post-entry biology remains poorly understood . Before the virus can integrate its genes into the host genome , it must reverse transcribe its RNA into DNA , travel to the nucleus , remove its protective capsid ( uncoating ) and target actively transcribing chromatin . However , the relationship between these processes and the order and location in which they occur remain hotly debated . Reverse transcription ( RT ) is the first enzymatic step in infection and likely occurs as soon as the viral capsid enters the cytosol . Evidence suggests that it precedes all other processes including uncoating and nuclear import . Depletion of dynein ( Fernandez et al . , 2015 ) and kinesin one heavy chain ( KIF5B ) ( Lukic et al . , 2014 ) , required for mechanical transport of viral cores to the nucleus , delays uncoating and nuclear entry but does not affect RT . Moreover , none of the capsid cofactors required for nuclear entry or integration are necessary for RT . Depletion of Nup153 , Nup358 , CPSF6 or TNPO3 inhibits nuclear entry and/or active chromatin targeting but not RT ( Marini et al . , 2015; Matreyek and Engelman , 2011; Matreyek et al . , 2013; Zhang et al . , 2010; Sowd et al . , 2016; Ocwieja et al . , 2011 ) . Capsid mutants at cofactor interfaces largely phenocopy cofactor dependence , reinforcing this notion . Mutants N57A , N74D and P90A have defective integration and/or infectivity in primary human macrophages ( Ambrose et al . , 2012; Rasaiyaah et al . , 2013; Schaller et al . , 2011 ) but can still synthesise DNA . The critical role the capsid plays in cellular trafficking and nuclear entry strongly suggests that it remains intact in some form until late in post-entry . Inducing premature uncoating in the cell via the restriction factor TRIM5α and the proteasome ( Stremlau et al . , 2006 ) , or the capsid drug PF74 , irreversibly blocks RT and inhibits infection ( Price et al . , 2014; Shi et al . , 2011 ) . Exactly where and when uncoating normally happens and whether this occurs in a single step , or in sequential partial uncoating steps , is unclear . Nevertheless , one implication of RT preceding all other steps is that complete uncoating cannot be immediate . Encapsidated reverse transcription ( ERT ) assays demonstrate that RT can take place inside purified HIV cores before uncoating when they are supplied with dNTPs ( Warrilow et al . , 2007 ) . Inside the viral capsid , RT enjoys a high enzyme and substrate concentration and synthesised DNA is protected from nucleic acid sensors and nucleases in the cytosol ( Rasaiyaah et al . , 2013; Lahaye et al . , 2013 ) . While the benefits of remaining encapsidated are clear , HIV capsids are known to be highly fragile in vitro ( Forshey et al . , 2002 ) and the accumulation of newly synthesised DNA increases their instability ( Rankovic et al . , 2017 ) . In contrast , it takes at least 10 hr for the virus to complete RT ( Butler et al . , 2001 ) and dock at the nucleus ( Arhel et al . , 2007 ) . Recently , we reported that HIV capsid hexamers possess an electropositive pore that is essential for RT and infection ( Jacques et al . , 2016 ) . A ring of arginine residues ( R18 ) within the pore avidly binds dNTPs , potentially explaining nucleotide transport into the capsid interior to fuel RT . However , the properties of the R18 pore and its consequences for encapsidated DNA synthesis and capsid stability remain largely unexplored . Crucially , dNTP is not the only polyanion in the cell nor the most abundant . Here , we show that the highly charged polyanion IP6 is specifically incorporated into HIV virions and directly binds the R18 hexamer ring . We demonstrate that IP6 greatly stabilises HIV cores , limiting their spontaneous disassembly and allowing efficient encapsidated RT . We propose that HIV uses IP6 to regulate capsid uncoating in a manner reminiscent of picornavirus pocket factors .
Our previous study on HIV capsid pores suggested but did not determine that they are unable to discriminate between ribo and deoxyribo nucleotides . This is an important question because ATP is typically present at cytosolic concentrations > 100 fold in excess of dNTPs and thus a potential competitor for dNTP recruitment by the capsid . We therefore investigated whether ATP competes with dNTPs for binding to HIV capsid hexamers and whether the pore can discriminate between them . We used differential scanning fluorimetry ( DSF ) to determine if ATP binds hexamers by testing for thermal stabilisation , as was previously observed for dNTPs ( Jacques et al . , 2016 ) . As previously , we used a mutant hexamer protein that is coordinated by disulphide bridges between monomers through the introduction of mutations A14C and E45C and with additional dimerisation mutants W184A and M185A ( Pornillos et al . , 2010 ) . ATP stabilised hexamers with an almost identical increase in melting temperature ( Tm ) as dATP ( Figure 1a ) . Moreover , other triphosphate ligands were also stabilising suggesting a shared mode of binding . Next , we solved a crystal structure of hexamer bound to ATP to determine its mode of binding ( Table 1 ) . We observed electron density indicating that ATP binds to hexamer at the centre of the sixfold axis and refinement of the six symmetrically equivalent ATP molecules with equal partial occupancy provided a good fit to the data ( Figure 1b , c ) . The structure shows that ATP is coordinated solely by the R18 ring confirming that hexamers do not discriminate ribo versus deoxyribo nucleotides ( Figure 1d , e ) . As capsid hexamers do not discriminate NTPs and dNTPs , we tested how the presence of ATP impacts on ERT . ERT reactions were carried out on capsid cores purified from HIV-1 virions by centrifugation through a triton layer and banding in a sucrose gradient ( Shah and Aiken , 2011 ) . To ensure that this procedure allows recovery of intact cores of the expected morphology , we carried out negative stain electron microscopy ( EM ) on capsid cores further concentrated by pelleting . We observed intact capsids similar to those described previously ( Welker et al . , 2000 ) ( Figure 2a ) . We investigated RT inside these cores by adding dNTPs , together with DNAse to remove DNA accumulating outside the capsid . Surprisingly , even at a 100-fold excess of ATP over dNTPs , not only did ATP fail to inhibit ERT but it actually increased the quantity of measured transcripts ( Figure 2b ) . Similar increases were observed with the non-hydrolysable analogue ATPγS , suggesting that this was not due to an undefined ATPase activity . Improvement in ERT efficiency was dose-dependent , with transcripts increasing over two orders of magnitude as ATP was increased from 10 to 1000 µM ( Figure 2c ) . Importantly , ATP had no impact on the activity of purified RT enzyme alone over the same concentration range ( Figure 2d ) . Furthermore , substantial ATP stimulation was only observed in ERT reactions carried out in the presence of nuclease ( Figure 2e ) . In the presence of DNase , any RT products that are made outside of the capsid , or become exposed as a result of uncoating , will be degraded . Thus , ATP does not alter the efficiency of RT directly but facilitates the accumulation of RT products within intact capsids . This is consistent with the fact that ATP hydrolysis is not required , as shown by the ATPγS data . We hypothesised that the ability of ATP to stabilise hexamers , as measured by DSF , may increase capsid stability during ERT . Consistent with this hypothesis , ATP only promoted ERT at concentrations of dNTPs < 100 µM ( Figure 2f ) . At higher dNTP concentrations , ATP presumably had no effect because dNTPs sufficiently stabilised hexamers themselves . This also suggests that different cellular dNTP concentrations ( such as during the cell cycle or between different cell types ) might alter HIV RT efficiency both directly , by increasing substrate concentration for DNA synthesis , and indirectly by stabilising the capsid . Given that ATP may stabilise hexamers we tested whether ATP is selectively incorporated into HIV particles . ATP was detectable in virions , although at levels approximately similar to those inside the cell ( 1 . 5 mM ) suggesting no substantial enrichment ( Figure 2g ) . The fact that ATP does not inhibit ERT suggests that only a few pores are sufficient for dNTP import or that there are alternative routes into the capsid . To investigate this further , we compared the ability of nucleotide-based reverse transcription inhibitor ( NRTI ) AZT and non nucleoside RT inhibitiors Rilpivirine ( RVP ) and Nevirapine ( NVP ) to block ERT . We hypothesised that if pores represent the only entry route into the capsid then NNRTIs might be less potent in an encapsidated assay because they lack a triphosphate and may fail to interact with the R18 ring . We first used DSF to test interaction with capsid hexamers and found that only AZT and not the hydrophobic compounds RVP or NVP matched the stabilisation of dATP ( Figure 3a ) . A small increase in hexamer stability was observed with RVP but equally in the presence and absence of DTT . This suggests any RVP interaction is not due to the R18 hexameric ring as this feature will be destroyed upon DTT addition because the disulphides holding the hexamer together become reduced . To corroborate these results we measured binding by fluorescence competition with BODIPY-ATP . Only AZT could compete with BODIPY-ATP and was capable of binding capsid hexamers ( Figure 3b ) . This data is consistent with pore interaction being dependent on the charged triphosphate group . Determining a structure of capsid hexamer in complex with AZT confirmed that interaction occurs in a similar manner to dATP and ATP ( Table 1 ) . AZT binds to hexamer at the centre of the six-fold axis and refinement of the six symmetrically equivalent AZT molecules with equal partial occupancy provided a good fit to the electron density ( Figure 3c ) . As with the previous complexes , while there is clear density for the phosphates the position of the base can only be inferred . The AZT triphosphates interact directly with R18 , which adopts alternative side-chain conformers to allow multiple hydrogen bonds and prevent steric clashes ( Figure 3d ) . Taken together with the ATP and dATP structures , this confirms that there are no specific hydrophobic interactions with the pore and is consistent with a lack of hexamer stabilisation by hydrophobic , uncharged NNRTIs . Next , we compared the activity of NRTI and NNRTIs in RT and ERT reactions to determine whether encapsidation markedly influences inhibitor potency . At the IC90 for in vitro RT inhibition , both AZT and the NNRTIs RVP and NVP blocked ERT with equivalent efficacy ( Figure 3e , f ) . The fact that NNRTIs are equally capable of preventing ERT despite showing no interaction with the charged pore suggests that specific binding is not required for entry into the capsid . NNRTIs are sufficiently small to pass through the pore and so may enter by simple diffusion . These experiments also do not rule out entry via a different route entirely . Our data with ATP suggested that polyanions may have an unappreciated but important role in stabilising HIV capsids and allowing encapsidated DNA synthesis to take place . Previously , we showed that the six-fold symmetric polyaninon hexacarboxybenzene interacts avidly with the R18 charged pore ( Jacques et al . , 2016 ) . Furthermore , it bound hexamers more tightly than dNTP and stabilised hexamers to a greater extent . Hexacarboxybenzene is not a cellular metabolite nor is it bioavailable; however , it is reminiscent of the fully phosphorylated inositol ring of inositol hexakisphosphate ( IP6 ) , a strongly negatively charged metabolite possessing six phosphate rather than carboxylic acid groups ( Figure 4a ) . Moreover , IP6 has previously been shown to interact with immature capsid and catalyse in vitro assembly of the immature lattice ( Campbell et al . , 2001 ) . We therefore tested by DSF whether IP6 can stabilise the mature capsid and promote ERT by facilitating the accumulation of RT products within the core . To obtain additional information as to stoichiometry of binding , we performed experiments at a range of IP6 concentrations with hexamer protein present at ~33 µM . To show the transition between stabilised species , the melt curves rather than change in stability ( ΔTm ) are shown ( Figure 4b ) . As predicted , IP6 strongly stabilised capsid hexamers , moreover to a substantially greater degree than ATP ( Figure 4c ) . A concentration of 20 µM IP6 gave approximately half maximal stabilisation to a solution of 33 µM hexamer . This suggests that hexamer stabilisation by IP6 is likely stoichiometric , with one IP6 molecule per hexamer . Importantly , addition of IP6 to an ERT reaction greatly increased the number of reverse transcripts in the presence of nuclease ( Figure 4d ) . This suggests that IP6 promotes the accumulation of RT products within intact capsids . To further test this hypothesis , we repeated our ERT experiment in the presence of the anti-capsid drug PF74 . At concentrations > 10 µM , PF74 is thought to destabilise the capsid and induce premature uncoating during infection ( Price et al . , 2014 ) , while in vitro it leads to a rapid failure of capsid integrity ( Marquez et al . , submitted ) . We measured no RT products in the presence of both PF74 and nuclease , consistent with induced uncoating and degradation of exposed transcripts ( Figure 4d ) . Importantly , in the presence of PF74 , IP6 no longer had any effect . Taken together our observations suggest that IP6 acts to stabilise capsids but that this stabilisation is not sufficient to prevent capsid failure induced by PF74 . Nevertheless , in the absence of PF74 , IP6 potently stabilised capsid cores at much lower concentrations than ATP . At concentrations where ATP showed no measurable effect , IP6 was able to maximally increase ERT ( Figure 4e ) . Next , we investigated whether IP6 intrinsically stabilises the HIV capsid or just during RT . Capsid cores are known to be unstable and uncoat after isolation . Consistent with this , addition of dNTPs after 24 hr incubation at room temperature resulted in little measurable ERT , in contrast to dNTP addition at time 0 which resulted in effective DNA synthesis ( Figure 5a ) . Experiments were carried out in the absence of nuclease , indicating that the absence of measured RT in pre-incubated samples reflects the inefficiency of transcription outside the capsid . Importantly , however , capsid cores that were incubated in the presence of IP6 for 24 hr before the addition of dNTPs supported a similar level of RT as capsids that had not been subjected to temperature-induced uncoating . Together with the previous data , this suggests that IP6 stabilises capsids both prior to and after RT . To investigate this further , we sought to define the intrinsic stability of an HIV capsid and how this is impacted by IP6 . To accomplish this , we employed a newly developed single molecule assay described in detail in an accompanying report . In summary , this method involves tethering HIV virions to glass cover slips and permeabilizing their lipid envelope using a pore-forming toxin ( Figure 5b ) . Chimeric virions were used in which GFP is cleaved from Gag during maturation of the particle , resulting in GFP molecules packaged within the mature capsid . The uncoating kinetics of individual capsids was then determined by monitoring the loss of GFP signal ( Figure 5c ) . Uncoating , here defined as sufficient core opening to allow the escape of GFP molecules , occurs as a single rapid event at the single-particle level ( Figure 5d ) . In the absence of IP6 , we observed that HIV capsids are highly unstable and rapidly uncoat with a half-life of ~7 min ( Figure 5d–f ) . Remarkably , addition of even 1 µM IP6 was sufficient to dramatically stabilise virions and increase their half-life well beyond an hour ( Figure 5e and f ) . Increasing the concentration of IP6 to 10 or 100 µM stabilised capsids so efficiently that too few virions uncoated during the timescale of the experiment for an accurate measurement of the half-life . However , estimates suggest that 10 or 100 µM IP6 stabilises capsid for 5 or 10 hr , respectively . We hypothesised that the remarkable capsid stabilisation achieved by IP6 is likely because the abundant negatively charged groups allow highly efficient coordination and charge neutralisation of the six arginine side-chains in the pore . To investigate this , we solved the structure of hexamer bound to IP6 ( Table 1 ) . The electron density for the complex confirmed that IP6 binds the R18 ring and six symmetry copies of IP6 at partial occupancy were an excellent fit to the data ( Figure 6a ) . This illustrates that IP6 can bind in a number of alternate conformations , with the axial phosphate in different orientations . The N-terminal β-hairpin of the hexamer adopts a ‘closed’ conformation in the IP6 complex , with IP6 itself located within the central chamber , above the R18 ring ( Figure 6b ) . Weak density was observed that may correspond to additional IP6 molecules in the solvent channel below R18 but their inclusion did not improve the fit of the model to the data . Previous hexamer structures have shown that the R18 side-chain can adopt multiple alternative rotamer conformations . Two alternate R18 conformers are necessary in the IP6 complex to prevent clashes with the ligand and maintain interaction . Despite the fact that IP6 is not planar and does not bind parallel to the R18 ring , alternate R18 conformers allow each arginine within the hexamer the potential for hydrogen bond or electrostatic interactions with a phosphate ( Figure 6c , d ) . The structure therefore indicates that IP6 is tightly bound by the R18 pore , would effectively neutralise the abundant positive charge and is likely to coordinate and maintain hexamer assembly . Inositol hexakisphosphate is an abundant polyanion , present in cells at ~50 µM concentrations ( Veiga et al . , 2006 ) . It is synthesised from Inositol pentakisphosphate ( IP5 ) , which can reach similar concentrations . We therefore tested whether IP5 or other precursor molecules are also capable of promoting ERT . DSF showed that other IP compounds , including IP5 and also to a lesser extent IP4 , were capable of stabilising capsid hexamers under assay conditions ( Figure 7a ) . Addition in ERT experiments at 1 µM concentration revealed a clearer distinction between IP compounds , with IP4 unable to stabilise capsids and promote RT product accumulation in the presence of nuclease ( Figure 7b ) . In contrast , IP5 promoted encapsidated RT indistinguishably from IP6 . Based on the above data , we hypothesised that IP5 or IP6 could be used by HIV in order to stabilise its capsid . IP5 and IP6 have previously been shown to catalyse the assembly of the immature lattice ( Campbell et al . , 2001 ) , suggesting that these molecules are recruited either during budding , in a producer cell , or following entry into a target cell . To distinguish between these two possibilities , we attempted to determine whether inositol phosphates are incorporated into HIV virions . We produced viruses from 3H-inositol labelled cells and carefully purified budded viral particles by filtering and multiple centrifugal pelleting steps through sucrose . Virions were acid extracted and strong anion exchange chromatography ( Sax-HPLC ) was employed to resolve the inositol phosphate present in the virions and in the correspondent infected cell extracts . Scintillation counting of Sax-HPLC eluted fractions revealed the specific presence of IP6 in HIV virions ( Figure 7c ) . Importantly , the levels of IP6 were enriched over those of IP5; indeed little IP5 was detected despite the fact that it is present within cells at concentrations around 25% that of IP6 . This suggests that it is IP6 selectively incorporated into virions rather than from potential co-purifying microvesicles that is being measured ( Bess et al . , 1997 ) . Subtilisin treatment is used to reduce microvesicle contamination in HIV-1 virion preparations ( Ott , 2009 ) . To ensure that IP6 detection is not from contaminating microvesicles , we repeated our experiments on subtilisin-treated samples . Subtilisin removed protein contaminants and abolished HIV-1 infectivity consistent with the digestion of envelope proteins , but did not reduce the levels of capsid protein or incorporated IP6 ( Figure 7d and Figure 7—figure supplement 1 ) . Finally , to directly demonstrate that IP6 is associated with the capsid core rather than the viral membrane , we stripped the envelope from produced virions using the same triton-layer centrifugation method as employed to prepare cores for ERT ( Shah and Aiken , 2011 ) . Incorporation in capsid cores was compared to virions by calculating the scintillation counts for IP6 per µg p24 , as determined by ELISA . To confirm that our core prep has removed enveloped particles , we also compared the relative levels of inositol . Inositol is an osmolyte found in the cytosol at ~mM concentrations and is therefore only expected in samples derived from enveloped virions and microvesicles . Consistent with this , we observed loss of the inositol peak in the capsid only sample ( Figure 7e ) . However , we observed the same levels of IP6 in both samples demonstrating that IP6 is associated with the capsid core ( Figure 7f ) . This result reinforces that measured IP6 is not derived from microvesicles but is enriched into HIV-1 virions and bound to the capsid core . We calculated the number of IP6 molecules per virion by determining the fraction of cellular IP6 that is incorporated . Parallel analysis of the cell extracts by PAGE-based separation revealed that 0 . 8% of total IP6 radioactivity was incorporated into the produced virions ( Figure 7g ) ( Losito et al . , 2009; Wilson et al . , 2015 ) . This corresponds to approximately 309 ± 41 IP6 molecules per viral particle , which is sufficient for 1:1 stoichiometry with hexamer pores present in an intact virion .
Here , we have shown that HIV virions package IP6 when they bud from a producer cell , that IP6 coordinates the electrostatic pores in HIV hexamers and that IP6 dramatically increases capsid stability both intrinsically and during ERT . The logic for requiring a compound such as IP6 to stabilise the capsid is in part due to the presence of an otherwise repulsive charged structural feature ( R18 in the mature capsid ) . However , if nullifying charge were the only reason for IP6 requirement , then it would be simpler to evolve these residues into apolar alternatives . Thus , using IP6 as a stabilising agent must confer benefits onto the HIV capsid that cannot be achieved by encoding stability via capsid residues alone . A central coordinating ligand like IP6 allows proteins to be stabilised independently of a hydrophobic core or surface and has been used extensively in evolution , giving rise to motifs like the zinc-finger . Small molecule ligands are also highly effective at catalysing multimerisation , something of particular advantage to an assembling viral capsid . Examples where IP6 has been exploited as a small molecule chaperone include HDAC co-repressor complexes , where it promotes stability ( Watson et al . , 2016 ) , and Bruton's tyrosine kinase , where it induces transient dimerisation of the PH-TH module ( Wang et al . , 2015 ) . A further attractive feature of using a small molecule ligand like IP6 to drive self-assembly is that it provides an inbuilt mechanism of disassembly: ligand binding is reversible and so its dissociation can be used as a disassembly trigger . Using IP6 to stabilise the capsid therefore avoids a fundamental problem in capsid design , namely that the capsid must be strong enough to protect its cargo but not too strong that it can never be uncoated . Based on our data , we propose that IP6 is the HIV equivalent of a picornavirus pocket factor , providing both stability and a mechanism to initiate uncoating . Pocket factors were originally described by Michael Rossmann , who observed a small molecule bound to the picornavirus capsid that he proposed stabilised it ( Hogle et al . , 1985; Rossmann et al . , 1985 ) . Pocket factors have subsequently been shown to stabilise picornavirus capsids until conformational changes induced by environmental cues during infection promote their dissociation and induce disassembly . We hypothesise that IP6 stabilises the HIV capsid in the cytosol , until its release triggers uncoating . Known capsid cofactors could provide the cues to induce a conformational change and provoke dissociation of IP6 . The dynamic N-terminal β-hairpin present in hexamers ( Jacques et al . , 2016 ) provides one possible mechanism; when closed IP6 will be retained in the pore but when open it would be free to dissociate . Testing this mechanism will be a key goal of future study , as will determining the importance of IP6 in post-entry HIV-1 infection . While we propose that IP6 serves as an HIV pocket factor , we cannot rule out that other cellular polyanions also play a role . For instance , although we observed no evidence for enrichment , ATP is highly abundant and was readily detected in virions . In our original discovery of the R18 pore and its importance in HIV RT and infection we postulated that it provided a channel to import nucleotides to drive ERT . The cellular data presented in that work is also consistent with the R18 pore being required to bind IP6 in order to stabilise the capsid during its transit through the cytosol . Further experiments are needed to dissect the relative importance of the R18 pore in attracting nucleotides for RT versus binding IP6 for capsid stabilisation . The fact that high concentrations of either ATP or IP6 , which would be competitors with dNTPs for pore binding , increase rather than decrease ERT efficiency argues against the R18 pore being essential for nucleotide import . However , as there is a multi-log increase in accumulated viral DNA as a result of capsid stabilisation by polyanions this could mask any decline due to reduced nucleotide import . Capsid permeability is not well understood , but the experiments presented here comparing inhibition by NRTIs and NNRTIs in ERT versus cellular infection suggest that either their efficiency of entry into the capsid is similar – and that interaction with the R18 ring is therefore not essential - or that this is not a limiting step . Either way , this indicates that small molecules not bound by the R18 ring can still enter the capsid through the pore , or through solvent channels as has previously been proposed . Importantly , as mature hexamers form only after budding , they cannot be responsible for recruiting polyanions such as IP6 into the virion . Of note , IP6 has previously been shown to catalyse immature lattice assembly ( Campbell et al . , 2001 ) , suggesting it has a role in the formation and maturation of capsid during viral production . In the immature hexamer , there are two lysine rings ( K227 and K158; equivalent to Gag residues 359 and 290 ) that are analogous to the R18 ring we have described in the mature hexamer ( Wagner et al . , 2016 ) . We speculate that it is these lysine rings that are responsible for the specific recruitment of IP6 into virions , where it drives immature lattice assembly . Upon protease cleavage and maturation , the six-helix bundle and lysine rings will be lost , making IP6 available to bind the R18 pore in mature hexamers and promote mature lattice formation . Thus , we propose that HIV capitalises on a readily available cellular polyanion to catalyse the assembly of immature and mature lattices , promote their stability and regulate the uncoating of capsid in newly infected cells . The identification of the inositol phosphate IP6 in HIV virions and its role in imparting onto the capsid a property of metastability provides a new dependence to be exploited in antiviral development .
The CA N-terminal domain and the disulfide-stabilised CA hexamer were expressed and purified as previously described , briefly: p24 capsid protein was disassembled in Tris ( pH 8 . 0 , 50 mM ) , NaCl ( 40 mM ) , 2-mercaptoethanol ( 20 mM ) , then reassembled in Tris ( pH 8 . 0 , 50 mM ) , NaCl ( 1 M ) , 2-mercaptoethanol ( 20 mM ) . This was followed by oxidation in Tris ( pH 8 . 0 , 50 mM ) , NaCl ( 1 M ) and redispersion in Tris ( pH 8 . 0 , 20 mM ) , NaCl ( 40 mM ) . Reassembled hexamers were observed by non-reducing SDS–PAGE . DSF measurements were performed using a Prometheus NT . 48 ( NanoTemper Technologies ) over a temperature range of 20–95°C using a ramp rate of 2 . 0°C / min . CA hexamer samples were prepared at a final concentration of 200 µM monomer in Intracellular Buffer in the presence or absence of 4 mM DTT . dNTPs or competitors were added at 200 µM . DSF scans are single reads of three replicates and were performed at least three times unless otherwise stated . Consistency between like points yields an uncertainty in Tm of no greater than 0 . 2°C . Lentiviral packaging plasmid psPAX2 encoding the Gag , Pol , Rev , and Tat genes and pMDG2 , which encodes VSV-G envelope , was obtained from Didier Trono ( Addgene plasmids # 12260 and # 12259 ) . HIV-1 Gag-Pol expression plasmid pCRV-1 ( Zennou et al . , 2004 ) and HIV-GFP encoding plasmid CSGW ( Naldini et al . , 1996 ) were kind gifts from Stuart Neil . The proviral construct pNL4 . 3-iGFP-ΔEnv was generated as described previously ( Aggarwal et al . , 2012 ) . It contains the open-reading frame for eGFP flanked by protease cleavage sites inserted into the Gag gene between the coding sequences for MA and CA . In addition the start codon of the Env gene is mutated to a stop codon to prevent expression of the envelope protein . Replication deficient VSV-G pseudotyped HIV-1 virions were produced in HEK293T cells using pMDG2 , pCRV GagPol and CSGW as described previously ( Price et al . , 2014 ) . Viral supernatant from HEK293T cells was pelleted over a 20% sucrose cushion in a SW28 rotor ( Beckman ) at 28 , 000 rpm at 4°C . Pellets were resuspended in PBS . For removal of microvesicle contaminants , samples were treated with subtilisin essentially as described by Ott ( 2009 ) , with virions subsequently purified by ultracentrifugation through 20% sucrose . Subtilisin efficiency was determined by SDS PAGE and infection assay . HIV-1 capsid cores were prepared using a protocol based on Shah and Aiken ( 2011 ) . 90 ml HEK293T supernatant containing VSV-G pseudotyped HIV-1 GFP was pelleted over 20% sucrose dissolved in core prep buffer ( CPB; 20 mM Tris ( pH 7 . 4 ) , 20 mM NaCl , 1 mM MgCl2 ) in an SW28 rotor ( Beckman ) at 25 , 000 rpm at 4°C . Pellets were gently resuspended at 4°C in CPB for 1 hr with occasional agitation . Resuspended pellets were treated with DNase I from bovine pancreas ( Sigma Aldrich ) for 1 hr at 200 µg/ml at room temperature to remove contaminating extra-viral DNA . Virus was subjected to spin-through detergent stripping of the viral membrane as follows . A gradient at 80–30% sucrose was prepared in SW40Ti ultracentrifuge tubes and overlaid with 250 µl 1% Triton X-100 in 15% sucrose , followed by 250 µl 7 . 5% sucrose . All solutions were prepared in CPB . 750 µl DNase-treated , concentrated virus was layered on top of the gradient and subjected to 32 , 500 rpm at 4°C for 16 hr . The preparation was fractionated and the location of cores was determined by ELISA for p24 ( Perkin Elmer ) . Core-containing fractions were pooled and snap frozen before storage at −80°C . Cores were prepared as above . Pooled core fractions were diluted to reduce the sucrose concentration to 20% , then samples were spun 2 hr at 4˚C at 45 K rpm in TLA55 rotor . Cores were resuspended in 8 µl CPB with 20% sucrose and loaded onto glow discharged carbon grids ( Cu , 400 mesh , Electron Microscopy Services ) for 5 min then stained for 3 min with 2% uranyl acetate . Micrographs were taken at room temperature on a Tecnai Spirit ( FEI ) operated at an accelerated voltage of 120 keV and Gatan 2k × 2 k CCD camera . Images were collected with a total dose of 30 e-/A ̊2 , between a defocus of 1–3 µm and a magnified pixel size of 0 . 93 nm/pixel . Viral cores were diluted to 50 ng/ml p24 with 60% sucrose in CPB . Final concentrations of dNTPs were 1 µM each ( unless otherwise indicated ) , DNase I and RNase A were at 100 mg/ml . 20 µl reactions were incubated at room temperature for 16 hr unless indicated otherwise . 4 µl of 5xMicrolysis Plus ( Microzone ) was added to each sample and processed according to manufacturers instructions . Reverse transcript products were detected using TaqMan Fast Universal PCR Mix ( ABI ) and RU5 primers to detect strong-stop DNA40 ( RU5 forward: 5'-TCTGGCTAACTAGGGAACCCA-3'; RU5 reverse: 5'-CTGACTAAAAGGGTCTGAGG-3'; and RU5 probe 5'- ( FAM ) TTAAGCCTCAATAAAGCTTGCCTTGAGTGC ( TAMRA ) −3' ) , GFP primers to detect first-strand transfer products ( described above ) and primers for second-strand transfer products40 ( 2 ST forward: 5'-TTTTAGTCAGTGTGGAAAATCTGTAGC-3'; 2 ST reverse: 5'-TACTCACCAGTCGCCGCC-3'; and 2 ST probe: 5'- ( FAM ) TCGACGCAGGACTCGGCTTGCT ( TAMRA ) −3' ) . Unless otherwise stated , ERT data is representative of at least three independent experiments . All crystals were grown at 17°C by sitting-drop vapour diffusion in which 100 nl protein was mixed with 100 nl precipitant and suspended above 80 µl precipitant . The structures were all obtained from 10 to 12 mg/ml protein mixed with PEG550MME ( 13–14% ) , KSCN ( 0 . 15 M ) , Tris ( 0 . 1 M , pH 8 . 5 ) and cryoprotected with precipitant supplemented with 20% MPD . For the ATP-bound structure , the protein was supplemented with 1 mM ATP or IP6 immediately before crystallisation . All crystals were flash-cooled in liquid nitrogen and data collected either in-house using Cu Ka X-rays produced by a Rigaku FR-E rotating anode generator with diffractionrecorded on a mar345 image plate detector ( marXperts ) , or at beamline I02 at Diamond Light Source . The data sets were processed using the CCP4 Program suite ( Winn , 2003 ) . Data were indexed and integrated with iMOSFLM and scaled and merged with AIMLESS ( Evans and Murshudov , 2013 ) . Structures were solved by molecular replacement using PHASER ( McCoy , 2007 ) and refined using REFMAC5 ( Murshudov et al . , 1997 ) . Between rounds of refinement , the model was manually checked and corrected against the corresponding electron-density maps in COOT ( Emsley and Cowtan , 2004 ) . Solvent molecules and bound ligands were added as the refinement progressed either manually or automatically within COOT , and were routinely checked for correct stereochemistry , for sufficient supporting density above a 2Fo − Fc threshold of 1 . 0 s and for a reasonable thermal factor . The quality of the model was regularly checked for steric clashes , incorrect stereochemistry and rotamer outliers using MOLPROBITY ( Chen et al . , 2015 ) . Final figures were rendered in The PyMOL Molecular Graphics System , Version 1 . 5 . 0 . 4 Schrödinger , LLC . Structures and data were deposited in the PDB database with codes 6ERM ( ATP complex ) , 6ERN ( AZT complex ) and 6ES8 ( IP6 complex ) . Viral particles were produced by transfecting HEK293T cells with a mixture of pNL4 . 3-iGFP-∆Env and psPAX2 ( 1 . 4:1 , mol/mol ) , collected 48 hr post transfection and viral membrane proteins were biotinylated using EZ-Link Sulfo-NHS-LC-LC-Biotin ( Thermo Scientific ) . Biotinylated viral particles were purified by size exclusion chromatography using a HiPrep 16/60 Sephacryl S-500 HR column ( GE Healthcare ) and captured on the surface of a glass coverslip modified with PLL ( 20 ) -g[3 . 4]-PEG ( 2 ) /PEG ( 3 . 4 ) -biotin ( Susos AG ) and streptavidin . The viral envelope was permeabilised by addition of perfringolysin O ( 200 nM ) in imaging buffer ( 50 mM HEPES pH 7 . 0 , 100 mM NaCl ) via microfluidic delivery and the diffraction-limited signal from the GFP-loaded viral particles was monitored by time-lapse total internal reflection fluorescence microscopy . Images were analysed with software written in MATLAB ( The MathWorks , Inc . ) . Fluorescence intensity traces were calculated for each viral particle in the field of view by integrating the fluorescence intensity in a 7 × 7 pixel region . Membrane permeabilisation was detected as a rapid drop in the signal resulting from the release of GFP contained in the viral particle in the volume outside the capsid ( not shown ) . This step mimics viral fusion in that the capsid is exposed to the surrounding medium and was defined as time zero for measuring capsid opening . Particles with intact cores were identified as those with residual GFP signal after permeabilisation , arising from GFP molecules trapped inside the closed capsid . The onset of capsid uncoating was then detected as the sudden release ( loss of the residual signal occurring typically within one frame ) of the encapsidated GFP molecules via a sufficiently large defect in the capsid lattice . One × 106 293 T cells were seeded into 2 × 10 cm dishes in inositol-free DMEM and left to adhere overnight . The media was replaced with 5 ml inositol-free DMEM supplemented with 5 µCi/ml 3H-inositol ( Perkin Elmer ) . After 3 days incubation , an additional 5 ml inositol-free media containing 5 µCi/ml 3H-inositol was added onto cells , which were then transfected with pMDG2 , pCRV GagPol and CSGW . Cells were left for a further 3 days to produce VSV-G pseudotyped HIV1 . Viral supernatants were topped up to 30 ml and pelleted over a 5 ml 20% sucrose cushion ) in a SW28 rotor ( Beckman ) at 28 , 000 rpm at 4°C . Pellets were resuspended in inositol-free media and pelleted as previously . After the second spin , pellets were resuspended in 1 ml PBS and spun at 13 , 000 rpm at 4°C in a bench top microfuge for 60 min . Pellets were frozen at −20˚C until processing . Cells were washed with PBS , then harvested by scraping , counted and pelleted for quantification of cellular IP6 labelling . Pellets were frozen at −20˚C until processing . For comparison of virion and purified capsid core samples , p24 levels were determined by ELISA for p24 ( Perkin Elmer ) . 293T CRL-3216 cells were purchased from ATCC and authenticated by the supplier . All cells are regularly tested and are mycoplasma free . Inositol phosphates extraction and analysis was performed modifying a previously described protocol ( Azevedo and Saiardi , 2006 ) . Cells and viral pellets were resuspended in 200 µl of extraction buffer ( 1M Perchloric acid , 5 mM EDTA ) and incubated on ice for 10 min . The samples were spun out at 13 , 000 rpm at 4°C for 5 min and the supernatant recovered . Viral pellet were re-extracted for 10 min at 100˚C using another 200 µl of extraction buffer and spun out as before . Supernatants from acid extractions were neutralised to pH6-8 using 1M Potassium carbonate , 5 mM EDTA ( approximately 100 µl ) and incubated on ice with the lids open for 1–2 hr . Samples were spun at 13 , 000 rpm at 4°C for 5 min and supernatant containing inositol phosphates was loaded onto HPLC or stored at 4°C . Inositol phosphates were resolved by strong anion exchange chromatography Sax-HPLC on a Partisphere SAX 4 . 6 × 125 mm column ( Hichrom ) . The column was eluted with a gradient generated by mixing buffer A ( 1 mM EDTA ) and buffer B ( 1 mM EDTA; 1 . 3 M ( NH4 ) 2HPO4 , pH 4 . 0 ) as follows: 0–5 min , 0% B; 5–10 min , 0–30% B; 10–85 min , 30–100% B; 85–95 min , 100% B . Fractions ( 1 ml ) were collected and analysed by scintillation counting after adding 4 ml of Ultima-Flo AP LCS-cocktail ( Perkin Elmer ) . The presence of ATP in virions was determined using a luciferase-based detection system following manufacturers instructions ( Abcam ab113849 ) . A sensitivity of 1 nM ATP detection was confirmed using a titration of 293 T cell lysate . The concentration of ATP estimated in 293 T cells assuming a volume of 1 . 77 × 10−12 L was ~8 mM . The concentration of ATP estimated in HIV virions , assuming a volume of 5 . 2 × 10−19 L , was ~1 . 5 mM . The level of IP6 in 293 T cells was determined by PAGE analysis of TiO2 purified inositol phosphates alongside polyphosphate standards ( Sigma Aldrich S6128 ) and known standards of IP6 ( Wilson et al . , 2015 ) . For 4 × 106 cells there is estimated to be approximately 0 . 75 nmol IP6 . The calculated figure of 309 ± 41 IP6 per virion is an average of three independent biological replicates . | Viruses like HIV invade cells and replicate their genome to create new viruses . To hide from components of our immune system that are active inside the cell , HIV uses a protein shell called a capsid , which protects its genome from detection and destruction . However , the capsid faces an engineering challenge beyond those faced by even the most complex man-made structures . This is because the capsid must be strong enough to survive for hours inside the cell but not so strong that it cannot quickly open when the virus needs to release its genome . How this process , called ‘uncoating’ , is achieved is one of the great unanswered questions in HIV biology . In 2016 , researchers made the unexpected discovery that the HIV capsid is decorated with hundreds of pores: one at the center of every subunit from which it is built . Each pore contains a ring of six positively charged amino acids that should destabilize the capsid and cause it to break apart . Yet similar pores are found on a diverse range of viruses . Mallery et al . – who include several of the researchers involved in the 2016 work – set out to investigate why the HIV capsid contains the positively charged pores . Initial experiments revealed that a molecule called IP6 , which is abundant in cells , can bind to the HIV capsid . To do so , six negatively charged phosphate groups in IP6 match up with the six positively charged residues in the pore . In a related study , Márquez et al . developed a new method that allows the fate of individual capsids to be visualized through time . Here , Mallery et al . use the method to show that IP6 increases how long the capsid remains intact from several minutes to over 10 hours . This allows HIV to copy its genome inside the capsid , meaning it remains protected while the virus prepares to produce new viruses . Mallery et al . also show that HIV packages more than 300 IP6 molecules into itself when it replicates . Other viruses called picornaviruses use small molecules called pocket factors to stabilize the capsid and to trigger uncoating . Mallery et al . propose that IP6 is an HIV pocket factor . Just as studies of pocket factors have stimulated the development of anti-picornavirus drugs , understanding the role of IP6 may help to develop new treatments for HIV . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | IP6 is an HIV pocket factor that prevents capsid collapse and promotes DNA synthesis |
Despite extensive scrutiny of the myosin superfamily , the lack of high-resolution structures of actin-bound states has prevented a complete description of its mechanochemical cycle and limited insight into how sequence and structural diversification of the motor domain gives rise to specialized functional properties . Here we present cryo-EM structures of the unique minus-end directed myosin VI motor domain in rigor ( 4 . 6 Å ) and Mg-ADP ( 5 . 5 Å ) states bound to F-actin . Comparison to the myosin IIC-F-actin rigor complex reveals an almost complete lack of conservation of residues at the actin-myosin interface despite preservation of the primary sequence regions composing it , suggesting an evolutionary path for motor specialization . Additionally , analysis of the transition from ADP to rigor provides a structural rationale for force sensitivity in this step of the mechanochemical cycle . Finally , we observe reciprocal rearrangements in actin and myosin accompanying the transition between these states , supporting a role for actin structural plasticity during force generation by myosin VI .
Myosin motor proteins are responsible for movement and force generation across multiple scales of biology ranging from muscle contraction to cell migration to intracellular transport ( Cheney and Mooseker , 1992; Huxley , 1969; Syamaladevi et al . , 2012 ) . Defects in myosin genes have been linked to muscular dystrophies , cardiac disease , cancer , and deafness , highlighting the critical role of myosins in cell function and human health ( Hirokawa and Takemura , 2003 ) . In efforts to better understand disease mechanisms and develop potential therapeutics , these motors have been the subject of extensive biophysical , biochemical , and structural characterization ( Cope et al . , 1996; Houdusse and Sweeney , 2016; Sweeney and Houdusse , 2010 ) . However , as the myosin superfamily features over 20 classes , a detailed understanding of each motor and its specific mechanisms remains incomplete ( Sellers , 2000 ) . Despite this diversity , the enzymatic mechanism of ATP-dependent force generation on filamentous actin ( F-actin ) is fundamentally conserved ( Figure 1A ) ( Geeves , 2016; Holmes , 1997; Lymn and Taylor , 1971; Sweeney and Houdusse , 2010 ) . The motor domain ( MD ) binds and hydrolyzes ATP , which allosterically produces conformational changes enabling low-affinity engagement with F-actin ( Figure 1A , Pre-power stroke state ) . The transient opening of the small switch II loop ( swII ) adjacent to the nucleotide binding cleft allows phosphate to escape ( Figure 1A , Pi release state ) , triggering the transition to a moderate F-actin binding affinity ADP state ( Figure 1A , ADP state ) accompanied by large-scale rearrangements in the converter region which are propagated through the lever arm to generate the power stroke ( Llinas et al . , 2015 ) . Subsequent ADP release results in the highest affinity actin-myosin interaction ( Figure 1A , rigor state ) . Re-binding of ATP into the nucleotide cleft then promotes myosin dissociation from the filament ( Figure 1A , post-rigor state ) and primes the motor for a successive cycle . Sequence divergence of the MD across the superfamily has modulated the kinetics of the various steps of this cycle to tune biophysical parameters including duty cycle , ATPase rate , and force sensitivity , and enabled regulation by post-translational modifications ( Uyeda et al . , 1994 ) . Significant sequence diversity is found on the surface of the MD which contacts F-actin , suggesting that modulation of this interface may enable optimization of these parameters for different cellular roles ( Berg et al . , 2001 ) . However , until very recently ( von der Ecken et al . , 2016 , 2015 ) , the inaccessibility of MD-F-actin complexes to near atomic-resolution structural characterization has been refractory to the detailed mechanistic dissection of this hypothesis . In this study , we focus on the MD of myosin VI , an unconventional myosin motor unique in its ability to walk ‘backwards’ towards actin filament pointed ends ( Wells et al . , 1999 ) . The large insert two in the myosin VI lever arm confers reverse directionality to the motor ( Bryant et al . , 2007; Park et al . , 2007 ) and the adjacent converter region adopts a unique conformation contributing to the large motor step size ( Ménétrey et al . , 2007; Ovchinnikov et al . , 2011 ) , while the rest of the MD , including the actin binding domains , retains a high degree of structural similarity to barbed-end directed motors ( Buss et al . , 2004 ) . In contrast to ‘rower’ myosins ( Leibler and Huse , 1993 ) that generate bulk contractile forces along actin filaments through assembly into filaments , myosin VI operates as either a processive dimeric transporter ( Dunn et al . , 2010; Sweeney et al . , 2007 ) or as a monomeric tether ( Lister et al . , 2004 ) . Myosin VI functions in endocytosis ( Altman et al . , 2007; Buss et al . , 2001; Morris et al . , 2002 ) , intracellular transport ( Inoue et al . , 2008 ) , and stereocilia maintenance ( Avraham et al . , 1995; Hertzano et al . , 2008; Melchionda et al . , 2001; Seiler et al . , 2004 ) , and it has been implicated in deafness ( Ahmed et al . , 2003; Melchionda et al . , 2001; Mohiddin et al . , 2004 ) and cancers ( Dunn et al . , 2006; Wang et al . , 2016; Yoshida et al . , 2004 ) . Extensive crystallographic analysis of myosin VI in the absence of actin has produced high-resolution snapshots of many key states ( Llinas et al . , 2015; Ménétrey et al . , 2005; Ménétrey et al . , 2008; Ménétrey et al . , 2007 ) , laying the groundwork for a complete structural description of the mechanochemical cycle of this motor , which serves as a model for the structural biochemistry of the myosin superfamily . Despite extensive structural and functional characterization , important details of myosin VI mechanism remain unresolved . It remains unclear how precisely phosphate release is coupled to an increase in actin-binding affinity in the ADP state , which is critical for ensuring the MD remains bound to the filament during the power stroke , and how subsequent ADP release further increases actin affinity . The conformation of the ADP state , which dominates the kinetic cycle of the motor and plays a central role in the basic mechanism of force generation ( De La Cruz et al . , 2001; Robblee et al . , 2004 ) , has not been characterized at high resolution . Additionally , the detailed mechanism by which force asymmetrically influences transitions between the ADP and rigor states is unknown . Mechanically gated acceleration of ADP binding has been reported to enable the motor to switch between anchor and transport functions ( Altman et al . , 2004; Chuan et al . , 2011; Robblee et al . , 2004 ) , and force-dependent inhibition of ADP release has also been reported based on single-molecule measurements of monomers ( Oguchi et al . , 2008 ) . Gating of ADP release has been considered as a mechanism for coordinating heads during processive walking ( Dunn et al . , 2010; Elting et al . , 2011; Oguchi et al . , 2008 ) , although kinetic measurements of dimers have favored an alternative ATP-gating model ( Sweeney et al . , 2007 ) . Pioneering low-resolution cryo-EM studies suggested that a minor repositioning of the lever arm accompanies the transition from ADP to rigor ( Wells et al . , 1999 ) , but it is unclear how this is coupled to nucleotide-dependent rearrangements within the motor domain , modulation of actin binding affinity , and force sensitivity . In addition to regulating the mechanochemical cycle , myosin-generated forces have been implicated in influencing actin conformation ( Anson et al . , 1995; Orlova and Egelman , 1997; Prochniewicz et al . , 2010; Prochniewicz and Thomas , 2001 ) , the functional implications of which remain unclear . Myosin II filaments induce severing events in F-actin ( Vogel et al . , 2013 ) , and the cryo-EM structure of rigor myosin IIC bound to actin reveals subtle actin conformational changes in response to myosin binding ( von der Ecken et al . , 2016 ) . Moderate-resolution cryo-EM reconstructions of myosin V bound to F-actin in nucleotide-free and ADP states suggest that binding by this motor may induce changes in actin twist , without further modulation of actin conformation during the mechanochemical cycle; however , the level of detail in the maps presented in this study precluded a detailed description of actin conformational changes ( Wulf et al . , 2016 ) . Actin structural rearrangements , such as altered helical twist , were also proposed to play a role in myosin VI motor activity and step size on the basis of early single-molecule tracking and negative-stain electron microscopy studies ( Nishikawa et al . , 2002 ) that predated our current structural understanding of the myosin VI dimer ( Houdusse and Sweeney , 2016 ) . It is unknown how myosin VI binding modulates the conformation of F-actin , and if actin assumes multiple conformations throughout the force generation cycle . Furthermore , it remains to be determined if myosin-induced conformational changes in actin are uniform among different myosin classes , or if this is an additional element of motor specialization . High-resolution structural snapshots of myosins in multiple actin-bound states are necessary to clarify this issue . Here we present cryo-EM reconstructions of myosin VI bound to F-actin in the rigor state at 4 . 6 Å resolution and the ADP state at 5 . 5 Å resolution along with corresponding atomistic models . Implementing novel adaptations of the Iterative Helical Real Space Reconstruction ( IHRSR ) and Molecular Dynamics Flexible Fitting ( MDFF ) approaches , we present a detailed model of the myosin VI-F-actin interface , and provide the first structure of myosin VI in the ADP state , to our knowledge the highest-resolution structure of any myosin in this state . We compare our rigor structure to the recent high-resolution structure of the myosin IIC-F-actin interface , finding that while the contact surface is conserved , the specific interactions differ substantially between the two myosins . By comparing our myosin VI-F-actin structures in the ADP and rigor states to each other and pre-existing crystal structures of the motor in actin-free states , we clarify the structural transitions of the force generation cycle and propose a structural mechanism for mechanical regulation of ADP affinity . Finally , by comparing the conformation of actin in the myosin VI-bound ADP and rigor state structures to bare filaments , we find that actin structural deformations accompany motor conformational changes during the force-generation cycle . This suggests that actin structural plasticity plays a role in actomyosin VI activity , an F-actin property which previous studies suggest is also likely to be exploited by other myosins , potentially by distinct mechanisms ( Anson et al . , 1995; De La Cruz et al . , 2001 , 1999; Drummond et al . , 1990; Kim et al . , 2002; Llinas et al . , 2015; Noguchi et al . , 2012; Oztug Durer et al . , 2011; Prochniewicz et al . , 2010; Prochniewicz and Thomas , 2001; von der Ecken et al . , 2016; Wulf et al . , 2016 ) .
We utilized an engineered myosin VI construct comprising the MD with 1 IQ fused to an RNA-binding L7Ae kink-turn domain ( Figure 1—figure supplement 1A , C ) . The L7Ae kink-turn domain is oriented such that RNA-binding extends the lever arm and can tune motor activity ( Omabegho et al . , 2017 ) . Combining datasets with and without RNA bound improved the resolution of our reconstruction considerably , suggesting that RNA binding does not alter motor conformation ( Figure 1—figure supplement 1B ) . Thus , we have excluded the engineered regions from our present structural analysis . For image analysis and 3D reconstruction , we developed a hybrid procedure consisting of initial alignment using an adapted EMAN2/SPARX ( Hohn et al . , 2007; Tang et al . , 2007 ) protocol for IHRSR ( Egelman , 2007 ) , which implements refinement and reconstruction of independent half-datasets to minimize noise bias in resolution estimation and alignment , followed by polishing refinement and reconstruction of the full dataset using FREALIGN ( Lyumkis et al . , 2013 ) . Utilizing this approach , we obtained a 3D reconstruction of the myosin VI MD in the nucleotide-free ( rigor ) state bound to F-actin at an average resolution of 4 . 6 Å in the actin filament and bound MD ( Figure 1B and Figure 1—figure supplement 1D ) . As is often the case with helically symmetric specimens , the level of detail in the map decays radially outward from the center of the filament ( Kucukelbir et al . , 2014 ) . Local resolution analysis suggests a gradient from slightly better than 4 Å in the actin region of the map , where large side-chains are definitively resolved , to around 6 Å resolution in the converter and lever arm , where only the contour of the backbone is visible ( Figure 1—figure supplement 1D , E ) . This presents a challenge for analysis , common with cryo-EM reconstructions , where heterogeneity in the map resolution necessitates caution in the generation and interpretation of atomistic models ( Kucukelbir et al . , 2014 ) . We therefore adapted the molecular dynamics flexible fitting ( MDFF ) ( Trabuco et al . , 2008 ) approach to generate a continuous atomistic model which captures high-resolution features in the best-resolved regions of the map by enabling fitting of large side-chains ( Figure 1C , blue ) while avoiding over-interpretation of lower-resolution areas , where the influence of the map was restricted to backbone conformation ( Figure 1C , green , see Materials and methods for details ) . To model the MD-actin interface , we assembled eight actin subunits from the cryo-EM structure of the actin-tropomyosin complex ( pdb 3J8A [von der Ecken et al . , 2015] ) and 6 MDs from the X-ray structure of nucleotide-free myosin VI ( pdb 2BKI [Ménétrey et al . , 2005] ) , which were truncated to exclude the converter and lever arm regions . No density was present for two regions of the MD , loop two and the Hypertrophic Cardiomyopathy ( HCM ) loop , consistent with flexibility , and only the molecular dynamics force field influenced their conformation ( Figure 1C , red ) . The resulting atomistic model ( ‘HR’ , high-resolution ) converged well with a molprobity score of 1 . 44 and a clash score of 0 . 41 ( Table 1 ) . Comparison of the HR MDFF rigor model to the crystal structure of myosin VI in rigor-like state in the absence of actin ( 2BKI ) demonstrates increased jaw closure to relieve a clash with the filament ( Figure 1—figure supplement 1F ) , highlighting the importance of visualizing the motor bound to actin to determine the structure of the rigor state . Cryo-EM structural studies and modelling analyses of diverse actomyosin complexes in strongly-bound states ( Behrmann et al . , 2012; Fujii and Namba , 2017; Lorenz and Holmes , 2010; Wells et al . , 1999; Wulf et al . , 2016 ) as well as hydroxyl-radical foot-printing studies ( Oztug Durer et al . , 2011 ) suggest that all myosins studied thus far engage essentially the same surface on F-actin . However , the lack of MD conservation in actin-binding regions suggests differences may exist in how specific interactions with this F-actin surface are formed by different classes of myosins , which could facilitate tuning of motor properties . To assess the level of conservation at the actomyosin interface , we undertook a detailed comparison of the myosin VI rigor HR MDFF model to the recent 3 . 9 Å structure of the myosin IIC-F-actin rigor complex , as this structure contains side-chain level resolution at the MD-actin interface ( von der Ecken et al . , 2016 ) . For this analysis , we present the superposition of all six actomyosin interfaces from the HR MDFF model , facilitating visualization of the clustering of side-chain positions and thereby providing a means of assessing confidence in specific contacts despite the limitations of the map resolution . Particularly well-resolved density regions , such as the actin nucleotide-binding cleft , demonstrate uniform positioning of large side-chains in density peaks , consistent with our resolution assessment ( Figure 1D ) . As with myosin IIC , the actomyosin interface is comprised of several myosin surface loops ( HCM loop , loop 2 , loop 3 , loop 4 , and helix-loop-helix ) located within the upper 50 KD ( U50 ) and lower 50 KD ( L50 ) domains of myosin which interact with subdomains 1 and 3 of one actin , and subdomain 2 of an adjacent actin ( Figure 2A ) , supporting overall conservation of the interface architecture ( Behrmann et al . , 2012; Holmes et al . , 2003; Rayment et al . , 1993; Várkuti et al . , 2012; von der Ecken et al . , 2016 ) . Regarded as central to all actomyosin interactions ( Kojima et al . , 2001; Sasaki et al . , 2002 ) , one of the initial contacts between myosin and actin is predicted to occur between the myosin helix-loop-helix ( HLH , I525-K550 ) motif in the L50 domain and an actin hydrophobic patch between actin SD1 and SD3 ( Figure 2B ) . In myosin VI we find that the hydrophobic residues P536 and L535 of the HLH are embedded in a groove comprised of I345 , L349 , and Y143 in actin SD1 ( Figure 2B ) , with clear density peaks to support positioning of these side chains . This interface is consistent with hydroxyl radical foot-printing studies demonstrating a hydrophobic interaction between this actin surface and skeletal muscle myosin ( Oztug Durer et al . , 2011 ) . R534 is oriented with its guanidinium group pointing away from the hydrophobic pocket , with the aliphatic portion potentially contributing to the hydrophobic interaction ( Figure 2B ) . The HLH for myosin IIC fits into a similar hydrophobic pocket in actin , with a conserved proline ( P561 ) contributing to this interaction . However , the other specific residues involved in the interaction differ substantially ( Figure 2—figure supplement 1B ) . In contrast to myosin VI , the myosin IIC HLH is comprised of aromatic side chains , with F560 playing a critical role in the interaction with actin ( von der Ecken et al . , 2016 ) . Interactions between myosin loop 3 ( H551-G576 ) in the L50 domain and actin SD1 and the D-loop of the adjacent actin subunit form the Milligan contact ( Milligan , 1996; Milligan et al . , 1990; Rayment et al . , 1993 ) , whose precise role in actin engagement is unclear . Studies of other myosins suggest that this interface is formed by complementary charged surfaces rather than specific salt bridges and thus plays only an ancillary role in the generating the high affinity interaction for the rigor state ( Houdusse and Sweeney , 2016; von der Ecken et al . , 2016 ) . Indeed , for myosin IIC , this seems to be the case ( Figure 2—figure supplement 1C ) . However , the size of loop three varies among myosins , which prior studies have suggested may relate to its prominence in the actomyosin interface ( Van Dijk et al . , 1999 ) . Consistent with this prediction , the large loop 3 of myosin VI likely makes more extensive contacts at this interface than myosin IIC , with MDFF suggesting probable interactions formed between D574 in loop three and K50 in the actin D loop , and E575 and R95 in actin SD2 ( Figure 2C ) . The E575 residue is not conserved among any other myosin isoforms , suggesting that this interaction may be specific to myosin VI ( Zhang and Liao , 2012 ) . For myosin IIC , the actin D loop interaction occurs through E570 in the HLH , whereas for myosin VI the interaction with the actin D loop is likely through D574 in loop 3 ( Figure 2C and Figure 2—figure supplement 1C ) . An additional unanticipated contact is made by myosin VI R561 , which forms a cation-π interaction with Y91 in actin in our model ( Figure 3C ) , discussed further in the next section . The actin residues R95 and Y91 have also been implicated myosin strong-binding interactions by hydroxyl radical foot-printing studies ( Oztug Durer et al . , 2011 ) . Some studies suggest the residue homologous to myosin VI S563 also interacts with actin in other myosins ( von der Ecken et al . , 2016; Zhang and Liao , 2012 ) , but in our model this residue points away from the interface and could instead potentially play a role in stabilizing loop 3 ( not shown ) . In addition to the Milligan contact interactions , MDFF suggests myosin VI makes another unique electrostatic interaction with F-actin . E354 in loop 4 ( A355-C362 ) of the U50 domain of myosin likely forms a salt bridge with K328 in actin SD3 ( Figure 2D ) as supported by clear density for side chains in this region . In contrast , D387 in myosin IIC is reported to interact with a similar charged region in actin comprised of K325 and K327 ( Figure 2—figure supplement 1E ) . However , our interpretation of the structure suggests that N385 is interacting with the charged actin pocket , since D387 is not oriented in a manner to make contacts with actin in this region ( Figure 2—figure supplement 1D ) . The hypertrophic cardiomyopathy ( HCM ) loop ( T392-P410 ) , which protrudes from the U50 domain , features numerous disease mutations ( Sellers , 2000 ) , highlighting the importance of this region for stabilizing interactions with actin . In our reconstruction , we observe both an ordered segment of the HCM loop , which forms an anti-parallel β-sheet comprising residues 392–396 and 406–410 , here referred to as the ‘base’ , as well as a flexible ‘tip’ for which no density was present in our reconstruction ( residues 397–405 ) . As no density was present for the tip , the HR MDFF model exhibits structural variability in this region , and we cannot confidently assign specific orientations to side chains . However , we find that tip residues A399-A402 lie adjacent to a small hydrophobic patch in actin between SD1 and SD2 ( Figure 2E ) , which mutagenesis studies in yeast actin have suggested contributes to the strong-binding myosin interface through residue I341 ( Miller et al . , 1996 ) . This is similar to the myosin IIC-actin interface , where the HCM loop docks on to the same hydrophobic patch in actin and is predominantly stabilized by hydrophobic interactions ( Figure 2- Figure Supplement E ) . While the myosin IIC HCM loop has weak electrostatic interactions at the tip with R424 fitting into a charged pocket of actin , the myosin VI HCM tip lacks charged residues ( Figure 2—figure supplement 1E ) ( von der Ecken et al . , 2016 ) . A similar electrostatic contact with actin could occur via T405 ( Figure 2E ) , a phosphorylation site implicated in regulating directional transport of endocytic clusters ( Buss and Kendrick-Jones , 2008; Naccache and Hasson , 2006 ) . In contrast with myosin IIC , the ordered myosin VI HCM base likely forms an electrostatic interaction with actin , as MDFF suggests a potential salt bridge between R393 and E334 of actin SD1 ( Figure 2F ) . An analogous arginine in myosin IIC , R419 , is a disease-related residue important for stabilizing interactions between actin and myosin ( Lorenz and Holmes , 2010 ) ; however , this residue does not interact with F-actin and instead stabilizes the HCM loop through interactions with Y426 on the opposing strand ( Figure 2—figure supplement 1F ) ( von der Ecken et al . , 2016 ) . Loop 2 ( F621-S642 ) bridges the U50 and L50 domains and has been implicated as the region responsible for initiating binding with actin ( Preller and Holmes , 2013 ) . While we cannot identify specific interactions due to the structural variability of this loop , for which density was not present , L638-I641 at the base of loop 2 are in close proximity to an actin hydrophobic patch , similar to myosin IIC ( Figure 2E and Figure 2—figure supplement 1E ) . Neighboring the hydrophobic base , charged loop residues K634 and K637 lie adjacent to an actin acidic patch comprised of D24-D25 and the acidic N-term ( Figure 2E ) which has been reported to be important for weak-binding actomyosin interactions in yeast actin based on mutagenesis analysis ( Miller et al . , 1996 ) . The homologous region in myosin IIC forms an electrostatic belt with this actin acidic patch that stabilizes the base of loop 2 , and similar interactions with D24-D25 are predicted for myosin V ( von der Ecken et al . , 2016; Wulf et al . , 2016 ) . Although the resolution of loop 2 is poor in our map , likely due to flexibility of this segment , myosin VI could potentially form similar types of electrostatic interactions in this region . While higher resolution reconstructions may clarify specific loop two and HCM tip interactions with actin , intrinsic disorder is also likely to limit visualization of these interfaces . Overall , our analysis reveals a notable lack of conservation at the actomyosin interface between myosin VI and myosin IIC . This is consistent with a model in which the enzymatic core of the MD has been preserved , while a mutable actin-binding surface provides a platform for tuning motor properties . Future structural studies of additional divergent myosin-actin complexes will facilitate the development of a theoretical framework linking specific interface features to biophysical parameters of the MD . To investigate the link between myosin nucleotide state , actin binding affinity , and force sensitivity , we obtained a reconstruction of actin bound to myosin VI in the ADP state at an average resolution of 5 . 5 Å ( Figure 1C , middle , Video 1 ) . The challenge of obtaining high-quality micrographs of this lower-affinity actin-bound state limited the number of segments incorporated into this reconstruction . This , along with the ADP state’s higher level of flexibility in the converter and lever arm regions suggested by biophysical and modelling studies ( Reifenberger et al . , 2009; Sun et al . , 2007; Wulf et al . , 2016 ) likely limited the overall resolution of this reconstruction . As with the rigor state reconstruction , the ADP state yielded a multi-resolution map ( Figure 1—figure supplement 1D , E ) with an estimated resolution of 4 . 7 Å at the actomyosin interface . We observe a clear density peak in the cleft ( Figure 3—figure supplement 1 ) that is absent from the rigor density map , as expected for bound ADP; however , the limited resolution precludes detailed modelling of the nucleotide . The ADP-bound Pi Release ( PiR ) state X-ray structure of myosin VI ( 4PFO [Llinas et al . , 2015] ) was used as the initial model for MDFF , as it contains ADP in the nucleotide-binding pocket . Because of the overall lower resolution at the interface , only backbone atoms were subject to positioning by the density map during the MDFF simulation , again excluding loop two and the HCM loop . The resulting atomistic model ( HR ) converged well with a molprobity score of 1 . 40 and a clash score of 0 . 33 ( Table 1 ) . To monitor global rearrangements of the MD between nucleotide states , we used low-pass filtered density maps and MDFF to extend our models for the ADP and rigor state actomyosin complexes to include the converter and lever arm regions ( details in Materials and methods ) . Due to the overall lower resolution of the filtered maps , we limit our analysis to backbone motions and represent these ‘LPF’ ( low-pass filtered ) MDFF models as backbone averaged structures instead of a superimposed ensemble ( Video 2 ) . As a control for bias imposed by the starting model ( 4PFO ) for the ADP MDFF structures , we also fit the rigor-like myosin VI ( PDB 2BKI ) structure , which we had previously used as the initial model for our rigor atomistic model , into the ADP state density map . This produced a final model ( ADP starting from 2BKI ) more closely resembling the ADP state starting from 4PFO ( Cα RMSD 0 . 8 Å ) than either the 2BKI starting model ( Cα RMSD 1 . 8 Å ) or our MDFF model of the rigor state ( Cα RMSD 1 . 1 Å ) ( Figure 3—figure supplement 2 , Table 2 ) . Regardless of the starting model , MDFF models of the ADP state more closely resemble each other than the rigor state , suggesting our fitting procedure is capturing structural differences between these states that are represented in the maps . Myosin VI affinity for actin increases as it progresses through the force generation cycle , with the rigor state exhibiting approximately 10-fold higher affinity for actin than the ADP state ( De La Cruz et al . , 2001; Robblee et al . , 2004 ) . However , it has been unclear how myosin nucleotide state affects actin affinity once the MD has engaged the filament . Prior comparisons of myosin VI crystal structures representing the states preceding ( Pre-power stroke , PPS and Pi Release , PiR ) and following ( rigor-like ) the ADP state demonstrated that major actin binding cleft rearrangements , reminiscent of a jaw closing , must occur between PiR and rigor to establish interactions with the actin filament ( Llinas et al . , 2015; Ménétrey et al . , 2005; Ménétrey et al . , 2008 ) . While pyrene quenching data ( Llinas et al . , 2015 ) indicated that cleft closure occurs immediately after Pi release , it remained possible that the ADP state displays an actin-binding cleft structure that is overall closed but distinct from the rigor state , which could be related to the lower affinity of the ADP state compared to the rigor state . We find very few changes in the actin binding cleft between the ADP and rigor state atomistic models , which are predominantly subtle local rearrangements which do not impact overall cleft closure ( Figure 3A and B , Figure 3—figure supplement 3 , and Figure 3—figure supplement 4 ) . The Cα RMSD between the U50 and L50 from the rigor and ADP states are 1 . 5 Å and 1 . 1 Å , respectively . Additionally , centroid distances of the U50 and L50 between the two states are 0 . 6 Å and 0 . 5 Å , demonstrating that there is minimal cleft movement ( Figure 3B ) . Atomistic models derived from intermediate-resolution cryo-EM reconstructions for myosin V bound to F-actin also showed minimal cleft rearrangements between the ADP and rigor states ( Wulf et al . , 2016 ) , in agreement with our findings that the major structural changes leading to actin-binding cleft closure must precede the ADP state . As actin binding cleft changes are minimal , alternative mechanisms may also be involved in increasing the affinity of the rigor state . In myosin V , the transducer , a large β-sheet linking the nucleotide-binding cleft to the actin-binding cleft , was reported to adopt a strained conformation in the ADP state , which is relieved upon nucleotide release ( Wulf et al . , 2016 ) . This motivated a model in which an effective increase in actin binding affinity resulted from relief of intramolecular strain in the MD as opposed to a conformational change which modifies contacts with actin . We observe a similar rearrangement of the transducer in myosin VI ( Figure 3—figure supplement 5 ) ; however , we reasoned that subtle modulation of the actin-binding interface could also contribute to differential binding affinity between these states . Analysis of side chain interactions suggested by MDFF shows that nearly all interactions are likely to be maintained between the two states ( Figure 3—figure supplement 3 and Figure 3—figure supplement 4 ) , with the exception of a single residue pair at the Milligan contact . The cation-π interaction between R561 in loop three with Y91 of actin SD2 is not present for the ADP state , suggesting that this contact is likely to form upon the transition from ADP to rigor ( Figure 3C ) . Supporting this model , density for R561 is present in the rigor state reconstruction , but notably absent in the ADP state , consistent with R561 being disordered in this state ( Figure 3C ) . This cation-π interaction is also absent in the ADP from 2BKI model we generated for validation purposes ( Figure 3—figure supplement 6 ) . Previous sequence analysis suggests R561 is conserved with only one other human myosin ( Zhang and Liao , 2012 ) , and a similar interaction is absent in both myosin IIC and myosin V structures ( Figure 3—figure supplement 6 ) , suggesting that this interaction may have evolved to support the specialized properties of myosin VI . Formation of this contact could play a role in increasing affinity for actin in the ADP to rigor transition of myosin VI; future high-resolution structural studies will be required to establish if analogous minor adjustments to the filament binding interface play a role in myosin V , as well as other myosins during this transition . The initiation of force generation occurs once the motor hydrolyzes ATP but has not yet released phosphate , leading to a weak interaction with actin termed the pre-power stroke state ( PPS , Figure 1A ) . Crystallographic analysis revealed a subsequent phosphate release state ( PiR ) representing the state immediately preceding the ADP state where phosphate has been released through a proposed escape tunnel ( Llinas et al . , 2015 ) , but the lever arm has not yet swung ( Figure 1A ) . Comparison of the PiR structure to our ADP structure thus facilitates a detailed analysis of the structural transitions accompanying the primary power stroke of myosin VI ( Figure 4A ) . The switch II loop ( swII ) plays an important role in arranging and stabilizing the myosin nucleotide-binding pocket . The PiR structure revealed that swII adopts an open conformation in this state , opening a path that would allow phosphate escape from the nucleotide binding pocket ( Llinas et al . , 2015 ) . This observation lead to the hypothesis that swII would transiently open only in the PiR state , closing immediately after to prevent phosphate re-binding and thereby enforcing the forward directionality of the mechanochemical cycle ( Llinas et al . , 2015 ) . Consistent with this model , we find that in both the ADP and rigor states , as with the PPS and post-rigor state structures , swII adopts a closed conformation when compared to the PiR state ( Figure 4B and C ) . To execute the power stroke , movement of the nucleotide binding cleft is propagated via the transducer , the relay helix , and the SH1 helix , leading to converter rearrangements which amplify these subtle motions into the swing of the lever arm . Unlike other myosins , the myosin VI PPS converter adopts an unusual conformation and must undergo rearrangements to transition into the rigor state ( Ménétrey et al . , 2007; Ovchinnikov et al . , 2011 ) . However , it has been unclear whether the ADP converter also adopts this unique PPS conformation and how converter rearrangements are propagated into lever arm movement prior to and after the ADP state . By comparing all previously crystalized converter conformations with our density maps ( Figure 4—figure supplement 1 ) , we confirm that the major converter rearrangement occurs from PiR to ADP , with the converter adopting a post-power stroke , rigor-like conformation in this state ( Figure 4D ) . Our structural data are thus consistent with a model in which the major power stroke is accomplished by a converter rearrangement licensed by cleft closure immediately upon phosphate release ( Figure 4E ) ( Llinas et al . , 2015 ) . Previous studies have predicted that forces propagated through the lever arm can allosterically control ADP release by gating conformational transitions in the motor domain required for nucleotide escape ( Altman et al . , 2004; Oguchi et al . , 2008 ) . Early low-resolution cryo-EM structures of myosin VI were consistent with this hypothesis , demonstrating that a small lever arm swing ( ~15–20° ) accompanies the transition from ADP to rigor , presumably due to nucleotide-dependent rearrangements in the MD ( Wells et al . , 1999 ) . However , the nature of these rearrangements and the mechanism coupling them to lever arm dynamics remain unclear . We observe a ~ 30° rotation of the converter around an axis nearly parallel to the actin filament upon the transition from ADP to rigor ( Figure 5—figure supplement 1 , Video 3 ) . This repositioning is sterically coupled to nucleotide cleft opening by opposing motions in the SH1 helix , which transitions from an extended to compact conformation , and the long relay helix , which exhibits winding at the end proximal to the converter ( Figure 5A and B , Figure 5—figure supplement 2 , and Video 4 ) . The relay helix contacts the transducer , which coordinates movement of the switch I ( residues 193–205 ) and N-terminal loops ( residues 96–106 and 305–312 ) away from the nucleotide-binding pocket ( Figure 5A , Video 2 , and Video 5 ) . The SH1 helix is connected to an unnamed loop we here refer to as the cleft loop ( residues 670–681 ) , which unexpectedly displays coherent displacement away from the cleft in the opposite direction ( Figure 5A and B ) . This remodeling is accompanied by smaller rearrangements in the P-loop ( residues 151–156 ) and insert 1 ( residues 278–303 ) , which likely do not play a major role in this step of the mechanochemical cycle ( Figure 5A , and Video 5 ) . To examine the coupling between these rearrangements and the lever arm , we grafted the X-ray structure of the ordered segment of the myosin VI lever arm ( PDB 3GN4 [Mukherjea et al . , 2009] ) on to the distal end of insert two present in our LPF models ( Figure 5C ) ( details in experimental procedures ) . As was observed in the rigor-like crystal structure ( Ménétrey et al . , 2005 ) , the rigor lever arm displays a prominent bend between insert 2 residues 784 and 785 , which our map and model reveals to be absent in the ADP state ( Figure 5C and D , Figure 4—figure supplement 1 ) . Based on these grafted models , bending produces a 30° reorientation of the lever arm , which protrudes off the filament axis in the ADP state but is almost perfectly parallel to the filament in the rigor state , providing a new explanation for early EM observations ( Wells et al . , 1999 ) . The converter rotation and lever arm bending results in a ~ 35 Å displacement of the tip of our modeled lever arm , with a ~ 10 Å projected displacement along the filament axis towards the pointed end ( Figure 5D ) . Although a similar magnitude displacement ( 9 Å ) was observed to accompany this sub-step in single-molecule optical trapping assays of full-length monomeric myosin VI ( Lister et al . , 2004 ) , the construct employed in this study featured additional sequence contributing to the lever arm and thus cannot be directly compared to our truncated model . While our maps and models do not contain atomistic detail in this region , it is tempting to speculate that bending is driven by an electrostatic interaction between negatively charged residue E14 in the proximal light chain bound to insert two and positive residues K736 and R732 , which can be seen in the high-resolution rigor-like crystal structure ( Figure 5—figure supplement 3 ) . A bent lever arm and this interaction are sterically incompatible with the ADP converter position , which would cause severe clashes between the proximal light chain and the MD ( Figure 5E ) . Our models suggest that nucleotide release is coupled to a converter rotation that licenses a lever arm bend in myosin VI , contributing to the displacement observed in previous structural and functional studies of the ADP to rigor transition ( Lister et al . , 2004; Wells et al . , 1999 ) ( Figure 5F ) . This mechanism is clearly not responsible for the small lever arm swing recently reported to be coupled to ADP release in myosin V , which lacks insert 2 ( Wulf et al . , 2016 ) ; rather , it provides a distinct , additional mechanism for myosin VI to reposition the lever arm between the ADP and rigor states . Myosin VI thus seems to have evolved unique conformational changes contributing to both the major power stroke , in which a rearrangement of the converter leads to a larger stroke size than would otherwise be obtained ( Ménétrey et al . , 2007 ) , and the subsequent ADP release sub-step , which amplifies converter rearrangements along the filament axis with a straight-to-bent transition in the lever arm . Furthermore , we propose that force could gate nucleotide engagement by regulating lever arm bending and the associated converter repositioning , with differential effects depending on the geometry ( see Discussion ) . F-actin has the capacity for structural polymorpshim ( Galkin et al . , 2010 ) , and has been observed to adopt distinct conformational states when in complex with several binding partners , notably becoming severely distorted when decorated with the severing factor cofilin ( McGough et al . , 1997 ) . Extensive biochemical and biophysical studies have suggested that myosin binding induces actin structural rearrangements and that actin structural plasticity is critical for proper myosin activity ( Anson et al . , 1995; Drummond et al . , 1990; Kim et al . , 2002; Nishikawa et al . , 2002; Noguchi et al . , 2012; Oztug Durer et al . , 2011; Prochniewicz et al . , 2010; Prochniewicz and Thomas , 2001 ) . Several recent structural studies have described subtle conformational changes in actin when bound by nucleotide-free myosin motor domain ( von der Ecken et al . , 2016; Wulf et al . , 2016 ) . Differences between conformations of actin filaments decorated with myosin V in the nucleotide-free and ADP states have also been described at intermediate resolution ( Wulf et al . , 2016 ) . Furthermore , an indirect reporter of actin conformation based on changes in pyrene fluorescence quenching ( De La Cruz et al . , 2001; De La Cruz et al . , 1999; Kim et al . , 2002; Llinas et al . , 2015; Prochniewicz et al . , 2010; Wulf et al . , 2016 ) has suggested that actin rearrangements accompany transitions between different states in the myosin mechanochemical cycle for myosin V and VI . However , the absence of high-resolution structures of the same myosin in multiple states bound to actin has hampered direct visualization of these rearrangements and an interpretation of their functional relevance during force generation . As we expected actin conformational changes to be subtle , we obtained a reconstruction of F-actin alone at 5 . 5 Å ( Figure 1B , right ) , as well as a corresponding MDFF model to control for error in micrograph pixel size calibration and differences in processing procedures with previously reported structures ( Galkin et al . , 2015; von der Ecken et al . , 2015 ) . The lower resolution of this reconstruction vs . those bound to myosin suggests that myosin binding may rigidify the filament and reduce inherent conformational flexibility ( Galkin et al . , 2012 ) . The myosin-bound density maps were then aligned in the reference frame of the actin-alone reconstruction , followed by re-docking of the corresponding MDFF models , a procedure which we found revealed regular patterns of actin protomer rearrangements which were masked when the MDFF models were superimposed based on the Cα coordinates of individual actin subunits ( data not shown ) . The refined helical parameters are essentially identical for all three filament states we report here ( Table 1 ) , in contrast to both myosin IIC and myosin V , where myosin binding has been reported to induce a 0 . 5–0 . 8° change in azimuthal rotation ( von der Ecken et al . , 2016; Wulf et al . , 2016 ) . Despite this preservation of filament architecture , the actin protomer adopts a unique conformation in each of the three states , with local deformations occurring at the actomyosin interface mediated by the D loop and , intriguingly , at distal lateral contacts in the interior of the filament mediating the interaction between the two strands of the actin filament ( Figure 6 , Figure 6—figure supplement 1 , Videos 6 and 7 ) . The D-loop is a flexible region of actin SD2 that can adopt a range of conformations to facilitate longitudinal contacts during actin polymerization ( Dominguez and Holmes , 2011; Galkin et al . , 2010; Otterbein et al . , 2001 ) and engage with actin binding partners ( Dominguez and Holmes , 2011 ) . As discussed earlier , the D-loop forms interactions with myosin VI loop three as a part of the Milligan contact . Consistent with what has previously been reported for myosin IIC ( von der Ecken et al . , 2016 ) and IE ( Behrmann et al . , 2012 ) , this flexible loop also shifts upon myosin VI binding , orienting slightly towards the MD ( Figure 6B–C and Video 3 ) . Both the ADP and rigor states exhibit excursions of the D-loop relative to the unbound state , with a Cα RMSD of 1 . 0 Å in both cases ( Figure 6D ) . The Cα RMSD between the ADP and rigor states is also of a similar magnitude ( 0 . 9 Å , Figure 6D ) , demonstrating that the D-loop adopts distinct conformational states as the myosin force generation cycle proceeds . The historically named hydrophobic plug ( H-plug , residues 263–273 ) adjoins three actin subunits within the filament lattice , fitting into the groove created by the interface between two actin subunits on the adjacent filament strand ( Chen et al . , 1993; Holmes et al . , 1990 ) . Unexpectedly , myosin VI engagement shifts the H-plug towards the two actins located on the opposite strand , with an increasing deviation of position from the unbound actin state as the motor proceeds from the ADP state ( Cα RMSD 0 . 9 Å ) to rigor ( Cα RMSD 1 . 6 Å , Figure 6B–D and Videos 6 and 7 ) . Although the D-loop and H-plug are distal from one another within a single subunit , they are brought into close proximity between laterally adjacent protomers within the filament , and a single side-chain pair , D-loop residue R39 and H-plug residue E270 , makes a direct contact between them across the interface ( Figure 6E ) . The geometry is incompatible with these residues forming a canonical head-to-head interaction with their charged groups due to their close proximity; instead , they pack together through what may be a mixed electrostatic/Van der Waals interaction . R39 additionally forms a canonical head-to-head salt-bridge interaction with D286 of subdomain 3 of a longitudinally adjacent subunit , placing this residue at a vertex that connects three actin subunits along and across the strands of the filament ( Figure 6E ) . We propose that this bi-partite interaction acts as an allosteric relay . As myosin VI remodels its binding site on actin , primarily through rearrangements of the D-loop , movement of the R39-D286 bridge necessitates repositioning of E270 due to steric exclusion . However , the electrostatic attraction between E270 and R39 prevents rotameric exchange of E270 , producing instead a distortion of the H-plug .
Despite the fundamental conservation of enzymatic mechanism across the myosin superfamily , our studies unambiguously demonstrate that the actomyosin interface is a highly evolvable interaction surface . Comparing myosin IIC and myosin VI , we observe only a single identical actin-binding residue , myosin IIC P561/myosin VI P536 in the HLH; all other interfacial residues have diverged despite the high degree of conservation of the nucleotide-binding pocket ( Figure 7A ) . The myosin VI interface residues we have identified are essentially completely conserved between species ( Figure 7A ) , and likely contribute to this motor’s specialization . The most notable difference is the large increase in the number of loop 3 – actin contacts in myosin VI vs . IIC . This includes a proliferation of specific electrostatic interactions and the unanticipated myosin R561 – actin Y61 cation-π interaction which forms upon the transition from ADP to rigor in our models . These residues are poorly conserved among other myosin isoforms , suggesting that this interaction may have evolved to support the specialized properties of myosin VI . Future comparative structural studies of diverse myosins in complex with actin will be required for a detailed dissection of how modulation of the actomyosin interface correlates with motor-specific biophysical properties . Furthermore , our data suggest that the mechanisms of disease-causing interfacial mutations in other myosins will be difficult to predict from sequence analysis alone , and thus will likely require structural characterization . Comparing our actin-bound reconstructions of the ADP and rigor states to previous structures of myosin VI in isolation in both pre- and post-power stroke states enables us to clarify the sequence of structural transitions which transduce ATP binding and hydrolysis into force production ( Video 8 ) . Nevertheless , the exact causal connections between nucleotide state , myosin conformation , and actin filament engagement remain to be fully resolved . Visualization of transient actin-binding interfaces in both pre-power stroke and post-rigor states that facilitate coordination of rearrangements in the nucleotide-binding cleft , actin-binding cleft , and converter will be necessary . The low-affinity of these high-energy states for actin renders this technically challenging; nevertheless , we optimistically anticipate that continued developments in cryo-EM methodology , including emerging methods for reconstructing filament – binding partner complexes with substoichiometric binding density ( Kim et al . , 2016; Liu et al . , 2017 ) , will render these as tractable structural targets in the future . While the major lever arm swing from the PiR to the ADP state generates directional motion , we propose that the second , smaller lever arm bend upon the ADP to rigor transition contributes to the force sensitivity of the motor . We hypothesize that the directional strain on the myosin head can regulate the lever arm bending observed in our models during transitions between ADP and rigor ( Figure 5D ) . Lever arm bending is coupled to converter rearrangements , which , through the SH1 and relay helices , promotes nucleotide cleft opening and ADP release ( Figures 5A , B and 7B ) . Rearward force applied to the lead head would disfavor the lever arm bend and thus promote the ADP-bound state; conversely , forward load would promote bending and thus favor the rigor state ( Figure 7B ) . This framework is consistent with a model wherein extreme rearward force locks the motor in the ADP state , facilitating a transition from processive transporter to actin-bound tether ( Altman et al . , 2004; Chuan et al . , 2011 ) . This proposal agrees with biochemical data demonstrating an increase in ADP dissociation rate under forward load and increased affinity for ADP under rearward load ( Altman et al . , 2004; Oguchi et al . , 2008 ) , and it is also consistent with recent simulation studies suggesting the converter can adopt a post-powerstroke conformation in the presence of force ( Mugnai and Thirumalai , 2017 ) . Kinetic measurements have suggested that ATP-binding is the force-sensitive step that coordinates the heads of a dimer via intramolecular strain ( Sweeney et al . , 2007 ) ; a structure of the post-rigor actomyosin VI transition complex would be required to develop a detailed mechanistic framework for force sensitivity in the rigor to post-rigor transition . Mechanoregulated ADP release has also been reported for the processive dimeric transporter myosin V ( Oguchi et al . , 2008; Wulf et al . , 2016 ) , as well as the monomeric tethering motor myosin IC ( Greenberg et al . , 2012 ) , and a small lever arm swing has also been reported for myosin V during this transition ( Veigel et al . , 2002 ) . However , these motors each have a distinct structural topology linking the lever arm to the converter , with insert 2 of myosin VI , the site of lever arm bending , being a unique feature of myosin VI . Thus , further studies will be required to reveal mechanistic consonance and dissonance between these motors and themes in myosin force sensitivity . Additionally , our high-resolution actomyosin VI structures provide a foundation for future motor engineering studies . Motor design represents a complementary route for investigating structural features conferring specific biophysical properties , including force sensitivity , which may additionally produce novel cytoskeletal motors with applications in biotechnology and biomedicine . Our studies suggest it may be feasible to alter stepping behaviors and force-sensitivity by exploring alternative lever-arm geometries which modulate bending and thereby the relative positioning of the lever-arm in ADP and rigor . The structural transitions we observe in actin may play an important role in the actomyosin VI force generation cycle . The anomalously high level of conservation in actin has been ascribed to a requirement for allosteric coordination between subunits ( Galkin et al . , 2010; Galkin et al . , 2015 ) , as we indeed observe , leading us to speculate at potential functional roles for these rearrangements . As the unbound actin state represents the conformation adopted by F-actin in the absence of exogenous factors , we propose that our free actin model represents a low-energy conformation of the H-plug and D loop in the context of the filament . As we observe the conformation of the H-plug becomes increasingly deformed as the motor proceeds from ADP to rigor , with increasing RMSD relative to the ground-state of unbound actin ( Figure 6D ) , we speculate that this segment is adopting an increasingly unfavorable conformation as the binding affinity of the actomyosin interface increases , suggesting strain energy may be stored in the filament as the force generation cycle proceeds ( Figure 7A ) . We propose that the D-loop acts as a ‘handle’ which enables myosin VI , and potentially other actin-binding proteins , to transmit conformational changes from the filament surface through an allosteric relay to the H-plug ( Figure 7A ) , converting binding energy into strain energy . As our data suggest that the filament is maximally strained when the motor is most stably bound in rigor , one possibility we envision is that strain energy stored in the filament facilitates MD displacement as ATP rebinds during the transition to post-rigor , with elastic recoil in the actin filament helping to drive this transition forward ( Figure 7A ) . Examination of the actomyosin IIC rigor complex suggests that D-loop remodeling by this motor does not produce H-plug distortion ( Figure 6—figure supplement 2 ) , indicating that this mechanism may be selectively employed by different myosins and represents another avenue for motor specialization . This model is consistent with previous functional data suggesting differential F-actin conformational dynamics in the presence of myosin V and muscle myosin S1 ( Prochniewicz et al . , 2010 ) An additional and non-exclusive possibility is that myosin VI-induced actin conformational states are also modulated through additional mechanisms to regulate the activity of the motor in a context-dependent manner . Actin nucleotide state , a marker of filament age within the cell , influences myosin VI processivity ( Zimmermann et al . , 2015 ) , and could exert its effects via such a mechanism . Finally , conformational changes generated at a single actomyosin VI interface could be allosterically communicated along a filament to influence the binding interactions or activity of other actin binding partners at distal sites , consistent with previous reports of myosin VI influencing the structural dynamics and mechanical rigidity of actin ( Prochniewicz et al . , 2011 ) . While our structural data clearly suggest that conformational changes should propagated across the lateral interface between strands , the technical necessity of saturating the filaments with myosin for high-resolution reconstruction is not compatible with visualizing rearrangements induced at a distance . Future structural studies of myosin VI bound to F-actin trapped in different nucleotide states , as well as filaments sparsely decorated with this motor , will facilitate experimental testing of these proposals .
KMEI: 50 mM KCl , 1 mM MgCl2 , 1 mM EGTA , 10 mM imidazole , pH 7 . 0 . G-Mg: 2 mM Tris , pH 8 , 0 . 5 mM DTT , 0 . 2 M ATP , 0 . 1 mM MgCl2 , 0 . 01% NaN3 . Myosin VI was engineered and purified as previously described ( Omabegho et al . , 2017 ) . Briefly , a DNA construct for protein expression was assembled from fragments encoding porcine myosin VI ( residues 1–817 ) and Archaeoglobus fulgidus L7Ae ( residues 9–118 ) , cloned into a pBiex-1 ( Novagen-Millipore , Burlington , MA ) expression vector modified to include codons for a C-terminal eYFP , and FLAG tag ( DYKDDDDK ) with intervening GSG repeats ( see Figure 1—figure supplement 1 ) . Proteins were expressed by direct transfection of SF9 cells and affinity purified as previously described ( Elting et al . , 2011; Liao et al . , 2009 ) . Rabbit skeletal muscle actin was prepared as previously described ( Pardee and Spudich , 1982 ) . F-actin was prepared by polymerizing 10 μM actin monomers in KMEI + G-Mg buffer overnight at 4°C . F-actin and myosin constructs were diluted to 0 . 3–0 . 6 μM and 2–4 μM , respectively , in KMEI . For nucleotide free conditions , myosin samples were supplemented with 10 U/mL apyrase ( Sigma ) ; for ADP conditions , myosin samples were supplemented with 5 mM Mg-ADP ( Sigma-Aldrich , St . Louis , MO ) pH 7 . 0 . F-actin ( 3 μL ) was applied to a plasma-cleaned 1 . 2/1 . 3 200-mesh C-flat holey carbon grid ( Protochips , Morrisville , NC ) in the humidified chamber of a Leica GP plunge freezer and incubated for 60 s at 25°C . Myosin ( 3 μL ) was then applied and incubated for 60 s . Solution ( 3 μL ) was then removed and an additional 3 μL of myosin was applied . After an additional 60 s , 3 μL of solution was removed , and then the grid was blotted for 2–3 s from the back with filter paper ( Whatman no 5 . ) and plunge-frozen in liquid ethane . Cryo-EM data were collected with the Leginon software on a Tecnai F20 operating at 200 kV using a Gatan K2 Summit direct electron detector in counting mode . Movies were collected with an exposure of 0 . 25 s/frame for a total of 6 . 0 s ( 24 frames ) at a dosage of 6 e-/Å2/s ( 7 . 6 e-/pixel/s ) yielding a total cumulative dose of 36 e-/Å2 . Data were collected at 1 . 5–3 μm underfocus at a nominal magnification of 29 , 000x , corresponding to a calibrated pixel size of 1 . 27 Å at the specimen level . For initial processing steps , image frames were aligned and summed with Unblur ( Grant and Grigorieff , 2015 ) without dose-weighting . Contrast transfer function ( CTF ) estimation and extraction of segments was performed in the Appion data-processing environment ( Lander et al . , 2009 ) . Unless otherwise specified , 2D image processing operations were carried out using proc2d from the EMAN processing package ( Ludtke et al . , 1999 ) . CTF parameters were estimated with CTFFIND3 ( Mindell and Grigorieff , 2003 ) . Segments were windowed in 512-pixel boxes with 81 Å of non-overlap corresponding to a step-size of 3 actin protomers , normalized with xmipp_normalize ( Scheres et al . , 2008 ) , then binned by 2 . Segments were extracted for each state: ADP ( 36 , 114 ) , rigor ( 56 , 116 ) , and actin alone ( 63 , 139 ) . For 3D refinement and reconstruction , we adapted the IHRSR protocol recently described in Kim et al . ( Kim et al . , 2016 ) , performing initial refinement and reconstruction using functions from the SPARX/EMAN2 ( Hohn et al . , 2007; Tang et al . , 2007 ) libraries and helical search using the program hsearch_lorentz ( Egelman , 2007 ) , followed by final refinement and reconstruction using FREALIGN ( Lyumkis et al . , 2013 ) . Briefly , segments were extracted from phase-flipped images , then refined against an initial model generated by low-pass filtering an actin reconstruction ( EMD-1990 [Behrmann et al . , 2012] ) to 35 Å . The reconstruction obtained from this refinement run was then low-pass filtered to 35 Å and used as the initial model for a second round of refinement , where poorly aligning segments were excluded using a cross-correlation cutoff of 1 . 5 σ . Segments with correlation scores above the cutoff were then divided into two random half-datasets , and independent refinement of these half-datasets ( to minimize noise bias [Scheres and Chen , 2012] ) was re-initialized using the same low-pass filtered initial model . After each round of refinement , the asymmetric reconstructions of the half-datasets were summed , and the sum was used to calculate new helical parameters . These helical parameters were then applied to each half-reconstruction independently , which were then compared and low-pass filtered based on the Fourier Shell Correlation ( FSC ) to provide the references for the next round of refinement . After refinement in EMAN2/SPARX , un-binned segments were generated using alignparts_lmbfgs ( Rubinstein and Brubaker , 2015 ) on all acquired frames to correct for non-uniform beam-induced drift ( motion correction ) and apply an exposure-dependent filter to maximize signal at all spatial frequencies ( Grant and Grigorieff , 2015 ) . Data from all frames were included in segments extracted at this stage . Parameters from the half-data sets were recombined , then final refinement and reconstruction was performed with FREALIGN v 9 . 11 using fixed helical parameters and a strict low-pass filter of 10 Å , as we found including higher-resolution information in the refinement did not improve the reconstructions ( data not shown ) . The final average resolutions reported were determined based on the FSC 0 . 143 criterion ( Rosenthal and Henderson , 2003 ) as 4 . 6 Å ( rigor ) , 5 . 5 Å ( ADP ) , and 5 Å ( actin alone ) ( Figure 1—figure supplement 1 ) . The maps were sharpened using a B-factor peaking at the nominal average resolution as indicated in Table 1 using the program BFACTOR . Local resolution assessment was performed in two independent fashions . Since it was clear that the resolution decayed radially from the core of the filament , we calculated a series of reconstructions with cylindrical masks of radii chosen to exclude certain portions of the map: 120 Å radius for the full map , 90 Å radius to exclude the converter and lever arm ( which was used to calculate the overall resolution reported ) , and 40 Å radius for the actomyosin interface ( Figure 1—figure supplement 1 ) . Resolutions were determined for each individual reconstruction based on the FSC 0 . 143 criterion . Local resolution was also estimated for strong-bound ADP and rigor states using ResMap ( Kucukelbir et al . , 2014 ) on the full density maps revealing a resolution gradient of better than ~4 Å ( actin ) to worse than 5–6 Å ( myosin lever arm ) . Atomistic models for the cryo-EM density maps were generated using the Molecular Dynamics Flexible Fitting ( MDFF ) procedure . Initial models were built from eight actin subunits ( 3J8A ) and six myosins ( 2BKI for rigor state , 4PFO ADP-strongbound ) . Two models were generated for each state: A high resolution ( HR ) model for actomyosin interface analysis and a low pass filtered ( LPF ) model for analysis of global MD rearrangements . For the HR model , the MD was truncated to exclude the lever arm and converter regions , and the electron density maps used were B-factor sharpened and filtered to nominal resolution as indicated in Table 1 . As there is no pre-existing structural information for the flexible loops , loop 2 and the HCM loop , we manually constructed these regions using Coot ( Emsley et al . , 2010 ) . Initial models were then assembled through rigid body docking in Chimera , followed by flexible fitting with DIREX ( Wang and Schröder , 2012 ) . MDFF was performed with explicit solvent , 50 mM KCl , and symmetry restraints imposed on Cα of actin . The simulation was run in three steps: a brief energy minimization step to remove severe clashes from the starting model , then molecular dynamics with low map weighting ( 250ns simulation ) , followed by a longer energy minimization ( 2000 steps ) using a higher map-weighting . To accommodate the multi resolution maps and the different resolutions of each density map , each state was subjected to MDFF differently: Due to the higher resolution of the rigor state density map , backbone atoms and large side chain atoms in actin ( Phe , Tyr , Trp , His , Arg , Gln , Lys , Met ) were subjected to fitting by the electron density map potential . For all other models ( ADP , actin alone , and ‘low resolution’ models ) , only backbone atoms were permitted to feel map potential . Loop two and the HCM loop were excluded from flexible fitting and only subjected to molecular dynamics , and the positions of ADP and magnesium ions were kept fixed during the molecular dynamics simulation due to the limited resolution of the reconstructions . We tested various values of the weighting factor ‘g’ for both the molecular dynamics and the long energy minimization stages and selected the optimal value by assessing quality using Molprobity ( Chen et al . , 2010 ) as implemented in Phenix ( Adams et al . , 2010 ) as described previously ( Kim et al . , 2016 ) . To generate the LPF models , for both ADP and rigor state , the lever arm and converter ( and CaM for the rigor structure ) from 2BKI were grafted onto the HR atomistic models , and electron density maps were B-factor sharpened and filtered to 7 . 5 Å resolution to accommodate the lower resolution portions of the map consisting of the converter and lever arm . Initial fitting was carried out through rigid body docking in Chimera and then flexible fitting with Direx . The MDFF was carried out in the same manner as the ‘high resolution’ models with the filtered density maps guiding fitting for only the protein main-chain atoms . Due to the overall lower resolution of the filtered maps , atomistic models were backbone averaged in Phenix ( Adams et al . , 2010 ) and side chains were truncated to poly-alanine . An analogous backbone averaging procedure was applied to actin subunits from the HR models to visualize conformational changes in actin between states in the force generation cycle ( Figure 6 ) . To create the extended lever arm models , the lever arm and 2 calmodulins from 3GN4 ( truncated at K848 ) were fit as a rigid body into the 7 . 5 Å resolution filtered density maps , which were segmented to only include the motor domain , converter , calmodulin and ordered region of the lever arm . After fitting , the lever arm was truncated , then grafted onto the LPF model at a site chosen to match the local path of the density . As the fits were dependent on the correspondence between the first IQ and calmodulin from the crystal structure and our density maps , only the lever arm and one calmodulin were fit into the density; thus , the second calmodulin and extension of the lever arm are extrapolations based on the crystal structure . The lever arm was grafted at V784 for the rigor model and D773 for the ADP model . Conservation analysis was carried out through sequence alignment using the EMBL-EBI Clustal Omega server ( Sievers et al . , 2011 ) of human myosin VI sequence with myosin VI sequences from 46 other organisms or 18 other human myosin isoforms obtained through a NCBI BLAST search . Conservation mapping onto the myosin VI structure was conducted in Chimera . Density map alignments , structural superpositions of atomistic models , RMSD calculations , centroid determinations , and displacement calculations were conducted in UCSF Chimera ( Pettersen et al . , 2004 ) . Inter-domain rotation axes and angles were calculated using DynDom3d ( Poornam et al . , 2009 ) . Cα displacement vectors were calculated using a Python script which has previously been described ( Alushin et al . , 2014 ) . Cryo-EM density maps and corresponding atomistic models for rigor , ADP , and actin alone reconstructions have been deposited in the Electron Microscopy Data Bank ( EMDB ) and Protein Data Bank ( PDB ) . Electron Microscopy Data Bank accession codes: EMD-7115 ( actin alone ) , EMD-7116 ( rigor ) , EMD-7117 ( ADP ) . Protein Data Bank accession codes: Actin alone: 6BNO ( HR MDFF ) , 6BNU ( averaged HR MDFF ) ; Rigor: 6BNP ( HR MDFF ) , 6BNV ( LPF MDFF ) ; ADP: 6BNQ ( HR MDFF ) , 6BNW ( LPF MDFF ) . All custom software utilized in structure determination and analysis are available at: https://github . com/alushinlab/goldhelix ( Alushin , 2017; copy archived at https://github . com/elifesciences-publications/goldhelix ) . | Like miniature motors , proteins called myosins generate the forces needed for cells to move and for muscles to contract . Myosins use the energy stored in a chemical called ATP to move along filaments made from another protein called actin and produce force . The same part of the myosin protein that binds to and uses ATP also contacts actin . As a myosin protein consumes ATP , it cycles through a series of shape changes to drive the motor protein forward , altering how it interacts with the actin filament in the process . Although all myosins use ATP in fundamentally the same way , individual members of this protein family have specialized properties that enable them to carry out different roles . It is not clear whether each type of myosin makes unique contacts with the actin filament , which could help determine these properties . Furthermore , mechanical forces can control the activity of myosin motors in ways that are poorly understood . Gurel et al . have now looked at a family member called myosin VI , which moves in the opposite direction along actin filaments relative to other myosins , to better understand the properties of these proteins . An imaging technique called cryo-electron microscopy ( cryo-EM ) was used to determine the three-dimensional structure of myosin VI bound to actin at two steps in its cycle . Gurel et al . found that myosin VI formed specific interactions with actin that were very different from another myosin family member called myosin IIc , whose structure bound to actin was already known . In addition , the structural changes observed between the two stages of myosin VI’s cycle provided insight into how force could be used to control the motor . Together these findings give a more detailed picture of how myosins work . They suggest that the surface of myosin that contacts actin can evolve to change the properties of a specific myosin . Studies of other myosins bound to actin will provide further insight into how distinct interactions relate to motor-specific properties . Future studies could also help scientists to understand how mutations in genes for myosins – which have been linked to a number of diseases in humans – alter the way in which myosins interact with actin filaments . This in turn could give insight into how these mutations disrupt the proteins’ activities . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Cryo-EM structures reveal specialization at the myosin VI-actin interface and a mechanism of force sensitivity |
Membrane protein biogenesis requires the coordinated movement of hydrophobic transmembrane domains ( TMD ) from the cytosolic vestibule of the Sec61 channel into the lipid bilayer . Molecular insight into TMD integration has been hampered by the difficulty of characterizing intermediates during this intrinsically dynamic process . In this study , we show that cotransin , a substrate-selective Sec61 inhibitor , traps nascent TMDs in the cytosolic vestibule , permitting detailed interrogation of an early pre-integration intermediate . Site-specific crosslinking revealed the pre-integrated TMD docked to Sec61 near the cytosolic tip of the lateral gate . Escape from cotransin-arrest depends not only on cotransin concentration , but also on the biophysical properties of the TMD . Genetic selection of cotransin-resistant cancer cells uncovered multiple mutations clustered near the lumenal plug of Sec61α , thus revealing cotransin’s likely site of action . Our results suggest that TMD/lateral gate interactions facilitate TMD transfer into the membrane , a process that is allosterically modulated by cotransin binding to the plug .
Most eukaryotic membrane proteins are cotranslationally integrated into the endoplasmic reticulum ( ER ) membrane ( Shao and Hegde , 2011 ) . This process begins when the first hydrophobic segment of a nascent membrane protein , often a transmembrane domain ( TMD ) , emerges from a translating ribosome and is recognized by the signal recognition particle ( SRP ) . The SRP system ( Akopian et al . , 2013 ) targets the ribosome-nascent polypeptide complex ( RNC ) to the ER membrane and transfers it to a translocation channel , or translocon , the central component of which is the Sec61 complex ( Osborne et al . , 2005 ) . Sec61 binds the ribosome near the polypeptide exit tunnel and mediates both cotranslational translocation and TMD integration . This means that the Sec61 channel opens not only toward the ER lumen , but also laterally toward the lipid bilayer . While lateral TMD release through the Sec61 channel was appreciated long ago ( Martoglio et al . , 1995; Do et al . , 1996; Mothes et al . , 1997 ) , the precise mechanism of this crucial step in membrane protein biogenesis remains unclear . During integration into the membrane , TMDs must move from the aqueous pore of the Sec61 channel into the surrounding lipid bilayer . Several models have been proposed to explain how this occurs . In one model , the TMD is thought to equilibrate between the aqueous Sec61 channel and the lipid bilayer ( Heinrich et al . , 2000; Hessa et al . , 2007; Ojemalm et al . , 2011 ) . This process would be facilitated by an intrinsically dynamic lateral gate in Sec61 that allows the polypeptide segment inside the channel to reversibly sample the membrane . A second model proposes that TMD integration is facilitated by direct interactions with Sec61 . This idea is supported by the analysis of stalled RNCs of membrane proteins at different lengths . Not only does the TMD of such stalled RNCs crosslink to Sec61 , but it also appears to be oriented in a specific manner , as judged by asymmetric crosslinks to adjacent TMD residues ( McCormick et al . , 2003; Sadlish et al . , 2005 ) . Molecular dynamics simulations further suggest that lateral gating is facilitated by the TMD and that TMD movement into the lipid bilayer is a kinetically irreversible event ( Zhang and Miller , 2010 , 2012 ) . Analysis of cotranslational integration in yeast has also revealed a kinetic barrier to TMD insertion ( Cheng and Gilmore , 2006; Trueman et al . , 2011 ) . Structural studies of the Sec61 complex have provided important clues to the mechanism of TMD integration . The crystal structure of the archaeal complex revealed that SecY ( homologous to eukaryotic Sec61α ) comprises a compact bundle of 10 transmembrane helices ( TM1–10 ) arranged in a pseudo-twofold symmetric structure , with TM1–5 and 6–10 forming two halves of a ‘clamshell’ ( Van den Berg et al . , 2004 ) . The interior of the clamshell forms an hourglass-shaped pore in the membrane , the center of which is occluded by a short α-helix termed the plug . The front of the clamshell , at the interface between TM2/3 and TM7/8 , forms a lateral gate that , when opened , could provide direct access to the lipid bilayer from the central pore . Translocation across the membrane requires movement of the plug to open the channel toward the ER lumen . This is thought to occur when the first TMD ( or signal sequence ) of an RNC intercalates between the lateral gate helices of Sec61α . When coupled with ongoing translation , this event would position the nascent polypeptide within the channel and provide access of the hydrophobic TMD to the lipid bilayer . This idea is supported by the observation that signal peptides can be crosslinked to Sec61α TM2 and TM7 ( Plath et al . , 1998; Wang et al . , 2004 ) and that mutations in the lateral gate can influence TMD integration ( Junne et al . , 2010; Trueman et al . , 2012 ) . While crystal structures of prokaryotic Sec61 have revealed the lateral gate in a continuum of partially open states ( Tsukazaki et al . , 2008; Zimmer et al . , 2008; Egea and Stroud , 2010 ) , how the gate transitions between open and closed conformations , and especially the role of the nascent TMD in this process , remains unclear . A major obstacle to understanding cotranslational TMD integration has been the inability to stabilize and interrogate discrete pre-integrated intermediates during this highly dynamic process . We and others previously described cotransins , a class of cyclic peptides that bind Sec61α and potently inhibit cotranslational translocation of a subset of secretory and membrane proteins ( Besemer et al . , 2005; Garrison et al . , 2005; MacKinnon et al . , 2007 ) . We reasoned that because Sec61 represents the final destination of a nascent TMD prior to its integration into the membrane , small-molecule modulators like cotransin could shed light on this poorly understood process . In this study , we have investigated the mechanism by which cotransin inhibits membrane protein insertion . We demonstrate that cotransin stabilizes an ephemeral pre-integrated intermediate in which the TMD docks against the cytosolic tip of the Sec61 lateral gate . Progression through this cotransin-arrested stage was found to depend on TMD hydrophobicity , helical propensity , and charge distribution , features that contribute to cotransin’s ability to discriminate among different substrates . Using genetic selection in human cancer cells , we identified multiple point mutations in the lumenal plug region of Sec61α that confer dominant resistance and prevent cotransin binding . These results support a model in which cotransin binding to the plug prevents lateral gating by susceptible TMDs , whereas TMDs with increased hydrophobicity and helical propensity can override the cotransin-imposed block . More generally , our characterization of a pre-integrated complex implies that dynamic interactions between the TMD and Sec61 can facilitate lateral gating and membrane integration .
To explore the effect of cotransin on TMD integration , we used a reconstituted system comprising a mammalian translation extract supplemented with microsomal ER membranes ( Sharma et al . , 2010 ) . When this system is programed with a truncated mRNA that lacks a stop codon , the ribosome translates to the end of the transcript and stalls , creating a synchronized population of ribosome nascent chain complexes ( RNCs ) of defined length that functionally engage the targeting and translocation machinery ( Gilmore et al . , 1991 ) . We used this approach to prepare translocation intermediates of TNFα , an integral membrane protein whose single TMD mediates targeting , membrane insertion , and translocation of the C-terminal ectodomain ( Figure 1A ) . Importantly , cotranslational integration of TNFα is potently inhibited by CT8 , a recently described cotransin variant ( Maifeld et al . , 2011 ) . 10 . 7554/eLife . 01483 . 003Figure 1 . CT8 blocks TMD integration into the membrane . ( A ) Schematic of TNFα primary structure and membrane orientation . Cys49 is indicated by an asterisk . ( B ) TNFα 126-mers containing an N-terminal FLAG- or HA-tag were translated in the presence of canine rough microsomes , solubilized with 1% Deoxy BigChap ( DBC ) , and immunoprecipitated ( IP ) with anti-FLAG affinity resin . IP eluates were analyzed by immunoblotting for FLAG-TNFα and Sec61α . ( C ) As in ( B ) , except that microsome-targeted 126-mers were prepared in the presence or absence of the photo-affinity probe CT7 and photolyzed ( hν ) either before ( lane 2 ) or after immunoprecipitation ( lane 3 ) . The photo-crosslinked CT7/Sec61α adduct was detected by click chemistry with TAMRA-azide , followed by in-gel fluorescence imaging . ( D ) Protease protection assays with 35S-labeled 126-mers translated in the presence or absence of canine rough microsomes ( CRM ) , followed by treatment with proteinase K ( PK ) and TX-100 ( Det ) as indicated . The nascent chain and protease-protected fragment are indicated ( closed and open triangle , respectively ) . An unidentified weak band ( asterisk ) was occasionally observed . ( E ) Microsome-targeted 126-mers with Cys49 or lacking all cysteines were assembled in the presence or absence of CT8 and treated with N-ethylmaleimide ( NEM ) as indicated . Samples were subsequently denatured with SDS , treated with PEG-maleimide ( PEG-Mal ) , and analyzed by SDS-PAGE/phosphorimaging . NEM inaccessibility is indicated by a shift to higher molecular weight , corresponding to nascent chains ( NC ) modified by PEG-Mal at Cys49 . ( F ) NEM accessibility of RNCs of varying length , defined as fraction of nascent chains modified by PEG-Mal , determined by phosphorimaging and normalized to a matched control reaction lacking NEM . Data represent the mean ± range of two independent experiments . For raw phosphorimaging data , see Figure 1—figure supplement 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 00310 . 7554/eLife . 01483 . 004Figure 1—figure supplement 1 . CT8 blocks TMD integration into the membrane . ( A ) Chemical structures of cotransin variants used in the study . CT7 contains a photo-reactive diazirine to enable covalent photo-crosslinking to Sec61α . All compounds block TNFα insertion into ER microsomes with similar potency ( Maifeld et al . , 2011 , and data not shown ) . ( B ) Scheme for CT7 photo-affinity labeling of Sec61α . The covalent CT7/Sec61α adduct ( formed after irradiation with UV-light , hν ) is detected using click chemistry between the alkyne group in CT7 and a rhodamine-azide ( TAMRA-N3 ) . ( C ) NEM accessibility assays of Cys49 RNCs of varying length in the presence and absence of CT8 . Gels are representative of the primary data used to derive the graph in Figure 1F . ( D ) Microsome targeted TNFα 126-mers with Cys49 or lacking all cysteines were treated with proteinase K ( PK ) as indicated . Samples were subsequently denatured with SDS , treated with PEG-maleimide ( PEG-Mal ) as indicated and analyzed by SDS-PAGE/phosphorimaging . PEG-Mal efficiently modifies the PK-protected TNFα fragment containing Cys49 . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 004 We first established the feasibility of producing a stable complex comprised minimally of CT8 , Sec61 , and TNFα RNCs . Direct detection of cotransin binding to Sec61 was aided by CT7 , an isosteric and equipotent analog of CT8 equipped with a diazirine for covalent photo-crosslinking to Sec61α ( MacKinnon et al . , 2007; Figure 1—figure supplement 1A , B ) . FLAG-tagged TNFα nascent chains of 126 residues ( 126-mers ) were assembled on ER microsomes in the presence or absence of CT7 , solubilized with digitonin , immunopurified with anti-FLAG beads , and analyzed for the presence of Sec61α and CT7 . Sec61α specifically copurified with FLAG-TNFα RNCs , as it was not detected on anti-FLAG beads recovered from control reactions programed with HA-tagged TNFα ( Figure 1B ) . Moreover , the amount of Sec61α copurifying with TNFα RNCs was identical in the absence or presence of CT7 ( Figure 1C , lane 1 vs 2 ) , suggesting that CT7 does not weaken the affinity of the RNC-Sec61 complex . Finally , the amount of CT7 covalently bound to Sec61α in the FLAG-TNFα immunoprecipitates was identical whether the photo-crosslinking step was carried out before or after the stringent immunopurification and washing steps ( Figure 1C , lane 2 vs 3 ) . Thus , CT7 , Sec61 , and TNFα RNCs assemble into a high-affinity ternary complex amenable to biochemical analysis . We employed a combination of protease protection and cysteine accessibility assays to probe the local environment of the TNFα nascent chain . At the 126-mer length , TNFα RNCs should have sufficient polypeptide beyond the TMD ( ∼75 amino acids ) to have stably inserted into the Sec61 channel in the ‘looped’ orientation , with the TMD at least partially exposed to the lipid bilayer ( Martoglio et al . , 1995; McCormick et al . , 2003 ) . Consistent with this expectation , 126-mer RNCs assembled in the absence of CT8 were predominantly shielded from protease , generating a large protected fragment that consists of the TMD plus the lumenal C-terminal domain ( Figure 1D , Figure 1—figure supplement 1D ) . This fragment was not observed in the absence of microsomes , suggesting that the TNFα TMD and ectodomain are protected by the Sec61 complex and the membrane . RNCs produced in the presence of CT8 generated sharply reduced amounts of the protected fragment , indicating that the nascent chain remained primarily in a protease-accessible compartment , despite binding tightly to the Sec61 complex ( Figure 1D ) . To probe the environment of the TMD itself , we analyzed TNFα RNCs containing a single native cysteine at position 49 , near the C-terminal end of the TMD . We monitored accessibility of Cys49 to N-ethylmaleimide ( NEM ) , which efficiently reacts with cysteine thiols exposed to an aqueous environment . In this assay , ER-targeted RNCs are first alkylated with NEM under native conditions , followed by quenching , denaturation with SDS , and modification of unreacted sulfhydryls with PEG-maleimide . Thus , PEG-modified nascent chains ( identified by a molecular weight shift ) represent RNCs whose Cys49 sulfhydryl was initially inaccessible to NEM . When 126-mers were assembled without CT8 , Cys49 was unreactive toward NEM , whereas the same nascent chains assembled in the presence of CT8 were strongly reactive and thus protected from subsequent reaction with PEG-maleimide ( Figure 1E ) . These results are consistent with the protease protection assay and demonstrate that CT8 causes 126-mers to occupy a distinct environment , one in which Cys49 in the TMD is exposed to aqueous solvent . Applying this assay to RNCs of varying length revealed that in the absence of CT8 , Cys49 progressed from a primarily NEM-accessible environment to one that was inaccessible ( Figure 1F ) . This transition occurred over a narrow range of nascent chain lengths , between 116- and 126-mers . By contrast , in the presence of CT8 , the TMD remained accessible to NEM at every nascent chain length tested ( Figure 1F , Figure 1—figure supplement 1C ) . Given that efficient alkylation with NEM requires an aqueous environment , the abrupt transition observed in the absence of CT8 likely represents TMD integration into the lipid bilayer . Not only is the polypeptide sufficiently long at this stage to exit the translocon , but this is also the length at which earlier studies have observed photo-crosslinking of the TMD to phospholipids ( Martoglio et al . , 1995; Mothes et al . , 1997; Urbanus et al . , 2001 ) . While the TMD may remain close to Sec61 at this stage ( McCormick et al . , 2003; Sadlish et al . , 2005; Devaraneni et al . , 2011; Frauenfeld et al . , 2011; Hou et al . , 2012 ) , it has most likely exited the central aqueous pore of Sec61 . Collectively , the above results demonstrate that CT8 prevents TMD insertion into the membrane at a step that occurs after RNC targeting to Sec61 . To gain insight into the topological relationship between the TMD and Sec61 along the integration pathway , we used bis-maleimidohexane ( BMH ) , which can covalently crosslink solvent-accessible cysteines that lie within ∼13 Å of each other . BMH crosslinking of wild-type TNFα 126-mers ( native cysteines at positions 30 and 49 in the TMD ) yielded no major crosslinked products ( Figure 2A ) , consistent with the solvent inaccessibility of the TMD at this length . In the presence of CT8 , the 126-mers crosslinked strongly to partners of ∼10 and ∼40 kDa ( Figure 2A ) , identified by immunoprecipitation as Sec61β and Sec61α , respectively ( Figure 2—figure supplement 1 ) . These crosslinks were diminished when a termination codon was introduced at position 126 , indicating that only tRNA-bound nascent chains crosslink to the Sec61 complex . Further supporting this interpretation , a portion of TNFα RNCs retained the tRNA during electrophoresis , and this peptidyl-tRNA was also found to efficiently crosslink to Sec61α and Sec61β ( Figure 2A , Figure 2—figure supplement 1B , arrowheads ) . Mutagenesis of each cysteine in TNFα showed that Cys49 was uniquely responsible for the crosslinks . 10 . 7554/eLife . 01483 . 005Figure 2 . CT8 stabilizes a transient pre-integrated intermediate . ( A ) Microsome-targeted TNFα 126-mers assembled in the presence or absence of CT8 ( 1 μM ) were treated with BMH as indicated ( Stop: termination codon at position 126 ) . Bands corresponding to the nascent chain ( NC ) and the NC crosslinked to Sec61α and Sec61β are indicated ( for IP confirmation , see Figure 2—figure supplement 1 ) . Residual NC-tRNA as well as crosslinks to Sec61α and Sec61β is also indicated ( closed and open triangle , respectively ) . ( B ) BMH crosslinking reactions with TNFα RNCs of varying length . ( C ) Microsome-targeted TNFα RNCs of the indicated length were assembled in the continuous presence of CT8 ( ‘Co-’ ) or treated with CT8 post-translationally ( ‘Post-’ ) , followed by BMH crosslinking . NC crosslinks to Sec61α were quantified by phosphorimaging and normalized to the cotranslationally CT8-treated control . Quantified data represent the mean ± standard deviation of three independent experiments . In ( B ) and ( C ) , TNFα RNCs contained a single cysteine ( Cys49 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 00510 . 7554/eLife . 01483 . 006Figure 2—figure supplement 1 . CT8 stabilizes a transient pre-integrated intermediate . ( A ) Denaturing immunoprecipitation ( IP ) of BMH crosslinking reactions . Cys49 126-mers were prepared in the presence or absence of CT8 and crosslinked with BMH . Reactions were then quenched with DTT , denatured in SDS and immunoprecipitated with antibodies directed against the indicated proteins . Eluates were analyzed by SDS-PAGE and autoradiography . ( B ) Microsome targeted wild-type TNFα 126-mers were assembled in the presence or absence of 1 µM CT8 and treated with BMH as indicated . The BMH treated samples were further treated with base or RNAse A and reactions analyzed on NuPage gels at pH 7 . 3 . Base and RNAse treatment collapses the nascent chain-tRNA bands . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 006 We next analyzed crosslinking patterns of TNFα RNCs containing a single cysteine ( Cys49 in the TMD ) at various chain lengths from 70 to 146 residues , in the absence and presence of CT8 . Without CT8 , crosslinks first appeared between the 80-mer and Sec61β , which contains a single cysteine in its presumably unstructured cytosolic tail ( Figure 2B ) . This demonstrates that the 80-mer was properly targeted to the Sec61 complex and that Cys49 , predicted to have just emerged from the ribosomal exit tunnel , was exposed to the cytosol . Crosslinking to Sec61β decreased as the nascent chain was extended to 96 residues , at which point strong crosslinks to Sec61α appeared ( Figure 2B ) . Given the ∼15 amino acid tether between the C-terminal end of the TMD and the polypeptide exit tunnel ( positions 51–65 of the 96-mer ) , Cys49 is likely situated within or near the cytosolic vestibule of Sec61α such that it can reach a solvent-exposed cysteine on Sec61α ( via the 13 Å crosslinker ) . As the nascent chain was extended beyond 96 residues , crosslinking to Sec61α decreased abruptly and was undetectable by 126 amino acids . This transition parallels the loss in NEM accessibility ( Figure 1F ) and further suggests that the TMD exits the Sec61α vestibule at this stage of nascent chain elongation . Integration of the TMD is therefore preceded by a step in which Cys49 of the TMD passes within ∼13 Å of an accessible cysteine in Sec61α . A strikingly different picture emerged when this analysis was performed in the presence of CT8 . Although short nascent chains ( up to 86-mers ) had a similar crosslinking profile , all RNCs beyond this length were clearly distinct from the matched samples lacking CT8 . Not only were the crosslinks to Sec61α and Sec61β more intense , but they also persisted beyond the point where TMD integration occurred in the absence of CT8 ( Figure 2B ) . Order-of-addition experiments indicated that TMD egress from the cytosolic vestibule ( operationally defined by the disappearance of TMD/Sec61α crosslinks ) becomes irreversible once the nascent chain has progressed from 106- to 126-mers ( Figure 2C ) . When 106-mers were first assembled in the absence of CT8 , isolated by sedimentation , and then incubated with CT8 post-translationally , enhanced TMD/Sec61α crosslinks were observed . By contrast , post-translational addition of CT8 to identically prepared 126-mers had no effect , indicating that the 126-mers had crossed a kinetic point-of-no-return that could not be reversed by CT8 . Taken together , these data argue: ( 1 ) CT8 stabilizes a configuration of the TMD/Sec61 complex that occurs just prior to a committed step of TMD integration , and ( 2 ) a similar pre-integrated configuration occurs in the absence of CT8 for nascent chain lengths of 96–106 amino acids . That CT8 actually enhances the initial interaction between Sec61α and the TMD ( as determined by BMH crosslinking ) suggests an allosteric mechanism of action . The CT8-arrested TMD/Sec61 complex provides a unique opportunity to characterize an otherwise transient , pre-integrated intermediate . We used BMH crosslinking to probe the environment of a cysteine residue scanned across the TMD and flanking regions of 126-mers assembled in the presence or absence of CT8 ( Figure 3A ) . As expected on the basis of protease and NEM accessibility experiments ( Figure 1D , E ) , 126-mers assembled in the absence of CT8 displayed no protein crosslinks to cysteines within the TMD ( Figure 3B , C ) . By contrast , 126-mers prepared with CT8 showed strong crosslinks to Sec61α , with peak intensities at positions 35 , 38 , 42 , 45 , and 49 ( Figure 3B , C ) . A helix-destabilizing proline mutation in the center of the TMD ( V41P ) abolished the periodic crosslinking pattern without significantly affecting RNC targeting to Sec61 ( Figure 3D , Figure 3—figure supplement 1D ) . An identical periodic crosslinking pattern was observed with both shorter ( 96-mers ) and longer ( 146-mers ) nascent chains ( Figure 3D , Figure 3—figure supplement 1B , C ) . The sharp periodicity of the crosslinks ( i , i+3 , i+7 , etc ) , along with the consistent pattern observed across multiple nascent chain lengths , indicates that the TMD helix is not oriented randomly relative to Sec61α . Instead , the TMD appears to have a defined orientation , docked to Sec61α in a manner that is presumably stabilized by CT8 . In the absence of CT8 , a periodic crosslinking pattern across the TMD was not observed ( Figure 2B ) , despite the strong crosslinks to Cys49 in the context of 96-mers ( Figure 3—figure supplement 1B ) . Thus , the pre-integrated TMD/Sec61α complex likely samples multiple conformations in the absence of CT8 . 10 . 7554/eLife . 01483 . 007Figure 3 . CT8 traps the TMD in a helical conformation with a defined orientation . ( A ) Positions of single cysteine mutations ( blue dashes ) analyzed by BMH crosslinking . ( B ) BMH crosslinking reactions of 126-mers containing a single cysteine at the indicated positions . ( C ) Crosslinking intensities ( NC × Sec61α ) were quantified by phosphorimaging of the gels shown in ( B ) and plotted as a function of cysteine position . Inset , TMD α-helix model ( red ) highlighting positions with strongest crosslinks to Sec61α ( yellow ) . ( D ) Overlay of BMH crosslinking profiles for the indicated constructs . Crosslinking intensities were normalized to an internal standard ( Cys49 126-mer ) included in each experiment . For raw phosphorimaging data , see Figure 3—figure supplement 1 . ( E ) Cartoon depicting approximate TMD disposition ( red ) of 126-mer assembled in the presence and absence of CT8 . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 00710 . 7554/eLife . 01483 . 008Figure 3—figure supplement 1 . CT8 traps the TMD in a helical conformation with a defined orientation . ( A–D ) BMH crosslinking profiles of RNCs of various lengths that carry single cysteines at the indicated positions of the TMD . The graphs on the right plot the intensity of the TMD/Sec61α crosslink ( ×Sec61α ) against the position of the cysteine along the TMD . For part ( D ) , the V41P position was not changed to cysteine . An internal standard construct ( Cys49 126-mer ) was included in each experiment ( last pair of lanes in each gel series ) to enable comparison of crosslinking efficiencies between different experiments . The values on the y-axes for plots in ( A–D ) are therefore comparable . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 008 Crosslinking experiments with the cysteine positioned either before or after the TMD helped define the overall topology of the nascent chain . Cysteines at positions 10 or 16 formed strong crosslinks to numerous proteins , including Sec61β ( Figure 3B ) , in a pattern that was nearly identical for RNCs assembled with or without CT8 . Thus , the N-terminal tail is highly accessible and localized to the cytosolic side of the membrane under both conditions . The polypeptide segment C-terminal to the TMD of integrated 126-mers is predicted to pass through the Sec61 channel and into the ribosomal exit tunnel . In the absence of CT8 , cysteines engineered along this segment in 126-mer RNCs showed strong crosslinks to Sec61α without concomitant crosslinks to Sec61β ( Figure 3B ) . By contrast , the presence of CT8 caused these same cysteines to crosslink strongly to Sec61β ( Figure 3B ) , again indicating that this region of the CT8-arrested nascent chain is exposed to the cytosol . We note that the N-terminal tail of CT8-arrested , pre-integrated TNFα is highly accessible to the cytosolic side of the membrane , consistent with the early establishment of a type II orientation . It is therefore unlikely that TNFα initially exposes its N-terminus to the ER lumen , as shown in a recent study of a different type II membrane protein ( Devaraneni et al . , 2011 ) . This difference may be due to the longer , charged N-terminal tail of TNFα relative to the aquaporin-4 construct used in the previous study . The protease protection , NEM accessibility , and BMH crosslinking data together reveal a distinct configuration of late-stage RNCs ( >96-mer ) assembled in the absence vs presence of CT8 ( Figure 3E ) . In the absence of CT8 , 126-mer RNCs are in the expected type II topology with a cytosolically disposed N-terminus and membrane-spanning TMD , followed by a C-terminal domain threading through the Sec61 channel to the ribosome . By contrast , the CT8-arrested state is distinguished by a TMD helix that has not integrated , but is instead accessible to the aqueous environment and docked to Sec61α in a specific orientation . This pre-integrated configuration is stabilized by CT8 from the 96-mer until at least the 146-mer . The degree to which the cotransin-stabilized Sec61 conformation resembles integration intermediates that form in the absence of cotransin is not clear . Cotransin may stabilize one of a continuum of conformational states adopted by Sec61 during the normal process of cotranslational integration ( e . g . , TMD docked to a lateral gate that is partially open toward the cytosol , but closed near the lumenal plug ) . To identify the docking site on Sec61α , we first sought to determine which cysteine ( s ) in Sec61α formed BMH-induced crosslinks to the pre-integrated TMD . Unfortunately , our efforts to produce recombinant mammalian Sec61αβγ complexes on which to perform mutagenesis were unsuccessful . In considering the available structural and functional data , we reasoned that the minimal unit for TMD docking is likely to be the Sec61α/γ sub-complex . This is because the β subunit is poorly conserved across species and does not contribute substantially to the structural core of the Sec61 complex . Moreover , a stable sub-complex of mammalian Sec61α and Sec61γ was found to function in RNC targeting and translocation assays , albeit only under certain conditions ( Kalies et al . , 1998 ) . We therefore investigated the Sec61α/γ sub-complex as an alternative system , using the criteria of stable co-association , cotransin binding , and RNC targeting as three independent indicators of correct folding and partial functionality . Expression of FLAG-tagged Sec61α and untagged Sec61γ subunits from a single baculovirus in Sf21 insect cells produced proteins that co-purified as a stable complex with ∼1:1 stoichiometry ( Figure 4—figure supplement 1A ) . Importantly , the photo-affinity probe CT7 labeled FLAG-Sec61α in a manner that was competed by CT9 ( Figure 4A , B , Figure 4—figure supplement 1B ) , a potent analog similar to CT8 ( Maifeld et al . , 2011 ) . Moreover , when TNFα 126-mers were translated in the presence of Sf21 microsomes containing recombinant Sec61α/γ , CT8 promoted the formation of BMH crosslinks between Cys49 and recombinant Sec61α ( Figure 4C , Figure 4—figure supplement 1C ) . In each of these experiments , recombinant FLAG-Sec61α behaved similarly to native Sec61α in canine pancreatic microsomes , suggesting that the recombinant complex is correctly folded and sufficiently functional to analyze its interaction with the TNFα 126-mer . 10 . 7554/eLife . 01483 . 009Figure 4 . The pre-integrated TMD docks to the cytosolic tip of the lateral gate . ( A ) Cotransin binding activity of recombinant FLAG-Sec61α . Microsomes from Sf21 cells expressing FLAG-Sec61α and Sec61γ were incubated with CT7 ( 50 nM ) in the presence or absence of excess CT9 ( 10 μM ) . Samples were photolyzed ( hν ) , and analyzed as in Figure 1C ( left panel ) . Samples were immunoprecipitated under denaturing conditions with anti-FLAG or anti-HA ( control ) beads and analyzed as previously ( right panel ) . ( B ) Photo-affinity labeling with CT7 as in ( A ) with microsomes from FLAG-Sec61α/γ mutant or control baculovirus-infected Sf21 cells . ( C ) 35S-labeled TNFα 126-mers ( Cys49 ) were translated in the presence of FLAG-Sec61α/γ microsomes and subjected to BMH crosslinking . After immunoprecipitation with anti-FLAG beads , eluates were analyzed by autoradiography and immunoblotting . ( D ) As in ( C ) , except that crosslinking reactions were performed with BMOE instead of BMH . See Figure 4—figure supplement 1 for further characterization of recombinant FLAG-Sec61α/γ . ( E ) Homology model of human Sec61α ( Erdmann et al . , 2009 ) ( based on PDB entry:1RH5 ) , highlighting positions of native and engineered cysteines ( yellow = strong , red = weak crosslinks to TNFα TMD Cys49 ) . Lateral gate helices are shown in blue ( TM2b/3 ) and green ( TM7/8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 00910 . 7554/eLife . 01483 . 010Figure 4—figure supplement 1 . The pre-integrated TMD docks to the cytosolic tip of the lateral gate . ( A ) Co-immunoprecipitation of recombinant FLAG-Sec61α and Sec61γ . Sf21 insect cell microsomes were solubilized with 1% Deoxy BigChap ( DBC ) and subjected to anti-FLAG immunoprecipitation ( IP ) . Eluted samples ( Sf21 ) were analyzed by immunoblotting next to a sample of canine microsomal ER membranes ( CRM ) . ( B ) Comparison of the relative CT7 photo-crosslinking efficiency of canine rough microsomes ( CRM ) and Sf21 microsomes containing wild-type FLAG-Sec61α/γ complex . CT7 crosslinking was performed as before . The relative crosslinking efficiency of recombinant Sec61α was normalized to the native protein . ( C ) BMH crosslinking reactions using recombinant FLAG-Sec61α/γ complex . Cys49 TNFα 126-mers were prepared in the presence of Sf21 microsomes , and BMH crosslinking reactions were conducted as described previously . Reactions were either analyzed directly ( Total ) or after denaturing IP with anti-FLAG affinity resin . IP with the anti-HA affinity resin served as a specificity control . ( D ) Western blots of Sf21 microsomes co-expressing either wild-type or mutant FLAG-Sec61α/γ . Insect cell microsomes from cells infected with a baculovirus that lacked Sec61 genes ( control virus ) served as a control . ( E ) CT7 crosslinking of mutant Sec61α/γ complexes containing the indicated cysteine substitutions . The region of the gel corresponding to FLAG-Sec61α is shown . Detection of photo-affinity labeling that is competed by CT9 indicates that the mutant Sec61α/γ complex is properly folded . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 010 To determine which of the eight native cysteines in Sec61α crosslinked to the TMD , each was mutated to alanine or serine . All Sec61α mutants co-expressed with Sec61γ at levels similar to the wild type and were properly folded , as assessed by specific photo-affinity labeling with CT7 ( Figure 4B , Figure 4—figure supplement 1D ) . A single mutation in Sec61α ( C13A ) abolished BMH crosslinking to the TMD , whereas each of the other cysteine mutants crosslinked efficiently in the presence of CT8 ( Figure 4C ) . Based on a homology model of the human Sec61 complex derived from the crystal structure of an archaeal ortholog ( Van den Berg et al . , 2004; Erdmann et al . , 2009 ) , Cys13 is predicted to be in a short α-helix that lies parallel to the cytosolic face of the membrane , just outside the cytosolic vestibule and adjacent to TM3 ( Figure 4E ) . To refine the location of the pre-integrated TMD , non-conserved residues throughout the cytosolic vestibule of Sec61α were individually mutated to cysteine in the background of the C13A mutation ( a mutant Sec61α construct lacking all 8 native cysteines was non-functional in the CT7 photo-crosslinking assay ) . Like the single mutants , all of the double mutants co-expressed with Sec61γ at similar levels and specifically bound CT7 ( Figure 4—figure supplement 1E ) . To enforce a more stringent distance constraint in the cysteine crosslinking experiments , we used bis-maleimidoethane ( BMOE ) , the shortest available bis-maleimide crosslinker ( ∼8 Å ) . Sec61α cysteine-substitution mutants were then analyzed for crosslinking to Cys49 in TNFα 126-mers . Unlike the longer BMH crosslinker ( ∼13 Å ) , BMOE failed to promote crosslinking of the TMD to Cys13 of wild-type Sec61α ( Figure 4D ) , suggesting that BMOE is a more stringent reporter of Cys–Cys proximity . Similarly , crosslinks to most of the engineered cysteines within the cytosolic vestibule of Sec61α were weak or undetectable ( Figure 4D ) . Notable exceptions were cysteine substitutions at Gly88 , Met91 , Ala95 , and Ser383 , each of which showed strong CT8-enhanced crosslinks . Mapping the location of these engineered cysteines onto a model of human Sec61α revealed a clear hotspot where the cytosolic face of TM2b meets the cytosolic tip of TM8 ( Figure 4E ) . This region of Sec61α represents the cytosol-exposed tip of the lateral gate , shown here to lie within ∼8 Å of the TMD ( Cys49 ) in the CT8-stabilized , pre-integrated configuration . Previous studies have shown that the cotransin sensitivity of a given secretory protein can be altered by mutations in the signal sequence ( Harant et al . , 2006 , 2007 ) . However , a mechanistic explanation for these effects has remained elusive . Given the intimate association between the TMD and the cytosolic face of TM2b and TM8 , we reasoned that certain TMD mutants should be able to overcome the CT8-imposed barrier to integration via productive interactions with the lateral gate helices and/or membrane lipids . Comparison of sensitive and resistant TMDs at multiple points along the integration pathway ( i . e . , at multiple nascent chain lengths ) should allow us to: ( 1 ) test whether they pass through a common intermediate and ( 2 ) define the point at which their fates diverge . Finally , analysis of diverse TMD mutants should provide insight into the biophysical basis of CT8 sensitivity and resistance . We identified TNFα TMD mutants with a broad range of CT8 sensitivity ( see below ) . We first present a detailed analysis of the double mutant , T45L/T46L . Inhibiting cotranslational integration of this mutant required 13-fold higher concentrations of CT8 as compared to wild-type TNFα ( IC50 of 1040 nM and 80 nM , respectively ) . BMH crosslinking of successively longer T45L/T46L RNCs assembled without CT8 showed a pattern that was similar to wild-type TNFα: crosslinks to Sec61α were observed over a narrow length range , 86- to 96-mers , beyond which the crosslinks abruptly disappeared ( Figure 5A ) . As with the wild type , low nanomolar concentrations of CT8 led to increased crosslinking between the T45L/T46L mutant and Sec61α , but this was true only at short nascent chain lengths ( Figure 5A , B ) . At longer chain lengths , crosslinking of the mutant TMD to Sec61α was not observed , consistent with its successful integration into the membrane beyond the 96-mer stage ( Figure 5A , C ) . A cysteine scan across the TMD of wild-type and mutant 96-mers revealed identical crosslinking patterns ( Figure 5D , Figure 5—figure supplement 1 ) , suggesting that the T45L/T46L TMD docks to Sec61α in a specific conformation and orientation similar to the wild type . Unlike the wild type , the mutant TMD escapes from the CT8-mediated arrest point and integrates into the membrane as the nascent polypeptide continues to elongate . 10 . 7554/eLife . 01483 . 011Figure 5 . Resistant and sensitive TMDs pass through a common pre-integrated intermediate . ( A ) BMH crosslinking reactions with T45L/T46L TNFα RNCs of varying length , assembled in the presence or absence of CT8 . ( B ) BMH crosslinking reactions of wild-type or T45L/T46L 96-mers in the presence of increasing concentrations of CT8 . Crosslink intensities ( × Sec61α ) were quantified by phosphorimaging and plotted as a function of [CT8] . ( C ) As in ( B ) , except 126-mers were used . ( D ) T45L/T46L 96-mers containing single cysteines at the indicated positions were assembled in the presence of CT8 ( 1 μM ) and subjected to BMH crosslinking . Crosslink intensities ( × Sec61α ) were quantified by phosphorimaging ( see Figure 5—figure supplement 1 for raw data ) and plotted as a function of cysteine position . Data from ‘wild-type’ 96-mer cysteine scan ( Figure 3D ) are shown for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 01110 . 7554/eLife . 01483 . 012Figure 5—figure supplement 1 . Resistant and sensitive TMDs pass through a common pre-integrated intermediate . ( A ) BMH crosslinking reactions of T45L/T46L 96-mers containing single cysteines at the indicated positions . ( B ) Quantified data from the gels shown in part ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 012 Given that CT8 arrests wild-type and T45L/T46L 96-mers in the same pre-integrated conformation , the question arises as to how the mutant TMD is able to escape and pass through the lateral gate . One potential explanation is the increased overall hydrophobicity of the T45L/T46L mutant , which is predicted to have a greater thermodynamic driving force for membrane integration ( Hessa et al . , 2007; Ojemalm et al . , 2011 ) . While mutating multiple polar residues to leucine resulted in increased CT8 resistance ( Figure 6B ) , further mutational analysis revealed that overall hydrophobicity is not the sole determinant . Hydrophobic to lysine mutations in the N-terminal region of the TMD , for example , led to increased CT8 resistance , whereas such mutations in the central or C-terminal regions had the opposite effect ( Figure 6C ) . This may be mechanistically related to the observation that positively charged flanking residues influence TMD orientation ( ‘positive-inside rule’ ) and facilitate membrane integration ( Hessa et al . , 2007; Lerch-Bader et al . , 2008 ) , potentially due to interactions with negative charges on or near the cytoplasmic face of Sec61 ( Goder et al . , 2004; Frauenfeld et al . , 2011 ) . Proline-scanning mutagenesis across the TMD resulted in similar position-dependent effects , conferring increased CT8 resistance when introduced into the N-terminal region and increased sensitivity in the central region ( Figure 6D ) . 10 . 7554/eLife . 01483 . 013Figure 6 . TMD sequence and biophysical properties determine CT8 sensitivity . ( A ) Amino acid sequence of the TNFα TMD . ( B ) Protease protection assays ( proteinase K , PK ) of full-length TNFα polar-to-leucine mutants as a function of increasing CT8 concentration . Translation reactions were carried out in the presence of increasing concentrations of CT8 or DMSO control , followed by digestion with PK as previously described ( Sharma et al . , 2010 ) . Full-length TNFα and the protease-resistant fragment are indicated by open and closed triangles , respectively . ( C ) As in ( B ) , except with hydrophobic-to-lysine mutants . ( D ) As in ( B ) , except with proline-scanning mutagenesis . IC50 values were plotted as a function of the position of the corresponding proline mutation . To compute IC50 values , protease-resistant fragments were quantified by phosphorimaging , normalized to the DMSO control , and fitted to a three-parameter equation using GraphPad Prism software . Representative curve fits in each set ( B–D ) are shown . Based on the online TMD prediction algorithm ( http://dgpred . cbr . su . se/ ) , no mutation was expected to change the N-terminal or C-terminal boundaries of the TMD . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 01310 . 7554/eLife . 01483 . 014Figure 6—figure supplement 1 . Mutations in the plug/lateral gate interface confer resistance to cotransin . ( A ) HCT-116 parental cells and resistant clones were treated with increasing concentrations of CT9 for 72 hr , and viability was assessed by the Alamar Blue assay ( mean ± S . D . , n = 4 ) . ( B ) Sensitivity of wild-type and resistant HCT-116 cells to CT8 and CT9 and the identified heterozygous S61α mutations . ( C ) Untagged Sec61α constructs are expressed in a tetracycline-dependent manner similarly as endogenous Sec61α . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 014 Collectively , the mutagenesis data suggest that helical propensity , hydrophobicity , and charge distribution can influence TMD movement through the lateral gate in opposition to CT8 binding . This apparent competitive relationship is indicated by shifts in the CT8 dose-response curves as a result of TMD mutations ( Figure 6 ) . Rather than attribute these differences solely to thermodynamic partitioning of the TMD into the lipid phase ( Hessa et al . , 2005 , 2007 ) , we propose that escape from the CT8 arrest point involves dynamic interactions between the TMD and the hydrophobic helices of the lateral gate ( TM2/3 and TM7/8 ) , in addition to interactions with membrane lipids . We speculate that these combined interactions , coupled with nascent chain elongation , ultimately cause the lateral gate to open , a conformational change that is opposed by cotransin binding to Sec61 . To gain further molecular insight into the mechanism of CT8-mediated arrest , we sought to determine its binding site on Sec61α . Repeated attempts to identify the CT7 photo-crosslinking site by mass spectrometry were unsuccessful , likely due to the challenge of analyzing heterogeneous , hydrophobic peptides derived from a CT7/Sec61α photo-conjugate . Given that CT8 and its more potent analog CT9 are cytotoxic to certain cancer cell lines ( Figure 7A and our unpublished results ) , we reasoned that it should be possible to identify resistance-conferring alleles of Sec61α by genetic selection . With few exceptions , dominant resistance mutations that reduce a drug’s binding affinity localize to the drug’s binding site or its immediate vicinity; confidence in this interpretation is especially high when multiple resistance mutations localize to the same site ( Campbell et al . , 2001; Shah et al . , 2002; Wacker et al . , 2012 ) , although remote allosteric effects on drug binding are also possible . 10 . 7554/eLife . 01483 . 015Figure 7 . Mutations in the lumenal plug of Sec61α confer resistance to cotransins . ( A ) Parental HCT-116 cells and resistant clones were treated with increasing concentrations of CT8 for 72 hr , and viability was assessed by the Alamar Blue assay ( mean ± S . D . , n = 4 ) . ( B ) HEK293-FRT cells stably expressing wild-type or mutant Sec61α were transfected with a plasmid encoding TNFα and treated with increasing concentrations of CT8 . After 24 hr , TNFα expression was analyzed by immunoblotting . ( C ) CT7 photo-crosslinking to recombinant wild-type and mutant FLAG-Sec61α . Microsomes were incubated with 250 nM CT7 in the presence or absence of excess CT9 ( 10 µM ) , photolyzed , and analyzed by in-gel fluorescence and immunoblotting as in Figure 1C . ( D ) Homology model of human Sec61α showing the location of cotransin resistance mutations ( red ) and the TMD docking site ( yellow ) . Lateral gate helices colored as in Figure 4E . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 015 We exposed the DNA repair-defective tumor cell line HCT-116 ( Wacker et al . , 2012 ) to cytotoxic concentrations of CT9 , our most potent cotransin variant ( see ‘Materials and methods’ for details ) . After 9–12 days , during which time most cells died , several colonies were isolated and found to display varying degrees of resistance to both CT8 and CT9 ( Figure 7A , Figure 6—figure supplement 1A , respectively ) . After amplifying the S61A1 coding sequence from total RNA , Sanger sequencing revealed that 11 of 11 resistant cell lines had one of five single-nucleotide transitions ( all heterozygous ) at four amino acid positions ( Figure 7A , Figure 6—figure supplement 1B ) . All five mutations associated with CT8 resistance cluster in the same region of Sec61α ( Figure 7D ) , at the interface between the plug ( R66I , R66G , G80V , S82P ) and the C-terminal end of TM3 ( M136T ) . This interface defines the side of the lateral gate that is closest to the ER lumen . The fact that five independent resistance mutations localize within ∼10 Å of each other to the lumenal plug region argues that this is the cotransin binding site . We characterized two mutations in greater detail , one in the plug ( R66I ) and the other at the lumenal end of TM3 ( M136T ) . To determine whether these mutants support TNFα integration in the presence of CT8 , we generated stable cell lines that express wild-type or mutant Sec61α constructs from a tetracycline-inducible promoter . For these experiments , we used HEK293 cells , whose viability is unaffected by CT8 treatment for at least 72 hr . Upon induction with tetracycline , both wild-type and mutant Sec61α transgenes ( untagged ) were expressed at similar levels as the endogenous protein ( Figure 6—figure supplement 1C ) . Whereas CT8 potently inhibited TNFα expression in cells with the wild-type Sec61α transgene ( IC50 ∼ 50 nM ) , it had little effect in cells carrying either the M136T or R66I mutant ( Figure 7B ) . These results indicate that the Sec61α mutants assemble into functional translocons and that the M136T and R66I mutations are sufficient to confer dominant resistance to CT8 . Because we could not easily distinguish endogenous Sec61α from the mutants expressed in HEK293 cells , we measured cotransin binding to recombinant Sec61α/γ overexpressed in Sf21 insect cells , as described in Figure 4 . CT7 photo-crosslinking assays revealed specific binding to wild-type Sec61α , but greatly reduced and undetectable binding to the M136T and R66I mutants , respectively ( Figure 7C ) . Although the mutations may have subtle effects on Sec61 function , the CT7 photo-crosslinking data argue that reduced cotransin binding causes resistance in the cell proliferation and TNFα expression assays . Structural , mutagenesis , and crosslinking analyses have all converged on the lateral gate as the site where hydrophobic segments exit the central pore of Sec61 and enter the lipid bilayer ( du Plessis et al . , 2009; Egea and Stroud , 2010; Frauenfeld et al . , 2011; Plath et al . , 1998; Trueman et al . , 2011; Tsukazaki et al . , 2008; Zimmer et al . , 2008 ) . However , the mechanism and timing of TMD egress , along with the role of the TMD itself in the integration process , have remained unclear . In this study , we have exploited a small-molecule inhibitor of cotranslational integration ( cotransin , CT8 ) to trap and interrogate a nascent TMD prior to its exit from the cytosolic vestibule . By analyzing recombinant cysteine mutants of Sec61α , we identified a TMD docking site near the cytosolic tip of the lateral gate . This intimate association suggests that the TMD helix may facilitate opening of the lateral gate . Indeed , such a gating transition may underlie the recently described ‘pulling force’ exerted by the translocon on a nascent TMD just before its integration into the membrane ( Ismail et al . , 2012 ) . Figure 8 depicts a model that places our biochemical data in the context of Sec61/SecY structures determined by x-ray crystallography and cryoelectron microscopy . In this model , RNC targeting to Sec61 allows partial opening of the lateral gate toward the cytosol , as observed in a crystal structure of SecYE bound to a Fab fragment ( Tsukazaki et al . , 2008 ) . In the SecYE/Fab structure , separation of TM2b from the cytosolic end of TM8 creates a notch in the lateral gate , which we propose to be the initial docking site for a nascent TMD after its release from SRP ( Figure 8 , middle ) . At the 96-mer stage , docking of the TMD to this site enables BMH crosslinking to Sec61α ( Figure 2B ) . As the nascent chain elongates , interhelical contacts that seal the lateral gate are progressively destabilized . This key transition , which is opposed by CT8 binding ( most likely to the plug ) , leads to complete intercalation of the TMD between helices TM2/3 and TM7/8 of the lateral gate , concomitant with exposure of the TMD to membrane lipids ( Figure 8 , right ) . As indicated by the pronounced rightward shifts in the CT8 dose-response curves ( Figure 6 ) , TMDs with greater hydrophobicity and helical propensity are better able to progress to this state , presumably in kinetic competition with CT8 . 10 . 7554/eLife . 01483 . 016Figure 8 . Model for cotransin-mediated inhibition of TMD integration . Following RNC targeting to Sec61 ( left panel ) , the TMD docks to a notch in the cytosolic tip of the lateral gate created by the separation of TM2b from TM8 ( middle panel ) . This otherwise transient , pre-integrated configuration is stabilized by CT8 binding to the lumenal plug region of Sec61α . Interactions between the TMD and the lateral gate facilitate gate opening and TMD movement to a second site on the lipid-exposed face of the lateral gate ( right panel ) . CT8 blocks TMD movement from the cytosolic docking site to the external , lipid-exposed site . Structural models ( bottom ) were adapted from PDB entries 1RH5 ( left panel , closed lateral gate ) , 2ZJS ( middle panel , pre-integrated configuration ) , and 3J00/3J01 ( right panel , integrated configuration ) . The model for the pre-integrated TMD configuration was approximated by manually placing the TMD ( modeled as an α-helix ) into the cytosolic vestibule between TM2b and TM7 ( PDB entry: 2ZJS ) , consistent with the BMOE crosslinking data . Lateral gate helices TM2b/3 and TM7/8 are blue and green , respectively , while the TMD helix is red . DOI: http://dx . doi . org/10 . 7554/eLife . 01483 . 016 Our discovery of five resistance mutations in Sec61α that localize to the lumenal plug region suggests that CT8 binds proximal to this site . This conclusion is further strengthened by the observation that R66I and M136T mutations abrogate CT7 binding ( Figure 7C ) , although we cannot exclude the possibility that mutations in the plug perturb cotransin binding to a remote site . Given that binding to the lumenal plug arrests the TMD in its cytosolic docking site >20 Å away , cotransins appear to exhibit characteristics of both allosteric and competitive inhibitors . We note that the closed conformation of the channel ( Van den Berg et al . , 2004 ) cannot accommodate a molecule the size of CT8 , whereas a cavity , outlined by the identified mutations and roughly large enough for CT8 , is apparent in structures with a partially open lateral gate ( Egea and Stroud , 2010 ) . We speculate that CT8 binding stabilizes interactions between the plug , TM3 , and TM7 , thereby sealing the lateral gate at its lumenal end in a manner similar to that proposed for the Q129L and N302L mutations in yeast Sec61p ( Trueman et al . , 2012 ) . In this view , CT8 acts as a molecular wedge that impedes TMD integration by blocking displacement of the plug . Alternatively , CT8 binding may itself result in the displacement or structural rearrangement of the plug; here again , CT8 would act as a molecular wedge that prevents TMD movement through the lumenal end of the lateral gate . This model is suggested by a recent electron crystallographic structure of the SecYEG complex in which a signal peptide bound to the lipid-exposed face of the lateral gate displaces the plug 10 Å relative to apo-SecYEG ( Hizlan et al . , 2012 ) . Further structural studies are needed to provide an atomic-resolution view of how cotransins engage Sec61 to arrest TMD integration .
The following commercially available primary antibodies and antibody resins were used: anti-FLAG M2 antibody and affinity matrix ( Sigma , St . Louis , MO ) , anti-HA affinity matrix ( Roche , Basel , Switzerland ) , anti-Sec61γ ( Proteintech Group , Chicago , IL ) , anti-Sec61α ( Novus Biologics , Littleton , CO ) , anti-human TNFα ( R&D Systems MAB2101 , Minneapolis , MN ) . The antibody directed against Sec61β was previously described ( Fons et al . , 2003 ) . Secondary antibodies conjugated to HRP ( Santa Cruz Biotechnology , Dallas , TX ) or infrared-dyes ( LI-COR Biosciences , Lincoln , NE ) were purchased . Preparations of rabbit reticulocyte lysate ( Sharma et al . , 2010 ) and canine rough microsomes ( Walter and Blobel , 1983 ) have been previously described . CT7 and CT8 were prepared as previously described ( MacKinnon et al . , 2007 ) . CT9 was prepared as previously described ( Maifeld et al . , 2011 ) . The sources of other reagents are noted in the text and were used without further purification . Site-directed mutagenesis was performed using the QuikChange method ( Stratagene , La Jolla , CA ) and all constructs were verified by DNA sequencing . The plasmid encoding human TNFα was previously described ( Maifeld et al . , 2011 ) . DNA templates encoding truncated TNFα constructs were prepared by PCR using a forward primer that contained a T7 promoter ( bold ) and a Kozak consensus sequence ( underlined ) followed by a region complementary to the 5′-end of TNFα ( 5′-GCCTAATACGACTCACTATAGGGAGACCATGAGCACTGAAAGCATGATCCGG-3′ ) . The reverse primer annealed to various regions of the TNFα coding region and introduced three additional methionines at the C-terminus to improve detection of translated products by autoradiography . The indicated length of the nascent chain includes the three terminal methionines . A similar PCR-based strategy was used for production of full-length TNFα DNA templates except the reverse primer included a stop codon . DNA templates encoding N-terminally tagged TNFα constructs were prepared by PCR using forward primers encoding either the 3x-FLAG or 3x-HA epitopes immediately upstream of the TNFα translational start site . Construction of the composite MultiBac baculovirus expression vector followed the previously described strategic guidelines ( Fitzgerald et al . , 2006 ) . Briefly , human Sec61α and Sec61γ genes were amplified by PCR using templates kindly provided by Professor Tom Rapoport ( Harvard Medical School ) , and a 3x-FLAG tag was introduced into the N-terminus of Sec61α . The Sec61α subunit was then cloned into the MCS-1 of vector pUCDM by In-Fusion cloning according to manufacturer’s instructions ( Clontech , Mountain View , CA ) . The Sec61γ subunit was cloned into SacI/HindIII sites ( MCS-2 ) of vector pKL . The two vectors were then fused by Cre-recombination yielding a ‘master plasmid’ , which was used to transform DH10α MultiBac cells for preparation of the MultiBac expression vector . Site-directed mutagenesis to create Sec61α mutants was performed on the master plasmid . All constructs were verified by DNA sequencing . Unless otherwise noted , SDS-PAGE was performed with 12% Tris/Tricine polyacrylamide gels . Prior to autoradiography , gels were stained with Coomassie brilliant blue to confirm equal protein loading and dried under vacuum . For quantitative autoradiography , dried gels were exposed to a storage phosphorous screen ( GE Healthcare , San Francisco , CA ) , imaged on a Typhoon 9400 phosphorimager ( Amersham ) , and the images were quantified using ImageJ software ( NIH ) . Dose-response data were normalized to the DMSO control and fitted with a three-parameter equation using GraphPad Prism ( GraphPad Software ) . For qualitative autoradiography , dried gels were exposed to Biomax MR film ( Kodak , Rochester , NY ) . For western blotting , proteins were transferred to nitrocellulose membranes . Following blocking of the membranes with 5% milk in TBST ( 50 mM Tris , 150 mM NaCl , 0 . 05% TX-100 , pH 7 . 6 ) , membranes were incubated with the appropriate primary antibodies at the following dilutions: 1:10 , 0000 ( anti-Sec61α ) , 1:20 , 000 ( anti-Sec61β ) , 1:500 ( anti-Sec61γ ) , 1:20 , 000 ( anti-FLAG ) . Blotting for Se61γ typically required overnight incubation at 4°C with the primary antibody . Following incubation with primary antibodies , membranes were washed and incubated with the appropriate HRP-conjugated secondary antibodies followed by chemiluminescent detection ( GE Healthcare ) . Alternatively , blots were incubated with the appropriate infrared dye-conjugated secondary antibodies followed by imaging on an Odyssey infrared fluorescent scanner ( LI-COR Biosciences ) . Cell-free transcription , translation , translocation , and Protease K ( PK ) protection assays were performed as previously described ( Sharma et al . , 2010 ) . Briefly , DNA templates encoding full-length or truncated TNFα constructs were transcribed with T7 Polymerase ( New England Biolabs , Ipswich , MA ) for 1 hr at 37°C and used immediately in the subsequent translation/translocation reactions . Translocation reactions were assembled at 0°C in the presence of CTs ( added from a 100 × stock in DMSO ) or an equivalent volume of DMSO . Unless otherwise indicated , reactions included 35S-Methionine ( Perkin Elmer , Waltham , MA , 2 μCi per 10 μl translation ) , and either canine pancreatic microsomes ( CRM , 1 ‘equivalent’ per 10 μl translation , as defined in Walter and Blobel , 1983 ) , or microsomal membranes derived from Sf21 insect cells that contained an equivalent amount of recombinant Sec61 complex . Translation was initiated by transferring the reactions to 32°C for 30 min and stopped by removing the reactions to ice . To isolate ER microsome-targeted RNCs , translation reactions were diluted with an equal volume of ice-cold high salt buffer ( 1M KOAc , 10 mM Mg ( OAc ) 2 , 50 mM Hepes , pH 7 . 8 ) and sedimented through a high salt sucrose cushion ( 0 . 5 M KOAc , 0 . 5 M sucrose , 5 mM Mg ( OAc ) 2 , 50 mM Hepes , pH 7 . 8 ) at 4°C in a TLA100 rotor ( Beckman ) for 10 min at 50 , 000 rpm . The membrane pellet was re-suspended to the original volume in membrane buffer ( 100 mM KOAc , 250 mM sucrose , 2 mM Mg ( OAc ) 2 , 0 . 1 mM TCEP , 50 mM Hepes , pH 7 . 8 ) . For analysis of tRNA-associated nascent chains following protease digestion ( as in Figure 2—figure supplement 1 ) , reactions were quenched with PMSF as previously described ( Sharma et al . , 2010 ) and then rapidly diluted into 10 volumes of boiling 1% SDS , 0 . 1 M Tris , pH 6 . 8 . Samples were then resolved on 4–12% NuPAGE Bis/Tris gels ( Invitrogen , Carlsbad , CA ) , which run at neutral pH and therefore help preserve the alkaline-sensitive peptidyl-tRNA bond . ER microsome-targeted RNCs in membrane buffer were isolated as described above and treated with 50 μM bis-maleimidohexane ( BMH , Pierce , Rockford , IL ) or bis-maleimidoethane ( BMOE , Pierce ) for 30 min at 0°C . To analyze total crosslinked products , reactions were quenched by addition of an equal volume of 2 × Laemmli sample buffer ( which contains a large molar excess of DTT over BMH or BMOE ) , heated for 1 min at 95°C , and analyzed by SDS-PAGE and autoradiography . To immunoprecipitate ( IP ) proteins from the reactions , the samples were quenched with 1 mM DTT for 10 min at 0°C , denatured with 1% SDS at 95°C for 1 min , and diluted 10-fold with IP buffer ( 1% Triton-X 100 , 100 mM NaCl , 50 mM Hepes , pH 7 . 8 ) . Anti-FLAG affinity resin or protein A sepharose ( GE Healthcare ) along with the appropriate primary antibodies were then added and the samples were rotated overnight ( ∼12 to 16 hr ) at 4°C . The resin was washed four times with IP buffer supplemented with 0 . 1% SDS , and bound proteins were eluted with 1 × Laemmli sample buffer at 95°C for 1 min . Eluted material was analyzed by SDS-PAGE and autoradiography . ER microsome-targeted RNCs in membrane buffer were divided into two equal portions ( 10 μl each ) and incubated with either 200 μM N-ethylmaleimide ( NEM , Sigma Aldrich ) or an equal volume of DMSO for 1 hr at 0°C . NEM was then quenched with 100 μl of quench buffer ( 5 mM DTT , 100 mM NaCl , 2 mM Mg ( OAc ) 2 , 50 mM Hepes , pH 7 . 8 ) for 10 min at 0°C , and the membranes were isolated by centrifugation at 4°C in a TLA100 rotor ( Beckman ) for 10 min at 70 , 000 rpm . The membrane pellet was then solubilized in a detergent-containing buffer ( 1% SDS , 0 . 25 mM TCEP , 50 mM Hepes , pH 7 . 8 ) and treated with an equal volume ( 10 μl ) of 16 mM 5 kDa PEG-maleimide ( Nektar , San Francisco , CA ) , prepared in 50 mM Hepes ( pH 7 . 8 ) . Reactions were incubated at 32°C for 1 hr and quenched with 20 mM DTT for 20 min at 32°C . To completely hydrolyze the nascent chain-tRNA bond prior to SDS-PAGE ( which simplified the appearance of autoradiograms ) , samples were treated with 100 μl of 200 mM Na2CO3 ( pH 12 ) for 30 min at room temperature , then diluted with 1 ml of IP buffer ( 1% TX-100 , 100 mM NaCl , 50 mM Hepes , pH 7 . 8 ) , and precipitated with 10% trichloroacetic acid ( TCA ) . Precipitated proteins were washed twice with ice-cold acetone , dissolved in 1 × Laemmli sample buffer and resolved by SDS-PAGE . The quantities of unmodified nascent chains ( NC ) and PEG-modified nascent chains ( NC × PEG ) were determined by phosphorimaging . The fraction of PEG-modified chains was defined as NC × PEG/ ( NC × PEG + unmodified NC ) . These values were normalized to control reactions that contained no NEM . NEM accessibility was defined as 1 − normalized fraction of PEG-modified chains . Sf21 microsomes containing 50 nM Sec61 were treated with either 10 μM CT9 ( Maifeld et al . , 2011 ) or DMSO for 30 min at 0°C , followed by incubation with 50 nM CT7 for an additional 30 min at 0°C . Samples were then photolyzed and crosslinked proteins were detected by click chemistry , SDS-PAGE , and in-gel fluorescent scanning as previously described ( MacKinnon et al . , 2007; MacKinnon and Taunton , 2009 ) . Sf21 insect cells were grown and maintained in SF-900 II serum-free media ( Gibco ) at 27°C following standard protocols ( Fitzgerald et al . , 2006 ) . The cells were transfected with the MultiBac vector using Fugene HD transfection reagent according to the manufacturer’s instructions ( Roche ) , and virus was propagated following published methods ( Fitzgerald et al . , 2006 ) . For expression of Sec61α/γ complexes , a 100 ml culture of cells at 0 . 5 × 106 cells/ml was infected with 2 ml of first generation virus ( V1 ) , which resulted in immediate arrest of cell growth . The cells were harvested 48 hr after growth-arrest by sedimentation at 800×g for 5 min . The cell were swollen in 20 ml hypotonic buffer ( 20 mM Hepes , pH 7 . 8 ) at 0°C for 20 min and broken with a microfluidizer ( Emulsiflex-C5 ) at 15 , 000 psi for 10 min . The lysate was immediately adjusted to 100 mM KOAc , 5 mM Mg ( OAc ) 2 , 1 mM EDTA , and 1 × EDTA-free protease inhibitor cocktail ( Roche ) and then clarified by centrifugation at 1000×g for 10 min . To isolate microsomal membranes , the clarified lysate was centrifuged at 45 , 000 rpm in a type 70 Ti rotor ( Beckman ) for 1 hr at 4°C and the resulting microsomal pellet was re-suspended with a glass dounce in 400 μl of buffer ( 50 mM Hepes , 250 mM sucrose , 1 mM CaCl2 , pH 7 . 8 ) . To remove endogenous RNA that co-purified with the membranes , the microsomes were treated with micrococcal nuclease ( New England Biolabs , 150 units/ml final concentration ) at 25°C for 10 min . The nuclease activity was quenched by adjusting the extract to 2 mM EGTA . Microsomes containing various Sec61 mutants were normalized for total protein by BCA protein assay ( Pierce ) and equivalent amounts of total protein resolved by SDS-PAGE next to a serial dilution of canine rough microsomes ( CRM ) containing a known concentration of the Sec61 complex . Proteins were then transferred to nitrocellulose and western blotted for the FLAG epitope , Sec61α , and Sec61γ . Following this standardized protocol , the expression level between different Sec61α mutants was found to be very similar and the concentration of recombinant FLAG-Sec61 in the final microsomal preparation was similar to the concentration of Sec61 in CRM . Translation reactions ( 100 μl ) were programed with mRNA templates encoding an N-terminal 3x-FLAG tag or 3x-HA tag ( control ) followed by the first 126 residues of TNFα . Reactions were assembled in the presence or absence of CT7 ( 1 μM ) . Following translation , ER microsome-targeted RNCs were isolated as described above , except the membrane pellet was brought up in two volumes of membrane buffer supplemented with 1% Deoxy BigChap ( DBC , Anatrace , Santa Clara , CA ) . The membranes were solubilized for 10 min at 0°C and insoluble material removed by centrifugation at 50 , 000 rpm in a TLA100 rotor at 4°C for 10 min . The supernatant was then incubated with anti-FLAG affinity resin and mixed with rotation at 4°C for 2 . 5 hr . The resin was then sedimented ( 600 × g , 3 min , 4°C ) and washed four times with 1 ml of membrane buffer containing 0 . 3% DBC . After the final wash , bound proteins were eluted with 250 μg/ml of 3x-FLAG peptide ( Sigma ) in membrane buffer containing 0 . 3% DBC . The eluted material was analyzed directly by SDS-PAGE and semi-quantitative western blotting , or was first photolyzed and then subjected to click chemistry and analyzed by SDS-PAGE and in-gel fluorescent scanning . Typically , ∼0 . 3 pmol of purified RNC-Sec61 complexes were obtained from a 100 μl translation reaction . HCT-116 cells and clonal lines were cultured in McCoy’s 5A medium ( Invitrogen ) at 37°C with 5% CO2 . HEK293-FRT cells were grown in Dulbecco’s Modified Eagle’s Medium ( Invitrogen ) and grown at 37°C with 10% CO2 . All cultures were supplemented with 10% FBS and penicillin-streptomycin . Cell viability assays were performed by plating 2500 HCT-116 cells in flat-bottomed 96-well plates and treating with CT8 or CT9 the following day . After 72 hr , cell viability was analyzed using the Alamar Blue assay ( Invitrogen ) according to the manufacturer’s instructions . HEK293 cells stably expressing wild-type or mutant Sec61α were generated using Flp-In 293 T-Rex cells ( Invitrogen ) as previously described ( Henise and Taunton , 2011 ) . For transient transfection experiments , 1 . 7 × 105 HEK293 T-Rex cells were seeded in 12-well dishes and transfected with a TNFα expression plasmid ( Maifeld et al . , 2011 ) using Lipofectamine 2000 ( Invitrogen ) for 2 hr , after which the culture media was changed to one containing increasing concentrations of CT8 . After 24 hr , cells were harvested and lysed in 0 . 5% Triton X-100 in PBS . The resulting lysates were normalized using the Bradford assay ( Bio-Rad ) and 20 µg of total protein was resolved on 12% Tris-Tricine gels and analyzed by immunoblotting . To derive resistant cell lines , HCT-116 cells were incubated with 28–41 nM CT9 for 9–12 days , after which cell colonies were isolated by ring cloning and cultured in drug-free media . Total RNA was isolated from HCT116 cells using the RNeasy Mini Kit ( Qiagen , Venlo , Netherlands ) according to the manufacturer’s instructions . Total cDNA was synthesized using anchored oligo-dT primers and Superscript III reverse transcriptase ( Invitrogen ) and S61A1 cDNA was amplified with Phusion polymerase ( Thermo Fisher Scientific ) and sequenced bi-directionally by Sanger sequencing . | Cells are surrounded by a plasma membrane that acts like a barrier around the cell—keeping the cell’s boundaries distinct from surrounding cells and helping to regulate the contents of the cell . This plasma membrane is made up mostly of two layers of fatty molecules , and is also studded with proteins . Some of these membrane proteins act as channels that allow nutrients and other chemicals to enter and leave the cell , while others allow the cell to communicate with other cells and the outside environment . Like all proteins , membrane proteins are chains of amino acids that are linked together by a molecular machine called a ribosome . The ribosomes that make membrane proteins are located on the outside of a membrane-enclosed compartment within the cell called the endoplasmic reticulum . To eventually become embedded within a membrane , a new protein must—at the same time as it is being built—enter a channel within the membrane of the endoplasmic reticulum . The newly synthesized protein chain enters this channel , called Sec61 , via an entrance near the ribosome and then threads its way toward the inside of the endoplasmic reticulum . However , there is also a ‘side-gate’ in Sec61 that allows specific segments the new protein to escape the channel and become embedded within the membrane . From here , the membrane protein can be trafficked to other destinations within the cell , including the plasma membrane . However , how the newly forming protein chain passes through the side-gate of Sec61 is not well understood . Now MacKinnon , Paavilainen et al . have used a small molecule called cotransin—which is known to interfere with the passage of proteins through Sec61—to observe the interactions between the Sec61 channel and the new protein . Cotransin appears to trap the new protein chain within the Sec61 channel by essentially ‘locking’ the side-gate . MacKinnon , Paavilainen et al . observed that the trapped protein interacts with the inside of the channel at the end closest to the ribosome—which is the likely location of the side-gate . In contrast , cotransin likely binds at the other end of the channel , to a piece of Sec61 that serves to plug the exit into the endoplasmic reticulum; and this plug is directly connected to the side-gate . By preventing the plug from moving out of the way , cotransin can somehow stop the new protein from passing through the side-gate . However , MacKinnon , Paavilainen et al . did find that some membrane proteins with certain physical and chemical properties could get through the gate , despite the presence of cotransin . The next challenge is to resolve exactly how interactions between cotransin and the Sec61 plug can block the escape of new proteins into the membrane . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2014 | An allosteric Sec61 inhibitor traps nascent transmembrane helices at the lateral gate |
The size and position of mitotic spindles is determined by the lengths of their constituent microtubules . Regulation of microtubule length requires feedback to set the balance between growth and shrinkage . Whereas negative feedback mechanisms for microtubule length control , based on depolymerizing kinesins and severing proteins , have been studied extensively , positive feedback mechanisms are not known . Here , we report that the budding yeast kinesin Kip2 is a microtubule polymerase and catastrophe inhibitor in vitro that uses its processive motor activity as part of a feedback loop to further promote microtubule growth . Positive feedback arises because longer microtubules bind more motors , which walk to the ends where they reinforce growth and inhibit catastrophe . We propose that positive feedback , common in biochemical pathways to switch between signaling states , can also be used in a mechanical signaling pathway to switch between structural states , in this case between short and long polymers .
The budding yeast kinesin Kip2 promotes microtubule growth in vivo . Deletion of this kinesin results in nuclear migration defects , and the phenotype is associated with shorter , less abundant cytoplasmic microtubules ( Cottingham and Hoyt , 1997; Huyett et al . , 1998; Miller et al . , 1998; Caudron et al . , 2008 ) . Conversely , Kip2 overexpression results in hyper-elongated cytoplasmic microtubules ( Carvalho et al . , 2004 ) . The stabilization of microtubules by Kip2 is thought to be indirect and a consequence of Kip2 transporting the growth regulator Bik1 ( Clip170 ) to microtubule plus ends ( Carvalho et al . , 2004; Caudron et al . , 2008 ) . To test whether Kip2 alone can promote microtubule growth , the activity of full-length , purified Kip2 was measured in dynamic microtubule assays using porcine tubulin in the presence of adenosine triphosphate ( ATP ) ( Gell , et al . , 2010; 2011 ) ( Figure 1A ) . Within 10 min , Kip2 ( Figure 1B , C ) , as well as Kip2-enhanced green fluorescent protein ( eGFP ) ( Figure 1—figure supplement 1A ) , strongly increased the length of freshly polymerized microtubules ( p<0 . 0001 , Welch’s unpaired t-test , please refer to Table 1 for porcine microtubule parameter values ) . The effect of Kip2 on microtubule length was almost completely inhibited when ATP was replaced by the non-hydrolyzable ATP analog adenylyl imidodiphosphate ( AMP-PNP ) ( Figure 1C , blue markers , p<0 . 0001 ) , showing that growth promotion requires ATP hydrolysis . To quantify how Kip2 influences microtubule dynamics , we drew kymographs from the time-lapse images of the dynamic microtubule assay ( Figure 1D ) . Kip2 increased the growth rate of microtubules ( the slope of the growing microtubule in the kymograph ) 2 . 9-fold ( Figure 1E ) . In addition , Kip2 reduced the frequency of catastrophe ( the transition between growth and shrinkage phases ) approximately 10-fold ( Figure 1F ) . Kip2 did not affect the shrinkage rate ( Figure 1—figure supplement 1B ) , or the frequency of rescue ( the transition between shrinkage and growth phases , Figure 1—figure supplement 1C ) . All dynamic data on porcine tubulin are contained in Table 1 . Note that rescue is not expected to have a large effect on microtubule length in our assays . This is because at lower Kip2 concentrations ( < 10 nM ) , the average distance shortened following catastrophe ( the shrinkage rate divided by the rescue frequency ) is greater than the average distance grown before catastrophe ( the growth rate divided by the catastrophe frequency ) , so microtubules usually shrink all the way back to the seed ( as expected by theory , Verde et al . , 1992 ) . On the other hand , at higher Kip2 concentrations , catastrophes are so rare that microtubules are expected to be very long before they catastrophe ( >18 μm for [Kip2] ≥ 10 nM ) . Consistent with the small contribution of rescue , the measured increase in microtubule length accorded with the effects of Kip2 on the growth rate and the catastrophe frequency alone ( Figure 1C , red line ) . The half-maximal stimulation of polymerization and inhibition of catastrophe occurred at ≈7 nM Kip2 . Given that the cellular concentration of Kip2 is ≈25 nM ( Ghaemmaghami et al . , 2003 ) , these results show that Kip2 affects microtubule dynamics in vitro at physiologically relevant concentrations . 10 . 7554/eLife . 10542 . 003Figure 1 . Kip2 is a microtubule polymerase and an anti-catastrophe factor for porcine tubulin . ( A ) Schematic of the experimental design: porcine tubulin ( green ) polymerizes onto stabilized microtubules ( red ) bound to the coverslip with antibodies ( blue ) , imaged using TIRF microscopy . ( B ) Microscopy images of dynamic microtubules grown from stabilized seeds without ( left ) and with 40 nM Kip2 ( right ) at t = 10 min . ( C ) Microtubule length as a function of Kip2 concentration in ATP ( black circles ) or AMP-PNP ( blue circles ) at t = 10 min . The red line indicates the expected microtubule length at t = 10 min , calculated from the measured growth rates and catastrophe frequencies in Table 1 according to the formula L = ( v+/f+- ) [1-exp ( -tf+- ) ] , where v+ is the growth rate and f+- is the catastrophe frequency ( ignoring rescues and assuming that regrowth occurs without delay ) . ( D ) Kymographs showing typical microtubule growth without ( left ) and with 5 nM Kip2 ( right ) in ATP . ( E ) Microtubule growth rate as a function of Kip2 concentration in ATP . ( F ) Catastrophe frequency as a function of Kip2 concentration in ATP . All error bars are standard errors of the mean . Please refer to Table 1 for values . AMP-PNP , adenylyl imidodiphosphate; ATP , adenosine triphosphate; TIRF , total internal reflection fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 00310 . 7554/eLife . 10542 . 004Figure 1—figure supplement 1 . Kip2 has no significant effect on microtubule shrinkage rate or rescue frequency . ( A ) Porcine microtubule length at 0 or 40 nM Kip2-eGFP in ATP , measured at t = 10 min . ( B ) Porcine microtubule shrinkage rate as a function of Kip2 concentration . The gray box indicates the Kip2 concentration regime in which microtubule catastrophe is very rare , and we did not quantify microtubule shrinkage rate . The data were fitted by linear regression , weighted by the SE . Slope = 0 . 09 ± 0 . 11 , y-intercept = 28 . 1 ± 0 . 58 μm/min . ( C ) Porcine microtubule rescue frequency ( shrinkage rate divided by rescue distance ) as a function of Kip2 concentration . The gray box indicates the Kip2 concentration regime in which microtubule catastrophe is very rare and we did not measure shrinkage rate or rescue frequency . The data were fitted by linear regression , weighted by the SE . Slope = 0 . 17 ± 0 . 07 , y-intercept = 0 . 18 ± 0 . 11 μm . The slope was not significantly different from zero ( p>0 . 05 ) . Error bars are SE . ATP , adenosine triphosphate; SE , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 00410 . 7554/eLife . 10542 . 005Figure 1—figure supplement 2 . Kip2 is a microtubule polymerase and an anti-catastrophe factor for yeast tubulin . ( A ) Kymographs from DIC microscopy showing typical microtubule growth with 4-μM unlabeled yeast tubulin without ( left ) and with Kip2 ( 5 nM , center , 20 nM , right ) in ATP . ( B ) Yeast microtubule growth rate as a function of Kip2 concentration in ATP . ( C ) Catastrophe frequency as a function of Kip2 concentration of yeast microtubules in ATP . Error bars are SE . Please refer to Table 2 for values . ATP , adenosine triphosphate; DIC , differential interference contrast; SE , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 00510 . 7554/eLife . 10542 . 006Figure 1—figure supplement 3 . Kip2 increases the growth rate in GTP-tubulin and lowers the off rate of GMPCPP tubulin . ( A ) Porcine microtubule growth rate from GMPCPP seeds as a function of tubulin concentration in ATP without Kip2 ( red ) and with 40 nM Kip2 ( black ) in solution . The data were fit using linear regression , weighted by the SE . The slope corresponds to a second-order association rate for free GTP-tubulin dimers of 0 . 70 ± 0 . 30 μM-1·s-1 without Kip2 and 1 . 48 ± 0 . 16 μM-1·s-1 with 40 nM Kip2 . The y-intercepts were –1 . 9 ± 4 . 0 s-1 without Kip2 and –0 . 21 ± 1 . 6 s-1 with 40 nM Kip2; the intercepts did not differ significantly from zero . ( B ) Shrinkage rates of GMP-CPP microtubules at 0 nM Kip2 ( light gray ) and 40 nM Kip2 ( dark gray ) in ATP and AMP-PNP . Error bars are SEs . AMP-PNP , adenylyl imidodiphosphate; ATP , adenosine triphosphate; GTP , guanosine-5'-triphosphate; SE , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 00610 . 7554/eLife . 10542 . 007Figure 1—figure supplement 4 . SDS–PAGE gels of Kip2 and Kip2-eGFP . Lane 1: Pooled Kip2 fractions after gel filtration ( 78 kDa ) . Lane 2: Pooled Kip2-eGFP fractions after gel filtration ( 105 kDa ) . Molecular weight markers ( Seeblue Plus2 ) are indicated by horizontal lines . eGFP , enhanced green fluorescent protein; SDS–PAGE , sodium dodecyl sulfate polyacrylamide gel electrophoresis . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 00710 . 7554/eLife . 10542 . 008Table 1 . Parameters of microtubule dynamics for 12 μM porcine tubulin ( mean ± SE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 008[Kip2] ( nM ) Length ( μm ) Growth rate ( μm/min ) Catastrophe frequency ( min-1 ) Catastrophe distancea ( μm ) Shrinkage rate ( μm/min ) Rescue frequency ( min-1 ) Rescue distanceb ( μm ) 01 . 0 ± 0 . 1 ( n = 54 ) 0 . 32 ± 0 . 02 ( n = 172 ) 0 . 166 ± 0 . 015 ( n = 126 ) 1 . 9 ± 0 . 2 27 . 6 ± 1 . 0 ( n = 130 ) 0 . 88 ± 0 . 33 ( n = 7 ) 32 ± 12 11 . 0 ± 0 . 2 ( n = 75 ) 0 . 37 ± 0 . 01 ( n = 152 ) 0 . 126 ± 0 . 012 ( n = 104 ) 2 . 9 ± 0 . 3 27 . 7 ± 0 . 9 ( n = 85 ) 0 . 20 ± 0 . 14 ( n = 2 ) 140 ± 100 22 . 1 ± 0 . 2 ( n = 88 ) 0 . 35 ± 0 . 01 ( n = 159 ) 0 . 135 ± 0 . 013 ( n = 115 ) 2 . 6 ± 0 . 3 29 . 7 ± 0 . 9 ( n = 110 ) 0 . 54 ± 0 . 24 ( n = 5 ) 55 ± 25 55 . 3 ± 0 . 2 ( n = 82 ) 0 . 62 ± 0 . 03 ( n = 77 ) 0 . 065 ± 0 . 011 ( n = 33 ) 10 ± 2 28 . 4 ± 1 . 8 ( n= 45 ) 2 . 2 ± 0 . 7 ( n = 9 ) 13 ± 4 106 . 4 ± 0 . 2 ( n = 75 ) 0 . 78 ± 0 . 03 ( n = 38 ) 0 . 043 ± 0 . 011 ( n = 16 ) 18 ± 5 29 ± 4 ( n = 18 ) 1 . 8 ± 0 . 7 ( n = 6 ) 16 ± 6 207 . 7 ± 0 . 3 ( n = 68 ) 0 . 99 ± 0 . 04 ( n = 36 ) 0 . 020 ± 0 . 006 ( n = 10 ) 50 ± 16 -- . -408 . 8 ± 0 . 6 ( n = 26 ) 0 . 94 ± 0 . 05 ( n = 18 ) 0 . 004 ( n = 1 ) 235 ---aThe catastrophe distance is the growth rate divided by the catastrophe frequency . bThe rescue distance is the shrinkage rate divided by the rescue frequency . To exclude potentially confounding effects introduced by using fluorescently labeled porcine brain tubulin , as well as to confirm that Kip2 , which is a yeast protein , has the same activity on its conspecific protein , we repeated the dynamic microtubule assays with unlabeled yeast tubulin ( Widlund et al . , 2012 ) and differential interference contrast ( DIC ) microscopy ( Figure 1—figure supplement 2A , please refer to Table 2 for yeast microtubule parameter values ) . Consistent with our porcine tubulin results , Kip2 increased the yeast microtubule growth rate by 2 . 3-fold and inhibited catastrophe 20-fold ( Figure 1—figure supplement 2B , C ) . The half-maximal stimulation of polymerization and inhibition of catastrophe for yeast tubulin occurred at ≈12 nM Kip2 , similar to the concentration at which Kip2 regulates porcine brain microtubules . In summary , Kip2 is a microtubule polymerase and anti-catastrophe factor in vitro and does not require additional proteins such as Bik1 for these activities . We will defer discussing a potential role for Bik1 until the end of the manuscript . 10 . 7554/eLife . 10542 . 009Table 2 . Parameters of microtubule dynamics for 4 μM yeast tubulin ( mean ± SE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 009[Kip2] ( nM ) Growth rate ( μm/min ) Catastrophe frequency ( min-1 ) 0 0 . 257 ± 0 . 004 ( n = 300 ) 0 . 234 ± 0 . 017 ( n = 191 ) 5 0 . 302 ± 0 . 005 ( n = 263 ) 0 . 137 ± 0 . 012 ( n = 141 ) 10 0 . 353 ± 0 . 008 ( n = 146 ) 0 . 103 ± 0 . 010 ( n = 116 ) 20 0 . 572 ± 0 . 01 ( n = 57 ) 0 . 020 ± 0 . 006 ( n = 13 ) 40 0 . 589 ± 0 . 012 ( n = 48 ) 0 . 009 ± 0 . 004 ( n = 5 ) 10 . 7554/eLife . 10542 . 010Figure 2 . Kip2 is a highly processive motor that dwells at plus ends . ( A ) Schematic of the experimental design . ( B ) Kymograph showing processive motility and plus end accumulation of individual Kip2-eGFP molecules on GMPCPP-stabilized microtubules in 1 mM ATP . The concentration of Kip2-eGFP was 0 . 085 nM . ( C ) Kymograph showing tightly bound Kip2-eGFP molecules in AMP-PNP . ( D ) Kymograph showing end residence of individual Kip2-eGFP molecules in ATP . ( E ) Kymograph showing end residence of 1 nM Kip2-eGFP spiked into 20 nM unlabeled Kip2 in the presence of 8 μM unlabeled tubulin in ATP . Arrow heads indicate microtubule plus end tracking events . ATP , adenosine triphosphate; AMP-PNP , adenylyl imidodiphosphate; eGFP , enhanced green fluorescent protein; GMPCPP , guanylyl ( a , ß ) methylene-diphosphonate . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 01010 . 7554/eLife . 10542 . 011Figure 2—figure supplement 1 . Kip2-eGFP velocity , run length and end residence time distributions . ( A ) Histogram showing velocities of single Kip2-eGFP molecules on GMPCPP-stabilized porcine microtubules in 1 mM ATP . ( B ) Histogram showing the distribution of Kip2-eGFP run lengths on GMPCPP-stabilized porcine microtubules in 1 mM ATP . The run length of Kip2 is predicted to be exponentially distributed , as dissociation from the microtubule lattice is expected to be a random process . The red line depicts a single exponential: f ( x ) = A exp ( -x/x0 ) , where A = 21 . 1 ± 2 . 1 and x0 = 3 . 6 ± 1 . 0 μm; run lengths between 0 and 0 . 5 μm are under-represented , likely due to the limited temporal resolution . ( C ) Histogram showing end residence times of Kip2-eGFP on GMPCPP-stabilized porcine microtubules . End residence times were included in the analysis only if single Kip2-eGFP molecules could be observed to arrive at , and dissociate from , a microtubule plus end . The red line depicts a single exponential: f ( t ) = A exp ( -t/t0 ) , where A = 18 . 2 ± 2 . 4 and t0 = 44 . 1 ± 14 . 5 s . ATP , adenosine triphosphate; eGFP , enhanced green fluorescent protein; GMPCPP , guanylyl ( a , ß ) methylene-diphosphonate . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 011 To gain insight into the mechanism of Kip2’s polymerase and anti-catastrophe activities , we determined how Kip2 affects microtubule assembly and disassembly kinetics . By measuring the rate of growth of porcine microtubules from guanylyl ( α , β ) methylene-diphosphonate ( GMPCPP ) seeds over a range of tubulin concentrations ( Figure 1—figure supplement 3A ) , we found that Kip2 doubled the effective tubulin association rate constant ( kon , the rate that tubulin is stably incorporated into the microtubule lattice ) to 1 . 5 μM-1 . s-1 ( at the plus end ) from 0 . 7 μM-1 . s-1 in the absence of Kip2 . In addition to accelerating the net addition of subunits , Kip2 also facilitated microtubule nucleation on the seeds , with robust growth observed at tubulin concentrations as low as 4 μM , compared with 10 μM in the absence of Kip2 ( Figure 1—figure supplement 3A ) . Thus , Kip2 acted like a nucleation factor in analogy to XMAP215 ( Wieczorek et al . , 2015 ) . The increased growth rate in the presence of Kip2 is expected to have only a modest effect on catastrophe because doubling the rate of microtubule growth by doubling the tubulin concentration only decreases the catastrophe frequency about twofold ( Gardner et al . , 2011; Walker , et al . , 1988 ) . Our observation that the catastrophe frequency decreased 10-fold might be explained by our finding that 40 nM Kip2 decreased the rate of dissociation of GMPCPP-tubulin subunits from GMPCPP microtubules ( koff ) approximately threefold ( Figure 1—figure supplement 3B ) . If GMPCPP-tubulin acts as an analog for guanosine-5'-triphosphate ( GTP ) -tubulin ( Hyman et al . , 1992 ) , then a decrease in koff is expected to stabilize the GTP cap and therefore inhibit catastrophe ( Bowne-Anderson et al . , 2013; Coombes et al . , 2103; Margolin et al . , 2011 ) . Thus , the increase in kon and the decrease in koff likely account for most of the decrease in the catastrophe frequency . To determine how Kip2 targets the plus ends of microtubules , we characterized its biophysical properties in single-molecule motility assays ( Figure 2A ) . Kymographs revealed that in 1 mM ATP , single Kip2-eGFP molecules associated with GMPCPP-stabilized porcine microtubules along the lattice and walked processively toward the plus end of the microtubule ( Figure 2B , Figure 2—figure supplement 1A ) . The velocity was 5 . 0 ± 0 . 9 μm/min at 28°C ( mean ± standard deviation [SD] , n = 674 traces ) . The average run distance before dissociating was 4 . 1 ± 0 . 3 μm ( mean ± SE , n = 217 , Figure 2—figure supplement 1B ) . A similar velocity was observed by Roberts et al . ( 2014 ) , though the run distance was shorter ( 1 . 2 μm ) . At the plus end , Kip2-GFP resided for 30 ± 26 s before dissociating ( mean ± SD , n = 40 , Figure 2D , Figure 2—figure supplement 1C ) , leading to an accumulation of up to 12 Kip2-eGFP molecules at the plus-end , based on the fluorescence intensity ( Figure 2B ) . When the ATP was replaced by the non-hydrolyzable analog AMP-PNP , Kip2-eGFP tightly bound to the lattice and did not translocate ( Figure 2C ) . Kip2-eGFP moved slower on dynamic microtubules ( 2 . 1 ± 0 . 89 μm/min ) , but this velocity is still greater than the microtubule’s growth speed , so Kip2-eGFP was able to catch up to the growing ends of dynamic microtubules and track them ( Figure 2E ) . Based on these properties , we conclude that Kip2’s mechanism differs from that of the well-studied microtubule polymerase XMAP215 . XMAP215 targets ends by diffusion and capture ( Brouhard et al . 2008; Widlund et al . 2011 ) , increases both kon and koff ( Brouhard et al . , 2008 ) and has little effect on catastrophe ( Zanic , et al . , 2013 ) . Furthermore , Kip2’s ATPase activity is necessary for its activity , while XMAP215 is not an ATPase . Thus , Kip2 is a unique regulator of microtubule dynamics . Two models for growth promotion can be envisaged . Kip2 may increase growth rates by shuttling tubulin to the microtubule plus end , locally increasing the tubulin concentration . Alternatively , it could promote microtubule growth by acting as a processive polymerase while at the plus end , similar to XMAP215 ( Brouhard et al . , 2008 ) . More work will be required to distinguish between these and other mechanisms . To probe the mechanical properties of Kip2 , we measured the stall force of single molecules using optical tweezers ( Jannasch et al . , 2013 ) . Positional tracking of single Kip2-powered microspheres moving along GMPCPP-stabilized porcine microtubules as a function of time under constant load revealed a zero-force speed of 4 . 0 ± 0 . 5 μm/min at 24 . 5°C , similar to that measured in the total internal reflection fluorescence ( TIRF ) assays . Kip2 stalled at a force of 0 . 81 ± 0 . 04 pN ( Figure 3A ) and showed a nearly linear force-velocity relation with increased velocity as the assisting force was increased ( Figure 3C ) . At high forces , the motor often slipped along the microtubule in the direction of the applied force without detaching ( Figure 3B ) . The ability to switch from the slip state to the normal translocation mode is thought to increase processivity by linking together several shorter run lengths ( Jannasch et al . , 2013 ) . Thus , Kip2 is a processive , low-force motor with long run lengths and end residence times . The low force supports the idea that individual Kip2 motors transport small cargos such as dynein , Bik1 , and other molecules ( Roberts et al . , 2014 ) rather than organelles , although it is possible that multiple Kip2s could cooperate to move larger cargos . The strong localization to the microtubule plus end accords with Kip2 being a regulator of microtubule dynamics . 10 . 7554/eLife . 10542 . 012Figure 3 . Kip2 is a low-force motor . ( A ) Stall force measurement tace . Sampling rate: 10 kHz , raw data ( gray ) , boxcar filtered to 50 Hz ( black ) . A force of 0 . 5 N corresponds to a displacement of about 17 nm . ( B ) Time trace for a slip event under 3 pN assisting force . Sampling rate: 20 kHz , raw data ( light cyan ) , boxcar filtered to 400 Hz ( dark cyan ) . ( C ) Kip2 force-velocity curve: positive is a hindering ( load ) force and negative is an assisting force . Open symbols include slip events . Error bars are standard errors of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 012 The low force and high processivity of Kip2 are reminiscent of the microtubule depolymerase Kip3 , in the kinesin-8 family ( Jannasch et al . , 2013; Varga et al . , 2006 ) . Kip3 is a length-dependent depolymerase that uses an antenna mechanism to preferentially localize to the plus ends of longer microtubules ( Varga et al . , 2009 ) . We therefore tested whether the promotion of microtubule growth by Kip2 is length-dependent ( Figure 4A , B ) . Without Kip2 , microtubules grew at a length-independent , constant rate ( black circles , p=0 . 06 , Student’s t-test on a linear fit to the raw data ) . By contrast , at low ( 1–2 nM ) and intermediate ( 5–10 nM ) Kip2 concentrations , long microtubules grew faster than short microtubules ( p<0 . 0001 ) . At high Kip2 concentrations ( 20–40 nM ) , microtubules again grew at a constant , length-independent rate ( green circles , p=0 . 36 ) ; however , at these high Kip2 concentrations , we expect all the length dependence to be in the first few microns , which is not well resolved in these experiments ( see green fitted line ) . An analysis of yeast microtubule growth rates as a function of microtubule length yielded similar results ( Figure 4—figure supplement 1A ) . Thus , Kip2 is a length-dependent microtubule polymerase . 10 . 7554/eLife . 10542 . 013Figure 4 . Kip2 promotes porcine microtubule growth in a length-dependent manner . ( A ) Kymograph showing acceleration of microtubule growth with increasing length at 40 nM Kip2 . ( B ) Porcine microtubule growth rate as a function of length without Kip2 ( black ) and binned for 1–2 nM Kip2 ( purple ) , 5–10 nM Kip2 ( blue ) and 20–40 nM Kip2 ( green ) . Lengths are binned for 0–2 μm , 2–3 μm , 3–4 μm , 4–6 μm , 6–8 μm , 8–12 μm , 12–16 μm and 16–24 μm . The data were fit with the equation , where v0 = 0 . 294 ± 0 . 009 μm/min is the initial growth rate; vmax = 1 . 03 ± 0 . 03 μm/min is the maximum growth rate; L is microtubule length and A = 39 . 8 ± 5 . 4 μm·nM . ( C ) Mean catastrophe length at various Kip2 concentrations for short ( light gray ) and long ( dark gray ) microtubules . In the short microtubule bins , we summed the total distance that microtubules grew while shorter than 4 μm and divided by the number of catastrophes that occurred at lengths < 4 μm . In the long microtubule bin , we summed the total distance that microtubules grew while longer than 4 μm and divided by the number of catastrophes that occurred at lengths > 4 μm ( Figure 4C , inset ) . The number of catastrophes was 120 ( 0 nM Kip2 ) , 102 ( 1 nM Kip2 ) , 111 ( 2 nM Kip2 ) and 23 ( 5 nM Kip2 ) . The number of catastrophes at higher Kip2 concentrations was too small to make statistically significant comparisons . Error bars are standard errors of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 01310 . 7554/eLife . 10542 . 014Figure 4—figure supplement 1 . Length dependence of growth and catastrophe for yeast tubulin . ( A ) Yeast microtubule growth rate as a function of length , without Kip2 ( black ) and binned for 5–10 nM Kip2 ( blue ) and 20–40 nM Kip2 ( green ) . Lengths are binned for 0–2 μm , 2–3 μm , 3–4 μm , 4–6 μm , 6–8 μm , 8–12 μm , 12–16 μm and 16–24 μm . The data were fit with the equation , where v0 = 0 . 238 ± 0 . 005 μm/min is the initial growth rate; vmax = 0 . 78 ± 0 . 02 μm/min is the maximum growth rate; L is microtubule length and A = 79 . 4 ± 8 . 7 μm·nM . ( B ) Mean length at catastrophe for yeast microtubules at 0 nM Kip2 ( n = 170 catastrophes ) , 5 nM Kip2 ( n = 101 ) and 10 nM Kip2 ( n = 80 ) . Bars depict dynamic microtubule lengths below 4 μm ( light gray ) or above 4 μm ( dark gray ) . Error bars are SE . SE , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10542 . 014 To test whether Kip2 also prevents catastrophe in a length-dependent manner , we measured porcine microtubule lengths at the moment of catastrophe . To compare the catastrophe frequency at short versus long microtubule lengths , we set a cut-off length at 4 μm , which equals the run length of Kip2 . Using data from the dynamic microtubule assays ( Figure 1 ) , we measured the catastrophe length for short microtubules as the total distance that microtubules grew while their length ( including seed ) was shorter than 4 μm divided by the number of catastrophes that occurred at lengths < 4 μm . For long microtubules , we summed the distance that microtubules grew while longer than 4 μm ( final length minus 4 μm ) and divided by the number of catastrophes that occurred at lengths > 4 μm ( Figure 4C , inset ) . In the absence of Kip2 , the catastrophe length of longer microtubules was less than that of shorter microtubules ( Figure 4C , 0 nM Kip2 , p<0 . 05 , Welch’s unpaired t-test ) . This reflects an increase in catastrophe frequency with length , as expected due to microtubule aging ( Gardner et al . , 2011 ) . By contrast , at 5 nM Kip2 , the catastrophe length of longer microtubules was greater than that of shorter microtubules ( Figure 4C , p<0 . 0001 ) . This indicates that the inhibition of catastrophe is length-dependent . Similar results were obtained for yeast microtubules at 10 nM Kip2 ( Figure 4—figure supplement 1B , p<0 . 0001 ) . Hence , in the absence of Kip2 , the catastrophe frequency increased with increasing microtubule length , whereas in the presence of Kip2 , the catastrophe frequency decreased with increasing microtubule length . Thus , both the increase in microtubule growth rate and the prevention of catastrophe by Kip2 increase with increasing microtubule length . Summarizing our results , we have found that budding yeast kinesin Kip2 promotes microtubule growth in vitro in a length-dependent manner . Because the rate at which Kip2 translocates exceeds the speed of microtubule growth , Kip2 catches up with the growing end of the microtubule ( Figure 2E ) where it promotes growth and inhibits catastrophe . As a consequence , this length dependence leads to positive feedback: the longer the microtubule , the greater the number of motors that land on it ( the microtubule acts as an antenna ) , the higher the number of motors that can reach the plus end , and the higher the growth rate and lower the catastrophe frequency . This , in turn , leads to longer microtubules , which attract more Kip2 and so on . Hence , we expect that once a microtubule is long enough , it will effectively “escape” catastrophe and keep growing almost indefinitely , switching from a catastrophe length of only a few microns in the absence of Kip2 to a length ≥ 40 μm at high Kip2 concentrations ( Figure 4C , Table 1 ) . In this sense , Kip2 “paves its own way” . Thus , by combining processivity with polymerase activity , Kip2 can perform an elementary ‘computation’ that switches short microtubules to long ones . This computation differs from that performed by kinesin-8 , a length-dependent depolymerase , which stabilizes microtubule length through negative feedback ( Gupta et al . , 2006; Mayr et al . , 2007; Stumpff et al . , 2008; Su , et al . , 2013;Widlund , et al . , 2011 Varga et al . , 2006; 2009 ) . We propose that this positive feedback mechanism may operate in vivo and account for the phenotype of Kip2 deletion , which is a reduction in the length and the number of cytoplasmic microtubules . First , in vivo the rate at which Kip2 translocates exceeds the rate of microtubule growth; the respective rates are 6 . 6 ( Carvalho et al . , 2004 ) and 2 . 3 μm/min ( Caudron et al . , 2008 ) . Second , the run length of Kip2 ( ≈4 μm ) exceeds the length of cytoplasmic microtubules ( ≈2 μm [Caudron et al . , 2008] ) . Taken together , these two observations imply that almost every Kip2 that lands on a microtubule will reach the growing plus end . By promoting growth and inhibiting catastrophe , Kip2 can deliver cytoplasmic dynein ( Roberts et al . , 2014 ) to the distal cortex of the growing daughter bud before the microtubules catastrophe . While the polymerase and anti-catastrophe activities can account for the deletion phenotype of Kip2 , it is not obvious why microtubule hyperelongation when Kip2 is overexpressed should require Bik1 ( Carvalho et al . , 2004 ) . We propose that Bik1 may be required to increase Kip2’s processivity in vivo . Feedback can only operate if the run length exceeds the microtubule length . In our in vitro assays , the Kip2 run length was ≈4 μm , whereas that measured by Roberts et al . ( 2014 ) was only ≈1 μm . We do not know why there was a difference , as the assay buffers were similar . Importantly , though , Roberts et al . ( 2014 ) found that Bik1 could increase Kip2’s run length 3–4 fold ( in the presence of Bim1 ) . Therefore , if the run length of Kip2 in vivo is short , then the requirement for Bik1 in the overexpression assays may be due to Bik1 acting as a processivity factor that increases the run length , thereby allowing more Kip2 to reach the end where it enhances microtubule growth .
Porcine brain tubulin was purified and labeled with tetramethylrhodamine or Alexa Fluor 488 ( Invitrogen , Carlsbad , CA ) according to the standard protocols , as previously described ( Gell et al . , 2011 ) . Preparation of GMPCPP-stabilized microtubule seeds was performed as previously described ( Gell et al . , 2010 ) . Full length 6xHis-Kip2 and 6xHis-Kip2-eGFP were expressed in SF+ cells using baculovirus expression and purified using affinity chromatography over 1 ml His-affinity columns ( GE Healthcare , Chalfont St . Giles , UK ) . Cells were lysed in 50 mM NaH2PO4 , 300 mM NaCl , 0 . 1% Tween-20 , 10 mM imidazole , protease inhibitors , 2 mM Mg-ATP , at pH = 8 . 0 . The wash buffer consisted of 50 mM NaH2PO4 , 300 mM NaCl , 100 mM imidazole , 2 mM Mg-ATP , at pH = 8 . 0 . The elution buffer consisted of 50 mM NaH2PO4 , 300 mM NaCl , 300 mM imidazole , 2 mM Mg-ATP , at pH = 8 . 0 . Affinity column purification success was checked by sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS–PAGE ) and Western blot using anti-6xHis antibody ( Genscript , Piscataway , NJ ) . Next , the 6xHis-tags were cleaved from the protein using PreScission protease ( GE Healthcare ) . The protease was added to the 300 mM imidazole elution fraction in a 1:50 dilution and incubated overnight on a rotary wheel at 4°C . Protein stability was confirmed by SDS–PAGE and enzymatic cleavage of the 6xHis-tag from the protein of interest by Western blot using anti-6xHis-antibody . Finally , Kip2 and Kip2-eGFP were purified to homogeneity by gel filtration over a Sephadex 200 column that was pre-washed with protein storage buffer: 1x BRB80 ( 80 mM PIPES , 1 mM MgCl2 , 1 mM EGTA , pH 6 . 8 ) supplemented with 10% glycerol , 1 mM Mg-ATP , 1 mM dithiothreitol ( Figure 1—figure supplement 4 ) . Final protein purity was checked by mass spectroscopy at the MPI-CBG in house mass spectroscopy facility . Protein concentration was determined by Bradford assay and purified proteins were snap-frozen using liquid nitrogen and stored at –80°C . The dynamic microtubule assay for dynamic growth of Alexa Fluor 488-labeled tubulin from tetramethylrhodamine-labeled GMPCPP-stabilized porcine tubulin seeds were imaged by TIRF microscopy as described previously ( Gell et al . , 2010 ) . The imaging buffer contained 1x BRB20 ( 20 mM PIPES , 1 mM MgCl2 , 1 mM EGTA , pH 6 . 8 ) supplemented with 100 mM KCl , 20 mM glucose , 20 μg/ml glucose oxidase , 8 μg/ml catalase , 0 . 1 mg/ml casein , 1 mM dithiothreitol , 0 . 001% tween-20 , 1 mM GTP and 1 mM Mg-ATP or AMP-PNP . The single-molecule motility assay on tetramethylrhodamine-labeled GMPCPP-stabilized tubulin seeds imaged by TIRF microscopy was described previously ( Gell et al . , 2010 ) . For all experiments , the imaging buffer contained no added GTP . Imaging was performed with an Andor iXon camera on a Zeiss ( Oberkochen , Germany ) Axiovert 200M microscope with a Zeiss ×100/1 . 46 plan apochromat oil objective and standard filter sets . An objective heater ( Zeiss ) was used to warm the sample to 28°C . The rate of photobleaching in our TIRF assays was low . In the AMP-PNP experiments ( e . g . Figure 2C ) , the mean time to bleaching of Kip2-eGFP was 249 ± 68 s ( mean ± SD , n = 10 ) . Given that the average run length of 4 . 1 μm corresponds to a run time of 82 s ( at 50 nm/s ) , we expect bleaching to have only a small effect on the measured run times . Similarly , bleaching will have little effect on the end residence times . The low rate of photobleaching accords with our earlier quantification of photobleaching ( Varga et al . , 2009 ) . DIC microscopy was described previously ( Bormuth et al . , 2007 ) . All experiments were performed at least three times on three different days . Image analysis was performed by creating kymographs of microtubule growth events in image J . For growth and shrinkage rates , typically > 20 microtubules were measured , and the mean and standard error of the mean ( SE ) are reported in the text and figures . For the catastrophe frequency , we divided the total number of events by the total observation time . For the rescue distance , we divided the total observed distance that microtubules shrank by the total number of rescue events . The relative error ( SE ) was estimated as the inverse of the square root of the number of events . This assumes that the catastrophe and rescue events are single-step ( Poisson ) processes . However , if the events are multistep ( e . g . from a gamma distribution ) , as is known to be the case for catastrophe ( Gardner et al . , 2011 ) , then the actual SE is smaller than the calculated one . Flow-cell construction and immobilization of GMPCPP-stabilized porcine microtubules were performed as previously described ( Jannasch et al . , 2013 ) . The imaging buffer for optical tweezer experiments contained 1xBRB20 supplemented with 100 mM KCl , 20 mM glucose , 20 μg/ml glucose oxidase , 8 μg/ml catalase , 0 . 1 mg/mlcasein , 0 . 5% b-mercaptoethanol , 1 mM Mg-ATP . The channels were rinsed with 20 μl imaging buffer with Kip2-functionalized microspheres . For the Kip2-functionalized microspheres , carboxylated polystyrene microspheres ( mean diameter 0 . 59 µm , Bangs Lab , Fishers , IN ) were bound covalently to anti-GFP antibody via a 3 kDa polyethylene glycol ( PEG ) linker , which , in turn , bound to the C-terminal eGFP of Kip2-eGFP-6xHis , as previously described ( Jannasch et al . , 2013 ) . The measurements were performed at 24 . 5°C and under single-molecule concentrations where only one out of four microspheres showed motility . Measurements were performed in a single-beam optical tweezers setup as previously described ( Schäffer et al . , 2007; Bormuth et al . , 2009; Jannasch et al . , 2013 ) . All measurements were done with a trap stiffness of 0 . 03 pN/nm . The optical trap was calibrated by analysis of the height-dependent power spectrum density as described previously ( Tolić-Nørrelykke et al . , 2006 ) . The force-velocity curve was measured using the constant-force mode . In this mode , the trapping laser was moved with a piezo mirror relative to the sample with an update rate of 200 Hz . Overall , we measured and analyzed the motion of 11 different single Kip2-eGFP-6xHis molecules . Data analysis was previously described ( Jannasch et al . , 2013 ) . | Cells contain an extensive network of long filaments called microtubules , which are made of a protein called tubulin and are essential for a wide variety of processes such as enabling cells to divide and move . Microtubules also serve as tracks along which motor proteins transport molecules from one part of the cell to another . In yeast cells , a motor protein called Kip2 transports its cargo to one end ( known as the “plus end” ) of the microtubules . This is the fastest-growing end of the microtubule , although it frequently switches between phases of growth and shrinkage . Previous research has shown that cells containing reduced amounts of Kip2 have much shorter filaments than normal cells . One suggested explanation of these results is that Kip2 controls microtubule growth by transporting a protein that regulates filament length to the plus end of the microtubule . However , by adding purified Kip2 to microtubules Hibbel et al . have now shown that Kip2 on its own can increase the length of filaments . The microtubules also switch much less frequently between growth and shrinkage in the presence of Kip2 . This is because Kip2 moves along filaments at a speed that is greater than the rate at which microtubules grow . This sets up a positive feedback loop that causes microtubule growth to accelerate , as more copies of Kip2 can bind to longer microtubules . Each of these motor proteins can then move to the plus end of the microtubule and help the filament to grow even longer . Several challenges remain . What is the molecular mechanism by which Kip2 increases the rate at which tubulin subunits are added to the microtubule end: does Kip2 carry tubulin dimers to the end in a shuttle-type mechanism ? How does Kip2 prevent other proteins from promoting shrinkage ? What stops microtubules from growing when they reach the end of the cell ? | [
"Abstract",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"short",
"report",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Kinesin Kip2 enhances microtubule growth in vitro through length-dependent feedback on polymerization and catastrophe |
Cytosolic hormone levels must be tightly controlled at the level of influx , efflux , synthesis , degradation and compartmentation . To determine ABA dynamics at the single cell level , FRET sensors ( ABACUS ) covering a range ∼0 . 2–800 µM were engineered using structure-guided design and a high-throughput screening platform . When expressed in yeast , ABACUS1 detected concentrative ABA uptake mediated by the AIT1/NRT1 . 2 transporter . Arabidopsis roots expressing ABACUS1-2µ ( Kd∼2 µM ) and ABACUS1-80µ ( Kd∼80 µM ) respond to perfusion with ABA in a concentration-dependent manner . The properties of the observed ABA accumulation in roots appear incompatible with the activity of known ABA transporters ( AIT1 , ABCG40 ) . ABACUS reveals effects of external ABA on homeostasis , that is , ABA-triggered induction of ABA degradation , modification , or compartmentation . ABACUS can be used to study ABA responses in mutants and quantitatively monitor ABA translocation and regulation , and identify missing components . The sensor screening platform promises to enable rapid fine-tuning of the ABA sensors and engineering of plant and animal hormone sensors to advance our understanding of hormone signaling .
Research by the Intergovernmental Panel on Climate Change ( IPCC ) indicates that the climate is becoming more variable and that the frequency of floods and droughts is increasing ( IPCC , 2013 ) . Elevated temperatures combined with periods of drought represent a major threat to food security ( Sreenivasulu et al . , 2007; Barnabas et al . , 2008; Brutnell and Frommer , 2012; Sreenivasulu et al . , 2012 ) . Thus , there is an obvious and urgent need to advance our understanding of plant tolerance to these stresses as well as the underlying mechanisms as a basis for engineering crops that can survive the anticipated environmental challenges ( Tester and Langridge , 2010; Schroeder et al . , 2013 ) . Abscisic acid ( ABA ) , a terpenoid plant hormone , has been found in all kingdoms of life , including plant saprophytic and pathogenic fungi , animals such as sponges ( Axinella polypoides ) and hydroids ( Eudendrium racemosum ) , and human parasites ( Toxoplasma gondii ) ( Crocoll et al . , 1991; Nagamune et al . , 2008; Li et al . , 2011 ) . Yet , we know little regarding the homeostasis and dynamics of this hormone in these systems . In plants , abscisic acid ( ABA ) serves as the key phytohormone produced during drought , and is a master regulator of water use efficiency , stomatal aperture , and other mechanisms of tolerance to drought and osmotic stress ( Sreenivasulu et al . , 2012 ) . ABA also controls seed dormancy and germination ( Finkelstein and Rock , 2002; Bentsink and Koornneef , 2008; Kanno et al . , 2010 ) . ABA is perceived by a family of ABA receptors ( PYR/PYL/RCAR proteins , Ma et al . , 2009; Park et al . , 2009 ) , which then bind to and prevent ABA co-receptor proteins ( protein phosphatase 2CA proteins , PP2CAs , Umezawa et al . , 2009; Vlad et al . , 2009 ) from deactivating Sucrose non-fermenting Related Kinase 2s ( SnRK2s , Fujii and Zhu , 2009; Fujita et al . , 2009; Nakashima et al . , 2009 ) . The resulting increase in SnRK2 activity triggers phosphorylation of transcription factors ( Yoshida et al . , 2010 ) and transporters ( Geiger et al . , 2009; Lee et al . , 2009; Sato et al . , 2009 ) , and affects ABA-dependent responses that vary with respect to cell type and developmental or environmental context . For example , only ∼25% of the ABA responsive transcripts in leaves are also ABA responsive in guard cells ( Wang et al . , 2011 ) , and ABA signaling specifically in the endodermis promotes lateral root quiescence in plants exposed to salt stress ( Duan et al . , 2013 ) . ABA responses affect many aspects of physiology and development , evidenced by pleiotropic phenotypes of mutants deficient in ABA metabolism ( Leon-Kloosterziel et al . , 1996 ) , perception ( Gonzalez-Guzman et al . , 2012 ) , or signaling ( Rubio et al . , 2009; Fujii et al . , 2011 ) . Hallmark responses include stomatal closure ( Leung and Giraudat , 1998 ) , and altered growth rates and metabolic adjustments during stress responses ( Ober and Sharp , 1994; Sharp and LeNoble , 2002 ) , senescence ( Hunter et al . , 2004 ) and seed dormancy ( Bentsink and Koornneef , 2008; Rodriguez-Gacio Mdel et al . , 2009 ) . Additionally , ABA affects other processes , including immunity ( Cao et al . , 2011 ) , crosstalk signaling with sugars ( e . g . , Arenas-Huertero et al . , 2000; Matiolli et al . , 2011 ) and other plant hormones ( Jaillais and Chory , 2010 ) . Thus , simple tuning of ABA levels broadly in order to change drought tolerance carries side effects such as increased seed dormancy ( e . g . , Lin et al . , 2007 ) . Understanding the diverse roles of ABA in multiple tissues requires quantitative knowledge of ABA levels and dynamics in individual cells . ABA accumulation under stress conditions is controlled at the level of biosynthesis , catabolism , and transport as well as through reversible conversion to ABA-glucose ester , an inactive form of ABA potentially used in both storage and long distance transport ( Cutler and Krochko , 1999; Seo and Koshiba , 2002; Wilkinson and Davies , 2002; Nambara and Marion-Poll , 2005 ) . Many of the enzymes involved in ABA metabolism have been identified , some of which are present as multi-gene families ( Tan et al . , 2003; Kushiro et al . , 2004; Lee et al . , 2006; Xu et al . , 2012 ) . The ubiquitous distribution of biosynthetic enzymes and receptors for hormones has raised the question whether they function locally or remotely from the site of synthesis; remote action is a hallmark of animal hormones . Long distance transport of auxin is critical ( Grieneisen et al . , 2007 ) , yet there has been a long-standing debate on whether and how hormones such as ABA are transported between plant organs . The weak acid properties of ABA ( pKa 4 . 7 ) led to the hypothesis that ABA diffuses freely across membranes in its undissociated lipophilic form ( Kaiser and Hartung , 1981 ) . Subsequent studies implicated cellular uptake via transport proteins ( Astle and Rubery , 1985b ) . Over the past few years , functional studies identified proteins that can mediate transport of ABA ( Kang et al . , 2010; Kuromori et al . , 2010 , 2011; Kanno et al . , 2012 ) . ABA levels can be monitored indirectly by expressing GUS , GFP or luciferase reporters under the control of ABA responsive promoters ( Christmann et al . , 2005; Kim et al . , 2011; Duan et al . , 2013; Geng et al . , 2013 ) , but the indirect nature of ABA detection of these reporters results in several potential limitations . For example , promoters may not respond in a linear fashion and can be subject to additional regulatory inputs . Furthermore , such reporters cannot detect rapid changes ( seconds to minutes ) and are susceptible to both variation in the activities of ABA signaling components and non-specific regulation of expression levels resulting from crosstalk with other signaling pathways ( common in hormone regulated gene expression ) ( Jaillais and Chory , 2010 ) . Improved direct measurement of ABA through biochemical methods such as in situ mass spectrometry ( Lorenzo Tejedor et al . , 2012 ) could increase the spatial resolution of ABA quantitation , but does not allow for dynamic measurement in live plants . Thus , direct sensors for ABA would represent a significant step forward in allowing the investigation of ABA levels with high spatial and temporal resolution . One of the most advanced technologies for high-resolution measurement of small molecules in living tissues is based on genetically encoded , ratiometric fluorescent sensors that bind to and report on the levels of the target molecule ( i . e . , sensors based on Förster Resonance Energy Transfer; FRET sensors [Okumoto et al . , 2012] ) . FRET sensors are fusion proteins that report target molecule interactions through changes in the conformation of intrinsic sensory domains . These conformational changes affect the efficiency of energy transfer from a fused FRET donor fluorescent protein to a fused FRET acceptor fluorescent protein . Changes in energy transfer can be detected by measuring changes in the relative intensity of the two fluorescent proteins ( ratio change ) after excitation of the donor; the ratio change reports target molecule concentration . FRET sensors have been used in plant tissues to study calcium and zinc dynamics with subcellular spatial and near real-time temporal resolution ( e . g . , Krebs et al . , 2012; Lanquar et al . , 2014 ) . Metabolite dynamics have also been studied using FRET sensors leading to , for example , the characterization of sugar transport in roots of intact seedlings ( Deuschle et al . , 2006; Chaudhuri et al . , 2008 , 2011; Grossmann et al . , 2011; Grossmann et al . , 2012 ) and the identification of novel sugar transporters ( Chen et al . , 2010 , 2012; Xuan et al . , 2013 ) . Here we developed a combinatorial and iterative engineering platform for FRET sensors that led to the identification of ABACUS1 , an Abscisic Acid Concentration and Uptake Sensor version 1 that reports ABA levels . The engineering platform is based on a series of Gateway destination vectors coding for FRET pairs and a library of Entry clones coding for potential ABA sensory domains . Destination vector series for expression of potential sensor fusions in bacteria , yeast or plants were generated . The combinatorial screening of the Destination vector series and the Entry clone library allowed for rapid testing of hundreds of constructs and led to the identification of ABACUS1 , which is specific for ABA and is optimized for reduced size , fluorophore brightness and photostability , and high ratio change ( i . e . high signal-to-noise ) . We then generated ABACUS1 variants covering a broad detection range for in vivo use . We studied the properties of a known ABA transporter through analysis of yeast cells coexpressing ABACUS1 . Using ABACUS1 sensors in the cytosol and nucleus of Arabidopsis , we detected reversible and concentration-dependent accumulation of exogenous ABA delivered to roots growing in the RootChip ( Grossmann et al . , 2011 , 2012 ) , a microfluidic device for continuous measurement of living roots in tightly controlled environmental conditions . ABA feedback on ABA levels was identified as a potentially critical aspect of ABA signaling as time-course analysis of ABACUS responses demonstrated negative regulation of cytosolic ABA levels by ABA .
Generation of high sensitivity FRET sensors is inherently empirical and requires multi-parameter optimization ( affinity , brightness , dynamic range , etc ) ( Okumoto , 2012 ) . Selection of the sensory domain is a critical step for designing FRET sensors , as it determines affinity and specificity . To generate FRET sensors for ABA , potential ABA sensory domains ( PAS ) were selected from members of ABA co-receptor complexes; specifically , nine members of the PYR/PYL/RCAR family of ABA receptors and three members of the PP2CA sub-family of ABA co-receptor phosphatases from Arabidopsis ( Table 1 , Supplementary file 1 ) . We generated two types of sensors: single domain sensors in which either a PYR/PYL/RCAR or PP2CA polypeptide alone ( sPAS ) is sandwiched between two fluorescent proteins , or double domain sensors , in which a PYL/PYR/RCAR fused via a linker to a PP2CA ( dPAS ) is sandwiched between two fluorescent proteins ( Figure 1A ) . To cover a large test space of ABA sensory domains , we engineered sensor constructs using >50 sPAS and dPAS variants sandwiched between a fluorescent protein FRET pair ( i . e . , multiple PAS Entry clones were recombined with a single destination vector ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 01741 . 003Table 1 . Guide to sensor component variants used for screening ( for details see Supplementary file 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 003Destination vector N-termGateway sitedPAS Entry cloneGateway siteDestination vector C-termpDR FLIPFRET acceptorN-term LinkerdPAS N-termdPAS LinkerdPAS C-termC-term LinkerFRET donor30Aphrodite . t9attB1 invariantHAB1 · ABI1aid · ABI1cd · PYR1 · PYL1 · PYL4 · PYL5 · PYL6 · PYL7 · PYL8 · PYL9 · PYL10L12 Flexible · L52 Spring · L65 Flexible · L71 α-helix · L118 Spring & α-helixHAB1 · ABI1aid · ABI1cd · PYR1 · PYL1 · PYL4 · PYL5 · PYL6 · PYL7 · PYL8 · PYL9 · PYL10attB2 invariantmCerulean32Aphrodite . t9t7 . eCFP . t934Aphrodite . t9t7 . TFP . t935Aphrodite . t9mTFP . t936Aphrodite . t9Cerulean37CitrineCerulean38edCitrineedCerulean39edAphrodite . t9t7 . ed . eCFP . t942CitrinemCerulean43edAphrodite . t9edCeruleanAbbreviations: t7 , t9: 7 or 9 amino acid terminal truncations; aid: ABA interaction domain; cd: catalytic domain; ed , m: enhanced dimerization or monomeric variant10 . 7554/eLife . 01741 . 004Figure 1 . ABA responses of potential FRET sensors expressed in yeast and tested in yeast cell lysates or as purified proteins . ( A ) Diagram of cloning strategy based on pDR FLIP Destination vectors encoding FRET fluorescent protein pairs and PAS Entry clones encoding sPAS or dPAS ABA sensory domains . Also shown is an example of fluorescence emission curves without and with ligand for a sensor [ABACUS1-80µ , see below] with a positive ratio change ( Δ DxDm/DxAm ) of 1 . 6 . ( B ) One linker variant ( L12 ) of one dPAS ( 110 ) combined with nine fluorescent protein pairs . ( C ) Five linker variants of one dPAS combined with one fluorescent protein pair . ( D ) Two linker variants of one dPAS combined with one fluorescent protein pair tested as purified proteins . DxAm/DxDm = acceptor emission with donor excitation over donor emission with donor excitation . dPAS = double putative ABA sensory domain . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 00410 . 7554/eLife . 01741 . 005Figure 1—figure supplement 1 . Fluorescence emission curves and ABA response of potential FRET sensors expressed in yeast and tested in yeast cell lysates or as purified proteins ( excitation wavelength = 428 nm ) . ( A ) Sensor designs with single Potential ABA Sensory domains ( sPAS ) and double Potential ABA Sensory domains ( dPAS ) combined with one fluorophore pair ( from pDR FLIP39 ) . ( B ) dPAS sensor designs using ABI1 truncations and two linker variants combined with one fluorophore pair ( from pDR FLIP39 ) . Five linker variants of one dPAS combined with nine fluorophore pair variants tested in cell lysates ( C ) or as purified proteins ( D ) . Ratio change = treated DxAm/DxDm/mock DxAm/DxDm . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 005 In parallel , we created an array of sensor constructs using a subset of the PAS domains combined with a wide spectrum of different fluorescent protein FRET pairs ( Table 1; Figure 1A ) . Use of different fluorescent proteins can have large effects on the brightness and ligand-induced FRET changes of a biosensor ( e . g . YFP vs improved variant Aphrodite ( Deuschle et al . , 2006 ) . Wild-type fluorescent proteins can form dimers or multimers , and mutations along the dimerization interface of the proteins can reduce ( e . g . , monomeric GFP , mGFP , Zacharias et al . , 2002 ) or promote dimerization ( enhanced dimerization ( ed ) fluorescent protein variants ) . Enhanced dimerization variants have successfully been used for improved sensor design ( e . g . , Vinkenborg et al . , 2007 , 2009 ) . To allow for testing a large combinatorial space of sensory domains and FRET pairs , we engineered a destination vector series encoding 10 different fluorescent protein FRET pairs ( Table 1 ) . Fluorescent protein variants included brightness variants and dimerization variants , as well as N- or C- terminal truncation variants ( ( indicated as t7 or t9 for truncation of 7 or 9 amino acids ) , Table 1 , Supplementary file 1 ) . Initial experiments , in which constructs were expressed in Escherichia coli and yeast , were unsuccessful , most likely due to proteolysis of the fusion proteins , as evidenced by maintenance of fluorescence but loss of FRET during cell lysis ( Figure 2 ) . Proteolysis appeared specific to PAS domain sensors and was not observed for other FRET sensors ( Fehr et al . , 2002 , 2003; Lager et al . , 2003; Okumoto et al . , 2005; Gu et al . , 2006; Lager et al . , 2006 ) . Sensory domain proteolysis was circumvented by using a protease-deficient yeast strain ( strain BJ5465 lacking Pep4 and Prb1 , Figure 2 ) for expression and purification . 10 . 7554/eLife . 01741 . 006Figure 2 . Fluorescence emission spectra of two sensors and individual fluorescent proteins expressed in yeast strain 23344C and protease deficient yeast strain BJ5465 ( excitation wavelength = 428 nm ) . Proteins were analyzed in washed cells and in the supernatant or pellet of cell lysates . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 006 A first screen was conducted by analyzing ABA-dependent changes in fluorescence emission spectra of sPAS sensors in cell lysates cleared of cellular debris by centrifugation ( Figure 1B , Figure 1—figure supplement 1A ) . FRET sensor responses were calculated as an ABA-induced change in the ratio of FRET acceptor emission ( Am ) to FRET donor emission ( Dm ) after excitation of the FRET donor ( Dx ) ; abbreviated ΔDxAm/DxDm . Four out of 15 sPAS sensors showed a relatively small ratio change ( ∼1 . 06–1 . 12 ) in response to addition of 200 µM ( ± ) -ABA ( Figure 1—figure supplement 1A ) . We hypothesized that dPAS sensors containing both partners of an ABA co-receptor complex might exhibit larger ABA-dependent ratio changes due to larger conformational changes triggered by ABA-dependent intra-molecular interactions between two ABA binding domains . The presence of both ABA interaction domains may also increase ABA affinity , as demonstrated for PYR/PYL/RCAR affinity in the presence of PP2CA co-receptors ( Santiago et al . , 2012 ) . An initial screen of dPAS containing full length PP2CA HAB1 linked via a flexible 12-amino acid linker ( L12 ) rich in alanine , glycine and serine to the N- or C-terminus of nine different PYR/PYL/RCARs ( dPAS19-dPAS36 , Supplementary file 1 ) in cleared cell lysates identified several ABA-responsive candidates ( e . g . , dPAS20: N-terminal HAB1-L12-C-terminal PYL1; Figure 1—figure supplement 1A ) . We also screened dPAS designs with truncations of the PP2CA ABI1 in place of full length HAB1 ( Figure 1—figure supplement 1B ) . Specifically ABI1 was truncated to its catalytic domain ( ABI1cd ) or a minimal ABA interaction domain ( ABI1aid; 49 amino acids in length; Supplementary file 1 ) . These variants ( dPAS97-120 , Supplementary file 1 ) constitute smaller proteins with potentially fewer interactions of the PP2CA domain with endogenous components in vivo; moreover , ABI1aid likely lacks PP2CA enzymatic activity . Among these combinations , ABI1aid fused to PYL1 ( dPAS110 ) consistently yielded comparatively large ratio changes in response to ABA ( ΔDxAm/DxDm ∼1 . 29 to 1 . 63; Figure 1—figure supplement 1B ) . The screen was repeated with variants in which an artificial linker consisting of a molecular spring domain coupled to an α-helix was inserted between the two domains ( L118; Table 1 ) . Among them , the ABI1aid/PYL1 combination again showed robust responses to ABA ( Figure 1—figure supplement 1B ) . Although dPAS110 yielded biosensors with consistently high ratio change , similar sensory domains are also promising and could exhibit complementary properties . For example , PYR1 and PYL1 share 77% amino acid sequence identity and the use of dPAS109 , containing PYR1 in an otherwise identical dPAS design , also led to ABA sensors with robust responses to ABA ( Figure 1—figure supplement 1B ) . Furthermore , the use of ABI1cd/PYL1 ( dPAS98 ) led to ABA responsive biosensors that could exhibit higher affinity for ABA compared to dPAS110 due to the presence of a complete PYL1 interaction interface in the ABI1 catalytic domain ( Figure 1—figure supplement 1B ) . Many other dPAS variants showed low or no detectable ABA response in this initial screen ( Figure 1—figure supplement 1B ) , underscoring the empirical nature of sensor development and the value of large-scale screens . Linkers can have dramatic effects on sensor responses ( Deuschle et al . , 2005b; Hires et al . , 2008; Takanaga et al . , 2008 ) . Therefore , five different linkers were inserted between the two PAS domains in dPAS110 . In addition , we screened the five dPAS110 linker variants in fusions to nine different fluorescent protein pair variants ( Figure 1—figure supplement 1C ) in cleared cell lysates . Many of the sensor variants did not show responses to ABA addition , however FRET pair variants containing enhanced dimerization versions of the fluorescent proteins were consistently ABA responsive ( Figure 1B , Figure 1—figure supplement 1C ) . As seen before , dPAS110 with L12 and L118 linkers showed large ABA-triggered responses when fused to the edAphrodite . t9/t7 . ed . eCFP . t9 pair; notably the other three linker variants also showed ABA responses ( Figure 1—figure supplement 1C ) . Similar effects of the linker variants were obtained for edCitrine/edCerulean ( Figure 1C , Figure 1—figure supplement 1C ) . To obtain high quality data , sensor proteins were affinity purified in high throughput ( Figure 1—figure supplement 1D ) or individually ( Figure 1D ) . dPAS110 fused by linkers L52 or L118 containing the elastic GPGGA-repeat ( i . e . molecular spring motif , Grashoff et al . , 2010 ) and sandwiched by edCitrine/edCerulean yielded the highest ratio change ( Figure 1D , Figure 1—figure supplement 1D ) . dPAS110 . L52 . DR38 , a fusion protein harboring edCerulean/edCitrine ( pDR FLIP38 ) , the ABA interaction domain of a PP2CA ABA co-receptor ( ABI1aid ) , an ABA receptor ( PYL1 ) , and three distinct linkers connecting the domains ( attB1 , L52 , attB2 ) , was named Abscisic Acid Concentration and Uptake Sensor version 1 ( ABACUS1 , Figure 3A ) . In ABACUS1 , the binding of ABA to PYR/PYL/RCAR receptor proteins and the ABA-dependent interaction of PYR/PYL/RCARs with PP2CA phosphatases is co-opted to drive an ABA-dependent structural change in the fusion protein , which increases energy transfer between donor and acceptor and results in a positive ratio change . This increase in energy transfer did not occur with the metabolic precursor xanthine , other plant hormones , or other ions and metabolites tested ( Figure 3B–D ) , but was also induced by the ABA analog pyrabactin ( Figure 3E ) . The ratio change was proportional to the ABA concentration ( Figure 4 ) ; the prototype ABACUS1 is characterized by an apparent affinity of ∼80 µM for ( + ) -ABA , and thus is referred to here as ABACUS1-80µ ( Figure 4 ) . 10 . 7554/eLife . 01741 . 007Figure 3 . ABACUS1 design and fluorescence response to ABA and related compounds . ( A ) Hypothetical model of ABACUS1 bound to ABA . The structure shown for dPAS110 is derived from a crystal structure of ABA bound to PYL1 and ABI1 ( PDB: 3JRQ ) ( Miyazono et al . , 2009 ) , and the structure shown for FRET donor enhanced dimer Cerulean ( edCerulean ) and FRET acceptor enhanced dimer Citrine ( edCitrine ) are derived from a crystal structure of Aequorea victoria GFP ( PDB:1EMA ) ( Ormö et al . , 1996 ) . The structures are visualized using MacPyMol cartoon representation except for the ABA interacting tryptophan 300 of ABI1 , which is shown in line representation and ( + ) –abscisic acid ( yellow , ABA ) , which is shown in stick representation . The N-termini are colored blue and C-termini are colored red . The linkers are represented as hypothetical cartoon models not derived from known structures . The expected distance between the C-terminus of ABI1aid and the N-terminus of PYL1 ( linked by L52 in ABACUS1 ) and the C-terminus of PYL1 and the N-terminus of ABI1aid ( linked to fluorescent proteins ) , is shown in angstroms . The overall domain order of ABACUS1 is N-terminus–edCitrine–attB1–ABI1aid–L52–PYL1–attB2–edCerulean–C-terminus . ABACUS1-2µ fluorescence emission at shown wavelengths in response to ( ± ) -ABA or xanthine ( B ) , other phytohormones ( C ) , glucose and various salts ( D ) , pyrabactin ( E ) . Excitation wavelength = 428 nm . Delta ratio ( ΔR ) = treatment DxAm/DxDm/mock DxAm/DxDm . ( F ) Titration curve for ABACUS1-2µ in response to equivalent concentrations of ( + ) -ABA supplied alone or as part of a racemic mixture with ( − ) -ABA . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 00710 . 7554/eLife . 01741 . 008Figure 4 . ABA response titration curves for wild-type ABACUS1 ( ABACUS1-80µ ) and four different point mutants . DxAm/DxDm values for purified sensor samples plus ABA were normalized to corresponding mock treated samples . Kd values were determined using Prism software ( GraphPad ) and are calculated based on the ( + ) -ABA concentration since ABACUS1 is stereospecific to ( + ) -ABA ( Figure 3F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 008 Since the cytosolic ABA concentration in individual cells is unknown , and because it is important to exclude artifacts caused by other parameters , an affinity series was generated . Structural studies of the ABA receptor protein complex guided the generation of ABACUS1-80µ affinity variants ( Melcher et al . , 2009; Miyazono et al . , 2009; Nishimura et al . , 2009; Santiago et al . , 2009; Hao et al . , 2010; Melcher et al . , 2010; Peterson et al . , 2010; Mosquna et al . , 2011; Santiago et al . , 2012; Soon et al . , 2012; Sun et al . , 2012; Zhang et al . , 2012 ) . Mutation of histidine-60 to proline switches PYR1 from a dimer to a monomer and increases its affinity for ABA ( Dupeux et al . , 2011 ) ; the corresponding mutation in the PYL1 moiety of ABACUS1-80µ produced ABACUS1-2µ , a variant with a higher affinity for ABA of ∼2 µM ( Figure 4 ) . Disruption of the ABA binding pocket of PYL1 by mutation of the ABA coordinating lysine-86 to alanine abrogates ABA binding ( Melcher et al . , 2009 ) ; the corresponding mutation in ABACUS1-80µ resulted in a non-responsive variant ABACUS1-nr ( Figure 4 ) . Mutation of ABI1 tryptophan-300 to alanine abrogates its interaction with ABA and the PYL1 binding pocket ( Melcher et al . , 2009 ) ; the respective mutation in ABACUS1-80µ ( ABACUS1-W300A ) showed a reduced ratio change in response to ABA , but surprisingly no reduction in affinity ( Figure 4 ) . This finding indicates that ABA binding to PYL1 determines the apparent Kd , whereas the formation of the ABA-PYL1-ABI1 ternary complex is required for a maximal ratio change in ABACUS1-80µ . Finally , a histidine-142 to alanine mutation in the ABA binding pocket of PYL1 , which is known to reduce ABA affinity ( Melcher et al . , 2009 ) , resulted in a drastically reduced affinity in ABACUS1-H142A ( Figure 4 ) . The linear detection range of FRET sensors with a single binding site covers approximately two orders of magnitude; ABACUS1-2µ and ABACUS1-80µ together cover an overlapping range from ∼200 nM to 800 µM ( Figure 4 ) . Further analyses focused on the use of these sensors . However , a more detailed analysis of the suite of sensors identified in the screen will allow for rapid optimization and expansion of the ABACUS toolbox , for example searches for combinations or variants with higher affinities and optimization against potential side effects of expression of the sensors on physiology . Yeast two-hybrid interaction screens can be used as an indirect way of monitoring ABA transport . When supplied exogenously with ≥10 µM ABA , the ABA uptake activity ( through endogenous transporters ) of yeast reaches high enough levels to allow detection of ABA-mediated yeast two-hybrid interactions between the ABA co-receptor components PYR/PYL/RCAR and PP2CAs ( Park et al . , 2009 ) . At reduced ABA levels , the ABA-mediated interaction is not detectable ( Kanno et al . , 2012 ) . A screen at low ABA levels had successfully been used to identify members of the nitrate/peptide transporter superfamily as putative ABA importers ( Kanno et al . , 2012 ) . Here we directly tested the ability of an ABA transporter to mediate ABA accumulation in yeast cells using ABACUS1 sensors . Yeast cells did not accumulate sufficient ABA to induce a ratio change in ABACUS1-2µ even at 20 µM exogenous ( ± ) -ABA ( Figure 5 ) . However , expression of the high affinity ABA importer AIT1/NRT1 . 2 conferred ABA transport activity as detected by DxAm/DxDm ratio change of ABACUS1-2µ and ABACUS1-80µ sensors ( Figure 5 ) . ABACUS1-2µ exhibited a near maximal ratio change in response to exogenous ABA supplied well below the sensor’s apparent Kd of 2 µM ( + ) -ABA , suggesting that AIT1/NRT1 . 2 concentrates ABA against an ABA gradient . 10 . 7554/eLife . 01741 . 009Figure 5 . ABA import by AIT1/NRT1 . 2 in protease deficient yeast expressing ABACUS1 variants . ABA was added to yeast cells suspended in 20 mM MES buffer ( pH4 . 7 ) and ABACUS1 fluorescence was measured after 5 min . DxAm/DxDm values for yeast samples plus ABA were normalized to corresponding mock treated samples . EV = empty vector control . Because ABACUS1 is stereospecific for ( + ) -ABA the effective ABA concentration available for sensing is ½ the ( ± ) -ABA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 009 FRET sensor expression under control of the CaMV 35S promoter has been shown to be subject to gene silencing ( Deuschle et al . , 2006; Krebs et al . , 2012 ) . Sensor silencing was reduced in the rdr6-11 gene silencing mutant of Arabidopsis ( Peragine et al . , 2004; Deuschle et al . , 2006 ) . When FRET calcium sensors were expressed from the UBQ10 promoter , silencing in wild type ( Col0 ) plants was reduced ( Krebs et al . , 2012 ) . Therefore , ABACUS1 was expressed under control of the UBQ10 promoter in both wild type and rdr6-11 mutants of Arabidopsis ( Supplementary file 1 ) . However , as observed for other FRET sensors ( Deuschle et al . , 2006; Chaudhuri et al . , 2008; Lanquar et al . , 2014 ) , significant levels of ABACUS1 fluorescence were detectable only in the rdr6 silencing mutant . Confocal microscopy showed bright fluorescence from the sensors , which lacked subcellular targeting sequences , in the cytosol of roots , hypocotyls and leaves ( root cell localization shown in Figure 6A ) . Seedlings of lines expressing ABACUS1 affinity variants grown on ½ × MS medium agar plates or on soil were phenotypically indistinguishable from each other and from untransformed controls ( Figure 7B , Figure 7—figure supplement 1 ) . Primary root growth of ABACUS1 lines was not hypersensitive to inhibition by salt or osmotic stress ( Figure 7C ) . In several experiments , ABACUS1-2µ lines showed increased anthocyanin accumulation in cotyledons under high salt conditions . However , primary root growth and germination of ABACUS1 lines was hypersensitive to inhibition by exogenous ABA , and hypersensitivity correlated with the affinity of the PYL1 domain of the biosensors ( Figure 7C–E ) . For example , at 20 µM exogenous ( ± ) -ABA , root growth of ABACUS1-2µ sensor lines was almost completely inhibited , while root growth of ABACUS1-80µ plants was strongly inhibited at 100 µM exogenous ( ± ) -ABA . For comparison , the ABA root inhibition phenotype of a PP2CA triple mutant ( abi1-2 , hab1-1 , pp2ca-1; Rubio et al . , 2009 ) , which is also hypersensitive to ABA , was similar to ABACUS1-80µ plants . The ABA hypersensitivity of ABACUS1 plants indicates that the PYL1 domain of ABACUS1 can inhibit endogenous PP2C co-receptor proteins as overexpression of PYR/PYL/RCAR proteins also results in ABA hypersensitivity phenotypes ( Mosquna et al . , 2011; Pizzio et al . , 2013 ) . These potential side effects can likely be overcome by making use of the suite of sensors developed in the sensor screen , specifically sensors that have a higher affinity and contain a less truncated PP2CA polypeptide ( e . g . , with dPAS20 or dPAS98 sensory domains ) , as well as variants that make use of other PYR/PYL/RCARs such as PYR1 in the dPAS109 sensory domain . Moreover , the existing sensors can be improved by mutagenesis of interaction domains for endogenous factors , as demonstrated by redesigning the binding interface of calmodulin and the calmodulin-binding peptide to produce calcium sensors that do not interact with endogenous calmodulin ( Palmer et al . , 2004 ) . For example , mutation of S112 in the ABACUS PYL1 domain would likely reduce interaction with endogenous PP2CAs without affecting interaction with the ABI1aid . 10 . 7554/eLife . 01741 . 010Figure 6 . Expression pattern and ABA responses of ABACUS1 in Arabidopsis roots . ( A ) Expression pattern and cytosolic localization of ABACUS1-2µ fluorescence ( AxAm ) in root cells in the root tip ( left ) , in the elongation zone ( center ) , and in the differentiation zone ( right ) . ( B ) ABA titrations of ABACUS1 in roots . Traces show ratio ( DxAm/DxDm ) changes of ABACUS1-2µ , ABACUS1-80µ and ABACUS1-nr roots in response to six 15-min pulses with increasing concentration of ( ± ) -ABA . ( ± ) -ABA pulses were raised in 5 × increments , from 0 . 2 µM to 625 µM . Inset shows the ABACUS1-2µ root at time zero with the region used for measurements outlined in yellow . ABA pulses are shown as grey areas and all ratios were normalized to time point 0 . ( C ) Spatial distribution of ABA mediated responses . Ratio images showing pattern of ABACUS1-2µ in response to six ( ± ) -ABA pulses as described above . Right: look up table used for false coloring of ratio images . Because ABACUS1 is stereospecific for ( + ) -ABA the effective ABA concentration available for sensing is ½ the ( ± ) -ABA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 01010 . 7554/eLife . 01741 . 011Figure 6—figure supplement 1 . ABA titrations of ABACUS1 roots from Figure 6B . Traces show FRET ( DxAm ) , CFP ( DxDm ) , and YFP ( AxAm ) fluorescence intensities of ABACUS1-2µ , ABACUS1-80µ and ABACUS1-nr roots in response to six 15-min pulses with increasing concentration of ( ± ) -ABA . ( ± ) -ABA pulses were raised in 5 × increments , from 0 . 2 µM to 625 µM . ( ± ) -ABA pulses are shown as grey areas and all ratios were normalized to time point 0 . Because ABACUS1 is stereospecific for ( + ) -ABA the effective ABA concentration available for sensing is ½ the ( ± ) -ABA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 01110 . 7554/eLife . 01741 . 012Figure 7 . ABA responses and plant growth of ABACUS1 plant lines ( homozygous , T3 generation ) . ( A ) Fluorescence response of ABACUS1 in roots to 5 µM ( ± ) -ABA . All lines were imaged simultaneously while growing in the same RootChip16 . Because ABACUS1 is stereospecific for ( + ) -ABA , the effective ABA concentration available for sensing is ½ the ( ± ) -ABA concentration . ( B ) 3 week old ABACUS1 plant lines ( homozygous T3 ) grown on soil in a long day light chamber . ( C ) Primary root growth of vertically grown Arabidopsis seedlings 4 days after transfer to ½ × MS agar medium plates and plates supplemented with 100 mM NaCl , 200 mM sorbitol or 20 µM ( ± ) -ABA . Values are normalized to mock treated roots and are presented as means and standard deviations of ∼15 roots . ( D ) Seed germination rate of Arabidopsis seeds sown on ½ × MS agar medium plates supplemented with 2 µM ( ± ) -ABA . Germination percentage is presented as mean and standard error of the mean of three independent experiments of 100–200 seeds each . ( E ) Primary root growth of vertically grown Arabidopsis seedlings 4 days after transfer to ½ × MS agar medium plates with increasing concentrations of ( ± ) -ABA . Root lengths are presented as means and standard deviations of ∼15 roots . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 01210 . 7554/eLife . 01741 . 013Figure 7—figure supplement 1:Expression and Plant growth of ABACUS1 plant lines ( homozygous T3 generation ) . ( A ) Top: 10 day old seedlings grown vertically on ½ × MS medium agar plates . Bottom: Fluorescence image of the same seedlings . ( B ) 3 week old ABACUS1 plant lines ( homozygous T3 ) grown on soil in a long day light chamber . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 013 Three ABACUS1 sensors were used to measure cytosolic ABA accumulation in response to addition of exogenous ABA to roots growing in the RootChip ( Grossmann et al . , 2011 , 2012 ) ( Figure 7A ) . The RootChip is a microfluidic device that permits environmental manipulations during continuous imaging of growing Arabidopsis roots . To analyze multiple treatments in different ABACUS1 lines simultaneously—required for side-by-side comparison of ABACUS1 behavior in different conditions–we used the second generation RootChip16 , in which the chip architecture was modified to accommodate 16 individual roots in 14 mm long observation chambers , and which allows perfusion with up to seven different solutions per chamber or parallel perfusion of 2 × 8 roots with two different solutions ( Figure 8 ) . 10 . 7554/eLife . 01741 . 014Figure 8 . Architecture of the RootChip16 . ( A ) Photograph of the RootChip16 with control layer microchannels and micromechanical valves stained in blue , flow layer with observation chambers in green , and 5-days old seedlings growing on medium-filled , cut pipette tips . ( B ) The device design accommodates 16 individual roots in 14 mm long observation chambers and features 6 solution inlets that address all chambers plus 2 inlets ( asterisks ) that allow perfusion of each half of the chip with two different solutions . To aid quick identification of chip features during imaging , labels are embossed as part of the control layer and visible under the microscope . Chambers are labeled with letters A–P and inlets are labeled with roman ( solution inlets ) or arabic numbers ( valve microchannels ) . On the left of each chamber the combination of valves is visible that is required to direct the flow to the respective chamber . Right: for comparison , the RootChip design is drawn to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 014 Time-courses under perfusion with pulses of increasing ABA concentration revealed that ABACUS1-80µ and ABACUS1-2µ responded with positive ratio changes—up to ∼1 . 3 for ABACUS1-2µ–to exogenous ABA in a reversible and concentration-dependent manner ( Figure 6B , C ) , indicating that basal endogenous ABA levels did not saturate the sensors . No dose dependent positive ratio change was observed in plants expressing the control sensor ABACUS1-nr ( Figure 6B ) . We observed that sensor brightness increased by about twofold within 6 hr after exposure of roots to ABA ( Figure 6—figure supplement 1 ) . Because sensors were expressed from the constitutive UBQ10 promoter , the most parsimonious interpretation is that sensor transcripts or polypeptides are stabilized by ABA . It is conceivable that PYL1 and/or the ABA interaction domain of ABI1 ( ABI1aid ) are subject to ABA-controlled turnover . Identification of the underlying signaling mechanism could guide the engineering of ABACUS sensors with increased stability of expression and potentially be exploited to engineer ABA signaling sensors whose intensity is ABA responsive . Additionally , the use of alternate PYR/PYL/RCAR isoforms ( e . g . , potentially PYR1 in dPAS109 ) or targeting to compartments that are inaccessible to the degradative mechanism might eliminate interference by degradation . Given that AIT1 as well as ABCG40 , an ABC transporter identified as a putative ABA importer , function as primary or secondary active systems , one may expect ABA to accumulate in the root cell cytosol against a concentration gradient . However , assuming that the affinity of the sensor is similar under in vitro and in vivo conditions , ABA does not appear to accumulate in the cytosol above externally supplied levels . Rather , at low micromolar concentrations , cytosolic levels approximate external levels ( the response K0 . 5 is ∼4 µM ( + ) -ABA with ABACUS1-2µ , Figure 6B ) . It is , however , important to note that steady state ABA levels in the cytosol also depend on the rates of compartmentation and degradation , and therefore the ability of a transport system to accumulate ABA is necessarily underestimated . Transport might also be underestimated by about twofold if transport activity is limiting and non-stereoselective for ( + ) -ABA ( i . e . , competition from ( − ) -ABA limits apparent ( + ) -ABA import ) since roots were treated with ( + ) -ABA as part of a racemic mix , and ABACUS1 is stereoselective for ( + ) -ABA ( Figure 3F ) . It is noteworthy that ABA levels accumulated much more slowly compared to glucose ( Chaudhuri et al . , 2008 ) ; saturation of sensor responses was reached only after ∼15 min in a typical experiment compared to less than 1 min for glucose ( Grossmann et al . , 2011 ) indicating that the uptake capacity for ABA is significantly lower relative to glucose . In contrast to ABACUS1-2µ , which saturated between 12 . 5 and 62 . 5 µM external ( + ) -ABA , ABACUS1-80µ continued to respond to higher levels of ABA ( i . e . , 312 . 5 µM ( + ) -ABA , Figure 6B ) . If we assume that in vivo the two sensors retain their relative in vitro ΔDxAm/DxDm properties ( Figures 4 ) , ABACUS1-80µ reaches ∼30% of ABACUS1-2µ maximal ratio change at 312 . 5 µM external ( + ) -ABA , which would correspond to an ( + ) -ABA concentration between 12 . 5 and 62 . 5 µM ( i . e . , <<312 . 5 µM ) . Thus , at higher concentrations of ( + ) -ABA ( i . e . , >62 . 5 µM ) , cytosolic levels do not approximate external levels and the import of ABA is partly saturated . Overall , the responses of cytosolic ABACUS1-2µ and ABACUS1-80µ indicate that micromolar levels of externally supplied ABA are imported slowly and not concentrated in root cells , characteristics that are not necessarily consistent with known mechanisms of ABA import ( i . e . , ion-trap , AIT1 , ABCG40 ) . Furthermore , the pattern of ABA import in different root regions and at various developmental stages appeared similar ( Figure 6C , Figure 9A; Video 1 ) , inconsistent with the observed expression patterns of known ABA importers ( Figure 10 ) . The ABA transport characteristics detected with ABACUS sensors are best explained by the activity of a broadly expressed ABA uniporter that has still to be identified . 10 . 7554/eLife . 01741 . 015Figure 9 . Dynamics of ABA uptake and elimination in roots as measured by ABACUS1-2µ . ( A ) Spatial differences in the dynamics of ABA response . Traces showing ratio ( DxAm/DxDm ) changes in different root zones of an ABACUS1-2µ expressing root in response to three 15-min pulses of 5 µM ( ± ) -ABA . The insert shows the root at time 0 with the regions used for the different zones outlined in the respective colors . ( B ) Temporal differences in ABA response dynamics . Red traces show ratio ( DxAm/DxDm ) changes of two ABACUS1-2µ roots from an experiment where roots were exposed to either three ( Root A , upper ) or two ( Root B , lower ) consecutive pulses of 5 µM ( ± ) -ABA . The blue lines show the derivative of the respective traces ( d[DxAm/DxDm]/dt ) to aid in interpreting the dynamics of uptake and elimination . For all traces , ( ± ) -ABA pulses are shown as grey areas and all ratios were normalized to time point 0 . Because ABACUS1 is stereospecific for ( + ) -ABA , the effective ABA concentration available for sensing is ½ the ( ± ) -ABA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 01510 . 7554/eLife . 01741 . 016Video 1 . The video shows an ABACUS1-2µ Arabidopsis root , growing in a RootChip16 , exposed to pulses of increasing concentrations of ABA . The RATIO and YFP videos show the root throughout the experiment whereas the central charts show the average ratio ( in magenta ) and intensity ( in yellow ) of three regions ( ROIs shown as overlaid on the root ) . The RATIO video ( left ) shows the ratio ( DxAm/DxDm ) using the color-code shown in the calibration bar on the leftmost side . The YFP video ( right ) shows acceptor emission intensities upon acceptor excitation ( AxAm ) . Grey areas in the charts indicate ABA pulses of increasing concentrations ( 0 . 1–312 . 5 µM ( + ) -ABA , in 5x increments and supplied as racemic mix ) . The red vertical line shows position of current time in the charts . Scale bar corresponds to 100 µm and the counter in the lower left corner shows time elapsed in minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 01610 . 7554/eLife . 01741 . 019Figure 10 . Pattern of ribosome-associated transcripts for AIT1 and ABCG40 transporter genes . Data are derived from microarray studies of RNA bound to polysomes ( http://efp . ucr . edu/ ) ( Mustroph et al . , 2009 ) . Cell-type-specific expression is based on coexpression with any of the six genes whose promoters were used for driving the ribosomal affinity tag: pGL . 2 for trichomes , pCER5 for epidermis , pRBCS for mesophyll , pSULTR2 . 2 for bundle sheath , pSUC2 for companion cells and pKAT1 for guard cells . While the cell-specificity of the pSUC2 promoter is unambiguous in companion cells with leakage into the sieve elements ( Truernit and Sauer , 1995 ) , bundle sheath expression of pSULT2 . 2 is not as well documented ( Takahashi et al . , 2000 ) . Further analysis is required , therefore the representation in this figures as taken from http://efp . ucr . edu/ may not accurately reflect the cell type specificity . The expression levels are color-coded with yellow indicating low levels of expression and red corresponding to high levels of expression . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 019 ABA synthesis , degradation and transport are highly regulated , for example the gene for the key enzyme for ABA catabolism CYP707A is induced by ABA ( Kushiro et al . , 2004 ) . The kinetics of FRET sensor responses can provide insights into the relevant metabolic fluxes . When pulsing roots with identical amounts of ABA , the slopes of ABACUS1-2µ responses changed for both accumulation and elimination phases ( Figure 9A ) . The apparent rate change was further tested directly by comparing two sets of roots growing in the same RootChip , one set that received an initial pulse after 15 min , followed by two additional pulses at 75 and 135 min , and a second set that was not exposed to the first pulse ( Figure 9B ) . Plots of the first derivatives show that during the first ABA pulse , the accumulation rate of a non-pre-exposed root was nearly twice that of a root pre-exposed to ABA . This difference may be caused by ABA triggering its own elimination from the cytosol , as the increase in elimination rate that occured gradually in the minutes following the first ABA pulse largely accounted for the decrease in accumulation rate at the start of the following ABA pulse . Interestingly , different zones of the root showed different elimination rates after a first ABA pulse , but all zones examined converged upon an accelerated elimination rate by a third ABA pulse ( Figure 9A ) . This difference in initial elimination rate could largely explain the initial differences in ABACUS1-2µ responses between different zones , for example after receiving an ABA pretreatment , the root tip and elongation zone showed similar maximal ABA levels ( Figure 9A ) . Since ABA degradation rates are expected to contribute to both decreased accumulation and increased elimination velocities , our data intimate the existence of an ABA-triggered induction of ABA degradation , modification or compartmentation , thus leading to reduced cytosolic ABA levels . ABA accumulation in leaf tissues regulates foliar responses to water deficit and osmotic stresses . Cell-specific transcriptomic analyses revealed that the expression of ABA biosynthetic enzyme genes after drought stress of Arabidopsis leaves was induced first in the vascular parenchyma followed by mesophyll cells ( Endo et al . , 2008 ) . However , transcriptional and genetic evidence supports a role for guard cell autonomous ABA in reducing stomatal aperture under low humidity ( Bauer et al . , 2013 ) . Dissection of water-stressed Vicia faba leaves and quantitative ABA analyses indicated that ABA accumulates in epidermal , mesophyll and guard cells to ∼10 µM when ignoring compartmentation ( Harris et al . , 1988 ) . To investigate ABA accumulation in the cytosol of leaf cells , we measured DxAm/DxDm ratios in ABACUS1-2µ Arabidopsis leaves 24 hr after transfer of the petioles of detached leaves to medium supplemented with 150 mM NaCl . When compared to leaves transferred to a control solution , salt treated ABACUS1-2µ leaves showed increased average ratio ( Δ ratio across five experiments between ∼1 . 14 to 1 . 59 , Figure 11A ) . In comparison , transfer to ABA medium ( 100 µM ( ± ) -ABA ) resulted in a greater increase in average ratio compared to controls ( Δ ratio across five experiments between ∼1 . 30 to 1 . 95 , Figure 11A ) . To control for potential artifacts resulting from changes to the cellular environment upon salt stress on ABACUS1 ( e . g . , by increased solute concentrations ) , we also measured DxAm/DxDm ratios in ABACUS1-80µ and ABACUS1-nr leaves 24 hr after transfer to osmotic stress , ABA and control solutions ( Figure 11B , C ) . Salt stress treatments of ABACUS 1–80µ leaves also increased average ratio ( Δ ratio between 1 . 11 and 1 . 41 ) while salt effects on ABACUS1-nr average ratio were variable ( Figure 11B , C ) . In contrast , the pattern of ABACUS1 responses to ABA treatments roughly corresponded to the Kd of the respective sensors . Overall , the variability of ABACUS1 leaf ratios in response to salt stress precludes interpretation of the NaCl induced Δ ratio in ABACUS1-2µ leaves as resulting from accumulation of ABA . Use of improved experimental approaches , ideally confocal fluorescence microscopy or light sheet microscopy of intact aerial tissues , as well as further optimized ABACUS1 lines , may help in elucidating spatiotemporal patterns of ABA levels in the leaf . 10 . 7554/eLife . 01741 . 017Figure 11 . ABACUS1 response to salt and ABA treatment of detached leaves . DxAm/DxDm ratios for Arabidopsis leaves after 24 hr of suspension of the petiole in ¼ × MS medium or the same medium supplemented with 150 mM NaCl or 100 µM ( ± ) -ABA . Average ratio are shown for ( A ) ABACUS1-2µ ( B ) ABACUS1-80µ and ( C ) ABACUS1-nr leaves . Because ABACUS1 is stereospecific for ( + ) -ABA the effective ABA concentration available for sensing is ½ the ( ± ) -ABA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 017 To be able to improve analysis of spatial differences in roots as well as potential differences in ABA levels between cytosol and nuclei , ABACUS1-2µ was fused to a nuclear targeting sequence . Analysis of plant lines expressing nuclear localized ABACUS1-2µNLS showed preferential nuclear localization as compared to ABACUS1-2µ ( Figure 12A–F ) . Initial experiments in the RootChip16 demonstrated that the predominantly nuclear-localized sensor also responded to external ABA ( Figure 12G ) . The nuclear-targeted sensor lines are expected to facilitate discrimination of ABA levels in neighboring cells . 10 . 7554/eLife . 01741 . 018Figure 12 . ABA response of nuclear localized ABACUS . ( A–F ) Confocal scanning microscopy images of the roots ( A and D ) , hypocotyl ( B and E ) and region around the shoot apical meristem ( C and F ) of seedlings expressing either the cytosolic ( A–C ) or the nuclear localized ( D–E ) version of the ABACUS1-2µ sensor . From left to right the images in each panel show ( Left ) ratios ( DxDm/DxAm , calculated as the sum of DxDm divided by the sum of DxAm in order to minimize artifacts due to pixel to pixel noise in z-direction ) , ( Middle ) maximum projections of the YFP signal and ( Right ) a bright field image of the middle plane in the z-stack with the maximum projected YFP signal overlaid . The calibration bar in ratio images ( C and F ) shows the look up table used for all ratio images . The scale bar in the bright field image with overlaid YFP shows a 40 µm × 40 µm square . ( G ) Trace showing ratio ( DxAm/DxDm ) of the root tip of a seedling expressing the nuclear localized ABACUS1-2µ sensor in response to ABA . The grey areas indicate two 15-min pulses of 5 µM ( ± ) -ABA and the ratio is normalized to the time point 0 . ( H ) The root imaged for the trace showed in ( G ) with the ROI used for measuring outlined in yellow and a scale bar showing the distance of 100 µm . Because ABACUS1 is stereospecific for ( + ) -ABA , the effective ABA concentration available for sensing is half the ( ± ) -ABA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 01741 . 018
The phytohormone ABA serves as a master regulator of plant stress avoidance and tolerance . Due to the extent and complexity of ABA signaling and ABA-dependent physiology , a more detailed understanding of the site and timing of ABA accumulation is critical . Here we show that FRET sensors for ABA , termed ABACUS , permit the investigation of ABA dynamics with high spatial and temporal resolution . Time course analysis of ABA perfusion of growing Arabidopsis roots expressing ABACUS1 reveals broad , concentration-dependent , and reversible cytosolic and nuclear accumulation of ABA , and indicates that ABA treatment accelerates ABA metabolism . The engineering of the ABACUS FRET sensors was a partially empirical process that required the iterative expression and testing of over 100 fusion proteins consisting of several variable component domains ( i . e . , a FRET pair of fluorescent proteins , ABA binding domain ( s ) , and variable linker regions ) . The first highly ABA responsive ABACUS consists of edeCFP as the FRET donor , edAFP as the FRET acceptor , and a composite ABA sensory domain ( dPAS110 ) consisting of a truncated ABI1 co-receptor ( limited to ABA interaction domain ) linked to the receptor PYL1 , Figure 3 ) . This specific composite ABA sensory domain ( ABI1aid , flexible 12 amino acid linker , PYL1 ) resulted in ABA sensors with higher ratio change compared to sensors with different sensory domain components ( e . g . , other PYR/PYL/RCARs or other truncations of ABI1 ) , or different sensory domain configuration ( e . g . , PYL1 , flexible 12 amino acid linker , ABI1aid ) . Truncation of ABI1 to a hypothetical minimal ABA interaction domain was selected as the 49 amino acid domain ( H279–D327 ) that is flanked by , but is structurally distinct from , two parallel arrays of β-sheets that form the PP2CA catalytic domain ( Miyazono et al . , 2009 ) . In addition to yielding large ratio changes relative to other dPAS component PP2CA domains , this truncation was deemed advantageous for its small size and for its presumed lack of PP2CA activity . Initial ABACUS designs were further optimized through screening additional FRET pairs and linkers . The screen identified fluorescent proteins with enhanced dimerization tendency as the most likely to produce highly responsive ABA sensors ( i . e . , large ABA-dependent ratio changes ) . This modification had proven effective for increasing the FRET ratio change , and thus signal-to-noise ratio , in other sensors ( Vinkenborg et al . , 2007 , 2009 ) . The L52 linker consisting of an elastic GPGGA repeat ( Grashoff et al . , 2010 ) showed improved response to ABA in combination with multiple fluorescent protein pair combinations ( Figure 1 , Figure 1—figure supplement 1 ) . The enhanced ratio change observed for the increased dimerization fluorescent proteins and spring linker could both be attributed to an increase in the change of the relative fluorophore positioning between unbound and ABA-bound states according to the following model . The tension provided by the spring linker could enhance an ‘open’ state in which ABI1aid and PYL1 domains are apart , thereby reducing the FRET efficiency of the fluorophore pair . Upon ABA binding , ABI1aid and PYL1 domains interact and effect a ‘closed’ state , bringing the fluorophores into closer proximity and increasing FRET efficiency . Upon closure , fluorescent proteins with enhanced dimerization tendencies will enhance the closed state of the sensor through dimerization . Furthermore , the parallel orientation of the heterodimer state is expected to result in high FRET efficiency ( Figure 3A ) . Verification of this model in which linker tension promotes an open state with lower FRET efficiency and fluorescent protein dimerization promotes a closed state with high FRET efficiency will require structural analyses . The ABACUS optimization process was accelerated using a cloning , expression and purification platform that is readily applicable to the engineering of FRET sensors for other small molecules or molecular events ( Table 1; Figure 1 ) . For example , the platform accelerated the development of the NiTrac nitrogen transensors ( Ho and Frommer , 2014 ) . A series of ABACUS sensors with differing affinities for ABA was generated through site directed mutagenesis of residues critical for ABA binding ( Figure 4 ) . The apparent Kd of 80 µM for the ABACUS1-80µ sensor and 2 µM for the ABACUS1-2µ sensor corresponds well with the Kd measured in vitro of the PYL1 components of their respective ABA sensory domains ( i . e . , PYL1 and PYL1H87P , Dupeux et al . , 2011 ) . However , ABA co-receptor complexes between a PP2CA and a PYR/PYL/RCAR can have >10-fold higher affinity according to a model derived from in vitro measurements ( Dupeux et al . , 2011 ) . Because the W300A mutation of the ABI1aid in ABACUS1-80µ did not affect the affinity ( Figure 4 ) , it is likely that , in contrast to ABI1 , the ABI1aid does not increase PYL1 affinity for ABA . Thus , biosensors with less truncated PP2CA domains , for example in the dPAS20 and dPAS98 sensory domains ( Figure 1 , Figure 1—figure supplement 1A , B ) , could have higher affinities for ABA and be highly useful for measurements of lower ABA concentrations . Furthermore , the ABA sensors developed in parallel by Waadt et al . ( 2014 ) are also be complementary to the ones described here . ABACUS1-80µ , ABACUS1-2µ and ABACUS1-nr were stably expressed in Arabidopsis plants and these plants were used to investigate the dynamics of cytosolic and nuclear ABA levels . It is noteworthy that these first generation ABA sensors have two potential drawbacks , that is their overexpression leads to ABA hypersensitivity and the sensor amount in roots was affected by ABA . Next generation ABACUS variants carrying mutations with reduced potential to interact with endogenous ABA signaling components and variants that are unaffected in sensor synthesis/turnover will be required for more detailed physiological analyses . In experiments carried out in the RootChip , the transport of exogenous ABA into the cytosol of Arabidopsis root cells , as determined using ABACUS responses , occurred in all root zones measured ( Figure 6C , Figure 9A; Video 1 ) . We did not observe large differences in ABA import across tissues , as described by Astle and Rubery ( Astle and Rubery , 1980 , 1983 ) , potentially because sensor measurements can accurately reflect cytosolic concentrations and are not susceptible to artifacts derived from cell to cell variation regarding the proportion of cytosol per gram fresh weight ( e . g . , variable relative volume of cytosol to vacuole ) . ABA is a weak acid with a pKa of 4 . 7 , and diffusion of the protonated lipophilic species from the acidic apoplasm to the alkaline cytosol had been suggested as a dominant transport mechanism ( i . e . , ion-trap model , Kaiser and Hartung , 1981; Daie et al . , 1984 ) . However , the finding that ABA transport into Arabidopsis root cells is not concentrative and relatively slow suggests that diffusion of the protonated lipophilic species is not the dominant transport mechanism in Arabidopsis roots under these growth conditions . ABACUS responses in root cells indicate that cytosolic ABA concentrations are at or below the concentration of ABA applied exogenously , rather than higher than outside as expected if transport were energy-driven . The ion-trap model is also not consistent with the ABA transport properties measured in yeast cells expressing ABACUS1-2µ , since ABA applied to yeast cells did not result in a ratio change of ABACUS1-2µ expressed in the yeast cytosol , even when supplied at 20 µM ( ± ) -ABA in a solution of pH 4 . 7 ( Figure 5 ) . An ion-trap transport mechanism would be expected to result in cytosolic ABA above the apparent Kd of ABACUS1-2µ at this concentration and pH; thus the lack of ABACUS1-2µ responses indicates that yeast requires carrier-mediated uptake for importing sufficient levels of ABA for detection by ABACUS1-2µ . Several ABA transporters have recently been identified ( Boursiac et al . , 2013 ) . Using ABACUS1 , we detected concentration of ABA into yeast cells co-expressing the high affinity ABA importer AIT1/NRT1 . 2 . AIT1/NRT1 . 2 and ABCG40 are high-affinity transporters that likely use active transport mechanisms and are expressed around vascular tissues in Arabidopsis roots ( Figure 12 ) , two characteristics that do not match ABA import into the cytosol of root cells expressing ABACUS1 . The presence of a broadly expressed uniporter for ABA in the plasma membrane of Arabidopsis root cells would best explain the characteristics of ABA transport activity measured using ABACUS1 . ABA is known to impact upon its own metabolism through various feedback mechanisms . Plant cells rapidly metabolize ABA ( Daie et al . , 1984 ) and ABA treatment accelerates ABA catabolism ( Ren et al . , 2007 ) . ABA treatment also induces the expression of ABA biosynthetic enzymes , potentially as part of a priming mechanism ( Barretto et al . , 2011; Pantin et al . , 2013 ) . In Arabidopsis roots growing in the RootChip16 , ABA treatment was not required for the induction of root cell transport activity , suggesting the transporter ( s ) involved are constitutively expressed . However , ABA does affect its own metabolism and/or compartmentation in root cells as evidenced by a reduction in the apparent accumulation rate and increase in the apparent elimination rate of ABA in root cells that have experienced a prior treatment with ABA ( Figure 6 ) . This observation is consistent with the induction of CYP707A , which encodes the key enzyme for the first step in ABA degradation ( Kushiro et al . , 2004 ) . Whether the acceleration by ABA of its own elimination from the cytosol is due to increased transport out of the cytosol to other compartments , increased conversion of ABA to ABA-glucose ester ( e . g . , through activation of ABA glycosyltransferases or inhibition of β-glucosidase 1 [BG1] ) , increased activity of ABA catabolic enzymes , or some combination of these changes remains to be determined . Biosensors such as ABACUS can be targeted to sub-cellular compartments to reveal sub-cellular analyte patterning and dynamics ( Jones et al . , 2013 ) undetectable with current methods . For example , three different locations ( apoplasm , Baier et al . , 1990; endoplasmic reticulum , Lee et al . , 2006; vacuole , Xu et al . , 2012 ) have been proposed for β-glucosidase activity on ABA-glucose ester pools , thus ABACUS could be targeted to these compartments to elucidate sub-cellular ABA distribution . Additionally , we expect that ABACUS1NLS will facilitate investigation of the spatial patterning of ABA dynamics ( Figure 12 ) and that future experiments performed with ABACUS sensors can shed further light on the spatiotemporal regulation of ABA transport and metabolism .
pFLIP vectors are Gateway Destination vectors carrying fluorescent proteins of a FRET pair flanking the Gateway cassette ( chloramphenicol resistance marker and the ccdB gene between attR1 and attR2 sites ) . In pFLIP2 ( Kaper et al . , 2008 ) , the eCFP gene , Gateway cassette , and Venus gene are inserted into the multiple cloning site of a pRSET-B bacterial expression vector backbone ( Invitrogen , Grand Island , NY ) . After recombination with a ligand-binding domain from a separate Gateway Entry clone , the fusion construct contains an N-terminal eCFP , the linker encoded by the attB1 site , the ligand-binding domain , the linker encoded by the attB2 site , and a C-terminal Venus . To generate a series of pFLIPi destination vectors for expression in E . coli with different fluorescent protein combinations , the 5′ fluorescent protein genes were exchanged using the XhoI and KpnI restriction sites and the C-terminal fluorescent protein genes were exchanged using the SpeI and PspOMI sites ( Table 1; Supplementary file 1 ) . For yeast expression , a similar series of Destination vectors , termed pDR-FLIPs , was constructed by mobilizing the entire FLIP cassette from pFLIPi vectors ( containing the two fluorescent proteins and the Gateway cassette ) into a pDR196 vector ( Loqué et al . , 2007 ) modified to contain XbaI and EagI sites at the 5′ end of the multiple cloning site ( pDR196-X ) . Specifically , the pFLIPi vectors were digested with XbaI , PspOMI and the fragment containing the FLIP cassette was ligated into the XbaI and EagI sites of the pDR196-X vector . To generate Destination vectors for plant expression under the control of the UBQ10 promoter , termed pPZP-FLIPs , the pFLIPi vectors were digested with XhoI and ApaI and the fragment containing the expression cassette was ligated into the XhoI and ApaI compatible SalI and BstX1 sites of the pPZP UBQ10-Kan vector . The pPZP UBQ10–Kan vector was created by replacing the CaMV 35S promoter of pPZP312 ( Hajdukiewicz et al . , 1994 ) with a fragment containing the UBQ10 promoter and a kanamycin resistance cassette flanked by SalI and BstXI sites ( XhoI and ApaI compatible ends , respectively ) . All destination vector sequences are provided in Supplementary file 1 . Putative ligand binding domains for ABA sensors ( i . e . , potential ABA sensing domains , or PAS ) were selected from the ABA co-receptor complex components , specifically nine of the PYR/PYL/RCAR proteins and three of the PP2C clade A phosphatases ( Supplementary file 1 ) . Single domain PAS ( sPAS ) were designed and PCR amplified with primers containing attB sites for recombination into pDONR221 ( Invitrogen ) in a Gateway BP reaction ( 5′ primers had attB1 and 3′ primers had attB2 sites ) . Double domain PAS ( dPAS ) were constructed by fusing a PYR/PYL/RCAR to a PP2C via a variable linker region . Double domain dPAS were designed and amplified in two PCR steps . In the first step , N-terminal domain sequences were amplified with 5′ primers containing the attB1 site and 3′ primers containing a 30 base pair sequence for overlapping PCR . C-terminal domain sequences were amplified with 5′ primers containing the same 30 base pair sequence for overlapping PCR and 3′ primers containing the attB2 site . In the second step , two products from the first reactions were amplified together in an overlap extension PCR reaction and the resulting product was recombined into pDONR221 in a BP reaction . The 30 base pair overlap sequence contains AscI and FseI restriction sites for subsequent insertion of additional linker sequences . Primers for amplification are listed in Supplementary file 1 . After the BP reaction , the resulting PAS Entry clones were digested with AscI and FseI ( NEB , Ipswich , MA ) and four additional linker sequences were inserted . Thus , for each unique double domain sensor construct a total of five dPAS Entry clones were generated , containing sequences coding for five distinct linkers between two ABA binding domains ( Table 1 ) . The PAS Entry clone for ABACUS1 was altered using QuikChange ( Agilent , Santa Clara , CA ) site-directed mutagenesis according to manufacturer’s instructions to generate the ABACUS affinity mutants . All primers for site-directed mutagenesis are included in Supplementary file 1 . To generate a pFLIPi38 destination vector for expression of a nuclear localized ABACUS , the 5′ edCitrine of pFLIPi38 was excised using the XbaI and KpnI restriction sites and replaced with an XbaI and KpnI digested PCR product containing a 5′ XbaI site followed by a BspHI site , a sequence coding for an SV40 derived nuclear localization signal ( NLS , LQPKKKRKVGG; Kalderon et al . , 1984 ) , and edCitrine . The primers used to amplify NLS-edCitrine are provided in Supplementary file 1 . The resulting vector was termed pFLIPiNLS38 . To generate a Destination vector for plant expression of nuclear localized ABACUS under the control of the UBQ10 promoter , the pFLIPiNLS38 vector was digested with BspHI and ApaI and the fragment containing the expression cassette was ligated into the BspHI and ApaI compatible BstX1 site of the pPZP UBQ10-Kan vector . All destination vector sequences are provided in Supplementary file 1 . LR reactions were performed in 96 well format using 1 µl of pFLIP Destination vector ( ∼25 ng/µl ) , 1 µl of PAS Entry clone ( ∼25 ng/µl ) and 0 . 5 µl LR clonase II ( Invitrogen ) . Samples were incubated 1–18 hr at 25°C and then added to 50 µl of chemically competent TOP10 E . coli cells ( Invitrogen ) . Bacterial transformation was performed according to manufacturer instructions ( Invitrogen ) except that after heat shock and dilution in SOC medium , 50 µl of the cell solution was again diluted in 800 µl Luria–Bertani ( LB ) medium containing 100 µg/ml carbenicillin ( Sigma , St . Louis , MO ) and cultured overnight . These cultures were then diluted again ( 20 µl in 2000 µl ) in LB medium with carbenicillin and cultured overnight . Expression plasmids were isolated in 96-well format using the NucleoSpin 96 Plasmid kit according to manufacturer’s instructions ( Machery-Nagel , Düren , Germany ) . Saccharomyces cerevisiae strain BJ5465 ( ATCC 208289 [MATa ura3-52 trp1 leu2- Δ1 his3-Δ200 pep4::HIS3 prb1-Δ1 . 6R can1 GAL] [Jones , 1991] ) was transformed with yeast expression plasmids in 96-well format using a lithium acetate transformation protocol ( De Michele et al . , 2013 ) . Transformed yeast was plated on synthetic complete ( SC ) medium with 0 . 8% agar ( Difco BD , Franklin Lakes , NJ ) supplemented with 240 mg/l leucine , and 20 mg/l tryptophan ( SC agar +Leu , Trp ) to select for complementation of uracil auxotrophy by the URA3 marker of the pDR FLIP based expression clones . Transformants were grown in 2 × 1 ml cultures in SC medium +Leu , Trp in 96-well culture blocks ( Greiner , Monroe , NC ) for preliminary fluorescence analysis of sensor expression and for high-throughput screening of sensors in cleared cell lysates . For metal-affinity chromatography purification of sensors , yeast strains expressing sensors were grown in 30 ml cultures in SC medium +Leu , Trp in 50 ml culture tubes . Although measurement of sensors in intact E . coli ( BL21-CodonPlus-RIL ) and yeast ( 23344C [MAT α ura3] , isogenic with Σ1278b [Bechet et al . , 1970] ) cells revealed that sensors were expressed and contained both fluorophores , we observed loss of FRET and protein cleavage upon cell lysis ( Figure 2 ) . Therefore , expression in these host systems were not further pursued . Yeast cell cultures ( OD600 ∼0 . 5 ) were centrifuged at 5000×g for 7 min , washed once in 1 ml PBS buffer ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 1 . 8 mMKH2PO4 , pH7 . 4 ) , transferred to a 96-well culture block in the case of 30 ml cultures , and centrifuged a second time at 5000×g for 7 min . The cell pellets were then frozen at −80°C . Frozen cell pellets were thawed and resuspended in 1 ml PBS buffer . A 50 µl aliquot was transferred to a clear bottom microtiter plate ( Greiner ) for analysis of whole cell fluorescence . Cells were then centrifuged at 5000×g for 7 min , supernatant was discarded and 700 µl of chilled glass bead slurry ( PBS buffer , 0 . 1% Triton X-100 and 50% vol/vol 0 . 5 mm Zirconia/Silica beads [BioSpec , Bartlesville , OK] ) was added to each cell pellet . Culture blocks were then sealed with aluminum foil tape ( 3M 439 Silver , 3-1/4 in width , 3M , Saint Paul , MN ) and vortexed to resuspend cells . The block was then loaded into a MM300 mixer mill ( Retsch , Haan , Germany ) , which was run for 15 min at a frequency setting of 20 . Cell lysate was centrifuged at 5000×g for 10 min at 4°C and 200 µl supernatant was transferred to a microtiter plate and centrifuged again ( 5000×g for 5 min ) to remove remaining cell debris . The resulting supernatant was then diluted in PBS buffer for use in fluorescence analysis in clear bottom microtiter plates using a Safire fluorimeter ( Tecan , Männedorf , Germany ) . Samples were diluted in 20 mM MOPS pH 7 . 0 in order to obtain fluorescence emission at 480 nm and 530 nm between 5000 and 50 , 000 relative fluorescence units ( RFU ) with an excitation wavelength of 428 nm and a gain between 70 and 100 . For all fluorescence measurements , bandwidth was set to 12 nm , number of flashes was 10 , and integration time was 40 µs . Fluorescence readings were acquired for donor fluorophore ( eCFP excitation 428 nm , eCFP emission 485 nm , abbreviated DxDm ) , acceptor fluorophore ( eYFP excitation 500 nm , eYFP emission 535 nm , abbreviated AxAm ) and for energy transfer from donor to acceptor ( eCFP excitation 428 nm , eYFP emission 530 nm , abbreviated DxAm ) . Additionally , a fluorescence emission scan reading from 470 to 550 nm ( step size 5 nm ) with an excitation wavelength of 428 nm was acquired . Cell lysates from BJ5465 yeast containing an empty vector were used as a negative control for background subtraction . To analyze sensor response to ABA , 100 µl of sensor samples were combined with 50 µl of an ABA in water solution ( diluted from a 20 mM [±]-ABA racemic mixture in 50 mM NaOH] ) or a NaOH in water mock control . ABACUS1-2µ response to ( + ) -ABA alone was indistinguishable from ( + ) -ABA as one half of a ( ± ) -ABA racemic mixture ( Figure 3F ) . Replicates were averaged and a ratio of DxAm/DxDm was calculated as an approximation of FRET ratio . ABA-dependent ratio change was calculated as the FRET ratio with ABA over FRET ratio with mock treatment . For high-throughput/low-yield sensor purification , sensors were purified from cleared cell lysates in 96-well format according to manufacturer’s instructions using the HisPur cobalt resin system ( Pierce ( Thermo Scientific ) , Rockford , IL ) . For low-throughput/high-yield sensor purification , lysates were diluted 1:2 in 50 mM Sodium phosphate , 300 mM NaCl , 5 mM Imidazole pH 7 . 4 and then filtered through a 0 . 45 µm PES filter and bound to Poly-Prep chromatography columns ( Bio-Rad ) containing His-bind resin ( Novagen ( EMD ) , Madison , WI ) . Columns were then washed twice with 50 mM Sodium phosphate , 300 mM NaCl , 5 mM Imidazole pH 7 . 4 and eluted in 50 mM Sodium phosphate , 300 mM NaCl , 200 mM Imidazole , pH 7 . 4 . Samples were diluted and analyzed on a Tecan Safire fluorometer as described for cleared cell lysates . Determination of the apparent Kd of the ABACUS variants was performed as described previously ( Deuschle et al . , 2005a ) . Testing ABACUS response to other compounds was performed as described above for ABA except that stock solutions were prepared as follows: pyrabactin was dissolved in DMSO , indole-3-acetic acid ( IAA ) , jasmonic acid ( JA ) , kinetin ( Kin ) , salicylic acid ( SA ) and gibberellic acid A3 ( GA3 ) were dissolved in 50% ethanol , NaCl , KCl and glucose were dissolved in water . Mock solutions contained the appropriate solvent . All chemicals were from Sigma unless otherwise noted . For low-throughput sample analysis , data are reported as means and standard deviations of 3–4 replicates and each experiment was performed at least three times with similar results . Entry clones for known or potential ABA transporters ( AIT1–AT1G69850 , ABCG40–AT1G15520 ) were recombined with the pXC-DR GWY Destination vector , and the yeast strain BJ5457 ( ATCC 208282 [MATalpha ura3-52 trp1 lys2-801 leu2-Δ1 his3-Δ200 pep4::HIS3 prb1-Δ1 . 6R can1 GAL , Jones , 1991] ) was transformed with the resulting expression clones . The pXC-DR GWY vector was created with SLIC ( Li and Elledge , 2007 ) by joining the 5713 bp KpnI and SacII fragment from the pMetYC GWY vector ( Lalonde et al . , 2010 ) with a PCR product containing the PMA1 promoter fragment and gateway cassette amplified from the pDR-GWY vector ( Loqué et al . , 2007 ) . pXC-DR GWY carries a Leu2 gene for complementation of BJ5457 leucine auxotrophy . Expression of transporters in the BJ5457 haploid allowed for combinatorial mating with BJ5465 yeast strains expressing ABACUS1-80µ , ABACUS1-2µ , and ABACUS1-nr . Diploid yeast strains expressing transporters and sensors were selected on SC agar medium +Trp . Diploid yeast strains were cultured in SC medium +Trp overnight and washed twice in SC medium +Trp and then resuspended in 20 mM MES buffer , pH 4 . 7 . 135 µl of yeast samples was mixed with 15 µl of 20 mM MES pH 4 . 7 containing 10 X ( ± ) –ABA solution or appropriate NaOH mock solution for 5 min . Fluorescence readings were collected using a Tecan Safire fluorometer and analyzed as described above for cleared cell lysates except that a diploid yeast strain with pDR-GWY and pXC-DR GWY empty vectors was used as a control for background subtraction . Data are reported as means and standard deviations of 3–4 replicates and each experiment was performed at least three times with similar results . Transgenic plant lines were generated using the Agrobacterium floral dip method with the Col0 rdr6-11 silencing mutant ( Peragine et al . , 2004 ) as described previously ( Deuschle et al . , 2006 ) . Transformants were selected on agar plates containing ½ × MS medium with BASTA . Fluorescence expression of transformants on agar plates was analyzed using a FluorChem Q imager ( Alpha Innotech , San Leandro , CA ) with CY2 excitation and emission and the following settings: 12 s exposure time , normal speed , ultra resolution and level 2 noise reduction . All ABACUS plant lines are summarized in Supplementary file 1 . For root length assays , plant lines were grown vertically on ½ × MS agar medium ( ½ × MS salts [PhytoTechnology Laboratories , Shawnee Mission , KS] , 0 . 8% Agar [BD] , 2 . 56 mM MES pH 5 . 7 ) for 5 days and then transferred to ½ × MS agar medium plates or plates supplemented with 100 mM NaCl , 200 mM sorbitol or ( ± ) -ABA as indicated . For ABA treatments , 20 mM ( ± ) -ABA was prepared freshly as a stock solution in 50 mM NaOH . Mock treatment plates were made using NaOH to keep pH consistent . Root tips were marked at the time of transfer and images were acquired after 4 days growth on vertical treatment plates . Root growth after transfer were measured using FIJI ( http://fiji . sc/ ) . Each experiment included ∼15 roots per genotype and was repeated three independent times with similar results . ABA sensitivity of seed germination rate of Arabidopsis seeds was assayed after sowing on ½ × MS agar medium plates supplemented with 0 , 2 or 20 µM ( ± ) -ABA . All genotypes showed 100% germination on ½ × MS agar medium plates and 0% germination on plates supplemented with 20 µM ( ± ) -ABA . Plates were stratified for 2 days at +4°C after which they were transferred to a growth chamber with standard long day conditions ( 16 hr high light , 22°C , 70% RH ) . Seeds were considered germinated at radicle emergence and germination rate was assayed 9 days post sowing in three independent experiments of 100–200 seeds per genotype . The abi1-2 , hab1-1 , pp2ca-1 triple mutant line was a gift from Pedro Rodriguez ( Rubio et al . , 2009 ) . The RootChip16 was designed using AutoCAD software ( Autodesk ) and fabricated essentially as described for the original RootChip ( Grossmann et al . , 2011 , 2012 ) . In short , designs were reproduced onto emulsion photomasks and molds for flow and control layers were fabricated with SU-8 photoresist on 4-inch silicon wafers ( Stanford Microfluidics Foundry ) ( Anderson et al . , 2000 ) . Height of control layer features was 20 µm; height of the flow channels and root observation chambers ( both on flow layer mold ) was 20 µm and 100 µm , respectively . Both chip layers were produced by pouring polydimethylsiloxane and RTV-615 onto chlorotrimethylsilane-treated molds . The control layer was spin-coated to a height of 30 µm . Both layers were baked 1 hr at 80°C , pealed off the wafers , trimmed to size , and holes were punched with 20 gauge ( 0 . 812 mm ) for solution inlets ( flow layer ) , 14 gauge ( 1 . 628 mm ) for solution outlets . Root inlets were punched in an angle of 30° to the normal with 18 gauge ( 1 . 024 mm ) . The assembled device was postbaked over night at 80°C and additional 20-gauge holes were punched for the control lines ( control layer ) . Completed chips were plasma bonded to optical glass ( No . 1 , thickness 0 . 15 mm ) and baked for 5 min at 80°C . The chip carrier was designed using AutoCAD and printed on a ProJet 3500HD ( 3D Systems , Rock Hill , SC ) with a central aperture for the chip ( 42 × 60 mm ) . Outer dimensions were 110 × 160 mm to fit into the insert aperture of a motorized microscope stage ( ASI , Carlsbad , CA ) . The carrier further contained water reservoirs to maintain high humidity during plant growth at the microscope . Mask designs and fabrication protocols will be made available upon request . The RootChip16 device was used as described in detail previously for the RootChip ( Grossmann et al . , 2011 , 2012 ) . Briefly , the device was sterilized by UV light exposure , immersed in liquid growth media ( ¼ × MS salts , 1 . 28 mM MES pH 5 . 7 supplemented with B vitamins at final concentrations of niacin 2 . 3 µM , thiamine 1 . 5 µM , pyridoxin 1 . 2 µM and myo-inositol 0 . 3 µM ) , and air in the root observation chambers was removed by applying suction at the solution outlets using a microliter pipette . 5-day old Arabidopsis seedlings , germinated on 5 mm long cut 10 µl pipette tips filled with solidified growth medium ( ½ × MS salts , 1% Agar , 2 . 56 mM MES pH 5 . 7 supplemented with B vitamins at final concentrations of niacin 4 . 6 µM , thiamine 3 . 0 µM , pyridoxin 2 . 4 µM and myo-inositol 0 . 6 µM , were transferred onto the chip by plugging the tips into the root inlets . The chip was incubated for 24 hr under standard long-day growth conditions ( 16 hr high light , 22°C , 70% RH ) to allow roots to grow into the observation chambers . Before imaging , the RootChip16 was inserted from below into the central aperture of the carrier and fixed with tape . Pressure lines for control over micromechanical valves and flow lines for perfusion of solutions were connected . To actuate the push-up valves on the chip , a closing pressure of 15 p . s . i ( 1 . 03 bar ) was applied to the water filled control lines . Solution vials were pressurized with 5–10 p . s . i . ( 0 . 34 bar ) to drive perfusion solutions through the flow lines of the device . The RootChip 16 was perfused with low strength growth medium ( ¼ × MS salts , 1 . 28 mM MES pH 5 . 7 ) . For experiments , 10 mM ( + ) -ABA was prepared freshly as a 20 mM ( ± ) -ABA stock solution in 50 mM NaOH and diluted in low strength growth medium . Mock solutions for the different experiments were made using NaOH as to keep pH constant . Vials with solutions were connected to root chip flow lines and pressurized . Perfusion of solutions was controlled by the RootChip 16 control valves ( Grossmann et al . , 2011 , 2012 ) connected to a valve controller controlled by a LabView ( National Instruments ) program . Arabidopsis plants were grown on ½ × MS agar medium ( ½ × MS salts , 0 . 8% Agar , 2 . 56 mM MES pH 5 . 7 ) for 5 days and then transferred to soil and grown in a growth chamber ( 16 hr light , 22/18°C , 50% RH ) . Leaves 5 and 6 were harvested at 21–25 days post sowing and their petioles were placed in solution containing ¼ × MS medium ( ¼ × MS salts , 1 . 28 mM MES pH 5 . 7 ) or the same medium supplemented with 150 mM NaCl or 100 µM ( ± ) -ABA . Leaves were incubated in a growth chamber ( 16 hr high light , 22°C , 70% RH ) for 24 hr and then imaged ( 60 µm Z-stacks were acquired for each position imaged ) . Data are presented as means and standard deviations of the average ratio value for five leaves with three images/leaf . RootChip and detached leaf imaging was performed using either a 5x 0 . 12 N PLAN , 10x 0 . 40 HC PL APO , or 20x 0 . 70 HC PL APO objective ( Leica , Wetzlar , Germany ) on an inverted epifluorescence microscope ( Leica DM IRE2 ) , equipped with a motorized stage ( Scan IM 127x83; Märzhauser Wetzlar , Wetzlar , Germany ) , a Polychrome V monochromator light source ( TILL Photonics , München , Germany ) ) , and an electron multiplying charge-coupled device camera ( QuantEM:512SC; Photometrics , Tucson , AZ ) . A DualView beam splitter ( Photometrics ) containing an ET470/24m and ET535/30m filter setup allowed simultaneous imaging of donor and acceptor emissions for FRET measurements as well as acceptor excitation/emission for normalization of sensor expression levels or for detection of artifacts that may affect the acceptor fluorescent protein . Imaging data were acquired using SlideBook 5 . 0 software ( Intelligent Imaging Innovations , Denver , CO ) . Data were acquired as multi-position time series with simultaneous acquisition of FRET donor and acceptor emission under donor excitation , followed by acquisition of acceptor emission under acceptor excitation . A typical acquisition used an intensification setting of 300 , a gain setting of 3 , binning of 2 × 2 , 400 ms exposures for DxDm and DxAm , 100 ms exposure for AxAm , and imaging intervals of 1 min . Confocal images were acquired on a Leica SP5 using a 20x 0 . 70 HC PLAN APO objective . 442 nm and 514 nm lasers were used for excitation of donor and acceptor , respectively . Fluorescence emission was detected by PMT detectors , set to detect 458–482 nm for donor emission and 520 to 550 nm for acceptor emission . Image processing and analysis were performed using FIJI ( http://fiji . sc/ ) and Matlab ( MathWorks ) . To correct for movement of regions of interest due to root growth , images were registered using the MultiStackReg v1 . 4 plug-in ( Thevenaz et al . , 1998 , modified by Brad Busse ) . When needed , multi-point time series were stitched together using the grid/collection stitching plug-in ( Preibisch et al . , 2009 ) using a maximum intensity fusion method . Mean intensity values for regions of interest were calculated as follows: Background was subtracted from all measured intensities; donor ( DxDm ) and acceptor ( DxAm ) intensities under donor excitation were corrected against acceptor intensity under acceptor excitation ( AxAm ) to correct for intensity fluctuation caused by focus drift , root movement , or changing sensor protein levels ( during long-term measurements ) . Ratios of DxAm/DxDm were calculated and data were normalized to time point zero . For detached leaf imaging , 60 µm Z-stacks of DxAm and DxDm images were maximum projected and then background subtracted . Background was calculated from average DxAm and DxDm values for rdr6-11 leaves ( n = 20 ) . Average ratios of DxAm/DxDm were calculated for each image . Data are presented as means and standard deviations of five leaves with three images per leaf . For calculating ratio images of the confocal aquisitions of ABACUS1-2µ , z-stacks of the DxDm and DxAm channels were summed before taking the ratio of the summed DxAm and summed DxDm . | Plants are able to respond to detrimental changes in their environment—when , for example , water becomes scarce or the soil becomes too salty—in ways that minimize stress and damage caused by these changes . Hormones are chemicals that trigger the plant’s response under these circumstances . Abscisic acid is the hormone that regulates how plants respond to drought and salt stress , and also controls growth and development . In the past , it was possible to measure the average level of this hormone in a given tissue , but not the level in individual cells in a living plant , nor in specific compartments within a cell . Moreover , it was difficult to follow directly how abscisic acid moved between the plant cells , tissues or organs . Now , Jones et al . ( and independently Waadt et al . ) have developed tools that can measure the levels of abscisic acid within defined compartments of individual cells in living plants and in real time . The plants were genetically engineered to produce sensor proteins with two properties: they can bind to abscisic acid in a reversible manner , and they contain two ‘reporters’ that fluoresce at different wavelengths . Shining light onto the plant at a specific wavelength that is only absorbed by one of the reporters causes both of the reporters on the sensor proteins to fluoresce . However , the two reporters fluoresce differently when the sensor binds to abscisic acid . Specifically , one reporter fluoresces more and the other less . Hence , measuring the ratio of these two wavelengths in the light that is given off by the sensor proteins can be used as a measure of the concentration of abscisic acid in a plant cell . Jones et al . used a high-throughput platform to engineer five sensor proteins that detect abscisic acid over a wide range of concentrations . Using these ‘ABACUS’ sensors in living plants could track the uptake of abscisic acid into root cells , and revealed that the concentration of the hormone inside the cell stayed below the levels provided on the outside . Since known abscisic acid-transporters are capable of raising the hormone concentration inside a cell above that provided on the outside , abscisic acid transport into plant roots may occur via as-yet-undiscovered transporter proteins . Jones et al . also show that root cells rapidly eliminate abscisic acid , and that adding extra abscisic acid to the roots increases the rate of elimination within minutes . Plants were also engineered to target the sensor proteins specifically to the cell nucleus . In the future , targeting these sensors to the cell wall should allow tracking of the cell-to-cell movement of this hormone . Further aims include using ABACUS to track abscisic acid in plants undergoing stress , and to use the high-throughput platform to develop new sensors to track other hormones in living organisms ( including animals ) . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"cell",
"biology"
] | 2014 | Abscisic acid dynamics in roots detected with genetically encoded FRET sensors |
Grid cells in the brain respond when an animal occupies a periodic lattice of ‘grid fields’ during navigation . Grids are organized in modules with different periodicity . We propose that the grid system implements a hierarchical code for space that economizes the number of neurons required to encode location with a given resolution across a range equal to the largest period . This theory predicts that ( i ) grid fields should lie on a triangular lattice , ( ii ) grid scales should follow a geometric progression , ( iii ) the ratio between adjacent grid scales should be √e for idealized neurons , and lie between 1 . 4 and 1 . 7 for realistic neurons , ( iv ) the scale ratio should vary modestly within and between animals . These results explain the measured grid structure in rodents . We also predict optimal organization in one and three dimensions , the number of modules , and , with added assumptions , the ratio between grid periods and field widths .
How does the brain represent space ? Tolman ( 1948 ) suggested that the brain must have an explicit neural representation of physical space , a cognitive map , that supports higher brain functions such as navigation and path planning . The discovery of place cells in the rat hippocampus ( O'Keefe , 1976; O'Keefe and Nadel , 1978 ) suggested one potential locus for this map . Place cells have spatially localized firing fields which reorganize dramatically when the environment changes ( Leutgeb et al . , 2005 ) . Another potential locus for the cognitive map of space has been uncovered in the main input to hippocampus , a structure known as the medial entorhinal cortex ( MEC ) ( Figure 1 , Fyhn et al . , 2004; Hafting et al . , 2005 ) . When rats freely explore a two-dimensional open environment , individual ‘grid cells’ in the MEC display spatial firing fields that form a periodic triangular grid which tiles space ( Figure 1A ) . It is believed that grid fields provide relatively rigid coordinates on space based partly on self-motion and partly on environmental cues ( Moser et al . , 2008 ) . The scale of grid fields varies systematically along the dorso–ventral axis of the MEC ( Figure 1A ) ( Hafting et al . , 2005; Barry et al . , 2007; Stensola et al . , 2012 ) . Recently , it was shown that grid cells are organized in discrete modules within which cells share the same orientation and periodicity but vary randomly in phase ( Barry et al . , 2007; Stensola et al . , 2012 ) . 10 . 7554/eLife . 08362 . 003Figure 1 . Representing place in the grid system . ( A ) Grid cells ( small triangles ) in the medial entorhinal cortex ( MEC ) respond when the animal is in a triangular lattice of physical locations ( red circles ) ( Fyhn et al . , 2004; Hafting et al . , 2005 ) . The scale of periodicity ( the ‘grid scale’ , λi ) and the size of the regions evoking a response above a noise threshold ( the ‘grid field width’ , li ) vary modularly along the dorso-ventral axis of the MEC ( Hafting et al . , 2005 ) . Grid cells within a module vary in the phase of their spatial response , but share the same period and grid orientation ( in two dimensions ) ( Stensola et al . , 2012 ) . ( B ) A simplified binary grid scheme for encoding location along a linear track . At each scale ( λi ) there are two grid cells ( red vs blue firing fields ) . The periodicity and grid field widths are halved at each successive scale . ( C ) The binary scheme in ( B ) is ambiguous if the grid field width at scale i exceeds the grid periodicity at scale i + 1 . For example , if the grid fields marked in red respond at scales i and i + 1 , the animal might be in either of the two marked locations . ( D ) The grid system is composed of discrete modules , each of which contains neurons with periodic tuning curves , and varying phase , in space . ( E ) For a simple winner-take-all decoder of the grids in panel D , decoded position will be ambiguous unless li ≤ λi + 1 , analogously to panel C ( see text ) . Variants of this limitation occur in other decoding schemes . DOI: http://dx . doi . org/10 . 7554/eLife . 08362 . 003 How does the grid system represent spatial location and what function does the modular variation in grid scale serve ? Here , we propose that the grid system provides a hierarchical representation of space where fine grids provide precise location and coarse grids resolve ambiguity , and that the grids are organized to minimize the number of neurons required to achieve the behaviorally necessary spatial resolution across a spatial range equal in size to the period of the largest grid module . Our analyses thus assume that there is a behaviorally defined maximum range over which a fixed grid represents locations . Our hypotheses , together with general assumptions about tuning curve shape and decoding mechanism , explain the triangular lattice structure of two-dimensional grid cell firing maps and predict a geometric progression of grid scales . Crucially , the theory further predicts that the ratio of adjacent grid scales will be modestly variable within and between animals with a mean in the range 1 . 4–1 . 7 depending on the assumed decoding mechanism used by the brain . With additional assumptions the theory also predicts that the ratio between grid scale and individual grid field widths should lie in the same range . These predictions naturally explain the structural parameters of grid cell modules measured in rodents ( Barry et al . , 2007; Giocomo et al . , 2011a; Stensola et al . , 2012 ) . Our results follow from general principles , and thus , we expect similar organization of the grid system in other species . The theory makes further predictions including: ( a ) the number of grid scales necessary to support navigation over typical behavioral distances ( i . e . , a logarithmic relation between number of modules and navigational range ) , ( b ) possible deficits in spatial behavior that will obtain upon inactivating specific grid modules , ( c ) the structure of one- and three-dimensional grids that will be relevant to navigation in , for example , bats ( Yartsev et al . , 2011 ) , ( d ) an estimate of the number of grid cells we expect in the mEC . Remarkably , in a simple decoding scheme , the scale ratio in an n-dimensional environment is predicted to be close to en . As we will explain , our results and their apparent experimental confirmation in Stensola et al . ( 2012 ) , suggest that the grid system implements a two-dimensional neural analog of a base-b number system . This provides an intuitive and powerful metaphor for interpreting the representation of space in the entorhinal cortex .
The key features of the grid system in the MEC are schematized in Figure 1A . Grid cells are organized in modules , and cells within a module share a common lattice organization of their firing fields ( Barry et al . , 2007; Stensola et al . , 2012 ) . These lattices have periods λ1 > λ2 >⋯λm , measured as the distance between nearest neighbor firing fields . It will prove convenient to define ‘scale factors’ ri = λi/λi+1 relating the periods of adjacent scales . In each module , the grid firing fields ( i . e . , the connected spatial regions that evoke firing ) are compact ( with a diameter denoted li ) after thresholding for activity above the noise level ( see , e . g . , Hafting et al . , 2005 ) . Within any module , grid cells have a variety of spatial phases so that at least one cell will respond at any physical location ( Figure 1B , D ) . Grid modules with smaller field widths li provide more local spatial information than those with larger scales . However , this increased spatial precision comes at a cost: the correspondingly smaller periodicity λi of these modules leads to increased ambiguity since there are more grid periods within a given spatial region ( e . g . , see scale 3 in the schematic one-dimensional grid in Figure 1B , D ) . By contrast , modules with large periods and field widths have less spatial precision , but also less ambiguity ( e . g . , in scale 1 in Figure 1B the red cell has only one firing field in the environment and hence no ambiguity ) . We propose that the entorhinal cortex exploits this trade-off to implement a hierarchical representation of space where large scales resolve ambiguity and small scales provide precision . Consistently with existing data for one- and two-dimensional grids ( Barry et al . , 2007; Brun et al . , 2008; Stensola et al . , 2012 ) , we will take the largest grid period λ1 to be comparable to the range over which space is represented unambiguously by a fixed grid without remapping ( Fyhn et al . , 2007 ) . ( An alternative view , that the range might greatly exceed the largest period , is addressed in the ‘Discussion’ . ) The spatial resolution of such a grid can be measured by comparing the range of spatial representation set by the largest period λ1 to the precision ( related to the smallest grid field width lm ) to quantify how many distinct spatial ‘bins’ can be resolved . We will assume that the required resolution is set by the animal's behavioral requirements . What are the advantages of a multi-scale , hierarchical representation of physical location ? Consider an animal living in an 8 m linear track and requiring spatial precision of 1 m to support its behavior . To develop intuition , consider a simple model where location is represented in the animal's brain by reliable neurons with rectangular firing fields ( e . g . , Figure 1B ) . The animal could achieve the required resolution in a place coding scheme by having eight neurons tuned to respond when the animal is in 1 m wide , non-overlapping regions ( see [Fiete et al . , 2008] for a related comparison between grid and place cells ) . Consider an alternative , the idealized grid coding scheme in Figure 1B . Here , the two neurons at the largest scale ( λ1 ) have 4 m wide tuning curves so that their responses just indicate the left and right halves of the track . The pairs of neurons at the next two scales have grid field widths of 2 m and 1 m respectively , and proportionally shorter periodicities as well . These pairs successively localize the animal into 2 m and 1 m bins . All told only six neurons are required , less than in the place coding scheme . This suggests that grid schemes that integrate multiple scales of representation can encode space more efficiently , that is , with fewer neural resources . In the sensory periphery , there is evidence of selection for more efficient circuit architectures ( e . g . , Simoncelli and Olshausen , 2001 ) . If similar selection operates in cortex , the experimentally measured grid architecture should be predicted by maximizing the efficiency of the grid system given a behaviorally determined range and resolution . Thus , we seek to predict the key structural parameters of the grid system—the ratios ri = λi/λi+1 relating adjacent scales ( which need not be equal ) . The need to avoid spatial ambiguity constrains the ratios ri . Again in our simple model , consider Figure 1C where the cells with the grid fields marked in red respond at scales i and i + 1 . Then the animal might be in either of the two marked locations . Avoiding ambiguity requires that λi+1 , the period at scale i + 1 , must exceed li , the grid field width at scale i . Variants of this condition will recur in the more realistic models that we will consider . Theoretically , one could resolve the ambiguity in Figure 1C by combining the responses of more grid modules , provided they have mutually incommensurate periods ( Fiete et al . , 2008; Sreenivasan and Fiete , 2011 ) . However , anatomical evidence suggests that contiguous subsets of the mEC along the dorso–ventral axis project topographically to the hippocampus ( Van Strien et al . , 2009 ) . While there is evidence that hippocampal place cells are not formed and maintained by grid cell inputs alone ( Bush et al . , 2014; Sasaki et al . , 2015 ) , for each of these restricted projections to represent a well-defined spatial map , ambiguities like the one in Figure 1C should be resolved at each scale . The hierarchical position encoding schemes that we consider below embody this observation by seeking to reduce position ambiguity at each scale , given the responses at larger scales . How should the grid system be organized to minimize the resources required to represent location unambiguously with a given resolution ? Consider a one-dimensional grid system that develops when an animal runs on a linear track . As described above , the ith module is characterized by a period λi , while the ratio of adjacent periods is ri = λi/λi+1 . Within any module , grid cells have periodic , bumpy response fields with a variety of spatial phases so that at least one cell responds at any physical location ( Figure 1D ) . If d cells respond above the noise threshold at each point , the number of grid cells ni in module i will be ni = dλi/li . We will take d , the coverage factor , to be the same in each module . In terms of these parameters , the total number of grid cells is N=∑i=1m ni=∑i=1m dλili , where m is the number of grid modules . How should such a grid be organized to minimize the number of grid cells required to achieve a given spatial resolution ? The answer might depend on how the brain decodes the grid system . Hence , we will consider decoding methods at extremes of decoding complexity and show that they give similar answers for the optimal grid . How do these results extend to two dimensions ? Let λi be the distance between nearest neighbor peaks of grid fields of width li ( Figure 3 ) . Assume in addition that a given cell responds on a lattice whose vertices are located at the points λi ( nu + mv ) , where n , m are integers and u , v are linearly independent vectors generating the lattice ( Figure 3A ) . We may take u to have unit length ( |u| = 1 ) without loss of generality , however |v| ≠ 1 in general . It will prove convenient to denote the components of v parallel and perpendicular to u by v∥ and v⊥ , respectively ( Figure 3A ) . The two numbers v∥ , v⊥ quantify the geometry of the grid and are additional parameters that we may optimize over: this is a primary difference from the one-dimensional case . We will assume that v∥ and v⊥ are independent of scale; this still allows for relative rotation between grids at different scales . At each scale , grid cells have different phases so that at least one cell responds at each physical location . The minimal number of phases required to cover space is computed by dividing the area of the unit cell of the grid ( λi2||u×v|| =λi2|v⊥| ) by the area of the grid field . As in the one-dimensional case , we define a coverage factor d as the number of neurons covering each point in space , giving for the total number of neurons N=d|v⊥|∑i ( λi/li ) 2 . 10 . 7554/eLife . 08362 . 006Figure 3 . Optimizing two-dimensional grids . ( A ) A general two-dimensional lattice is parameterized by two vectors u and v and a periodicity parameter λi . Take u to be a unit vector , so that the spacing between peaks along the u direction is λi , and denote the two components of v by v∥ , v⊥ . The blue-bordered region is a fundamental domain of the lattice , the largest spatial region that can be unambiguously represented . ( B ) The two-dimensional analog of the ambiguity in Figure 1C , E for the winner-take-all decoder . If the grid fields in scale i are too close to each other relative to the size of the grid field of scale i − 1 ( i . e . , li − 1 ) , the animal might be in one of several locations . ( C ) The optimal ratio r between adjacent scales in a hierarchical grid system in two dimensions for a winner-take-all decoding model ( blue curve , WTA ) and a probabilistic decoder ( red curve ) . Nr is the number of neurons required to represent space with resolution R given a scaling ratio r , and Nmin is the number of neurons required at the optimum . In both decoding models , the ratio Nr/Nmin is independent of resolution , R . For the winner-take-all model , Nr is derived analytically , while the curve for the probabilistic model is derived numerically ( details in Optimizing the grid system: winner-take-all decoder and Optimizing the grid system: probabilistic decoder , ‘Materials and methods’ ) . The winner-take-all model predicts r=e≈1 . 65 , while the probabilistic decoder predicts r ≈ 1 . 44 . The minima of the two curves lie within each others' shallow basins , predicting that some variability of adjacent scale ratios is tolerable within and between animals . The green and blue bars represent a standard deviation of the scale ratios of the period ratios between modules measured in Barry et al . ( 2007 ) ; Stensola et al . ( 2012 ) . ( D ) Contour plot of normalized neuron number N/Nmin in the probabilistic decoder , as a function of the grid geometry parameters v⊥ , v∥ after minimizing over the scale factors for fixed resolution R . As in Figure 3C , the normalized neuron number is independent of R . The spacing between contours is 0 . 01 , and the asterisk labels the minimum at v∥=1/2 , v⊥=3/2; this corresponds to the triangular lattice . DOI: http://dx . doi . org/10 . 7554/eLife . 08362 . 006 As before , consider a situation where grid fields thresholded for noise lie completely within compact regions and assume a simple decoder which selects the most activated cell and does not take tuning curve shape into account ( Coultrip et al . , 1992; Maass , 2000; de Almeida et al . , 2009 ) . In such a model , each scale i simply serves to localize the animal within a circle of diameter li . The spatial resolution is summarized by the square of the ratio of the largest scale λ1 to the smallest scale lm: R2 = ( λ1/lm ) 2 . In terms of the scale factors r~i=λi/λi+1 , we write R2=∏i=1m r~i2 , where we also define r~m=λm/lm . To decode the position of an animal unambiguously , each cell at scale i + 1 should have at most one grid field within a region of diameter li . We therefore require that the shortest lattice vector of the grid at scale i has a length greater than li − 1 , in order to avoid ambiguity ( Figure 3B ) . We wish to minimize N , which will be convenient to express as N=d|v⊥|∑i r~i2 ( λi+1/li ) 2 . There are two kinds of contributions here to the number of neurons—the factors r~i2 are constrained by the overall resolution of the grid , while , as we will see , the combination |v⊥| ( λi + 1/li ) 2 measures a packing density of discs placed on the grid lattice . This suggests that we should separate the minimization of neuron number into first optimizing the lattice and then optimizing ratios . After doing so , we can check that the result is the global optimum . To obtain the optimal lattice geometry , we can ignore the resolution constraint , as it depends only on the scale factors and not the grid geometry . We may then exploit an equivalence between our optimization problem and the optimal circle-packing problem . To see this connection , consider placing disks of diameter li on each vertex of the grid at scale i + 1 . In order to avoid ambiguity , all points of the grid i + 1 must be separated by at least li: equivalently , the disks must not overlap . The density of disks is proportional to li2/ ( λi+12|v⊥| ) , which is proportional to the reciprocal of each term in N . Therefore , minimizing neuron number amounts to maximizing the packing density; and the no-ambiguity constraint requires that the disks do not overlap . This is the optimal circle packing problem , and its solution in two dimensions is known to be the triangular lattice ( Thue , 1892 ) , so v∥=1/2 and v⊥=3/2 . Furthermore , the grid spacing should be as small as allowed by the no-ambiguity constraint , giving λi+1=li . We have now reduced the problem to minimizing N=d32 ∑ir~i2 , over the scale factors r~i , while fixing the resolution R2 . This optimization problem is mathematically the same as in one dimension if we formally set ri≡r~i2 . This gives the optimal ratio r~i2=e for all i ( Figure 3C ) . We conclude that in two dimensions , the optimal ratio of neighboring grid periodicities is e≈1 . 65 for the simple winner-take-all decoding model , and the optimal lattice is triangular . The optimal probabilistic decoding model from above can also be extended to two dimensions with the posterior distributions P ( x|i ) becoming sums of Gaussians with peaks on the two-dimensional lattice . In analogy with the one-dimensional case , we then derive a formula for the resolution R2 = λ1/δm in terms of the standard deviation δm of the posterior given all scales . The quantity δm may be explicitly calculated as a function of the scale factors r~i and the geometric factors v∥ , v⊥ , and the minimization of neuron number may then be carried out numerically ( Optimizing the grid system: probabilistic decoder , ‘Materials and methods’ ) . In this approach , the optimal scale factor turns out to be r~i≈1 . 44 ( Figure 3C ) , and the optimal lattice is again triangular ( Figure 3D ) . Attractor network models of grid formation readily produce triangular lattices ( Burak and Fiete , 2009 ) ; our analysis suggests that this architecture is functionally beneficial in reducing the required number of neurons . Even though our two decoding strategies lie at extremes of complexity ( one relying just on the most active cell at each scale and another optimally pooling information in the grid population ) their respective ‘optimal intervals’ substantially overlap ( Figure 3C; see Figure 5B in 'Materials and methods' for the one-dimensional case ) . This indicates that our proposal is robust to variations in grid field shape and to the precise decoding algorithm ( Figure 3C ) . The scaling ratio r may lie anywhere within a basin around the optimum at the cost of a small number of additional neurons . Such considerations also suggest that these coding schemes have the capacity to tolerate developmental noise: different animals could develop grid systems with slightly different scaling ratios , without suffering a large loss in efficiency . In two dimensions , the required neuron number will be no more than 5% of the minimum if the scale factor is within the range ( 1 . 43 , 1 . 96 ) for the winner-take-all model and the range ( 1 . 28 , 1 . 66 ) for the probabilistic model . These ‘optimal intervals’ are narrower than in the one-dimensional case and have substantial overlap . In summary , for 2-d case , the theory predicts that ( 1 ) the ratios between adjacent scales should be a constant; ( 2 ) the optimal scaling constant is e≈1 . 65 in a simple WTA decoding model , and it is ≈1 . 44 in a probabilistic decoding model; ( 3 ) the predictions for the optimal grid field width depends on the specific decoding method , ( 4 ) The grid lattice should be a triangular lattice . Our predictions agree with experiment ( Barry et al . , 2007; Giocomo et al . , 2011a; Stensola et al . , 2012 ) ( see Reanalysis of grid data from previous studies , ‘Materials and methods’ for details of the data re-analysis ) . Specifically , Barry et al . ( 2007 ) ( Figure 4A ) reported the grid periodicities measured at three locations along the dorso–ventral axis of the MEC in rats and found ratios of ∼1 , ∼1 . 7 and ∼2 . 5 ≈ 1 . 6 × 1 . 6 relative to the smallest period ( Barry et al . , 2007 ) . The ratios of adjacent scales reported in Barry et al . ( 2007 ) had a mean of 1 . 64 ± 0 . 09 ( mean ± std . dev . , n = 6 ) , which almost precisely matches the mean scale factor of e predicted from the winner-take-all decoding model , and is also consistent with the probabilistic decoding model . In another study ( Krupic et al . , 2012 ) , the scale ratio between the two smaller grid scales , measured by the ratio between the grid frequencies , is reported to be ∼1 . 57 in one animal . Recent analysis based on a larger data set ( Stensola et al . , 2012 ) confirms the geometric progression of the grid scales in individual animals over four modules . The mean ratio between adjacent scales is 1 . 42 ± 0 . 17 ( mean ± std . dev . , n = 24 ) in that data set , accompanied by modest variability within and between animals . These measurements again match both our models ( Figure 4A ) . 10 . 7554/eLife . 08362 . 007Figure 4 . Comparison with experiment . ( A ) Our models predict grid scaling ratios that are consistent with experiment . ‘WTA’ ( winner-take-all ) and ‘probabilistic’ represent predictions from two decoding models; the dot is the scaling ratio minimizing the number of neurons , and the bars represent the interval within which the neuron number will be no more than 5% higher than the minimum . For the experimental data , the dot represents the mean measured scale ratio , and the error bars represent ± one standard deviation . Data were replotted from Barry et al . ( 2007 ) ; Stensola et al . ( 2012 ) . The dashed red line shows a consensus value running through the two theoretical predictions and the two experimental datasets . ( B ) The mean ratio between grid periodicity ( λi ) and the diameter of grid fields ( li ) in mice ( data from Giocomo et al . , 2011a ) . Error bars indicate ± one S . E . M . For both wild-type mice and HCN knockouts ( which have larger grid periodicities ) , the ratio is consistent with e ( dashed red line ) . ( C ) The response lattice of grid cells in rats forms an equilateral triangular lattice with 60° angles between adjacent lattice edges ( replotted from Hafting et al . , 2005 , n = 45 neurons from six rats ) . Dots represent outliers , as reported in Hafting et al . ( 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08362 . 007 The optimal grid was triangular in both of our models , again matching measurements ( Figure 4C ) ( Hafting et al . , 2005; Moser et al . , 2008; Stensola et al . , 2012 ) . However , the minimum in Figure 3D is relatively shallow—the contour lines indicating equally efficient grids are widely spaced near the minimum . This leads us hypothesize that the measured grid geometries will be modestly variable around the triangular lattice , as reported in Stensola et al . ( 2012 ) . A recent study measured the ratio between grid periodicity and grid field size to be 1 . 63 ± 0 . 035 ( mean ± S . E . M . , n = 48 ) in wild-type mice ( Giocomo et al . , 2011a ) . This ratio was unchanged , 1 . 66 ± 0 . 03 ( mean ± S . E . M . , n = 86 ) , in HCN1 knockout strains whose absolute grid periodicities increased relative to the wild type ( Giocomo et al . , 2011a ) . Such measurements are consistent with the prediction of the simple winner-take-all model , which predicts a ratio between grid period and grid field width of λi/li=e≈1 . 65 ( Figure 4B ) .
We have shown that a grid system with a discrete set of periodicities , as found in the entorhinal cortex , should use a common scale factor r between modules to represent spatial location with the fewest neurons . In other words , the periods of grid modules should be organized in a geometric progression . In one dimension , this organization may be thought of intuitively as implementing a neural analog of a base-b number system . Roughly , the largest scale localizes the animal into a coarse region of the environment and finer scales successively subdivide the region into b ‘bins’ . For example , suppose that the largest scale has one firing field in the environment and that b = 2 , so that subsequent scales subdivide this firing field into halves ( Figure 1B ) . Then , keeping track of which half the animal occupies at each scale gives a binary encoding of location . This is just like a binary number system being used to encode a number representing the location . Our problem of minimizing neuron number while fixing resolution is analogous to minimizing the product of the number of digits and the number of decimal places ( which we can term complexity ) needed to represent a given range R of integers in a base-b number system . The complexity is approximately C ∼ b logb R . What ‘base’ minimizes the complexity of the representation ? We can compute this by evaluating the extremum ∂C/∂b=0 and find that the optimum is at b = e ( details in Optimizing a ‘base-b’ representation of one-dimensional space , ‘Materials and methods’ ) . Our full theory is a generalization of this simple fixed-base representational scheme for numbers to noisy neurons encoding two-dimensional location . It is remarkable that natural selection seems to have reached such efficient solutions for encoding location . Our theory quantitatively predicted the ratios of adjacent scales within the variability tolerated by the models and by the data ( Figure 4 ) . Further tests of our theory are possible . For example , a direct generalization of our reasoning says that in n-dimensions the optimal ratio between grid scales for winner-take-all decoding is en ( as compared to e in two dimensions ) . The three-dimensional case is possibly relevant to the grid system in , for example , bats ( Yartsev et al . , 2011; Yartsev and Ulanovsky , 2013 ) . Robustly , for any given decoding scheme , our theory would predict a smaller scaling ratio for 3d grids than for 2d grids . The packing density argument given above for two-dimensional lattice structure , when generalized to three dimensions , would predict a face center cubic lattice or hexagonal close packing , which share the highest packing density . Bats are known to have 2d grids when crawling on surfaces ( Yartsev et al . , 2011 ) and if they also have a 3d grid system when flying , similar to their place cell system ( Yartsev and Ulanovsky , 2013 ) , our predictions for three-dimensional grids can be directly tested . In general , the theory can be tested by comprehensive population recordings of grid cells along the dorso–ventral axis for animals moving in one- , two- , and three-dimensional environments . Our theory also predicts a logarithmic relationship between the natural behavioral range and the number of grid modules . To estimate the number of modules , m , required for a given resolution R2 via the approximate relationship m=logR2/logr~2 . Assuming that the animal must be able to represent an environment of area ∼ ( 10 m ) 2 ( e . g . , Davis et al . , 1948 ) , with a positional accuracy on the scale of the rat's body size , ∼ ( 10 cm ) 2 , we get a resolution of R2 ∼ 104 . Together with the predicted two-dimensional scale factor r~ , this gives m ≈ 10 as an order-of-magnitude estimate . Indeed , in Stensola et al . ( 2012 ) , 4–5 modules were discovered in recordings spanning up to 50% of the dorsoventral extent of MEC; extrapolation gives a total module number consistent with our estimate . How many grid cells do we predict in total ? Consider the simplest case where grid cells are independent encoders of position in two dimensions . Our likelihood analysis ( details in Optimizing the grid system: probabilistic decoder , ‘Materials and methods’ ) gives the number of neurons as N = mc ( λ/σ ) 2 , where m is the number of modules and c is constant . In detail , c is determined by factors like the tuning curve shape of individual neurons and their firing rates , but broadly what matters is the typical number of spikes K that a neuron emits during a sampling time , because this will control the precision with which location can be inferred from a single cell's response . General considerations ( Dayan and Abbott , 2001 ) indicate that c will be proportional to 1/K . We can estimate that if a rat runs at ∼50 cm/s and covers ∼1 cm in a sampling time , then a grid cell firing at 10 Hz ( Stensola et al . , 2012 ) gives K ∼ 1/5 . Using our prediction that the number of modules will be ∼10 and that λ/σ ≈ 5 . 3 in the optimal grid ( see Optimizing the grid system: probabilistic decoder , ‘Materials and methods’ ) , we get Nest ≈ 1400 . This estimate assumed independent neurons and that the decoder of the grid system will efficiently use all the information in every grid cell's response . This is unlikely to be the case . Given homogeneous noise correlations within a grid module , which will arise naturally if grid cells are formed by an attractor mechanism , the required number of neurons could be an order of magnitude higher ( Sompolinsky et al . , 2001; Averbeck et al . , 2006 ) . ( Noise correlation between grid cells was investigated in Mathis et al . ( 2013 ) ; Dunn et al . ( 2015 ) —they found positive correlation between aligned grids of similar periods and some evidence for weak negative correlation for grids differing in phase . ) Thus , in round numbers , we estimate that our theory requires something in the range of ∼1400–14000 grid cells . Are there so many grid cells in the MEC ? In fact , we need this number of grid cells separately in layer II and layer III of the MEC since these regions likely maintain separate grid codes . ( To see this , recall that layers II and III project largely to the dentate gyrus and CA1 , respectively [Steward and Scoville , 1976; Dolorfo and Amaral , 1998] , while the place map in CA1 survives lesions of the dentate input to CA1 via CA3 [Brun et al . , 2002] . ) Physiological studies ( Sargolini et al . , 2006 ) have shown that only about 10% of the cells in MEC are layer II grid cells and another 10% are layer III grid cells . Cells that have weak responsiveness during spatial tasks are probably undersampled in such experiments and so the real proportion of grid cells is likely to be somewhat smaller . Other studies ( Mulders et al . , 1997 ) have shown that MEC has ∼105 neurons . Thus , we can estimate that layer II and layer III each contain something in the range of 5000–10000 grid cells . This is well within the predicted theoretical range . Our analysis assumed that the grid code is hierarchical , with large grids resolving the spatial ambiguity created by the multiple firing fields of the small grids that deliver precision of location . Recall that place cells are thought to provide one readout of the grid system . Anatomical evidence ( Van Strien et al . , 2009 ) shows that the projections from the mEC to the hippocampus are restricted along the dorso-ventral axis , so that a given place cell receives input from perhaps a quarter of the mEC . The data of Stensola et al . ( 2012 ) show additionally that the dorsal mEC is impoverished in large grid modules . If place cells were formed from grids via summation as in the model of ( Solstad et al . , 2006 ) , the anatomy ( Van Strien et al . , 2009 ) and the hierarchical view of location coding that we have proposed would together predict that dorsal place cells should be revealed to have multiple place fields in large environments because their spatial ambiguities will not be fully resolved at larger scales . Preliminary evidence for such a multiplicity of dorsal place fields appears in Fenton et al . ( 2008 ) ; Rich et al . ( 2014 ) . However , a naive model where place cells are sums of grid cells would also suggest that the multiple place fields would be arranged in an orderly , possibly periodic , manner . To the contrary , the data ( Fenton et al . , 2008; Rich et al . , 2014 ) show that the multiple place fields of dorsal hippocampal cells are organized in a disorderly fashion . On the other hand , real grid fields show significant variability in period , orientation , and ellipticity even within a module ( Stensola et al . , 2012 ) —this variability would disorder any linearly summed place fields , changing the prediction of the naive model . We have not attempted to investigate this in detail because there is also significant evidence ( summarized in Bush et al . , 2014; Sasaki et al . , 2015 ) that place cells are not formed and maintained via simple summation of grid cells alone , although they are influenced by them . It would be interesting for future work to integrate the accumulating information about the complex interplay between the hippocampus and the mEC to better understand the consequences of hierarchical grid organization for the hippocampal place system . We assumed that the largest scales of grid modules should be roughly comparable to the behavioral range of the animal . This is consistent with the existing data on grid modules ( Stensola et al . , 2012 ) and with measurements in the largest environments tested so far ( Brun et al . , 2008 ) ( periods at least as large as 10 m in an 18 m track ) . To accommodate very large environments , grids could either increase their scale ( as reported at least transiently in Barry et al . , 2007; Stensola et al . , 2012 ) or could segment the environment into large sections ( Derdikman et al . , 2009; Derdikman and Moser , 2010 ) across which remapping occurs ( Fyhn et al . , 2007 ) . These predictions can be tested in detail by exploring spatial coding in natural environments of behaviorally appropriate size and complexity . In fact , ethological studies have indicated a typical homing rate of a few tens of meters for rats with significant variation between strains ( Davis et al . , 1948; Fitch , 1948; Stickel and Stickel , 1949; Slade and Swihart , 1983; Braun , 1985 ) . Our theory predicts that the period of the largest grid module and the number of modules will be correlated with homing range . In our theory , we took the coverage factor d ( the number of grid fields overlapping a given point in space ) to be the same for each module . In fact , experimental measurements have not yet established whether this parameter is constant or varies between modules . How would a varying d affect our results ? The answer depends on the dimensionality of the grid . In two dimensions , if neurons have weakly correlated noise , modular variation of the coverage factor does not affect the optimal grid at all . This is because the coverage factor cancels out of all relevant formulae , a coincidence of two dimensions ( see Optimizing the grid system: probabilistic decoder , ‘Materials and methods’ , and p . 112 of Dayan and Abbott , 2001 ) . In one and three dimensions , variation of d between modules will have an effect on the optimal ratios between the variable modules . Thus , if the coverage factor is found to vary between grid modules for animals navigating one and three dimensions , our theory can be tested by comparing its predictions for the corresponding variations in grid scale factors . Similarly , even in two dimensions , if noise is correlated between grid cells , then variability in d can affect our predicted scale factor . This provides another avenue for testing our theory . The simple winner-take-all model assuming compact grid fields predicted a ratio of field width to grid period that matched measurements in both wild-type and HCN1 knockout mice ( Giocomo et al . , 2011a ) . Since the predicted grid field width is model dependent , the match with the simple WTA prediction might be providing a hint concerning the method the brain uses to read the grid code . Additional data on this ratio parameter drawn from multiple grid modules may serve to distinguish and select between potential decoding models for the grid system . The probabilistic model did not make a direct prediction about grid field width; it instead worked with the standard deviation σi of the posterior P ( x|i ) . This parameter is predicted to be σi = 0 . 19λi in two dimensions ( see Optimizing the grid system: probabilistic decoder , ‘Materials and methods’ ) . This prediction could be tested behaviorally by comparing discrimination thresholds for location to the period of the smallest module . The standard deviation σi can also be related to the noise , neural density and tuning curve shape in each module ( Dayan and Abbott , 2001 ) . Previous work by Fiete et al . ( 2008 ) proposed that the grid system is organized to represent very large ranges in space by exploiting the incommensurability ( i . e . , lack of common rational factors ) of different grid periods . As originally proposed , the grid scales in this scheme were not hierarchically organized ( as we now know they are Stensola et al . , 2012 ) but were of similar magnitude , and hence it was particularly important to suggest a scheme where a large spatial range could be represented using grids with small and similar periods . Using all the scales together ( Fiete et al . , 2008 ) argued that it is easy to generate ranges of representation that are much larger than necessary for behavior , and Sreenivasan and Fiete argued that the excess capacity could be used for error correction over distances relevant for behavior ( Sreenivasan and Fiete , 2011 ) . However , recent experiments tell us that there is a hierarchy of scales ( Stensola et al . , 2012 ) which should make the representation of behaviorally plausible range of 20–100 m easily accessible in the alternative hierarchical coding scheme that we have proposed . Nevertheless , we have checked that a grid coding scheme with the optimal scale ratio predicted by our theory can represent space over ranges larger than the largest grid period ( ‘Range of location coding in a grid system’ , Appendix 1 ) . However , to achieve this larger range , the number of neurons in each module will have to increase relative to the minimum in order to shrink the widths of the peaks in the likelihood function over position . It could be that animals sometimes exploit this excess capacity either for error correction or to avoid remapping over a range larger than the period of the largest grid . That said , experiments do tell us that remapping occurs readily over relatively small ( meter length ) scales at least for dorsal ( small scale ) place cells and grid cells ( Fyhn et al . , 2007 ) in tasks that involve spatial cues . Our hierarchical grid scheme makes distinctive predictions relative to a non-hierarchical model for the effects of selective lesions of grid modules in the context of specific models where grid cells sum to make place cells ( details in ‘Predictions for the effects of lesions and for place cell activity’ , Appendix 1 ) . In such a simple grid to place cell transformation , lesioning the modules with small periods will expand place field widths , while lesioning modules with large periods will lead to increased firing at locations outside the main place field , at scales set by the missing module . Similar effects are predicted for any simple decoder of a lesioned hierarchical grid system that has no other location related inputs—that is , animals with lesions to fine grid modules will show less precision in spatial behavior , while animals with lesions to large grid modules will confound well-separated locations . In contrast , in a non-hierarchical grid scheme with similar but incommensurate periods , lesions of any module lead to the appearance of multiple place fields at many scales for each place cell . Recent studies which ablated a large fraction of the mEC at all depths showed an increase in place field widths ( Hales et al . , 2014 ) , as did the more focal lesions of Ormond and McNaughton ( 2015 ) along the dorso–ventral axis of the mEC . However , there are multiple challenges in interpreting these experiments . First , the data of Stensola et al . ( 2012 ) shows that there are modules with both small and large periods at every depth along the mEC—the dorsal mEC is simply enriched in modules with large periods . So Hales et al . ( 2014 ) ; Ormond and McNaughton ( 2015 ) are both removing modules that have both small and large periods . A simple linear transformation from a hierarchical grid to place cells would predict that removing large periods increases the number of place fields , but Hales et al . ( 2014 ) did not look for this effect while in Ormond and McNaughton ( 2015 ) the reported number of place fields decreases after lesions ( including complete dirsruption of place fields of some cells ) . The underlying difficulty in interpretation is that while place cells might be summing up grid cells , there is evidence that they can be formed and maintained through mechanisms that may not critically involve the mEC at all ( Bush et al . , 2014; Sasaki et al . , 2015 ) . Thus , despite the interpretation given in Kubie and Fox ( 2015 ) ; Ormond and McNaughton ( 2015 ) in favor of the partial validity of a linearly summed grid to place model , it is difficult for theory to make a definitive prediction for experiments until the inter-relation of the mEC and hippocampus is better understood . Mathis et al . ( 2012a ) and Mathis et al . ( 2012b ) studied the resolution and representational capacity of grid codes vs place codes . They found that grid codes have exponentially greater capacity to represent locations than place codes with the same number of neurons . Furthermore , Mathis et al . ( 2012a ) predicted that in one dimension a geometric progression of grids that is self-similar at each scale minimizes the asymptotic error in recovering an animal's location given a fixed number of neurons . To arrive at these results the authors formulated a population coding model where independent Poisson neurons have periodic one-dimensional tuning curves . The responses of these model neurons were used to construct a maximum likelihood estimator of position , whose asymptotic estimation error was bounded in terms of the Fisher information—thus the resolution of the grid was defined in terms of the Fisher information of the neural population ( which can , however , dramatically overestimate coding precision for neurons with multimodal tuning curves [Bethge et al . , 2002] ) . Specializing to a grid system organized in a fixed number of modules , Mathis et al . ( 2012a ) found an expression for the Fisher information that depended on the periods , populations , and tuning curve shapes in each module . Finally , the authors imposed a constraint that the scale ratio had to exceed some fixed value determined by a ‘safety factor’ ( dependent on tuning curve shape and neural variability ) , in order reduce ambiguity in decoding position . With this formulation and assumptions , optimizing the Fisher information predicts geometric scaling of the grid in a regime where the scale factor is sufficiently large . The Fisher information approximation to position error in Mathis et al . ( 2012a ) is only valid over a certain range of parameters . An ambiguity-avoidance constraint keeps the analysis within this range , but introduces two challenges for an optimization procedure: ( i ) the optimum depends on the details of the constraint , which was somewhat arbitrarily chosen and was dependent on the variability and tuning curve shape of grid cells , and ( ii ) the optimum turns out to saturate the constraint , so that for some choices of constraint the procedure is pushed right to the edge of where the Fisher information is a valid approximation at all , causing difficulties for the self-consistency of the procedure . Because of these limits on the Fisher information approximation , Mathis et al . ( 2012a ) also measured decoding error directly through numerical studies . But here a complete optimization was not possible because there are too many inter-related parameters , a limitation of any numerical work . The authors then analyzed the dependence of the decoding error on the grid scale factor and found that , in their theory , the optimal scale factor depends on ‘the number of neurons per module and peak firing rate’ and , relatedly , on the ‘tolerable level of error’ during decoding ( Mathis et al . , 2012a ) . Note that decoding error was also studied in Towse et al . ( 2014 ) and those authors reported that the results did not depend strongly on the precise organization of scales across modules . In contrast to Mathis et al . ( 2012a ) , we estimated decoding error directly by working with approximated forms of the likelihood function over position rather than by approximating decoding error in terms of the Fisher information . Conceptually , we can think of the winner-take-all analysis as effectively approximating the likelihood in terms of periodic boxcar functions; for the probabilistic analysis , we treat the likelihood as a periodic sum-of-Gaussians . Since at least scores of cells are being combined within modules , the Gaussian approximation to local likelihood peaks is valid , allowing us to circumvent detailed analysis of tuning curves and variability of individual neurons . These approximations allow analytical treatment of the optimization problem over a much wider parameter range without requiring arbitrary hand-imposed constraints . Our formulation of grid resolution then simply estimates the number of distinct regions that a fixed range can be divided into . We then fix this resolution as being behaviorally determined and minimize the number of required neurons while allowing the periods of the modules , and , crucially , the number of modules , to vary to achieve the minimum . All told , our simpler , and more intuitive , formulation of grid coding embodies very general considerations trading off precision and ambiguity with a sufficiently dense population of grid cells . The simplicity and generality of our setting allows us to make predictions for structural parameters of the grid system in different dimensions . These predictions—scaling ratios in 1 , 2 , and 3 dimensions; the ratio of grid period to grid field width; the number of expected modules; the shape of the optimal grid lattice; an estimate of the total expected number of grid cells—can be directly tested in experiments . There is a long history in the study of sensory coding , especially vision , of identifying efficiency principles underlying neural circuits and codes starting with Barlow ( 1961 ) . Our results constitute evidence that such principles might also operate in the organization of cognitive circuits processing non-sensory variables . Furthermore , the existence of an efficiency argument for grid organization of spatial coding suggests that grid systems may be universal amongst the vertebrates , and not just a rodent specialization . In fact , there is evidence that humans ( Doeller et al . , 2010; Jacobs et al . , 2013 ) and other primates ( Killian et al . , 2012 ) also have grid systems . We expect that our predicted scaling of the grid modules also holds in humans and other primates .
Suppose that we want to resolve location with a precision l in a track of length L . In terms of the resolution R = L/l , we argued in the ‘Discussion’ that a ‘base-b’ hierarchical neural coding scheme will roughly require N = b logb R neurons . To derive the optimal base ( i . e . , the base that minimizes the number of the neurons ) , we evaluate the extremum ∂N/∂b=0: ( 1 ) ∂N/∂b=∂ ( b logb R ) ∂b=∂ ( b ln Rlnb ) ∂b=ln R ln b−1 ( ln b ) 2 . Setting ∂N/∂b=0 gives lnb − 1 = 0 . Therefore , the number of neurons is extremized when b = e . It is easy to check that this is a minimum . Of course , the base of a number system is usually taken to be an integer , so the argument should be taken as motivating the more detailed treatment of neural representations of space above . Neurons are of course not constrained to organize the periodicity of their tuning curves in integer ratios . Consider a probabilistic decoder of the grid system that pools all the information available in the population of neurons in each module by forming the posterior distribution over position given the neural activity . In this general setting , we assume that the firing of different grid cells is weakly correlated , that noise is homogeneous , and that the tuning curves in each module i provide dense , uniform , coverage of the interval λi . With these assumptions , we will first consider the one-dimensional case , and then analyze the two-dimensional case by analogy . We reanalyzed the data from Barry et al . ( 2007 ) and Stensola et al . ( 2012 ) in order to get the mean and the variance of the ratio of adjacent grid scales . For Barry et al . ( 2007 ) , we first read the raw data from Figure 3B of their paper using the software GraphClick , which allows retrieval of the original ( x , y ) -coordinates from the image . This gave the scales of grid cells recorded from six different rats . For each animal , we grouped the grids that had similar periodicities ( i . e . , differed by less than 20% ) and calculated the mean periodicity for each group . We defined this mean periodicity as the scale of each group . For four out of six rats , there were two scales in the data . For one out six rats , there were three grid scales . For the remaining rat , only one scale was obtained as only one cell was recorded from that rat . We excluded this rat from further analysis . We then calculated the ratio between adjacent grid scales , resulting in 6 ratios from five rats . The mean and variance of the ratio were 1 . 64 and 0 . 09 , respectively ( n = 6 ) . For Stensola et al . ( 2012 ) , we first read in the data using GraphClick from Figure 5D of their paper . This gave the scale ratios between different grids for 16 different rats . We then pooled all the ratios together and calculated the mean and variance . The mean and variance of the ratio were 1 . 42 and 0 . 17 , respectively ( n = 24 ) . Giocomo et al . ( 2011a ) reported the ratios between the grid period and the radius of grid field ( measured as the radius of the circle around the center field of the autocorrelation map of the grid cells ) to be 3 . 26 ± 0 . 07 and 3 . 32 ± 0 . 06 for Wild-type and HCN KO mice , respectively . We halved these measurements to the ratios between grid period and the diameter of the grid field to facilitate the comparison to our theoretical predictions . The results are plotted in a bar graph ( Figure 4B ) . Finally , in Figure 4C , we replotted Figure 1C from Hafting et al . ( 2005 ) by reading in the data using GraphClick and then translating that information back into a plot . | In the 1930s , neuroscientists studying how rodents find their way through a maze proposed that the animals could construct an internal map of the maze inside their heads . The map was thought to enable the animals to navigate between familiar locations and also to identify shortcuts and alternative routes whenever familiar ones were blocked . In the 1960s , recordings of electrical activity in the rat brain provided the first clues as to which nerve cells form this spatial map . In a region of the brain called the hippocampus , nerve cells called ‘place cells’ are active whenever the rat finds itself in a specific location . However , place cells alone are not able to support all types of navigation . Some spatial tasks also require cells in a region of the brain called the medial entorhinal cortex ( MEC ) , which supplies most of the information that the hippocampus receives . Cells in the MEC called ‘grid cells’ represent two-dimensional space as a repeating grid of triangles . A given grid cell is activated if the animal is located at a particular distance and angle away from the center of any of these triangles . The size of the triangles in these grids varies systematically throughout the MEC . Individual grid cells at one end of the structure encode space in finer detail than grid cells at the opposite end . Wei et al . have now used mathematical modeling to explore how grid cells are organized . The model assumes that the brain seeks to encode space at whatever resolution an animal requires using as few nerve cells as possible . The model successfully reproduces several known features of grid cells , including the triangular shape of the grid , and the fact that the size of the triangles increases in steps of a specific size across the MEC . In addition to providing a mathematical basis for the way that grid cells are organized in the brain , the model makes a number of testable predictions . These include predictions of the number of grid cells in the rat brain , as well as the pattern that grid cells adopt in three-dimensions: a question that is currently being studied in bats . Wei et al . 's findings suggest that the code used by the grid to represent space is an analog of a decimal number system—except that space is not subdivided by factors of 10 to form decimal ‘digits’ , but by a quantity related to a famous constant in the field of mathematics called Euler's number . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | A principle of economy predicts the functional architecture of grid cells |
Epigenetic clocks for mice were generated based on deep-sequencing analysis of the methylome . Here , we demonstrate that site-specific analysis of DNA methylation levels by pyrosequencing at only three CG dinucleotides ( CpGs ) in the genes Prima1 , Hsf4 , and Kcns1 facilitates precise estimation of chronological age in murine blood samples , too . DBA/2 mice revealed accelerated epigenetic aging as compared to C57BL6 mice , which is in line with their shorter life-expectancy . The three-CpG-predictor provides a simple and cost-effective biomarker to determine biological age in large intervention studies with mice .
Age-associated DNA methylation ( DNAm ) was first described for humans after Illumina Bead Chip microarray data became available to enable cross comparison of thousands of CpG loci ( Bocklandt et al . , 2011; Koch and Wagner , 2011 ) . Many of these age-associated CpGs were then integrated into epigenetic age-predictors ( Hannum et al . , 2013; Horvath , 2013; Weidner et al . , 2014 ) . However , site-specific DNAm analysis at individual CpGs can also provide robust biomarkers for aging . For example , we have described that DNAm analysis at only three CpGs enables age-predictions for human blood samples with a mean absolute deviation ( MAD ) from chronological age of less than five years ( Weidner et al . , 2014 ) . Such simplistic age-predictors for human specimen are widely used because they enable fast and cost-effective analysis in large cohorts . Recently , epigenetic clocks were also published for mice by using either reduced representation bisulfite sequencing ( RRBS ) or whole genome bisulfite sequencing ( WGBS ) ( Petkovich et al . , 2017; Stubbs et al . , 2017; Wang et al . , 2017 ) . For example , Petkovich et al . described a 90 CpG model for blood ( Petkovich et al . , 2017 ) , and Stubbs and coworkers a 329 CpG model for various different tissues ( Stubbs et al . , 2017 ) . Nutrition and genetic background seem to affect the epigenetic age of mice – and thereby possibly aging of the organism ( Cole et al . , 2017; Hahn et al . , 2017; Maegawa et al . , 2017 ) . In analogy , epigenetic aging of humans is associated with life expectancy , indicating that it rather reflects biological age than chronological age ( Lin et al . , 2016; Marioni et al . , 2015 ) . However , DNAm profiling by deep sequencing technology is technically still challenging , relatively expensive , and not every sequencing-run covers all relevant CpG sites with enough reading depth .
Therefore , we established pyrosequencing assays for nine genomic regions of previously published predictors ( Petkovich et al . , 2017; Stubbs et al . , 2017 ) . These regions were preselected to have multiple age-associated CpGs in close vicinity . DNAm was then analyzed in 24 blood samples of female C57BL/6 mice that covered a broad range of 12 different age groups ( 11 to 117 weeks old ) . The nine amplicons covered a total of 71 CpG sites ( Supplementary file 1 ) and we used machine learning to identify the best fitted model for epigenetic age-predictions using cross-fold validation on the training set . The best results were observed for 15 CpGs from five different amplicons that provided an extremely high correlation with chronological age in the training set ( R2 = 0 . 99; mean absolute deviation [MAD]=2 . 76 weeks; Supplementary file 2 ) , albeit the training set might be too small for this approach . To make the method more easily applicable and more cost-effective , we wanted to focus on less CpGs . When we varied the regularization parameters for models with less CpGs , the precision declined significantly . For example the best model with three CpGs comprised the three CpGs of Hsf4 ( CpGs# 3 , 4 , 5 ) that also revealed the overall highest Pearson correlations with chronological age ( R2 = 0 . 95; MAD = 5 . 24 weeks ) . However , combination of different hypo- and hypermethylated amplicons might be advantageous to facilitate better assessment of plausibility of the results . Therefore , we alternatively selected those three CpGs that revealed the highest Pearson correlation with chronological age in different amplicons . These three CpGs were associated with the genes Proline rich membrane anchor 1 ( Prima1: chr12:103214639; R2 = 0 . 71 ) , Heat shock transcription factor 4 ( Hsf4: chr8:105271000; R2 = 0 . 95 ) and Potassium voltage-gated channel modifier subfamily S member 1 ( Kcns1: chr2:164168110; R2 = 0 . 83; Figure 1A–C; Figure 1—figure supplement 1 ) . Notably , all three CpGs were derived from the epigenetic age-predictor for blood samples ( Petkovich et al . , 2017 ) . A multivariable model for age-predictions was established for DNAm at the CpGs in Prima 1 ( α ) , Hsf4 ( β ) , and Kcns1 ( γ ) : Predicted ageC57BL/6 ( in weeks ) = −58 . 076 + 0 . 25788 α + 3 . 06845 β + 1 . 00879 γ Age-predictions correlated very well with the chronological age of C57BL/6 mice in the training set ( R2 = 0 . 96; MAD = 4 . 86 weeks; Figure 1D ) . Our three CpG age-predictor was subsequently validated in a blinded manner for 21 C57BL/6J mice ( 7 to 104 weeks old ) from the University of Ulm ( validation set 1 ) and 19 C57BL/6J mice ( 14 to 109 weeks old ) from the University of Groningen ( validation set 2 ) . The results of both validation sets revealed high correlations with chronological age ( R2 = 0 . 95 and 0 . 91 , respectively; Figure 1E–H ) with relatively small MADs ( 6 . 9 and 7 . 1 weeks ) and median absolute errors ( MAE; 5 . 0 and 5 . 9 weeks ) . Thus , our age-predictions seem to have similar precision as previously described for multi-CpG predictors based on RRBS or WGBS data ( Petkovich et al . , 2017; Stubbs et al . , 2017; Wang et al . , 2017 ) . Gender did not have significant impact on our epigenetic age-predictions for mice ( Figure 2 ) , as described before ( Maegawa et al . , 2017; Petkovich et al . , 2017; Stubbs et al . , 2017 ) . In contrast , the human epigenetic clock is clearly accelerated in male donors ( Hannum et al . , 2013; Horvath , 2013; Weidner et al . , 2014 ) . This coincides with shorter life expectancy in men than woman , whereas in mice there are no consistent sex differences in longevity ( Goodrick , 1975 ) . To address the question if our three CpG signature was also applicable for other tissues than blood we analyzed the DNAm in skin , kidney , intestine , lung , liver , heart , brain , testis , and pancreas of 3 young ( 9 . 6 weeks old ) and three old mice ( 56 . 9 weeks old ) . In all tissues tested the samples of old mice were predicted to be older using our three CpG signature . However , the different DNAm levels clearly demonstrate that the model needs to be retrained to be applied for these tissues ( Figure 3 ) . Subsequently , we analyzed epigenetic aging of DBA/2 mice that have a shorter life expectancy than C57BL/6 mice ( Goodrick , 1975 ) ( 33 mice from Ulm and Groningen; 6 to 109 weeks old ) . The three CpGs in Prima1 , Hsf4 and Kcns1 revealed high correlation with chronological age ( R2 = 0 . 91 , 0 . 88 and 0 . 83 , respectively ) , albeit the offset in DNAm between DBA/2 and C57BL/6 mice indicated that the signature needs to be retrained for different mouse strains ( Figure 4a–c ) . Notably , the slopes were higher in DBA/2 mice , particularly for the CpG in Prima1 . Furthermore , DNAm of Hsf4 increased at a higher rate in young DBA/2 mice , indicating that it is more accurately modelled as a function of logarithmic age . This has also been described in human for many age-associated CpGs in pediatric cohorts ( Alisch et al . , 2012 ) . In fact , epigenetic age-predictions in DBA/2 mice seemed to follow a logarithmic model of age ( R2 = 0 . 89; Figure 4d ) rather than a linear association ( R2 = 0 . 86 ) . These results provided evidence for accelerated epigenetic aging of DBA/2 mice . Either way , epigenetic age-predictions were overall significantly overestimated in the shorter-lived DBA/2 mice , suggesting that age-predictors need to be adjusted for different inbreed mice strains . To this end , we have retrained a multivariate model for DBA/2 mice: Predicted ageDBA/2 ( in weeks ) = 87 . 54294–1 . 22221 α + 0 . 991558 β + 0 . 355444 γ This adjusted model facilitated relatively precise age-predictions for DBA/2 mice ( R2 = 0 . 95; MAD = 7 . 1 weeks; MAE = 5 . 3 weeks; Figure 4e ) .
Generation of confined epigenetic signatures is always a tradeoff between integrating more CpGs for higher precision and higher costs for analysis ( Wagner , 2017 ) . It was somewhat unexpected that with only three CpGs our signature facilitated similar precision of epigenetic age-predictions as the previously published signatures based on more than 90 CpGs . This can be attributed to the higher precision of DNAm measurements at individual CpGs by bisulfite pyrosequencing , which is one of the most precise methods for determining DNAm at single CpG resolution ( BLUEPRINT consortium , 2016 ) . Particularly in RRBS data not all CpG sites are covered in all samples and a limited number of reads notoriously entails lower precision of DNAm levels at these genomic locations . Thus , genome wide deep sequencing approaches facilitate generation of robust large epigenetic age-predictors , while site specific analysis may compensate by higher precision of DNAm measurement at individual CpGs . The ultimate goal of epigenetic age-predictors for mice is not to develop near perfect age predictors , but to provide a surrogate for biological aging that facilitates assessment of interventions on aging . In fact , using deep sequencing approaches ( RRBS or WGBS ) several groups already indicated that relevant parameters that affect aging of the organism - such as diet , genetic background , and drugs - do also impact on epigenetic aging ( Cole et al . , 2017; Hahn et al . , 2017; Maegawa et al . , 2017 ) . It is yet unclear if epigenetic aging signatures can be specifically trained to either correlate with chronological age or biological age . For humans , recent studies indicate that this might be possible ( Levine et al . , 2018 ) and we have previously demonstrated that even individual age-associated CpGs can be indicative for life expectancy ( Zhang et al . , 2017 ) . Further studies will be necessary to gain better understanding how epigenetic age predictions are related to the real state of biological aging , and how it is related to alternative approaches to quantify biological aging , such as telomere length ( Belsky et al . , 2018 ) . Our three CpG model has been trained for blood samples – a specimen that is commonly used in biochemical analysis and the small required volume can be taken without sacrificing the mice . However , epigenetic aging may occur at different rates in different tissues . It is difficult to address this question in humans because it is difficult to collect samples of various tissues in large aging cohorts , whereas this is feasible in mice . We demonstrate that age-associated DNAm changes occur in multiple tissues in our three CpGs albeit they were initially identified in blood ( Petkovich et al . , 2017 ) . Furthermore , DNAm levels may vary between different hematopoietic subsets ( Frobel et al . , 2018; Houseman et al . , 2014 ) . In the future , sorted subsets should be analyzed to determine how the three CpG signature is affected by blood counts . The results of our three CpG signature suggest that epigenetic aging is accelerated in DBA/2 mice . Notably , in elderly DBA/2 mice the epigenetic age predictions revealed higher ‘errors’ from chronological age , which might be attributed to the fact that the variation of lifespan is higher in DBA/2 than C57BL/6 mice ( de Haan et al . , 1998; Goodrick , 1975 ) . It will be important to validate the association of the epigenetic age-predictions with biological age by additional correlative studies , including life expectancy in mice . Taken together , we describe an easily applicable but quite precise approach to determine epigenetic age of mice . We believe that our assay will be instrumental to gain additional insight into mechanisms that regulate age-associated DNAm and for longevity intervention studies in mice .
Blood samples of C57BL/6J mice of the training set and of the validation set one were taken at the University of Ulm by submandibular bleeding ( 100–200 μl ) of living mice or postmortem from the vena cava . C57BL/6J samples of the validation set two were taken at the University of Groningen from the cheek . DBA/2J samples were taken at the University of Ulm ( n = 14 ) and Groningen ( n = 19 ) . All mice were accommodated under pathogen-free conditions . Experiments were approved by the Institutional Animal Care of the Ulm University as well as by Regierungspräsidium Tübingen and by the Institutional Animal Care and Use Committee of the University of Groningen ( IACUC-RUG ) , respectively . To analyze age-associated changes in different tissues we used three young ( 9 . 6 weeks old ) and three old mice ( 56 . 9 weeks old ) C57BL/6J mice ( JaxMice ) in accordance with relevant Spanish and European guidelines after approval by the Biodonostia Animal Care Committee . These mice were sacrificed and dissected immediately . 25 mg of tissue ( 10 mg in the case of spleen ) or 200 µl of blood were used for DNA extraction . Genomic DNA was isolated from 50 µl blood using the QIAamp DNA Mini Kit ( Qiagen , Hilden , Germany ) . Kidney and liver DNA extractions were digested with Ribonuclease A ( 100 mg/ml , Sigma R4875 ) . DNA concentration was quantified by Nanodrop 2000 Spectrophotometers ( Thermo Scientific , Wilmington , USA ) . 200 ng of genomic DNA was subsequently bisulfite-converted with the EZ DNA Methylation Kit ( Zymo Research , Irvine , USA ) . Bisulfite converted DNA was subjected to PCR amplification . Primers were purchased at Metabion and the sequences are provided in Supplementary file 3 . 20 µg PCR product was immobilized to 5 µl Streptavidin Sepharose High Performance Bead ( GE Healthcare , Piscataway , NJ , USA ) , and then annealed to 1 µl sequencing primer ( 5 μM ) for 2 min at 80°C . Amplicons were sequenced on PyroMark Q96 ID System ( Qiagen , Hilden , Germany ) and analyzed with PyroMark Q CpG software ( Qiagen ) . We used a penalized regression model from the R package glmnet on the training dataset to establish a predictor of mouse age based on CpG methylation . The alpha parameter of glmnet was set to 1 ( lasso regression ) and the lambda parameter was chosen by cross-fold validation of the training dataset ( 10-fold cross validation ) . Alternatively , we trained our multivariable model with preselected CpGs based on location in three different amplicons , high Pearson correlation ( R ) of DNAm with chronological age , and combination of hyper- and hypomethylated sites . Linear regressions , MAD and MAE were calculated with Excel . Statistical significance of the deviations between predicted and chronological age was estimated by Mann–Whitney U test or Student´s t-test as indicated . | Epigenetic marks are chemical modifications found throughout the genome – the DNA within cells . By influencing the activity of nearby genes , the marks govern developmental processes and help cells to adapt to changes in their surroundings . Some epigenetic marks can be gained or lost with age . A lot of aging research focuses on one type of mark , called “DNA methylation” . By measuring the presence or absence of specific methyl groups , scientists can estimate biological age – which may differ from calendar age . Recent studies have developed computer models called epigenetic aging clocks to predict the biological age of mouse cells . These clocks use epigenetic data collected from the entire genomes of mice , and are useful for understanding how the aging process is affected by genetic parameters , diet , or other environmental factors . Yet , the genome sequencing methods used to construct most existing epigenetic clocks are expensive , labor-intensive , and cannot be easily applied to large groups of mice . Han et al . have developed a new way to predict biological aging in mice that needs methylation information from just three particular sections of the genome . Even though this approach is much faster and less expensive than other epigenetic approaches to measuring aging , it has a similar level of accuracy to existing models . Han et al . use the new method to show that cells from different strains of laboratory mice age at different rates . Furthermore , in a strain that has a shorter life expectancy , aging seems to be accelerated . The new approach developed by Han et al . will make it easier to study how aging in mice is affected by different interventions . Further studies will also be needed to better understand how epigenetic marks relate to biological aging . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"tools",
"and",
"resources",
"genetics",
"and",
"genomics"
] | 2018 | Epigenetic age-predictor for mice based on three CpG sites |
During cortical development , the identity of major classes of long-distance projection neurons is established by the expression of molecular determinants , which become gradually restricted and mutually exclusive . However , the mechanisms by which projection neurons acquire their final properties during postnatal stages are still poorly understood . In this study , we show that the number of neurons co-expressing Ctip2 and Satb2 , respectively involved in the early specification of subcerebral and callosal projection neurons , progressively increases after birth in the somatosensory cortex . Ctip2/Satb2 postnatal co-localization defines two distinct neuronal subclasses projecting either to the contralateral cortex or to the brainstem suggesting that Ctip2/Satb2 co-expression may refine their properties rather than determine their identity . Gain- and loss-of-function approaches reveal that the transcriptional adaptor Lmo4 drives this maturation program through modulation of epigenetic mechanisms in a time- and area-specific manner , thereby indicating that a previously unknown genetic program postnatally promotes the acquisition of final subtype-specific features .
The mammalian cerebral cortex is tangentially subdivided into several functional areas allowing effective interactions with the external world by organizing sensory information into a coherent perceptual model of the environment . All neocortical areas are radially organized into six neuronal layers that show prominent diversities in features and functions despite similarities in their laminar organization . This is mainly due to differences in the molecular identity , morphology and long-range connectivity of residing neurons ( Greig et al . , 2013; Huang , 2014; O'Leary et al . , 2007 ) . Cortical projection neurons ( PNs ) can be subdivided into three major classes: ( i ) the intra-telencephalic ( IT ) neurons projecting to ipsilateral and/or , through the corpus callosum , to contralateral regions of the telencephalon ( i . e . callosal projection neurons – CPN ) ; ( ii ) the cortico-thalamic ( CT ) neurons projecting to different dorsal thalamic nuclei; and ( iii ) the pyramidal-tract ( PT ) neurons , also defined as subcerebral projection neurons ( SCPN ) , that innervate different subcerebral targets , such as the striatum , the brainstem , and the spinal cord ( Harris and Shepherd , 2015; Shepherd , 2013 ) . Major molecular determinants of CPN and SCPN neuronal classes ( Greig et al . , 2013; Leone et al . , 2008; Molyneaux et al . , 2007; Srinivasan et al . , 2012 ) act by both promoting the identity of a given neuronal class and repressing alternative ones ( Fame et al . , 2011; Greig et al . , 2013; Kiritani et al . , 2012; Molnar and Cheung , 2006; Molyneaux et al . , 2007; Reiner et al . , 2010; Sohur et al . , 2014 ) . For example , Satb2 , a chromatin remodeling protein , drives CPN specification and axonal pathfinding , and represses the expression of the transcription factor Ctip2 ( gene name Bcl11b ) , which controls the connectivity of SCPN ( Alcamo et al . , 2008; Arlotta et al . , 2005; Baranek et al . , 2012; Britanova et al . , 2008; Srivatsa et al . , 2014 ) . However , Satb2 was shown to control also subcerebral connectivity ( Leone et al . , 2014 ) , indicating that the final acquisition of a given cell type is not based on the function of a unique transcriptional regulator but most probably on the combination of more determinants expressed at different levels . This is in agreement with observations performed on early embryonic stages , where neuronal types are not yet fully specified and factors with mutually exclusive functions largely overlap . Notably , expression of transcriptional determinants tends to segregate after birth in a cell- or time-specific manner when neurons differentiate and the major neocortical classes become more distinct from each other ( reviewed in Greig et al . , 2013 ) . Nevertheless , it is still not clear whether the transcriptional regulators acting during early stages of neuronal specification play any additional roles during postnatal maturation when PNs acquire their final properties . Even if antithetic factors promoting callosal or subcerebral PNs can co-localize in a small subset of neurons after birth ( Azim et al . , 2009; Baranek et al . , 2012; Leone et al . , 2008; Tomassy et al . , 2010 ) , it is not known whether their co-expression has a functional meaning , nor whether this hybrid molecular population corresponds to a permanent subgroup of cortical PNs . In particular , the persistence of few double Ctip2/Satb2-positive neurons at early postnatal stages of corticogenesis represents a conundrum , since Satb2 is a strong repressor of Ctip2 ( Alcamo et al . , 2008; Baranek et al . , 2012; Britanova et al . , 2008; Leone et al . , 2014 ) . Moreover , Satb2 and Ctip2 regulate the expression of two netrin1 receptors ( Unc5C and DCC , respectively ) with opposite effects on axon guidance ( Srivatsa et al . , 2014 ) . Here , we show that Ctip2/Satb2 co-expression does not delineate a transient stage of cortical development but instead distinct subpopulations of major PN classes , whose number progressively increases in the postnatal somatosensory mouse cortex . Double Ctip2/Satb2 ( C/S+ ) -expressing cells define at least two neuronal subclasses , which project either to the brainstem or to the contralateral cortex and retain unique molecular , morphological , and electrophysiological features that distinguish them from single Ctip2- or Satb2-expressing cells . Moreover , we demonstrate that the transcriptional adaptor Lmo4 allows the co-localization of Ctip2 and Satb2 by competing with Satb2 for the binding to Hdac1 , a histone deacetylase normally recruited by the Satb2-NuRD complex bound to the Ctip2 locus ( Alcamo et al . , 2008; Britanova et al . , 2008 ) . Notably , the distribution of both Lmo4+ and C/S+ cell populations in lower layers change over time in the cortical motor and somatosensory areas during postnatal stages of development . Overall , we show that common molecular pathways may regulate different specification programs during postnatal maturation and/or refinement of functionally distinct classes of neocortical neurons ( such as IT and PT neurons ) in an area- and time-specific manner .
Previous studies reported transient embryonic Ctip2/Satb2 ( C/S+ ) co-expression in the developing cortical plate; however , whether this co-expression persists at postnatal stages was still unclear ( Alcamo et al . , 2008; Baranek et al . , 2012; Britanova et al . , 2008; Leone et al . , 2014 ) . We thus examined the number and distribution of C/S+ cells from E14 . 5 to P21 to investigate whether they constitute a transient or rather a permanent population of cortical PNs . We first observed that while the number of single Ctip2+ and double C/S+ cells strongly decreases in the cortical plate from E14 . 5 to E16 . 5 , as previously reported ( Leone et al . , 2014 ) , double C/S+ ( but not single Ctip2+ ) cells resurge between E16 . 5 and P0 ( Figure 1A and B ) and constitute 18% of lower layer ( LL ) neurons of the somatosensory cortex ( S1 ) by P21 ( Figure 1B ) . Then , to verify whether the postnatal increase of C/S+ neurons was due just to a general increment in Ctip2 and Satb2 expression , we calculated the percentage of C/S+ cells on the total of Ctip2+ or Satb2+ cells and found an increase with respect to both populations ( Figure 1C ) . 10 . 7554/eLife . 09531 . 003Figure 1 . Temporal and areal distribution of Ctip2/Satb2+ neurons in the neocortex . ( A ) . Coronal sections from prospective ( pS ) and primary ( S1 ) somatosensory areas of E14 . 5 , E16 . 5 , P0 , P7 , and P21 cortices immunolabeled for Ctip2 and Satb2 . ( B ) . Percentage of Satb2+ , Ctip2+ , and double Ctip2/Satb2 ( C/S+ ) neurons on the total number of DAPI+ neurons in the cerebral cortex at different ages . The counting was performed in the cortical plate ( CP ) from E14 . 5 to E16 . 5 and only in lower layers ( LL ) from P0 to P21 . ( C ) . Number of C/S+ neurons calculated as a percentage of DAPI+ , Satb2+ , and Ctip2+ cortical neurons at different ages . ( D ) . Immunostaining for Satb2 and Ctip2 on P0 brain coronal sections from frontal/motor ( F/M ) and pS areas . Top right panels represent high-magnification views of boxes in layer V depicted in left panels . Arrowheads indicate C/S+ neurons . In bottom right panels , quantification and laminar distribution of double C/S+ neurons . ( E ) . Immunostaining for Satb2 and Ctip2 on P7 brain coronal sections from primary motor ( M1 ) and S1 cortices . Top right panels represent high-magnification views of boxes in layer V depicted in left panels . Arrowheads indicate C/S+ neurons . In bottom right panels , quantification and laminar distribution of double C/S+ neurons . Data are represented as means ± SEM . *p≤0 . 05 , **p≤0 . 01 , ***p≤0 . 001 . SEM , standard error of the mean; UL , upper layers . Scale bars: A , D , E , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 00310 . 7554/eLife . 09531 . 004Figure 1—figure supplement 1 . Time- and area-specific variations in the number and distribution of Ctip2/Satb2+ cells in early postnatal brains . ( A ) . Quantification and laminar distribution of double C/S+ neurons on the total of Ctip2+ or Satb2+ populations in frontal/motor ( F/M ) and prospective somatosensory ( pS ) cortices . ( B ) . Quantification and laminar distribution of double C/S+ neurons on the total of Ctip2+ or Satb2+ populations in primary motor ( M1 ) and somatosensory ( S1 ) areas . Data are represented as means ± SEM . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . SEM , standard error of the mean; UL , upper layer neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 004 Next , we investigated whether this population of C/S+ neurons was differentially distributed between frontal/motor ( F/M ) and parietal/somatosensory cortices . Interestingly , we observed that the highest percentage of C/S+ cells is primordially localized in layer V of the F/M cortex at P0 , whereas the corresponding layer of the prospective somatosensory ( pS ) region has five times less double-positive cells ( Figure 1D ) . When calculated on the total of Ctip2+ cells , the percentage of C/S+ neurons only slightly changes between F/M and pS , whereas it results four times higher in the F/M cortex with respect to the total of Satb2+ cells ( Figure 1—figure supplement 1A ) . This trend is inverted at P7 , where the highest percentage of double C/S+ cells in LL is localized in the primary somatosensory area ( S1 ) ( Figure 1E and Figure 1—figure supplement 1B ) , suggesting the existence of a time- and region-specific mechanism allowing Ctip2 expression in Satb2+ cells . Taken together , our data reveal the progressive postnatal and area-specific development of a molecularly hybrid neuronal subpopulation co-expressing two transcriptional regulators with mutually exclusive functions in cortical development . Since double C/S+ neurons are maintained at least until P21 , we next investigated the molecular features and connectivity distinguishing these cells from single Ctip2+ or Satb2+ PNs . To analyze their connectivity , we injected cholera toxin subunit B ( CTB ) -conjugated fluorophores ( Conte et al . , 2009 ) into the cervical spinal cord , the rostral pontine region and ithe contralateral S1 cortex of P2/P3 pups ( Figure 2A–C ) . Injections performed at the level of the spinal cord exclusively labeled corticospinal PNs ( CSpPNs ) ( Figure 2A ) , whereas injections in the rostral pons marked all layer V subcerebral PNs ( SCPN ) , including CSpPN , cortico-pontine ( CpPN ) , and other cortico-brainstem ( CBPN ) neurons projecting through the pyramidal tract ( Figure 2B ) . Immunolabeling of retrograde-traced cells at P7 revealed that C/S+ neurons constitute 31% of all subcerebral neurons; however , they do not reach the spinal cord indicating that C/S+ subcerebral neurons most probably project to the brainstem nuclei ( Figure 2A and B ) . CTB injections in the contralateral S1 cortex labeled callosal projection neurons ( CPN ) and showed that 32% of these neurons co-expressed Ctip2 and Satb2 ( Figure 2C ) . 10 . 7554/eLife . 09531 . 005Figure 2 . Ctip2 and Satb2 co-localization describes subgroups of cortico-brainstem and callosal projection neurons ( A-C ) . Top left , schematic representations of cholera toxin subunit B ( CTB ) injections at cervical level to trace corticospinal projection neurons ( CSpPNs ) alone ( A ) , at the midbrain/hindbrain junction to label subcerebral projection neurons ( SCPN ) including CSpPNs ( B ) , and in the contralateral somatosensory cortex ( S1 ) to label callosal projection neurons ( CPN . ( C ) . On bottom left of ( A-C ) , immunostaining for Satb2 and Ctip2 on retrogradely labeled PNs of P7 S1 coronal sections . Squared panels in ( A-C ) represent high magnification views of boxes in layer V depicted in left panels . Filled arrowheads indicate retrogradely labeled neurons double labeled for Satb2 and Ctip2 , while empty arrowheads indicate retrogradely labeled neurons labeled either for Ctip2 or Satb2 . On top right of ( A-C ) , quantification of Ctip2+ and double Ctip2/Satb2+ ( C/S+ ) retrogradely labeled neurons on the total number of retrogradely labeled PNs in layer V . ( D ) . On the top , schematic representation of the labeling paradigm to simultaneously double-label SCPN and CPN . Red retrobeads or Green IX retrobeads were injected into the midbrain/hindbrain junction at P2 and into the contralateral S1 at P3 , respectively . On the bottom , panels representing labeled CPN ( green ) and SCPN ( red ) on P7 S1 coronal sections . ( E ) . Left panel columns represent coronal sections traced for CPN and SCPN and immunostained for Sox5 , Bhlhb5 , and Er81 . Panels on the right represent high-magnification views of boxes in layer V depicted in left panels . Filled white and yellow arrowheads indicate CPN or SCPN positive for the used marker , whereas empty arrowheads indicate negative staining . Scale bars: A , B , C , low-magnification images , 200 μm; D , E , low-magnification images , 100 μm; A , B , C , E , high magnifications , 50 µm . Data are represented as mean ± SEM . SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 00510 . 7554/eLife . 09531 . 006Figure 2—figure supplement 1 . Analysis of the molecular code of postnatal Ctip2/Satb2+ neurons . ( A ) . To the left , detailed primary somatosensory area ( S1 ) coronal sections from P7 brains co-immunolabeled for Ctip2 , Satb2 and either Bhlhb5 , Er81 or Sox5 . On the right , squared panels represent high-magnification views of boxes depicted in left panels . Arrowheads indicate triple labelled cells . ( B ) Quantification of triple-positive cells in layer V of S1 . Data are represented as means ± SEM . Scale bar: 100 µm . SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 006 The observation that C/S+ neurons project to both contralateral and subcerebral targets raised the question of whether these cells constitute a hybrid population of dual-projecting neurons or two distinct subpopulations . To address this issue , CTB-coated beads were co-injected into the pons and the contralateral S1 of early postnatal brains ( Figure 2D ) . No co-labeled callosal and subcerebral cells were observed in three independent experiments ( Figure 2D ) , indicating that neurons co-expressing Ctip2 and Satb2 constitute two independent subpopulations: one projecting to the brainstem and the other to the contralateral hemisphere through the corpus callosum . We then tested whether double C/S+ cells could be further distinguished by the expression of other molecular markers . We used the transcription factors Bhlhb5 , highly expressed in CSpPNs of the sensorimotor cortex ( Joshi et al . , 2008 ) , Sox5 present in the major classes of corticofugal PNs ( subplate , corticothalamic and all subcerebral PNs ) ( Lai et al . , 2008 ) , and Er81 ( also named Etv1 ) expressed throughout layer V in both CPNs and SCPNs ( Molnar and Cheung , 2006; Yoneshima et al . , 2006 ) . Analyzing the distribution of these factors in C/S+ cells of P7 S1 cortices indicated that nearly 90% of C/S+ cells also express Bhlhb5 , 58% Er81 and almost 70% Sox5 in layer V ( Figure 2—figure supplement 1A and B ) . To further assess whether they are selective for callosal and/or subcerebral C/S+ cell subpopulations , we combined retrograde labeling with immunostaining for Bhlhb5 , Er81 , or Sox5 and found that Sox5 and Bhlhb5 are exclusively expressed in SCPNs , whereas Er81 solely labels CPNs in layer V ( Figure 2E ) . Taken together , these data indicate that Ctip2/Satb2 co-expression describes at least two layer V subpopulations: callosal PNs co-expressing Er81 , and cortico-brainstem PNs co-expressing Bhlhb5 and/or Sox5 . To unravel specific morphological and electrophysiological features of C/S+ cells , we exploited the Thy1-eYFP-H transgenic line , which labels layer Vb neurons from P14 onwards ( Feng et al . , 2000; Porrero et al . , 2010 ) . We first verified that YFP+ neurons did express Ctip2 and Satb2 in the S1 area of P21 Thy1-eYFP-H cortices ( Figure 3—figure supplement 1A ) . While 19 . 3% of GFP+ neurons resulted positive for Ctip2 but not for Satb2 ( +/- ) and 76 . 5% co-expressed Ctip2 and Satb2 ( +/+ ) , only 1 . 4% was positive for Satb2 alone ( -/+ ) and 2 . 8% were negative for both markers ( Figure 3—figure supplement 1A and B ) . In addition , YFP+/ ( C/S+ ) cells in layer V represent 55 . 7% of double C/S+ neurons , indicating overall that this mouse line represents an appropriate tool to undertake a detailed morphological and electrophysiological analysis of C/S+ neurons . Comparison of different morphological features including soma shape , dendritic complexity , and apical dendrite length of YFP+ 3D-reconstructed neurons allowed the classification of C/S+ and single Ctip2+ neurons into two major subpopulations ( Figure 3A and B ) . Overall , the soma of C/S+ neurons is significantly smaller in terms of diameter , area , and volume when compared to single Ctip2 neurons; moreover , it occupies on average deeper regions of layer Vand shows earlier bifurcation of the apical tuft . However , K-means clustering of all these parameters revealed that the C/S+ cells are constituted by at least three different subtypes , whereas Ctip2+ neurons by at least two ( Figure 3B–D ) . Whereas subtype 1 ( orange ) is unique to C/S+ neurons , subtype 2 ( magenta ) and subtype 3 ( green ) are common to both groups , even if subtype 2 is prevalent in Ctip2+ cells and subtype 3 is mainly represented in C/S+ neurons . Thus , C/S+ and Ctip2+ neurons can be mainly subdivided into two distinct morphological subgroups in the P21 S1 cortex . 10 . 7554/eLife . 09531 . 007Figure 3 . Morphometric and electrophysiological characterization of Thy1-eYFP-H layer V neurons . ( A ) . The bar charts represent comparisons between double C/S+ ( black ) and single Ctip2+ ( grey ) neurons for different morphological features in YFP+ cells of P21 Thy1-e-YFP-H transgenic brains . The asterisks indicate statistical significance . *p≤0 . 05 , **p≤0 . 01 , ***p≤0 . 001 . ( B ) . Pictograms used to schematically illustrate morphological features and qualitative differences among YFP+ neurons . ( C ) . 3D reconstructions of representative neurons of the three distinct morphological profiles . Upper part of the image illustrates length ( expressed as soma distance from the pial surface ) and bifurcation of the apical dendrite . The bottom part of the images indicates soma reconstruction and basal dendrite data . ( D ) . The bar charts on the left represent the relative number of cells belonging to the morphological profiles identified by K-mean clustering analysis performed separately on each of the two molecular classes ( double C/S+ and single Ctip2+ cells ) . The line graphs on the right represent morphological features for profile 1 ( orange ) , unique to double C/S+ cells , profile 2 ( magenta ) and 3 ( green ) , shared by both groups . ( E ) . Immunofluorescence for Satb2 , Ctip2 , and biocytin on S1 coronal sections from P21 Thy1-eYFP-H transgenic cortices . ( F ) . Table showing the input resistance ( Rin reflecting the membrane resistance ) , the sag ( difference of voltage between peak and steady-state potentials ) , the first interspike interval ( ISI ) and the difference of amplitude between the first and second action potential ( AP ) , and the fast after-hyperpolarization ( fAHP ) of the three identified subpopulations . ( G ) . Traces showing the variation in membrane potential when a hyperpolarizing current was injected ( -0 . 2 nA; bottom ) and the trains of action potentials when a depolarizing current was injected to the cell to reach the AP threshold ( top ) . ( G’ ) . Magnifications of first and/or second APs . Scale bars: G , 20 mV - 50 ms; G’ , 20 mV - 5ms . Statistics ( Mann-Whitney ) : a = difference between Ctip2+ and C/S+ type 1 cells; b = difference between Ctip2+ and C/S+ type 2; c = difference between C/S+ type 1 and type 2 cells . Data are represented as means ± SEM . 1p<0 . 05; 2p<0 . 01; 3p<0 . 001 . SEM , standard error of the mean . Scale bars: C , 10x mag . , 100 µm; E , 40x mag . , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 00710 . 7554/eLife . 09531 . 008Figure 3—figure supplement 1 . Morphometric properties of YFP-positive neurons from the Thy1-eYFP-H transgenic line . ( A ) . To the left , coronal section corresponding to the primary somatosensory area ( S1 ) of a P21 Thy1-eYFP-H transgenic brain immunolabeled for YFP , Satb2 , and Ctip2 . To the right , square panels representing high-magnification views of neurons residing in the cortical region delimited by the red box in the left panel . These neurons are either positive for Ctip2 alone or for Ctip2/Satb2 ( C/S+ ) . ( B ) . Distribution of C+/S+ ( light blue ) , C+/S- ( orange ) , C-/S+ ( grey ) , and C-/S- ( yellow ) neurons among YFP+ cells . Light blue depicts neurons with small soma size ( majority of C+/S+ ) , whereas orange and dark blue indicate neurons with a big soma size ( almost all Ctip2+ and a small percentage of C+/S+ YFP+ neurons ) . Scale bars: A , left panel , 1 mm; A , right panel , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 00810 . 7554/eLife . 09531 . 009Figure 3—figure supplement 2 . Electrophysiological analyses of YFP-positive neurons from the Thy1-eYFP-H transgenic line . ( A ) . Immunofluorescences for GFP and biocytin performed on YFP+ cells from Thy1-eYFP-H transgenic brains . ( B ) . Trains of action potentials ( AP ) triggered by a depolarizing current injected for 2 s to reach the threshold of AP ( left ) or twice this threshold ( right ) in cells expressing Ctip2 only ( top ) or both Ctip2 and Satb2 . Scale bars: 20 mV – 250 ms for top panels and 20 mV – 50 ms for bottom panels ( insets ) . ( C ) . Table showing the main features analyzed in C/S+ and single Ctip2+ cells . Statistics ( Mann-Whitney ) : a = difference between Single Ctip2 and Double Satb2 Ctip2 type 1 cells; b = difference between Single Ctip2 and Double Satb2 Ctip2 type 2; c = difference between Double Satb2 Ctip2 type 1 and type 2 . Data are represented as means ± SEM . 1p<0 . 05; 2p<0 . 01; 3p<0 . 001 . SEM , standard error of the mean . Scale bar: A , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 009 Since neurons with a large soma constituted broad fractions of both C/S+ and Ctip2+ populations ( Figure 3—figure supplement 1B ) , we investigated whether they would differ for their respective electrophysiological characteristics . We recorded the activity of large soma YFP+ cells on P21 Thy1-eYFP-H brain slices and labeled recorded neurons with biocytin to subsequently test them for the expression of Ctip2 and/or Satb2 ( Figure 3E , Figure 3—figure supplement 2A ) . Steps of hyperpolarizing currents were first applied and the input resistance was measured , based on the I-V curves ( Figure 3F , G and G’ ) . Cells only expressing Ctip2 ( n = 9 ) had a greater resistance compared to C/S+ neurons ( n = 23 ) ( Ctip2+: Rpeak = 109 . 8 ± 11 . 7 MΩ and Rss = 84 . 8 ± 7 . 9 MΩ; C/S+: Rpeak = 73 . 3 ± 3 , 9 MΩ and Rss = 57 . 1 ± 2 . 3 MΩ; respectively; p<0 . 05 ) as well as a greater sag ( difference between the voltage at peak and at steady-state: 4 . 6 ± 1 . 2 mV and 2 . 1 ± 0 . 3 mV , respectively , p<0 . 05 ) ( Figure 3F ) , indicating that these two populations with large soma can be discriminated by their intrinsic electrical properties . Interestingly , the analysis of trains of action potentials generated by a step of depolarizing current distinguished again two distinct subpopulations within the C/S+ group . The first type of C/S+ neurons produces a train of single action potentials similar to those obtained in Ctip2+ cells , while the second type generates doublets or even triplets of action potentials ( Figure 3G , G’ and Figure 3—figure supplement 2B ) . Analysis from the I-V curves or from the action potentials generated at threshold showed further differences between the two types of C/S+ neurons , such as the cell resistance and size of the sag or the characteristics of action potentials and inter-spike intervals ( Figure 3F and Figure 3—figure supplement 2C ) . Taken together , these data support the existence of two subtypes of C/S+ neurons that differ from Ctip2+ cells , and confirm that at P21 neurons co-expressing Ctip2 and Satb2 represent distinct subclasses of cortical PNs in layer V of the S1 cortex . We next aimed at deciphering the mechanisms responsible for the co-expression of Ctip2 and Satb2 in the postnatal somatosensory cortex . Satb2 is known to repress Ctip2 expression by recruiting the Nucleosome Remodeling and Deacetylase ( NuRD ) complex , which in turn deacetylates the Ctip2 locus by interacting with the histone deacetylase 1 ( Hdac1 ) ( Alcamo et al . , 2008; Britanova et al . , 2008 ) . The protooncogene Ski was shown to play a key role in the Hdac1-NuRD complex interaction ( Baranek et al . , 2012 ) ; hence , we hypothesized that the resurgence of C/S+ cells at postnatal stages might be due to the down-regulation of Ski . On the contrary , Ski expression remained high from P0 to P21 and was observed also in several C/S+ cells after birth ( Figure 4—figure supplement 1A–C ) , suggesting that a Ski-dependent mechanism is unlikely to contribute to the postnatal increase of C/S+ neurons in the S1 cortex . Among other candidate genes that might interfere with the Satb2-mediated Ctip2 repression , we selected the transcriptional adaptor Lmo4 . This factor interacts with several components of the NuRD complex , is highly expressed in the rostral F/M region , where C/S+ are more abundant , and only in scattered cells of the pS cortex at P0 ( Figure 4A ) ( Cederquist et al . , 2013; Gomez-Smith et al . , 2010; Huang et al . , 2009; Singh et al . , 2005 ) . Lmo4 expression gradually increases in S1 at postnatal stages , reaching its peak in LL at P7 , then in all layers at P21 ( Figure 4B ) . The timing of Lmo4 expression is thus consistent with the increase of C/S+ cells in S1 from P0 to P21 ( Figure 1B-C and 4B ) , and accordingly , the number of triple Lmo4/Ctip2/Satb2-expressing cells progressively increases from P0 to P21 in layers V and VI ( Figure 4C ) . Thus , the temporal and spatial dynamics of Lmo4 expression in C/S+ neurons indicate that this factor might favor their specification at peri- and postnatal stages of corticogenesis . 10 . 7554/eLife . 09531 . 010Figure 4 . Lmo4 controls the number of Ctip2+ and double C/S+ neurons in the somatosensory cortex . ( A ) . Whole-mount in situ hybridization for Lmo4 on P0 brain ( top left panel ) and immunofluorescence for Lmo4 on coronal sections . ( Below ) Expression of Lmo4 in high-magnification views of frontal motor ( F/M ) and prospective somatosensory ( pS ) coronal sections . ( B ) . Coronal sections from pS and primary somatosensory area ( S1 ) of brains from E14 . 5 to P21 immunostained for Lmo4 . On the right , time course of the percentage of Lmo4+ neurons on the total of DAPI+ cells ( cortical plate -CP- for prenatal brains and lower layers –LL- for postnatal brains ) . ( C ) . Triple immunostaining for Lmo4/Ctip2/Satb2 on coronal sections from pS and S1 E14 . 5 to P21 brains . Bottom squared panels represent high-magnification views of boxes in layer V depicted in top panels . On the right , quantification of Lmo4+ ( blue line ) and Ctip2/Satb2+ ( C/S+ ) cells ( green line ) on the total of DAPI+ cortical cells , and of triple Lmo4/C/S+ cells ( orange ) on the total number of C/S+ neurons . ( D ) . Double immunofluorescence for Satb2 and Ctip2 on coronal sections of P7 control and Lmo4 CKO somatosensory ( S1 ) areas . Bottom squared panels represent high-magnification views of boxes in layers V and VI of top panels . Arrowheads point to double C/S+ neurons . ( E and F ) Quantification and layer distribution of double C/S+ or single Ctip2+ neurons on the total of DAPI+ cells ( E ) , and of double C/S+ on the total of Satb2+ ( top panel F ) or Ctip2+ ( bottom panel F ) neurons . NCX: neocortex , pV: prospective visual , Hip: hippocampus , Str: striatum , UL , upper layer neurons . Scale bars: B , C , D and lower panels of A , 100 μm , upper panels of A , 1mm . Data are represented as means ± SEM . *p≤0 . 05 , **p≤0 . 01 , ***p≤0 . 001 . SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 01010 . 7554/eLife . 09531 . 011Figure 4—figure supplement 1 . Ski is expressed by several C/S+ neurons at postnatal stages . ( A-C ) . Left panel , coronal sections of S1 from P0 ( A ) , P7 ( B ) , and P21 ( C ) triple immunolabeled for Satb2 , Ctip2 , and Ski . Square panels on the right in ( A-C ) represent high-magnification views of boxes depicted in left panels . Arrowheads indicate triple-positive cells for Satb2/Ctip2/Ski . Scale bars: A-C , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 01110 . 7554/eLife . 09531 . 012Figure 4—figure supplement 2 . Unaltered distribution of Satb2+ neurons in the absence of Lmo4 . ( A ) . Coronal sections of pS regions from controls and Lmo4 CKO ( Lmo4 flox/floxEmx1Cre ) P0 brains immunolabeled for Lmo4 . Yellow arrow and asterisk indicate expression and absence of Lmo4 in the neocortex ( NCX ) , respectively . White arrows indicate maintenance of Lmo4 expression in the striatum ( Str ) and thalamus ( Th ) in both genotypes . ( B ) . Quantification and laminar distribution of the total number of Satb2+ cells show no significant changes in Lmo4 CKO S1 cortices compared to control ones at P7 . Data are represented as means ± SEM . SEM , standard error of the mean . Scale bar: A , 600 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 012 To determine whether Lmo4 is required in the specification of C/S+ cells , we first exploited a mouse mutant line ( Lmo4 CKO ) , in which Lmo4 is specifically inactivated in the cortex under the control of the Emx1 promoter ( Huang et al . , 2009 ) ( Figure 4—figure supplement 2A ) . We examined Ctip2 and Satb2 expression in P7 Lmo4 CKO cortices , when normally a high number of S1 lower layer cells and 92% of double C/S+ cells express Lmo4 ( Figure 4B and C ) . Notably , the total number of both C/S+ and Ctip2+ cells is decreased in Lmo4 CKO S1 cortices ( Figure 4D and E ) , whereas the number of Satb2-expressing cells is not particularly affected by the absence of Lmo4 ( Figure 4—figure supplement 2B ) . This might indicate that the alteration in C/S+ cells is mainly due to a general decrease in Ctip2 expression . However , the ratio of C/S+ cells decreased also compared to the total number of Ctip2+ ( in upper layer V and VI ) and Satb2+ cells ( in layer VI ) ( Figure 4F ) . Since the number and distribution of Satb2+ cells does not significantly change , the reduced number of C/S+ cells is most probably due to an increased repression of Ctip2 in Satb2+ neurons . Next , we investigated whether Lmo4 is cell-autonomously required in the specification of double C/S+ cells in layer V by overexpressing Lmo4 in the S1 of WT cortices . To this aim , we cloned the coding sequence of Lmo4 into the pCdk5r1-IRES-EGFP vector ( Figure 5A ) , which drives the selective expression of a given transcript in postmitotic neurons ( Wang et al . , 2007b ) . To evaluate the efficacy of this new construct ( pCdk5r1-Lmo4-EGFP ) , we electroporated mouse brains at E13 . 5 and analyzed the somatosensory cortex at P0 when Lmo4 is only faintly expressed ( Figure 4A ) . Electroporated ( GFP+ ) cells expressed high levels of Lmo4 , whereas similar regions in the contralateral cortex contained just few Lmo4+ cells ( Figure 5B , C ) . 10 . 7554/eLife . 09531 . 013Figure 5 . Lmo4 overexpression increases the number of Ctip2+ and C/S+ neurons in lower layers . ( A ) . Schematic representation of the vector used to overexpress Lmo4 in postmitotic neurons . ( B ) . Immunostaining for Lmo4 on a coronal section of the contralateral ( non-electroporated ) hemisphere of P0 electroporated brains . ( C ) . Coronal sections of a pCdk5r1-Lmo4-IRES-GFP electroporated E13 . 5 hemisphere immunolabeled for Lmo4 and GFP at P0 . Bottom squared panels represent high-magnification views of boxes depicted in upper panels . ( D ) . Immunostaining for Satb2 , Ctip2 , and GFP on coronal sections from P7 S1 cortices electroporated at E13 . 5 with pCdk5r1-IRES-GFP ( on the top ) or pCdk5r1-Lmo4-IRES-GFP ( on the bottom ) . Squared panels represent high-magnification views of boxes depicted in top panels . Filled arrowheads indicate double C/S+ GFP+ cells , whereas empty arrowheads indicate GFP+ cells not expressing Ctip2 . ( E ) . Quantification of Ctip2+/GFP+ cells on the total number of GFP+ cells ( on the top ) , and of ( C/S+ ) /GFP+ cells on the total number of GFP+ cells ( on the bottom ) in layer V of electroporated brains . Data are represented as means ± SEM . *p≤0 . 05 . ( IUE ) In utero electroporated; ( Hip ) hippocampus; ( F/M ) frontal motor area; ( pS ) prospective somatosensory area; SEM , standard error of the mean . Scale bars: B , C 300 µm; C , high magnification: 50 µm , D , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 013 The pCdk5r1-Lmo4-EGFP construct was then electroporated at E13 . 5 , and brains were collected at P7 , when the laminar specification of PNs is nearly completed ( Figure 5D ) . Control ( pCdk5r1-EGFP electroporated ) cells were predominantly Satb2+ and Ctip2- in layer V ( Figure 5D ) , whereas the overexpression of Lmo4 increased the ratio of Ctip2+ neurons among GFP+ cells from 7 . 6 ± 4 . 7% to 31 . 3 ± 3 . 8% ( n = 3 , p = 0 . 03 ) and the percentage of C/S+ neurons from 5 . 2 ± 3 . 4% to 16 . 4 ± 1 . 0% in LL ( n = 3 , p<0 . 05 ) ( Figure 5E ) , supporting a cell-autonomous role for Lmo4 in inducing Ctip2 expression in both Satb2+ and Satb2- cells . Overall , our Lmo4 loss- and gain-of-function approaches both demonstrate that Lmo4 acts in the specification of C/S+ cells primarily by modulating Ctip2 expression in layer V . To further confirm that Lmo4 plays a role in the specification of C/S+ cells , we analyzed the distribution of these cells in mice lacking the transcription factor Nr2f1 ( also called COUP-TFI ) in cortical neurons ( Nr2f1 fl/flEmx1-Cre , from now on Nr2f1 CKO ) ( Armentano et al . , 2007 ) . Nr2f1 CKOs exhibit a remarkable upregulation of Lmo4 in layer V and in the lower part of UL of the somatosensory cortex at P0 ( Alfano et al . , 2014 ) ( Figure 6A–B ) . In agreement with our previous studies ( Alfano et al . , 2014; Tomassy et al . , 2010 ) , the number of Ctip2+ cells is significantly increased in the mutant motorized somatosensory ( mS ) cortex , whereas Satb2+ cells are only slightly augmented ( Figure 6C–D and Figure 6—figure supplement 1A ) . As expected , the number and distribution of C/S+ neurons are also increased on the total of cells and relative to the number of Satb2+ and Ctip2+ cells in Nr2f1 CKOs ( Figure 6C–D and Figure 6—figure supplement 1A ) . Finally , mutant brains showed a much higher number of triple Lmo4/Ctip2/Satb2+ cells in layer V and upper layer VI than controls ( Figure 6—figure supplement 1C ) . While 68% of C/S+ cells express Lmo4 in control cortices at P0 , this ratio rises to 92% in mutant brains ( Figure 6—figure supplement 1D ) , supporting a correlation between increased Lmo4 expression and higher number of C/S+ cells in LL of mutant cortices . 10 . 7554/eLife . 09531 . 014Figure 6 . Increase of Lmo4- and double C/S-expressing neurons in the motorized somatosensory cortex of Nr2f1 CKO brains . ( A ) . Whole-mount in situ hybridization for Lmo4 ( top panels ) and Lmo4 immunofluorescence on coronal sections from the prospective somatosensory ( pS ) cortex ( bottom panels ) of P0 control and Nr2f1 CKO brains . ( B ) . Quantification and layer distribution of Lmo4+ neurons in pS of P0 control and Nr2f1 CKO brains . ( C ) . Coronal sections from the pS of P0 control and Nr2f1 CKO brains immunolabeled for Ctip2 and Satb2 . Bottom squared panels represent high-magnification views of layer V neurons in boxes depicted in top panels . Arrowheads indicate C/S+ neurons . ( D ) . Quantification of Ctip2+ and C/S+ neurons in the pS of P0 control and Nr2f1 CKO brains as a percentage of the total number of cells ( DAPI+ ) or of Satb2+ neurons . ( E ) . Coronal sections of S1 from controls and Nr2f1 CKO P7 brains retrogradely-labeled in the pontine region and in the cervical spinal cord . ( F ) . Immunofluorescence for Satb2 and Ctip2 on P7 Nr2f1 CKO S1 retrogradely-labeled cortices . Filled arrowheads in high-magnification views indicate retrogradely-labeled subcerebral projection neurons ( SCPNs ) double positive for Satb2 and Ctip2 , whereas empty arrowheads indicate retrogradely-labeled SCPNs positive for Ctip2 . ( G ) . Quantification of Ctip2+ and C/S+ retrogradelylabeled SCPNs on the total number of labeled PNs in layers Va , Vb , and VIa of Nr2f1 CKO brains . F/M , frontal motor area; pS , prospective primary somatosensory area; pV , prospective primary visual area; CSpPN , corticospinal projection neurons . UL , upper layers; VI , layer VI . Data are represented as means ± SEM . *p≤0 . 05 , **p≤0 . 01 , ***p≤0 . 001 . SEM , standard error of the mean . Scale bars: A , 1 mm; lower panel A , C , 100 μm , E , F , 200 μm; high-magnification views in F , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 01410 . 7554/eLife . 09531 . 015Figure 6—figure supplement 1 . Increase of double Ctip2/Satb2+ and triple Lmo4/Ctip2/Satb2+ neurons in the motorized Nr2f1 CKO somatosensory cortex . ( A ) . Quantification and layers V and VI distribution of the total population of single Satb2+ cells ( left ) and of double C/S+ cells on the total of Ctip2+ neurons ( right ) in P0 control ( Ctrl ) and Nr2f1 CKO ( CKO ) pS cortices . ( B ) . Triple immunostaining for Lmo4/Ctip2/Satb2 on P0 control and Nr2f1 CKO pS coronal sections;on the right , high-magnification views of the boxes depicted in left panels . Arrowheads indicate triple Lmo4+/ ( C/S+ ) neurons . ( C ) . Distribution of triple Lmo4+/ ( C/S+ ) neurons in layers V and VI of pS in controls and mutant brains . The quantity of neurons for each layer was indicated as an absolute number counted in a 600-μm wide region of P0 control and Nr2f1 CKO cortices . ( D ) . Pie charts representing the percentage of Lmo4+/ ( C/S+ ) neurons on the total of C/S+ neurons in the pS of P0 control and Nr2f1 CKO brains . Data are represented as means ± SEM . *p≤0 . 05 , **p≤0 . 01 , ***p≤0 . 001 . SEM , standard error of the mean . Scale bars: B , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 01510 . 7554/eLife . 09531 . 016Figure 6—figure supplement 2 . Decrease of Ctip2 expression in the motorized Nr2f1 CKO somatosensory cortex after Lmo4 downregulation . ( A , C ) Immunostaining for Lmo4 and GFP on the motorized somatosensory cortex ( mS1 ) of P0 Nr2f1 CKO mice electroporated at E13 . 5 with a control shRNA ( A ) or Lmo4 shRNA ( C ) . To the right , details of boxes depicted in left panels . ( B , D ) On the top , coronal sections from the mS1 cortex of P0 Nr2f1 CKO mice electroporated at E13 . 5 with a control shRNA ( B ) or Lmo4 shRNA ( D ) immunolabeled for Ctip2 and GFP . On the bottom , high-magnification views of boxes depicted on top panels . Arrowheads indicate GFP cells electroporated with the control shRNA or Lmo4 shRNA expressing Ctip2 . UL , upper layer neurons; Str , striatum; WM , white matter; mS1 , motorized somatosensory cortex; NCX , neocortex . Scale bars: ( A-D ) , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 016 Next , we investigated whether ectopic Lmo4 expression and increased number of C/S+ cells in the mutant somatosensory cortex coincided with altered neuronal connectivity of layer V neurons . Previous work showed that corticospinal projection neurons ( CSpPNs ) were abnormally located in layer VIa of Nr2f1 CKO ( Tomassy et al . , 2010 ) . By injecting CTB-coated beads either into the brainstem or the spinal cord regions of control and Nr2f1 mutant brains , we observed an expansion of cortico-brainstem projection neurons ( CBPNs ) in layer Va and confirmed the mispositioning of the CSpPNs in upper layer VI of mutant cortices ( Figure 6E ) . Interestingly , the highest percentage ( 66 . 4 ± 6 . 7% ) of CBPNs co-expressing Ctip2 and Satb2 was in layer Va , whereas none of the labeled cells co-expressed these genes in layer VIa ( Figure 6F and G ) . This suggests that Lmo4 increase may underlie both the strong increase in C/S+ cells and the shift of connectivity from corticospinal to corticobrainstem targets observed in mutant layer V . Thus , to demonstrate that the upregulation of Lmo4 in the Nr2f1 CKO brains was directly involved in layer V Ctip2 radial expansion , we electroporated an Lmo4-specific shRNA construct ( Qin et al . , 2012 ) into E13 . 5 mutant brains and analyzed Lmo4 and Ctip2 laminar distribution in P0 electroporated mS1 cortices ( Figure 6—figure supplement 2 ) . While GFP+ control cells co-localize with Lmo4 and Ctip2 ( Figure 6—figure supplement 2A , B ) , Lmo4 shRNA-expressing cells show a drastic reduction not only in Lmo4 expression , as expected , but also in Ctip2 levels in layer Va ( Figure 6—figure supplement 2C , D ) . This suggests that Ctip2 overexpression in layer Va is mainly due to Lmo4 increase and that Nr2f1 loss not only upregulates Lmo4 levels but also favors its action on Ctip2 expression . In addition , cells electroporated with the control shRNA show a delayed migration to the apical region of the cortex , a phenotype that was already described in our previous publication ( Alfano et al . , 2011 ) . Notably , downregulating Lmo4 rescues the radial migratory defect and allows electroporated cells to reach the proper radial position in the upper regions of P0 cortices . These data demonstrate that Lmo4 is functionally involved in the abnormal upregulation of layer V Ctip2 expression observed in the absence of Nr2f1 , and more in general , support a key role for Lmo4 in Ctip2 de-repression . Next , we investigated whether Lmo4 was able to positively regulate Ctip2 by directly interfering with the molecular machinery underlying Satb2-mediated Ctip2 repression ( Alcamo et al . , 2008; Britanova et al . , 2008 ) . Lmo4 was shown to bind Histone deacetylases 1 and 2 ( Hdac1 and 2 ) to repress its downstream target genes ( Singh et al . , 2005 ) . Since the Hdac1-NuRD-Satb2 complex assembly is fundamental for Ctip2 repression ( Alcamo et al . , 2008; Baranek et al . , 2012; Britanova et al . , 2008 ) , we examined whether Lmo4 was able to compete with Satb2 for Hdac1 interaction . To this aim , we performed immunoprecipitation with Lmo4- and Satb2-specific antibodies on nuclear proteins extracted from control and Nr2f1 CKO P1 cortices , where Lmo4 and Ctip2 are strongly upregulated ( Figure 7A ) . Immunoprecipitated protein fractions were analyzed by Western blot using an antibody specific for Hdac1 ( Alcamo et al . , 2008; Britanova et al . , 2008; Gyorgy et al . , 2008 ) . Comparable levels of Hdac1 were co-immunoprecipitated ( Co-IP ) with Lmo4 and Satb2 antibodies in control conditions , whereas Hdac1-Lmo4 interaction increased at the expense of Hdac1-Satb2 in Nr2f1 CKO brains . These changes in the amount of each complex were not due to dramatic changes in the total amount of Satb2 and Lmo4 proteins between wt and Nr2f1 mutant brains . Indeed , Satb2 protein levels resulted unaltered between the two conditions , whereas the amount of Lmo4 resulted only slightly increased in Nr2f1 mutants ( Figure 7A ) . Since the overall Lmo4 upregulation is not remarkable and takes place particularly in the mS1 cortex , the higher amount of Hdac1-Lmo4 complex in mutant Co-IP fractions might be due to a higher binding affinity between these factors in somatosensory regions . Accordingly , the number of C/S+ cells is remarkably higher in lower layers of S1 than in those of M1 in P7 wt brains ( Figure 1E ) despite similar Lmo4 levels between S1 and M1 areas ( our data and ( Huang et al . , 2009 ) . 10 . 7554/eLife . 09531 . 017Figure 7 . Lmo4 interacts with Hdac1 and prevents its binding to the Ctip2 locus . ( A ) . On the top , Western blot on nuclear extracts from controls and Nr2f1 CKO P1 cortices immunoprecipitated with antibodies specific for Lmo4 ( IP Lmo4 ) , Satb2 ( IP Satb2 ) , or an unrelated epitope ( IP control ) . IP fractions were analyzed using an antibody against Hdac1 , whereas the corresponding input fractions were analyzed with an antibody specific for β-actin . On the bottom-left , Western blot on nuclear extracts from controls and Nr2f1 CKO P1 cortices for Satb2 , Hdac1 , Lmo4 , and β-actin . To the right , ratio between Hdac1-Lmo4 and Hdac1-Satb2 complexes immunoprecipitated from control and Nr2f1 CKO extracts . ( B ) . Western blot performed on nuclear extracts from controls and Nr2f1 CKO P1 cortices immunoprecipitated with specific antibody for Lmo4 . IP fractions were analyzed using an antibody against Ski , Satb2 or an unrelated epitope ( control ) . ( C ) . On the left , semi-quantitative PCR performed on Chromatin-immunoprecipitation ( ChIP ) samples from controls and Nr2f1 CKO P1 cortices . The assay was performed using antibodies against Hdac1 , the anti-Histone H4 ( acetyl K12 ) ( H4K12 ) and primers amplifying a MAR1 sequence on Ctip2 and Rnd2 loci . On the right , QPCR performed on ChIP samples from controls and Nr2f1 CKO cortices . ( D ) . On the top , Western blot on nuclear extracts from COS7 cells transfected with an equal amount of Satb2- and an increasing amount of Lmo4-expressing vectors and immunoprecipitated with antibodies specific for Satb2 ( IP Satb2 ) , Lmo4 ( IP Lmo4 ) , or an unrelated epitope ( IP control ) . IP fractions were analyzed using an antibody against Hdac1 . On the bottom , Western blot on nuclear extracts from the transfected COS7 cells with specific antibodies for Satb2 , Lmo4 , and β-actin showing increase of Lmo4 and similar Satb2 and β-actin protein levels . ( E ) . Schematic model of the putative mechanism by which Lmo4 de-represses Ctip2 expression . . **p≤0 . 01 . MAR1: Matrix attachment region 1; NuRD complex: Nucleosome Remodeling and Deacetylase complex . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 017 Finally , we investigated whether Lmo4 binds Hdac1 by interacting with the Ski-Hdac1 complex and preventing its interaction with Satb2 ( Baranek et al . , 2012 ) . To this aim , we analyzed the Lmo4-immunoprecipitated nuclear fractions from control and mutant cortices with the antibodies for Ski and Satb2 . We found that Lmo4 binds Ski consistently , whereas it interacts very limitedly with Satb2 , suggesting that Lmo4 normally interacts with Ski before it binds to the Satb2-NuRD complex . However , the increased Hdac1-Lmo4 binding does not seem to be related to higher interaction between Lmo4 and Ski , since the amount of bound Ski does not change in the mutant extracts ( Figure 7B ) . We next verified whether the chromatin state in the Ctip2 genetic locus varies between control and mutant cortices ( Figure 7C ) . We carried out a chromatin immunoprecipitation ( ChIP ) assay with the anti-Hdac1 antibody and with an antibody specific for the acetylated form of histone 4 ( H4K12ac ) , whose levels are proportional to the rate of transcriptional activity . Immunoprecipitated chromatin fractions from P1 cortices were analyzed by semi-quantitative PCR and confirmed by QPCR using primers amplifying part of the matrix attachment region ( MAR ) in the Ctip2 locus ( Britanova et al . , 2008 ) . As a control , we used primers amplifying a sequence in the last intron of the Rnd2 gene containing a previously described Nr2f1-binding site ( Alfano et al . , 2011 ) . Our data show that a lower amount of Hdac1 binds the Ctip2 locus , which accordingly appears more acetylated and thereby more active ( Figure 7C ) . No Hdac1 binding and acetylation differences were observed with control primers , confirming specificity of its effect on the Ctip2 locus ( Figure 7C ) . Since changes in the Hdac1-Satb2 interaction and in Ctip2 de-repression observed in Nr2f1 mutant brains might be not only or not directly related to the observed Lmo4 increase in the mS1 region , we repeated the CoIP experiments by overexpressing Lmo4 and Satb2 in COS7 cells . To confirm the competition between Lmo4 and Satb2 for Hdac1 binding , COS7 cells were transfected with increasing quantities of a vector expressing Lmo4 under the control a CMV enhancer ( pCIG2-Lmo4 ) together with a constant amount of the pCAG-Satb2 plasmid ( Britanova et al . , 2008 ) ( Figure 7D ) . Cell extracts , analyzed by immunoprecipitating proteins with an anti-Hdac1 antibody , showed a progressive increase in Hdac1-Lmo4 interaction , in line with the Lmo4 increase , at the expense of Hdac1-Satb2 binding ( Figure 7D ) . This confirmed the competition model and put in direct relation the Lmo4 increase with the decreased Hdac1-Satb2 interaction . Overall , our analysis indicates that Lmo4 progressively interferes with the Satb2-mediated Ctip2 repression by sequestering Hdac1 before it interacts with the Satb2-NuRD complex on the Ctip2 locus and hence , favoring Ctip2 and Satb2 co-localization in cortical LL during postnatal stages of development ( Figure 7E ) .
After birth , Satb2 and Ctip2 are not any more an exclusive hallmark of callosal ( CPN ) or subcerebral ( SCPN ) projection neurons , respectively , but are co-expressed in distinct subclasses of CPNs and SCPNs with specific connectivity profiles , morphological , and electrophysiological characteristics . In line with our results , Satb2 expression was shown to be associated with the SCPN markers Fezf2 and Sox5 in CPN populations of motor areas ( Sohur et al . , 2014; Tantirigama et al . , 2014 ) . In addition , Satb2 is not only required for CPNs but also for the proper differentiation and axon pathfinding of SCPNs ( Leone et al . , 2014 ) . It is thus possible that early-born CPNs , which originate at a similar time as SCPNs and reach lower layers in S1 , are more similar to subcerebral than late-born CPNs of upper layers . We also show that C/S+ neurons expressing Er81 define a distinct subpopulation of CPNs residing in layer V of S1 . This subpopulation is unlikely to project to the contralateral striatum , since callosal-striatal neurons are almost absent in S1 after P15 and fail to express Ctip2 but are instead positive for Sox5 ( Sohur et al . , 2014 ) . Accordingly , none of the C/S+ CPNs identified here were positive for Sox5 , whereas a group of C/S+ SCPNs clearly expressed Sox5 . Our molecular analysis thus identified novel subclasses of SCPNs and CPNs major neuronal groups with distinct features , which can be specifically identified by Ctip2 andSatb2 co-expression . The morphological and electrophysiological characterizations revealed hybrid but also unique features of C/S+ subpopulations compared to single Ctip2+ cells . Although a subset of both populations projects to subcerebral targets , the C/S+ neuronal axons do not reach the spinal cord and show typical morphological features of CPNs , such as a small soma , long , and thin apical dendrites and late bifurcation of the apical tuft ( Hattox and Nelson , 2007 ) . However , a small group of C/S+ neurons also share characteristics of SCPNs with single Ctip2+ cells , such as a large soma , high number of secondary dendrites , thick apical dendrites and an early bifurcation of the apical tuft ( De la Rossa et al . , 2013; Hattox and Nelson , 2007; Tantirigama et al . , 2014 ) . Thus , C/S+ cells include different subsets of PNs sharing some typical CPN and SCPN morphological features . Electrophysiological recordings also revealed that C/S+ neurons exhibit a lower sag value than single Ctip2+ cells , indicating that C/S+ cells might acquire some UL electrical features , since normally these neurons have a lower hyperpolarization-activated current ( Ih ) than LL neurons ( De la Rossa et al . , 2013; Sheets et al . , 2011 ) . This is also reminiscent of the observations made in Satb2 mutant or Fezf2-overexpressing brains , where UL neurons acquire electrical features of LL neurons ( De la Rossa et al . , 2013; Leone et al . , 2014 ) . Moreover , Ctip2+ neurons show higher ISI ( first interspike interval ) value than C/S+ cells . A similar variation in ISI values was previously described between UL and LL neurons ( De la Rossa et al . , 2013 ) , suggesting that the co-expression of Satb2 in Ctip2+ layer V neurons may result in electrical features characteristic of UL neurons . Finally , since these variations in electrophysiological properties were observed by analyzing cells with similar morphologies , C/S+ and Ctip2+ neurons might diverge in the expression of ion channels and pumps ( Oswald et al . , 2013; Staff et al . , 2000 ) . In the future , it might be interesting to investigate eventual correlations between the expression of Satb2 and/or Lmo4 and the function/expression of specific ion channels in layer V neurons . Our time-course analysis of C/S+ neurons confirms that only a small population of cells maintains the expression of these two antithetic factors during perinatal stages of corticogenesis ( Baranek et al . , 2012; Leone et al . , 2014 ) ( Figure 8 ) . This is consistent with the paradigm that competing molecular programs direct the differentiation of major PN classes during late embryonic stages of corticogenesis ( reviewed in Greig et al . , 2013 ) . However , we found that distinct subpopulations of C/S+ PNs are maintained and gradually increase postnatally in lower layers of the S1 cortex . 10 . 7554/eLife . 09531 . 018Figure 8 . Hypothetical model of layer V neuronal subtype refinement . After birth , Lmo4 expression progressively increases in Satb2+ and in Ctip2+ cells of S1 cortex . Lmo4 de-represses Ctip2 expression in Satb2+ cells and probably Satb2 in Ctip2+ cells . According to the different dynamics leading to the refinement of distinct C/S+ subclasses , these cells will project either to subcerebral or to contralateral targets . Vertically oriented arrows indicate connectivity at postnatal stages . The blue dashed line and blue circles in different cells delimit territories of Lmo4 expression . Blue marine circles surrounding cells depict subcerebral projection neuron ( SCPN ) subpopulations , whereas orange ones delineate callosal neuron ( CPN ) subgroups . Black lines delimit different neuronal populations of the somatosensory cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 09531 . 018 This study also demonstrates that Lmo4 progressively increases and co-localizes with double C/S+ neurons after birth in the somatosensory area . Lmo4 is a widely expressed transcriptional modulator known to regulate several key biological processes , from cell growth to fate determination ( Sang et al . , 2014 ) . It is well known that Lmo4 acts as a scaffolding protein for the assembly of multi-protein complexes and interacts with several co-factors of the NuRD complex ( Gomez-Smith et al . , 2010; Singh et al . , 2005; Wang et al . , 2007a ) . Although other factors might be involved in Satb2/Ctip2 co-expression , our work demonstrates that Lmo4 de-represses Ctip2 by sequestering Hdac1 , a critical component of the NuRD complex recruited by Satb2 on the Ctip2 locus to inactivate its transcription . Overall , our data constitute first direct evidence that the control of epigenetic mechanisms may underlie area-specific variations in neuronal features . Most importantly , such processes take place after birth and seem to contribute to the maturation and refinement rather than to the initial specification of neuronal subtypes of the somatosensory area . Obviously , this study does not constitute an exhaustive investigation on the molecular mechanisms regulated by Lmo4 , which seems to be also able to induce Satb2 expression in Ctip2-positive cells , since a notable portion of C/S+ cells seems to derive from a subpopulation of Ctip2+ cells ( model in Figure 8 ) . This is not totally unexpected in light of previous reports that correlated Lmo4 expression with callosal ( Satb2+ ) development ( Azim et al . , 2009; Molnar and Cheung , 2006; Ye et al . , 2015 ) . Accordingly , we found that the number of C/S+ cells decreased in Lmo4 mutants not only on the total of the Satb2+ subpopulation but also on the total of the Ctip2+ one , suggesting that C/S+ cells may also derive from Lmo4-mediated de-repression ( and/or activation ) of Satb2 in Ctip2+ cells . Finally , we noticed that Ctip2 was poorly upregulated in upper layers of Nr2f1 CKO motorized somatosensory cortices and of Lmo4-electroporated cortices . This could be explained by the presence of a laminar-specific transcription factor , which would act synergistically with Lmo4 to efficiently activate Ctip2 expression in lower layers . More experiments are required to test this hypothesis . Lmo4 had already been implicated in specifying SCPN and CPN identities in the rostral motor cortex ( Cederquist et al . , 2013 ) , but a similar function in sensory areas was not previously unveiled . Here , we show that Lmo4 acts preferentially in layers V and VI of the S1 cortex in which the majority of neurons ( SCPN ) projects subcerebrally and expresses Ctip2 at different levels . Lmo4 levels gradually increase from layer VI to upper layers at postnatal stages , together with the progressive specification of C/S+ cells . Interestingly , we report that C/S+ neuron projections target the brainstem , but not the spinal cord . If Lmo4 and/or Satb2 and Ctip2 co-localization modulates axon targeting , what are their mechanisms of action ? Ctip2 and Satb2 inhibit the expression of DCC and Unc5C , respectively , two receptors of the guidance cue Netrin-1 ( Srivatsa et al . , 2014 ) . Since these receptors are involved in the midline crossing of corticospinal axons ( Finger et al . , 2002 ) , one conceivable hypothesis would be that co-expression of Ctip2 and Satb2 inhibits both receptors , thus preventing subcerberal PNs to reach the spinal cord . Another explanation could be that Lmo4 inhibits neurite outgrowth by repressing the expression of the receptor tyrosine kinase Alk that plays key roles in neuritogenesis ( Lasek et al . , 2011 ) . Alternatively , Lmo4 might interact with Lhx2 , another LIM homeobox transcription factor known to regulate the expression of the guidance receptors Ephrin-A5 as well as Robo1 and 2 , which in turn control different steps of corticospinal axon pathfinding ( Marcos-Mondejar et al . , 2012; Shetty et al . , 2013 ) . Thus , either Lmo4 expression or Ctip2 and Satb2 co-expression might control several neuronal processes allowing the correct maturation and/or refinement of PT and IT subtypes during postnatal stages of corticogenesis . We envisage that also other transcription factors normally involved in determining distinct classes of PNs during development , might take a similar postnatal function during cortical maturation .
Overall , we show that area- and time-specific changes in common molecular mechanisms modulate the final features of both CPN and SCPN subpopulations in an area- and time-specific manner . More in general , our analysis provides first direct evidence of a common developmental program directing the molecular and cellular maturation of both IT and PT projection neurons . This might have important implications in the study of neocortical development , since 'serial homologies' ( i . e . similar connectivity or molecular codes ) among neocortical areas might be due to variations on a 'common theme' , as previously suggested ( Harris and Shepherd , 2015 ) , rather than to multiple independent and area-specific genetic programs .
All mouse experiments were conducted according to national and international guidelines and have been approved by the local ethical committee ( CIEPAL NCE/2011-23 ) . Nr2f1 fl/fl and Lmo4 fl/fl mice were crossed to Emx1-Cre mice to inactivate either Nr2f1 or Lmo4 in cortical cells . WT , Nr2f1 fl/fl , or Lmo4 fl/fl were taken as controls . Mouse lines were genotyped as previously described ( Armentano et al . , 2007; Armentano et al . , 2006; Goebbels et al . , 2006; Huang et al . , 2009 ) . Thy1-eYFP-H mice were obtained from The Jackson Laboratory and genotyped as described in ( Feng et al . , 2000 ) . Midday of the day of the vaginal plug was considered as embryonic day 0 . 5 ( E0 . 5 ) . Mice at P0 , P7 , and P21 were intracardially perfused with paraformaldehyde ( PFA ) 4% . Embryonic and postnatal brain samples were fixed either for 2 hr ( for immunohistochemistry ) or for over-night ( forin situ hybridization [ISH] ) at 4°C in PFA 4% . Brain slices used for patch clamp experiments were fixed 4 hr at 4°C in 4% PFA after recording . Samples were , then , either embedded in Optimal Cutting Temperature ( OCT ) medium ( JUNG , Germany ) after being equilibrated to 30% sucrose , and cut on a Leica cryostat , or gradually dehydrated to 96% ethanol for whole-mount ISH . Samples to be used for floating immunofluorescences and cholera toxin-injected brains were embedded in 4% agarose after fixation and then cut on a Leica vibratome at 200 µm . No samples were excluded in this study and for each experiment at least three animals from different litters were used . Immunofluorescences on cryosections ( CI ) were performed as previously described ( Armentano et al . , 2007; Armentano et al . , 2006 ) . Floating immunofluorescence ( FI ) was performed on vibratome sections , which were blocked with 10% goat serum , 3% bovin serum albumin ( BSA ) , and 0 . 3% triton X-100 over night at 4°C . Primary and secondary antibodies were carried out over night at 4°C . The following primary antibodies were used: mouse anti-Satb2 ( dil . CI = 1:20 , FI = 1:80 , Abcam , UK ) , rat anti-Ctip2 ( dil . CI = 1:300 , FI = 1:400 , Abcam ) , rabbit anti-Ctip2 ( 1:500 , Abcam ) , rabbit anti-Sox5 ( 1:300 , Gentaur , France ) , rabbit anti-Ski ( 1:50 , Santa Cruz Biotechnology , Dallas , Texas ) , rat anti-Lmo4 ( 1:500 , gift from Valsvader’s lab ) , rabbit anti-Er81 ( 1:1000 , gift from Arber’s Lab ) , guinea pig anti-Bhlhb5 ( 1:50000 , gift from Novitch’s Lab ) , rabbit anti-GFP ( 1:1000 , Molecular Probes , Eugene , Oregon ) , chicken anti-GFP ( 1:800 , Abcam ) . The following secondary antibodies were used: goat anti-rabbit FC ( 488 , 594 , 633 ) , goat anti-rat FC ( 488 , 594 , 633 ) , goat anti-mouse FC ( 488 , 594 , 633 ) , and goat anti-guinea pig FC ( 488 , 594 , 633 ) ( dil . CI = 1:300 , FI = 1:400 , Life Technologies , Thermo Fisher Scientific , USA ) , goat anti-rabbit FC 350 ( 1:150 , Life Technologies ) and donkey anti-mouse FC 405 ( 1:300 , Abcam ) . To reveal biocytin injected in patch-clamped cells , it was used Texas Red avidin D ( 1:500 , Vector Laboratories , Burlingame , California ) . Slices were mounted with the following mounting solution: 80% glycerol , 2% N-propyl gallate , 1 µg/ml Hoechst ( Invitrogen , Thermo Fisher Scientific , USA ) . Whole-mount ISH was performed as described in ( Alfano et al . , 2014; Alfano et al . , 2011; Armentano et al . , 2007 ) . The antisense Lmo4 RNA probe was labeled using the DIG RNA labelling Kit ( Roche , Switzerland ) following the manufacturer’s instructions . To synthesize the pCdk5r1-Lmo4-IRES-GFP , the Lmo4 ORF was amplified using available cDNA with the following primers: Mlu1-Lmo4 . fw ( 5’GGACGCGTTGAGAGCAGCTC3’ ) and MluI-Lmo4 . rev ( 5’GGACGCGTTTCTGCATTACTC3’ ) . These primers were designed with an Mlu1 restriction cassette at their 5’ end . Once amplified , the Lmo4 amplicon was purified using the QIAGEN PCR Purification Kit ( following manufacturer’s protocols ) and digested with Mlu1 ( Biolabs , Ipswich , Massachusetts ) . The digested Lmo4 ORF was cloned into the empty pCdk5r1-IRES-GFP vector ( digested as well with Mlu1 ) . The plasmid was validated and sequenced using the following primer: pCDK5C . fw ( 5’-AGGACTAAACGCGTCGTGTCC-3’ ) . To generate the pCIG2-Lmo4-IRES-GFP , the Lmo4 variant 2 mRNA sequence was amplified using available P0 cDNA and the following primers: Lmo4_VAR2 . FW ( 5’-GAAGTCCCCGAGCTGGTTTG-3’ ) and Lmo4 . REV ( 5’- CCATACTAGAGCAAATGTCTCTG-3’ ) . Lmo4 amplicon was cloned into the pCRII-TOPO vector ( Invitrogen ) according to manufacturer's instructions and , then , excised by using SpeI and EcoRV restriction enzymes ( Biolabs ) . The SpeI ends was made blunt by fill-in with Klenow fragment ( Biolabs ) . Finally , the Lmo4 fragment was cloned into a pCIG2 plasmid ( Heng et al . , 2008 ) previously digested with SmaI ( Biolabs ) . The pCIG2 ( pCAGGS-IRES-GFP2 ) is a modified version of the pCIG vector ( Megason and McMahon , 2002 ) , which was obtained by inserting an IRES-GFP sequence into the pCAGGS vector ( Niwa et al . , 1991 ) . Positive clones were amplified and purified by the QIAGEN Endofree Maxiprep Kit . In utero electroporations of pCdk5r1-IRES-EGFP , pCdk5r1-Lmo4-IRES-EGFP , control or Lmo4-specific shRNA were performed as previously described ( Alfano et al . , 2011; Tabata and Nakajima , 2001 ) . Briefly , after a 3-cm laparotomy on deeply anesthetized pregnant females and once extroflected uteri , the DNA mix ( 1 mg/ml ) was injected into the lateral ventricle of E13 . 5 embryos using a Femtojet microinjector ( Eppendorf , Germany ) . The electroporations were performed on whole heads using a Tweezertrode electrode ( diameter 7 mm; BTX ) connected to a NEPA21 electroporator ( NEPAGENE , Japan ) with the following parameters: four 37 V pulses , P ( on ) 50 ms , P ( off ) 1 s , 5% decay . Then uteri were reallocated in the abdominal cavity , and both peritoneum and abdominal skin were sewn with surgical sutures ( B . Braun Surgical , Germany ) . COS7 cells were cultured in DMEM ( 4 . 5 g/L; Invitrogen ) containing 10% FCS . Transient transfections of Satb2 and Lmo4 were performed using Lipofectamine TM 2000 ( Invitrogen ) according to manufacturer’s instructions at a cell confluence of 60-70% . An equal amount of pCAG-Satb2 plasmid ( gift from V . Tarabykin’s lab ) was added to the cells with increasing amount of pCIG2-Lmo4-IRES-GFP plasmid . A pCIG2-IRES-GFP plasmid was used to compensate any variations in the amount of the pCIG2-Lmo4-IRES-GFP by maintaining the total quantity of transfected DNA constant . Cells were harvested for co-immunoprecipitation and immunoblotting 48 hr after transfection . Co-immunoprecipitation was performed as described in ( Britanova et al . , 2008 ) with some modifications . Nuclear proteins were extracted from P1 control and Nr2f1 CKO cerebral cortices , or from harvested cells , using the NE-PER kit ( Thermo Fisher Scientific ) according to manufacturer’s instructions . Nuclear extracts were then dialyzed against buffer D ( 20% glycerol , 20 mM HEPES ( pH = 7 . 9 ) , 100 mM KCl , 0 . 2 mM EDTA , 0 . 5 mM DTT , 0 . 5 mM PMSF , all from Sigma-Aldrich ) in Slyde-A-Lyzer Mini Dialysis Units ( 3500 M . W . C . O . from Fisher Scientific , Thermo Fisher Scientific ) . For pre-clearing , 50 μg of the nuclear extracts were incubated with 100 μl Protein A Sepharose 50% bead slurry ( Sigma-Aldrich ) . The pre-cleared nuclear extracts were then immunoprecipitated with either 2 μg of mouse anti-Satb2 ( Abcam ) or of rat anti-Lmo4 antibody ( a gift from Jane Visvader’s lab ) . A mouse anti-Brdu antibody ( Sigma-Aldrich ) was used as control . Immunocomplexes were collected by adding 100 μl of 50% Protein A-Sepharose bead slurry to the mix . The bound fraction was separated by pulse centrifugation and pelleted beads and input were re-suspended in 1x Nupage loading buffer ( Invitrogen ) . Samples were loaded on a 10% SDS polyacrylamide gel and subjected to standard SDS-PAGE electrophoresis on Mini-Protean tetra cell ( Biorad , Hercules , California ) . Then , immunocomplexes were transferred to Hybond-P membrane ( Amersham , GE Healthcare , UK ) via a Trans-Blot SD Semi-Dry Transfer Cell ( Biorad ) . Immunoblotting on total nuclear extracts , bound and unbound fractions was performed with the following antibodies: rat anti-Lmo4 ( 1/500 , gift from J . Visvader ) , rabbit anti-Hdac1 ( 1:500 , Millipore , Merck , Germany ) , rabbit anti-Ski ( 1:50 , Santa Cruz ) , mouse anti-Satb2 ( 1:50 , Abcam ) , and rabbit anti-β-actin ( 1:500 , Abcam ) . Primary antibodies were then detected by embedding the membrane in anti-rabbit , anti mouse or anti-rat biotinylated antibodies ( 1:500 , Vector ) and successively in ABC mix ( Vector Laboratories ) . Revelation of the signals was performed by SuperSignal West Pico Chemiluminescent Kit ( Thermo Scientific ) , and images were taken by Luminescent Image Analyzer LAS-3000 ( Fujifilm , Japan ) . Chromatin-immunoprecipitation ( ChIP ) assay on genomic DNA from controls and Nr2f1 CKO cortices was performed as described in ( Kuo and Allis , 1999 ) . Neocortices were dissected from 7 controls and Nr2f1 CKO P1 pups , and diced in ice cold Hanks Buffered Saline Solution . Proteins were crosslinked to DNA by adding 1% formaldehyde to the solution . The tissue was than disrupted by homogenization in lysis buffer ( 20 mM HEPES pH7 . 4 , 1 mM EDTA , 150 mM NaCl , 1% SDS , 125 mM Glycine , PMSF 0 . 2 mg/ml ) . Nuclei were collected by centrifugation , re-suspended in sonication buffer ( 20 mM HEPES pH7 . 4 , 1 mM EDTA , 150 mM NaCl , 0 . 4% SDS , PMSF 0 . 2mg/ml ) and disrupted by 6 pulses of 10 µ amplitude in a Soniprep150 Sonicator ( Sanyo ) . Before immunoprecipitation , samples were pre-cleared 1 hr in 50% ProteinA-Sepharose slurry and then incubated ON at 4°C with 3 µg of the following antibodies: rabbit anti-Hdac1 ( Millipore ) , rabbit anti-H4K12 ( Abcam ) , and a control antibody ( rabbit anti-GFP , Molecular probe ) . An aliquot of DNA was not immunoprecipitated and used as a control; 0 . 5 μl of DNA from each sample were used to perform a PCR for semiquantitative analysis of the ChIP experiment using the following primers: for the Ctip2 locus , Ctip2MAR . fw 5’-GCTTGGACTCAGTGTACCTC-3’ and Ctip2MAR . rev 5’-CAAGAAAGCACACACCGAGA-3’ and for the Rnd2 locus , BsA . fw 5’-CGTTTGACCTTCCACCTTAG-3’ and BsA . rev: 5’-TCCCACTTGCTTGGCCAGC-3’ . PCR bands were acquired by FUJI 3000 LAS intelligent dark box equipped with a CCD camera . Hdac1 and H4k12 fold enrichment on Ctip2 and Rnd2 loci were tested by QPCR analysis of ChIP samples using the LightCycler 480 Real-Time PCR System ( Roche ) and the above-mentioned primers . For retrograde labeling , anesthetized P2 pups were placed on a stereotaxic apparatus and injected with the cholera toxin subunit B ( CTB- 1 mg/ml; Invitrogen ) conjugated fluorophores ( Alexa Fluor , Thermo Fisher Scientific ) in different brain regions . CPN in the somatosensory cortex were retrogradely labeled via injection of 92 nl of Alexa Fluor 488-conjugated CTB . Coordinates ( in mm ) were: AP: +1 . 2; and ML: 1 . 3 from the lambda; DV: 0 . 2 from the pial surface . Subcerebral injections were performed under ultrasound guidance using a Vevo 770 ultrasound backscatter microscopy system ( Visual Sonics , Canada ) at cervical vertebral level 1 ( C1 ) to C2 to label corticospinal projection neurons ( CSpPN ) , or at the midbrain-hindbrain junction to label subcerebral projection neurons ( SCPNvia 92 nl injections of Alexa Fluor 555-conjugated CTB . Dual retrograde labeling of SCPN and CPN was performed by injecting Red Retrobeads or Green IX Retrobeads ( Lumafluor Inc , Durham , North Carolina ) , respectively , in P2 and P3 pups brains in the same regions described above . Injected pups were perfused at P7 and brains were collected as described in previous sections . Images of immunostained cryosections were acquired using a DM6000 microscope ( Leica , Germany ) equipped with LEICA DFC 310 FX camera , while images of immunofluorescences on floating-thick sections were taken with a Zeiss 710 confocal microscope . ISH and whole-mount ISH were acquired by a Zeiss Imager Z1 microscope equipped with AXIOCAM MRm camera and a Leica Spot microscope , respectively . Images from optical and confocal microscopes were then processed using Photoshop and Zen-lite 2012 softwares , respectively . Images of P0 coronal sections from the prospective somatosensory ( pS ) and the frontal/motor ( F/M ) regions were subdivided into 6 bins . Bins 1 and 2 represent layer VI , bins 3 and 4 represent layer V , while bins 5 and 6 represent the upper layers . At P7 the radial surface of analyzed brain regions was subdivided into 10 bins: bins 1–3 represent layer VI , bins 4–6 represent layer V , and bins 7–10 represent upper layers . Counting of single or double-labeled cells was normalized to the total number of DAPI cells in each bin . For triple immunofluorescences , the counting was performed on cortical images with a constant width of 600 μm . Each counting performed on electroporated and choleratoxin-labeled neurons was normalized to the total number of GFP- or choleratoxin-labeled cells in the layer of interest . All the data were statistically analyzed and graphically represented using Microsoft Office Excel software . The error bars represent the standard error of the mean ( SEM ) . Two-tailed Student’s t-test was used for the analysis of statistical significance ( *p≤0 . 05 , **p≤0 . 01 , ***p≤0 . 001 ) between to different groups . Neuron morphology was reconstructed in BitPlane Imaris from confocal 3D image stacks . Soma shape features were obtained using built-in Volume Rendering functionality on images with 40x magnification ( criteria: surface background subtraction of 15 μm and detail smoothening of 2 μm ) . Extracted features of each soma shape include X , Y , and Z position of the center of mass , soma area , soma volume , oblate ellipticity , and sphericity . Calculation of these feature values was performed automatically as described in BitPlane Imaris technical documentation ( http://www . bitplane . com/download/manuals/ReferenceManual6_1_0 . pdf ) . Each soma was separated in a region of interest ( ROI ) of approximately 30 × 30 × 10 μm . In order to effectively render soma volume , the ROI was used to automatically compensate differences in local contrast . Limits of the soma volume were calculated automatically within ROI in most of the cases . For more detailed information , see the SI section . Apical and basal dendrites features were obtained using Imaris Filament Tracer plugin . Features include apical dendrite diameter , total apical dendrite length ( also referred as neuron radial position with respect to pial surface ) , number of basal dendrites , angle of each basal dendrite compared to apical one , apical dendrite ramification and bifurcation point . Apical dendrite diameter was measured at 5 μm from soma limit . Radial position and Bifurcation features were obtained from 10x image stacks . Radial position was measured as the total distance from the soma center to the pial membrane . Bifurcation was measured as the distance from the soma to the first bifurcation of the apical dendrite . Apical dendrite ramification was calculated on confocal Images of 40x as the number of secondary dendrites emerging from the apical dendrite , within a length of approximately 70 μm starting from the soma limit . Ctip2+/Satb2+ and Ctip2+/Satb2- populations were compared using non-parametric Wilcoxon Mann–Whitney U-test . Comparison of populations , based on the data collected from 40x images , was performed on n = 145 neurons . Comparison based on Radial Position and Bifurcation was performed on n = 52 neurons ( 52 matched neurons on 10x and 40x images ) . Custom clustering analysis was performed using MATLAB and to diminish scale impact on clustering , z-scale standardization was applied across all feature variables . The unsupervised k-means++ clustering algorithm was applied using squared Euclidean as distance measure . The optimal number of clusters was defined by comparing the quality of cluster separation and neuron features similarity within each cluster . The silhouette values were calculated for outcomes for each testing cycle starting from k = 2 to k = 5 . After the optimal cluster number has been defined , clustering was performed with large number of iterations ( iterations = 1000 ) . MATLAB Scripts and their dependencies are available for download at https://github . com/nikiluk/signalife-moo-clust/ and could be used according to their licensing terms . Whole-cell patch clamp recordings were performed on the soma of layer V YFP+ cortical neurons from 350 µm live slices of somatosensory cortices of the Thy1-eYFP-H transgenic brains . For whole-cell experiments , the slices ( 350 µm ) were perfused with artificial cerebrospinal fluid ( ACSF ) which comprised ( mM ) : 124 NaCl , 3 KCl , 26 NaHCO3 , 1 . 25 NaH2PO4 , 2 CaCl2 , 1 MgSO4 , 15 D-glucose , 0 . 05 picrotoxin , 0 . 05 2-amino-5-phosphonovaleric acid ( APV ) and 0 . 02 6 , 7-dinitroquinoxaline-2 , 3-dione ( DNQX ) , bubbled with O2:CO2: 95:5% . Visually guided , whole-cell recordings were obtained at 29°C from the soma of GFP-positive cortical neurons in layer V of somatosensory area , using patch electrodes ( 3–5 MΩ ) filled with ( in mM ) : 20 KCl , 100 Kgluconate , 10 HEPES , 4 Mg-ATP , 0 . 3 Na-GTP , 10 Na-phosphocreatine and 0 . 1% biocytin ( Sigma Aldrich , Merck , Germany ) . Recordings were performed using an AxoPatch 200B amplifier ( Axon Instruments , Foster City , CA ) , filtered at 10 kHz and were not corrected for liquid junction potentials . Data were collected using ClampEx software and analyses of recorded responses were performed using Clampfit . The cells usually had a holding potential between -65 and -70 mV or were hold at these potentials . To calculate cell properties ( resistance , time to peak , sag ) , currents from -200 pA to 80 pA were injected for 500 ms . The voltages were measured at the peak amplitude and at steady-state and the I-V curves were plotted . The slope of these curves corresponded to the resistance at peak ( Rpeak ) and at steady-state ( Rss ) . The sag was measured as the difference of voltage at peak to the voltage at steady state when -200 pA was injected . To measure firing and action potential properties , depolarasing currents were injected for 2 s by steps of 20 pA . The relationship between the amount of injected current and the firing frequency was then plotted . The ratio of the first interspike interval ( ISI ) to the last ISI was analyzed using responses recorded at two times the threshold current . To calculate action potential ( AP ) characteristics , we analyzed responses of cells at threshold current . Firing threshold is calculated as the interpolated membrane potential at which the derivative ( dV/dt ) equals 20 V/s . The firing threshold was set as the baseline to calculate characteristics of each AP ( amplitude , half-width , duration , rise time ) . The fast afterhyperpolarization ( fAHP ) is the difference between the firing threshold and the minimum value found within 3 ms of the spike . The time of fAHP is the time between the peak of the spike and the fAHP . The depolarizing afterpolarization ( DAP ) following the fAHP ( fDAP ) is the difference between the maximum value obtained within 5 ms after the fAHP and the minimum value measured to calculate the fAHP . When bursts of 2 or 3 AP were fired at once , the threshold of the second AP was chosen as the maximum value to calculate the fDAP . The medium AHP ( mAHP ) is the difference between the threshold of AP and the minimum value found within 50 ms of the spike . The medium DAP ( mDAP ) is the difference between the maximum value found withnin 70 ms after the minimum value used to calculate the mAHP and this minimum value . When 2 or more APs were fired at once , the mAHP and mDAP were measured after the last AP of the burst . | The cerebral cortex is part of the outer layer of the mammalian brain , and it is important for a range of processes , including sensing , movement and conscious thought . The cerebral cortex is subdivided into several areas that are deputed to different functions . Each area is composed of an astounding variety of cells called projection neurons , which send information from the cerebral cortex to distant parts of the brain . There are three main types of projection neurons , which each connect to a different brain region . However , when projection neurons first form in the embryo , they are all broadly similar . They then activate a combination of genes that determine their identity and behaviour through the activity of a vast range of transcription factors ( proteins that control gene expression ) . At first , most of these transcription factors are active in more than one type of cortical neuron , but after the animal is born , these proteins become increasingly restricted to just one type of neuron . In this way , the major classes of projection neurons are specified . However , the mechanisms defining the remarkable variety of projection neuron subtypes in the cerebral cortex are still largely unknown . Two of the transcription factors that act in the development of the major classes of projection neurons are called Satb2 and Ctip2 . Satb2 prevents the activity of Ctip2 , since the two proteins have opposite effects . However , some neurons in newborn animals produce both of these transcription factors . Using mouse models , Harb et al . found that just after birth the number of projection neurons that express both Ctip2 and Satb2 increases in the cortical area that processes touch sense information – called the somatosensory cortex . These neurons are divided into two subclasses , each of which communicates with a different part of the brain . This suggests that Ctip2/Satb2 co-expression defines two subgroups of the major projection neuron classes rather than specifying new cell types . Further investigations revealed that after birth Satb2 and Ctip2 are co-expressed in neurons due to another protein ( called Lmo4 ) that modifies the structure of the DNA region that contains the Ctip2 gene . This prevents Satb2 from repressing Ctip2 . This work demonstrates that the great variety of projection neurons in the mammalian cerebral cortex is not due to the existence of several genetic programs directing the development of each single neuronal subtype . Instead , this variety is due to mechanisms that modify and refine , after birth , the processes that specify the major projection neuron classes . The main challenge in the future will be deciphering all the mechanisms that tilt the balance toward a given neuron subtype , and investigating whether and how this balance can be altered . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Conclusion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2016 | Area-specific development of distinct projection neuron subclasses is regulated by postnatal epigenetic modifications |
The SNAREs SNAP25 and SNAP23 are proteins that are initially cytosolic after translation , but then become stably attached to the cell membrane through palmitoylation of cysteine residues . For palmitoylation to occur , membrane association is a prerequisite , but it is unclear which motif may increase the affinities of the proteins for the target membrane . In experiments with rat neuroendocrine cells , we find that a few basic amino acids in the cysteine-rich region of SNAP25 and SNAP23 are essential for plasma membrane targeting . Reconstitution of membrane-protein binding in a liposome assay shows that the mechanism involves protein electrostatics between basic amino acid residues and acidic lipids such as phosphoinositides that play a primary role in these interactions . Hence , we identify an electrostatic anchoring mechanism underlying initial plasma membrane contact by SNARE proteins , which subsequently become palmitoylated at the plasma membrane .
Palmitoylation is a post-translational modification of a protein which causes its stable attachment to a cellular membrane . Examples of proteins that follow this paradigm are the homologous SNARE ( soluble N-ethylmaleimide-sensitive factor attachment receptor ) proteins SNAP25 and SNAP23 , which after translation are initially cytosolic proteins . In order to function in vesicle fusion , they relocate to the plasma membrane . SNAP23 is ubiquitously expressed , whereas the neuronal SNAP25 is highly abundant in the synapse and in the plasma membrane of neuroendocrine cells ( Jahn and Fasshauer , 2012; Wilhelm et al . , 2014; Knowles et al . , 2010 ) . Stable attachment to membranes is achieved after palmitoylation of a cysteine cluster , which is most probably catalyzed by the plasma membrane resident palmitoyl acyltransferase DHHC2 . DHHC2 is characterized by the presence of a conserved DH ( H/Y ) C motif and can palmitoylate SNAP25 and SNAP23 ( Greaves et al . , 2010 ) . The majority of SNAP25 molecules reside in the plasma membrane , while 20% are located in a perinuclear recycling endosome-trans-Golgi network ( Aikawa et al . , 2006 ) . A two-compartment model for SNAP25 trafficking has been proposed which speculates that the endocytic recycling of SNAP25 might be coupled to its depalmitoylation , followed by its repalmitoylation and recycling back to the plasma membrane ( Aikawa et al . , 2006 ) . In any case , in steady-state , the large majority of SNAP25 molecules are stably attached to the cell membrane . The minimal domain necessary for SNAP25 plasma membrane targeting has been mapped to amino acids 85–120 ( Gonzalo et al . , 1999 ) , comprising the cysteine cluster at the N-terminus ( for relevant SNAP25 region see Figure 1 ) , while the C-terminus interacts with DHHC proteins ( Gonzalo et al . , 1999; Greaves et al . , 2010 ) . For DHHC interactions to occur , proximity to the membrane is an important factor and could even be rate limiting in the attachment process . 10 . 7554/eLife . 19394 . 003Figure 1 . SNAP25 plasma membrane targeting in live cells . ( a ) Diagram of the amino acid sequences from position 51 to 125 for wild type SNAP25 and for several constructs in which the net charge of the cysteine rich region ( CRR; box ) is increased or decreased . Numbers associated with the constructs’ names refer to the net charge of the CRR . Cysteines are highlighted in yellow; negatively and positively charged amino acids are highlighted in red and blue , respectively . A net charge of +3 is found in all mammalian SNAP25 proteins whose encoding genes are available in the UniProtKB/Swiss-Prot-data bank ( see Figure 1—figure supplement 1 ) . ( b ) Confocal micrographs from live PC12 cells expressing wt-SNAP25 ( +3 ) , SNAP25–5 and SNAP25+10 as N-terminally GFP-tagged constructs . Also shown are two more constructs , based on wt-SNAP25 ( +3 ) and SNAP25+10 , in which the four cysteines for palmitoylation are exchanged for glycines ( SNAP25 ( C-to-G ) and SNAP25+10 ( C-to-G ) ) . Red elongated boxes mark the regions of interest ( ROIs ) in which the fluorescence distribution at the cell periphery was analysed by linescans . White graphs illustrate the corresponding fluorescence traces . ( c ) For one experiment several traces were averaged . ( d–f ) Ratio between cell periphery and cytosol signal for ( d ) the variants lacking cysteines for palmitoylation , ( e ) constructs with an altered charge around the cysteine cluster , and ( f ) a construct with an eliminated polybasic cluster located at the C-terminus of SNAP25 . Values are given as means ± S . E . M . ( n = 3–19; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ns = not significant ) . For the constructs exhibiting weakest and strongest targeting in ( c ) , Figure 1—figure supplement 2 shows that the ratio between cell periphery and cytosol signal is independent of the expression level . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 00310 . 7554/eLife . 19394 . 004Figure 1—figure supplement 1 . Cysteine-rich regions from different species . In the UniProtKB/Swiss-Prot-data bank , 15 hits are found for the SNAP25 gene . From those hits , we eliminated a fragment of rabbit SNAP25 that lacks the cysteine-rich region ( CRR ) , and added the SNAP25a isoform of rat SNAP25 ( P60881-2 ) . Note that in P36976 and P36975 , some of the included positively charged amino acids are a few residues further downstream than is the case in mammalian proteins . Species include Rattus norvegicus ( rat ) , Mus musculus ( mouse ) , Homo sapiens ( human ) , Bos taurus ( bovine ) , Pan troglodytes ( chimpanzee ) ; Macaca mulatta ( rhesus macaque ) , Pongo abelii ( Sumatran orangutan ) , Gallus gallus ( chicken ) , Torpedo marmorata ( marbled electric ray ) , Danio rerio ( zebrafish ) , Carassius auratus ( goldfish ) , and Drosophila melanogaster ( drosophila ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 00410 . 7554/eLife . 19394 . 005Figure 1—figure supplement 2 . No correlation between the periphery/cytosol signal ratio and the expression level . The periphery/cytosol-ratio in individula cells is plotted versus fluorescence peak intensity . This figure is based on experimental data included in Figure 1e . In Figure 1e ratios for each experimental day are averaged , whereas individual values are shown here . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 005 SNAP25 and SNAP23 are not modified in the cytosol by isoprenyl- or myristoyl groups , which would increase their membrane affinity and facilitate initial membrane contact; neither receptors nor membrane-targeting motifs have been identified previously and thus there has been some debate over how this initial contact may be mediated . Polybasic amino acid patches are known to mediate non-specific interactions with anionic lipids . For instance , plasma membrane targeting of myristoylated K-Ras requires an N-terminal polybasic domain ( Cadwallader et al . , 1994; Wright and Philips , 2006 ) , which localizes G protein α subunits to the plasma membrane , although this region is not required for subunit palmitoylation ( Pedone and Hepler , 2007; Crouthamel et al . , 2008 ) . Other examples suggest that phosphorylation of basic residues located upstream of palmitoylated cysteines regulates the palmitoylation of a potassium channel through an electrostatic switch ( Jeffries et al . , 2012 ) . In some instances , polybasic patches are the main driving force for protein attachment to the negatively charged plasma membrane ( Cho and Stahelin , 2005 ) . SNAP25 has a net negative charge: of its 206 amino acids , 21% are negatively and 14% are positively charged . Still , a modest excess of three positive charges around the cysteine cluster might be available for non-specific interactions with anionic lipids . Nevertheless , this charge accumulation appears very small when compared to the number of positive charges that mediate electrostatic contacts in other cellular processes ( Heo et al . , 2006 ) . Here , we set out to investigate whether the subcellular distribution of SNAP25 and SNAP23 can be regulated through such a small accumulation of positive charges , despite the proteins’ overall negative net charge .
We wondered whether the above mentioned small charge accumulation close to the cysteine cluster is exclusively found in rat SNAP25 . Comparison of the cysteine-rich region of the SNAP25 protein ( UniProtKB/Swiss-Prot-data bank ) in a variety of species revealed that this region carries a net positive charge in all species except Drosophila ( Figure 1—figure supplement 1 ) . The distribution of charged amino acids adds up to +3 in all seven mammalian species , and also in chicken , zebrafish and goldfish ( although for zebrafish and goldfish there is also an isoform with a charge of +1 ) . The cysteine-rich region has four cysteines for palmitoylation in the centre , which are flanked by four lysines on each side ( Figure 1a ) . The sequence around the cysteines also contains five acidic amino acids: four upstream and one downstream of the cysteines . Hence , eight basic and five acidic amino acids yield a net charge of +3 located downstream of the cysteine cluster . If positive charges close to the cysteine cluster ( Figure 1a ) constitute the main driving force for initial plasma membrane association , their elimination should diminish plasma membrane targeting . To test this hypothesis , we overexpressed GFP-tagged SNAP25 in neuroendocrine PC12 cells and analysed how plasma membrane targeting depends on these charges . Equatorial optical sections were imaged in live cells ( Figure 1b ) . Linescans perpendicular to the plasma membrane reveal the GFP-SNAP25 distribution at the cell periphery ( plasma membrane + cytosol that optically cannot be resolved ) and the cytosol ( Figure 1b and c ) . Relating these values to each other yields a ratio >1 if there is a plasma membrane-associated fraction . A variety of SNAP25 mutants with increased or decreased charge in the cysteine-rich region were analysed . As control for non-targetable protein , cysteine palmitoylation sites were substituted by glycines ( Figure 1d ) . As expected , SNAP25 proteins that lacked cysteines ( construct SNAP25 ( C-to-G ) ) did not locate to the plasma membrane ( Figure 1d ) . We then reduced the charge from wt-SNAP25 ( +3 ) to −1 by substituting four lysines with alanines ( for construct details , see Figure 1a ) . In SNAP25-1distal , the outer four lysines up- and downstream of the cysteine cluster are exchanged , and in SNAP25-1proximal the four inner ones are exchanged . Both of these constructs showed strongly diminished plasma membrane targeting ( Figure 1e ) . Further reducing positive charges to a net charge of −5 , by substituting all eight lysines with alanines ( SNAP25-5; see Figure 1a ) , abolishes targeting almost completely ( Figure 1e ) . We also tested a construct termed SNAP25-5hydrophob which is identical to SNAP25-5 except that the lysines were not replaced by alanines but instead by more hydrophobic leucines ( Figure 1a ) . The reasoning for increasing the hydrophobicity is a previous report suggesting that the hydrophobicity of this domain plays a role in initial membrane association ( Greaves et al . , 2009 ) . Compared to SNAP25-5 , SNAP25-5hydrophob showed no increase in targeting; instead , targeting was further reduced and hardly detectable . We next tested whether more positive charges would increase targeting efficiency . Here , the finding was ambiguous , as increasing the charge to +7 ( SNAP25+7 ) or +10 ( SNAP25+10; for constructs see Figure 1a ) diminishes or promotes targeting , respectively ( Figure 1e ) . Substituting the cysteines by glycines in the SNAP25+10 mutant ( SNAP25+10 ( C-to-G ) ) causes a cytosolic distribution ( Figure 1d ) . Hence , an increase in positive charges cannot substitute for attachment by palmitoylation . There is another small cluster of positive charges downstream of the cysteine-rich region in the C-terminal part of the SNAP25 linker . Elimination of these charges by introducing the mutations R191A , R198A and K201A has no effect on membrane targeting ( Figure 1f ) . This indicates that the mere presence of positive charges is not sufficient for membrane targeting . Rather , the position of charged residues within the protein structure determines their effect . SNAP23 is 60% identical to SNAP25 . Like SNAP25 , it carries three positive charges in the cysteine-rich region ( Figure 2a ) . Reduction of the positive charges by six units in SNAP23-3 diminishes membrane association ( Figure 2 ) . As observed for SNAP25 , increase of positive charges can promote ( see SNAP23+10 in Figure 2 and SNAP23+16 in Figure 2—figure supplement 1 ) or diminish ( see SNAP23+11a and SNAP23+11b in Figure 2—figure supplement 1 ) membrane targeting . Finally , exchange of cysteines for glycines ( SNAP23 ( C-to-G ) ) yields a cytosolic distribution ( Figure 2 ) , showing that palmitoylation is also required for stable attachment of SNAP23 . 10 . 7554/eLife . 19394 . 006Figure 2 . SNAP23 plasma membrane targeting . ( a ) Amino acid sequences from positions 45 to 119 shown for SNAP23+10 , wt-SNAP23 ( +3 ) and SNAP23-3 . The box indicates the cysteine-rich region ( CRR ) in which mutations were introduced . Red , blue and yellow , respectively , highlight negatively charged amino acids , positively charged amino acids and cysteines . ( b , c ) The periphery/cytosol signal ratio from confocal micrographs was analysed as described in Figure 1 . Values are given as means ± S . E . M . ( n = 3–8; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . For more SNAP23 constructs , see Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 00610 . 7554/eLife . 19394 . 007Figure 2—figure supplement 1 . Correlation between SNAP23 targeting and charge of the cysteine-rich region . Extended Figure 2a and c showing the additional constructs SNAP23+11a , SNAP23+11b and SNAP23+16 . ( a ) The sequence of rat SNAP23 from position 45 to 119 is depicted for wt-SNAP23 ( +3 ) and for several constructs carrying different charges in the cysteine-rich region ( CRR , from 64 to 100 , box ) . The net charge in this region is given by the numbers associated with the constructs’ names . Cysteines for palmitoylation are highlighted in yellow , negatively and positively charged amino acids are highlighted in red and blue , respectively . ( b ) Periphery/cytosol signal ratio . Values are given as means ± S . E . M . ( n = 3–8; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ns = not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 007 In conclusion , the analysis of SNAP25/SNAP23 constructs shows that membrane targeting is dependent on positive charges located close to the cysteine cluster . Charges > +3 can increase or decrease targeting , suggesting that primary structure is not the only determinant for stronger targeting and that secondary structural elements are also important . Structural features may define whether the charges can be exposed to the membrane environment . Alternatively , they may determine the geometry of the electrostatic contact ( and hence the accessibility of the cysteines ) that plays a role in the palmitoylation reaction . In addition to immunofluorescence imaging , we employed an independent method to analyse the subcellular distribution . SNAP25 constructs that showed strong effects in live cells and the constructs without palmitoylation sites were studied by cell fractionation ( Figure 3 ) . In contrast to the imaging analysis , the biochemical cell fractionation experiment makes it possible to relate the absolute amounts of the membrane and cytosol fraction to each other . In line with the imaging experiments , we observed that SNAP25-5 has a weaker and SNAP25+10 a stronger membrane association than wt-SNAP25 ( +3 ) . In the imaging experiment , no membrane-associated fraction of SNAP25 ( C-to-G ) or SNAP25+10 ( C-to-G ) was visible . This is different in the membrane fractionation assay where we detected a small membrane-associated fraction for SNAP25+10 ( C-to-G ) ( Figure 3 ) . This fraction might have been overlooked by confocal microscopy , which is unable to resolve small plasma membrane-associated pools in the presence of a strong cytosolic background . For a better microscopic analysis , it was necessary to eliminate the cytosolic background . We therefore turned to ‘unroofed cells’ ( Heuser , 2000 ) , also called plasma membrane sheets . To achieve this , cells were exposed to a brief ultrasound pulse that exerts a shearing force to the upper cellular parts , leaving behind the basal plasma membranes . Then , the membrane sheets were imaged to quantify the recruited protein in the absence of the cytosolic background . Wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 were readily detected on the membrane sheets ( Figure 4 ) in amounts that would be expected from the confocal microscopic subcellular distribution analysis . SNAP25 ( C-to-G ) was not detectable , either because it is present at undetectable amounts after membrane sheet generation ( see also the hardly detectable membrane fraction of SNAP25 ( C-to-G ) in the cell fractionation assay in Figure 3 ) , or because it binds weakly and washes off during sample mounting and imaging ( which takes up to ≈ 35 min in total ) . By contrast , SNAP25 +10 ( C-to-G ) was clearly present , despite the absence of palmitoylation sites , documenting a higher plasma membrane affinity when compared to SNAP25 ( C-to-G ) . 10 . 7554/eLife . 19394 . 008Figure 3 . Subcellular distribution of SNAP25 constructs analysed by cell fractionation . PC12 cells expressing the indicated constructs were mechanically homogenised followed by centrifugation , yielding supernatant and pellet that contain the cytosolic and the membrane fraction , respectively . Fractions were analysed by Western blotting using an antibody against GFP . Left: immunoblots of one representative experiment . For each construct , the respective cytosol and membrane fractions are shown at arbitrary scaling ( for the entire blot see Figure 3—figure supplement 1; for the average expression levels see Figure 3—figure supplement 2 ) . Right: the ratio between membrane-associated and cytosolic protein was quantified from the band intensities . Values are given as means ± S . E . M . ( n = 4; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 00810 . 7554/eLife . 19394 . 009Figure 3—figure supplement 1 . Entire Western blot . Blot used for Figure 3 . Equal amounts of proteins were loaded . C , cytosol fraction; M , membrane fraction . In this experiment , SNAP25 ( C-to-G ) and SNAP25+10 ( C-to-G ) tended to be expressed at higher levels . On average , the expression levels of the constructs were similar ( see Figure 3—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 00910 . 7554/eLife . 19394 . 010Figure 3—figure supplement 2 . Variation of expression levels . The expression levels in the experiments performed for Figure 3 were calculated by adding up the signal intensities from the cytosol and the membrane fraction of each construct . The variability in expression levels between experiments is considerable . However , we do not note a correlation between membrane-association and expression levels . Values are given as means ± s . d . ( n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 01010 . 7554/eLife . 19394 . 011Figure 4 . Association of wt-SNAP25 ( +3 ) , SNAP25-5 , SNAP25+10 and SNAP25+10 ( C-to-G ) with isolated plasma membranes . ( a ) Plasma membrane sheets were generated from cells expressing the indicated GFP-SNAP25 constructs by mechanical shearing forces , followed by direct imaging . During imaging , the sample was screened for green fluorescence and all membranes exhibiting green fluorescence were imaged in the green channel , followed by imaging of the blue channel . We also analysed sheets from cells transfected with SNAP25 ( C-to-G ) , but in these samples , no green fluorescence was visually detectable in the screening process . Top , blue fluorescent dye ( TMA-DPH ) visualizing the location and shape of the membrane sheets; bottom , GFP fluorescence of the membrane sheet associated SNAP25 variants . The same images are shown at two different lookup tables ( LUT ) . ( b ) Quantification of GFP-fluorescence on membrane sheets , normalized to wt-SNAP25 ( +3 ) . Values are given as means ± S . E . M . ( n = 3–7; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 011 Cell fractionation and membrane sheet experiments both show that an increase in positive charge produces a membrane-associated fraction even in the absence of palmitoylation . This poses the question of whether the small increase in membrane targeting of SNAP25+10 is , in part , achieved independently from stable membrane-attachment through palmitoylation . To clarify this issue , we examined whether the extent of plasma membrane targeting correlates with the degree of palmitoylation for wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 , using SNAP25 ( C-to-G ) as a negative control . Freshly transfected cells were incubated with a palmitate analogue carrying an alkyne-group to allow click-labelling with a fluorophore . Hence , the newly synthesized SNAP25 constructs should be palmitoylated by the clickable palmitate . Cells were lysed and the constructs were immunoprecipitated using their GFP-tags . A Cy5-fluorophore was covalently attached to the alkyne group through click-chemistry and the immunoprecipitate was subjected to western blot analysis . The membrane was immunostained for GFP , and the palmitate-Cy5 fluorescence was related to the GFP signal . As expected , the data show that SNAP25-5 is much less palmitoylated than wt-SNAP25 ( +3 ) , and SNAP25 ( C-to-G ) shows no detectable traces of clickable-palmitates ( Figure 5 ) . SNAP25+10 is not significantly more palmitoylated than wt-SNAP25 ( +3 ) ( Figure 5 ) . 10 . 7554/eLife . 19394 . 012Figure 5 . Assessment of the palmitoylation of GFP-SNAP25 constructs . PC12 cells were transfected with wt-SNAP25 ( +3 ) , SNAP25-5 , SNAP25+10 , or SNAP25 ( C-to-G ) and fed with alkyne-palmitate overnight . Cells were then lysed , and the GFP-tagged constructs were immunoprecipitated . Subsequently , a click reaction with Cy5-azide was performed to label incorporated palmitate , and the samples were subjected to SDS-PAGE and Western blotting . The amount of protein was quantified by immunolabelling ( anti-GFP antibody/IRDye-labelled secondary antibody ) and used for normalization of the palmitate-Cy5 signal , yielding the palmitate/GFP ratio . The panels show the fluorescence of the palmitate-Cy5 ( top ) and the GFP signal ( bottom ) of one representative immunoblot . The bar chart shows palmitate/GFP ratios of n = 5 independent experiments ( mean + S . E . M . ; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ns = not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 012 The outcome of this experiment is well in line with ( i ) the linescan analysis ( Figure 1 ) , ( ii ) the biochemical cell fractionation assay ( Figure 3 ) and ( iii ) the membrane sheet association assay ( Figure 4 ) . The four different test systems indicate that SNAP25-5 is less targeted to the plasma membrane and less palmitoylated than wt-SNAP25 ( +3 ) . Moreover , SNAP25 ( C-to-G ) cannot associate with the membrane . Finally , SNAP25+10 shows a trend towards increased targeting . However , the data suggest that SNAP25+10 associates with the plasma membrane not only through palmitoylation but also by pure protein electrostatics . Initial plasma membrane association via protein electrostatics may not be the only mechanism affected by the introduced mutations . A disturbed interaction with palmitoyl transferases is unlikely as all of the substitutions are located distal to the QPARV motif which is important for the interaction of SNAP25 and DHHC palmitoyl transferases ( Greaves et al . , 2010 ) . However , some mutations map to the N-terminal SNARE motif of SNAP25 that interacts with plasmalemmal syntaxin 1A . Early studies on SNAP25 plasma membrane targeting proposed syntaxin as the receptor for initial contact establishment ( Vogel et al . , 2000; Washbourne et al . , 2001 ) , although a subsequent study provided compelling evidence against this hypothesis ( Loranger and Linder , 2002 ) . To test whether our mutations affect the formation of a complex with syntaxin , we employed a fluorescence recovery after photobleaching ( FRAP ) assay . This assay is capable of probing interactions between SNAP25 and syntaxin by measuring the mobility of SNAP25 ( Halemani et al . , 2010 ) . Mobility diminishes the stronger SNAP25 binds to syntaxin and the more syntaxin is present . The slow-down is dependent on a syntaxin–SNAP25 interaction , as it requires the N-terminal SNARE motif of SNAP25 . Moreover , a mutant carrying introduced prolines in the N-terminal SNARE-motif of SNAP25 ( prolines interfere with the alpha-helix formation , which in turn is required for SNARE-complex formation ) moves almost independently of the syntaxin concentration ( Halemani et al . , 2010 ) . We tested our constructs on membrane sheets with or without co-expressed syntaxin 1A-RFP . Syntaxin 1A-RFP slows down the mobility of wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 in a similar fashion , arguing against an altered capability of complex formation caused by the introduced mutations ( Figure 6 ) . Therefore , we conclude that the differences in initial membrane association are not due to an altered affinity of binding to syntaxin . 10 . 7554/eLife . 19394 . 013Figure 6 . Probing the interaction of SNAP25 constructs with syntaxin 1A . ( a ) Illustration of a fluorescence recovery after photobleaching ( FRAP ) experiment that measures SNAP25 interactions with syntaxin on membrane sheets generated from PC12 cells expressing the respective constructs . GFP-SNAP25 mobility was analysed in the absence ( top row ) or presence ( bottom row ) of co-expressed syntaxin 1A-RFP ( for RFP fluorescence see images on the left shown at the same scaling ) . Right ( from left to right ) : membrane sheets before bleaching of a square region of interest ( ROI ) , the first image immediately after bleaching , and 5 s and 40 s after bleaching . The ROI refills with GFP-signal faster in the absence of syntaxin 1A-RFP . ( b ) Averaged fluorescence recovery traces from one experiment , in the absence ( grey ) or presence ( red ) of syntaxin 1A-RFP . Values are given as means ± S . D . ( n = 7–12 membrane sheets ) . Hyperbola functions are fitted to the averaged traces yielding the half time of recovery . ( c ) Average half times of recovery for wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 , in the absence and presence of overexpressed syntaxin 1A . Values are given as means ± S . E . M . ( n = 3–4; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ns = not significant ) . Please note that , in this experiment , large pixels were used to keep bleaching low . Therefore the spatial resolution is lower than that in the other experiments and does not allow for resolving the SNAP25 micropatterning . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 013 Next , we tested whether membrane association of SNAP25 is directly mediated by negatively charged lipids . We used reconstituted liposomes containing distinct lipid compositions but lacking any proteins ( thus eliminating the role of potential SNAP25 binding partners such as syntaxin 1 ) ( Figure 7 ) . This assay also clarifies whether the few amino acids located in the relatively small SNAP25 segment do indeed affect lipid binding . GST-tagged wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 were expressed in bacteria and thus lack the palmitoylation modification . The constructs were purified , immobilized on glutathione beads and incubated with Atto647N-PE-labelled liposomes containing either PC/PS or PC/PS/PE/PI/cholesterol in the absence or the presence of distinct phosphoinositides . SNAP25-bound liposomes were quantified by their Atto647N-PE fluorescence and the results were corrected for unspecific binding to GST-beads . All SNAP25 constructs show only weak liposome binding in the absence of phosphoinositides , which excludes phosphatidylserine ( PS ) as a major/sole binding partner ( Figure 7 ) . 10 . 7554/eLife . 19394 . 014Figure 7 . Binding of purified SNAP25 constructs to reconstituted liposomes . Atto647N-labeled liposomes containing either POPC or a complex lipid mixture ( POPC/DOPS/POPE/cholesterol/PI ) and the indicated amounts of PI ( 4 , 5 ) P2 or PI ( 3 , 4 , 5 ) P3 were added to immobilized GST-wtSNAP25 ( +3 ) , GST-SNAP25-5 and GST-SNAP25+10 and incubated 1 hr at 4°C . After washing the beads , the amount of bound liposomes was measured by their Atto647N fluorescence ( excitation: 639 nm , emission: 669 nm ) . The amounts of liposomes specifically bound to the different GST-SNAP25 constructs were determined by subtracting the values derived from the GST controls . Values are given as means ± S . E . M . ( n = 3; t-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ns = not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19394 . 014 The presence of 4% PI ( 4 , 5 ) P2 significantly enhanced SNAP25 binding in the PC samples and in the complex lipid mixture . As in the cellular assays , the reduction or addition of positive charges reduced or increased the SNAP25 interaction , respectively . Replacing PI ( 4 , 5 ) P2 by PI ( 3 , 4 , 5 ) P3 profoundly increased binding , but largely diminished the differences between SNAP25 constructs . This suggests that the presence of additional negative charges creates additional binding contacts , which are probably outside of the cysteine-rich region . In order to keep the number of negative charges comparable to those provided by 4% PI ( 4 , 5 ) P2 , PI ( 3 , 4 , 5 ) P3 was reduced to 2 . 8% . This condition yielded a stronger difference in binding in the range between the cell fractionation ( Figure 3 ) and the membrane sheet assay ( Figure 4 ) . Although the magnitudes of the effects are difficult to compare because of the different assay systems ( the complex composition of the plasma membrane versus the simple lipid mix in liposomes , concentrations of binding partners , and the presence of SNAP25-palmitoylation in the cellular experiments may modulate the outcome ) , the observations point to primary interactions that occur in the cysteine-rich region . Since PS is not capable of recruiting SNAP25 , and as PI ( 3 , 4 , 5 ) P3 is much less abundant in the cellular membrane than PI ( 4 , 5 ) P2 ( Balla , 2013 ) , these findings point to PI ( 4 , 5 ) P2 as the most likely intracellular binding partner . However , the effect of PI ( 3 , 4 , 5 ) P3 on wt-SNAP25 ( +3 ) binding was very prominent . Thus , membrane microdomains that are locally enriched in PI ( 3 , 4 , 5 ) P3 at the plasma membrane or in an endosomal compartment ( Wang and Richards , 2012 ) could also be preferential sites for SNAP25 targeting . The liposome binding assay thus identifies phosphoinositides as primary membrane-targeting factors interacting with positive charges in the vicinity of the cysteine cluster . However , we cannot exclude the possibility that other sites in SNAP25 contribute to phosphoinositide-dependent binding .
In conclusion , we suggest that after physically contacting the membrane , a SNAP25/SNAP23 protein increases its dwell time at the plasma membrane through an electrostatic mechanism . The basic residues either directly bind to acidic lipids or produce a local positive electrostatic potential which attracts acidic PIPs . The latter mechanism has been described for MARCKS , where 13 basic residues laterally sequester three PIP2 molecules ( Gambhir et al . , 2004 ) . The increased membrane binding observed after introducing additional positive charges ( SNAP25+10 ) may reflect a MARCKS-protein-like PIP2-binding behaviour not physiologically relevant for the targeting of SNAP25/SNAP23 . Once formed , such a protein-lipid aggregate seems difficult to dissolve , as SNAP25+10 ( C-to-G ) remains attached to the membrane for more than 30 min ( Figure 4 ) . Palmitoylated cysteine residues are often preceded and/or followed by basic amino acids ( Bizzozero et al . , 2001 ) . It has been speculated that these positively charged residues bind to the negatively charged acyl-coenzyme A , and thus augment the acylation rate . Our data do not argue against this hypothesis but show that these residues have a distinct function in anchoring the protein to the cell membrane . This is supported by the in vitro liposome binding assay , which shows the charge-dependent binding in the absence of palmitoylation ( Figure 7 ) . Elimination of positive charges distal from the cysteine cluster has no effect on targeting ( Figure 1f; construct SNAP25 ( R191A , R198A , K201A ) ) . This suggests that a random electrostatic contact is not sufficient for targeting . Rather , electrostatic anchoring needs to position the cysteine ( s ) in a way that facilitates palmitate attachment . The electrostatic contact thus needs to be established by amino acids close to the cysteine residues . Such a mechanism would also explain why the diminishment of positive charges is always accompanied by a loss of targeting , because fewer positive charges will decrease the binding affinity and/or change the geometry of the established contact . On the other side , not all mutants with an increased positive charge show increased membrane targeting . Some actually exhibit a trend towards lesser binding . Perhaps the contact can also become too tight for the attachment of palmitates . Hence , the short targeting motif seems to be optimized to mediate electrostatic anchoring , while still allowing for access to the cysteines for the attachment of palmitates . Previous reports on plasma membrane-localized , small GTPases showed that most of them contain two or three polybasic subclusters , each spanning about four to five amino acids ( Heo et al . , 2006 ) . Removal of one subcluster abolished plasma membrane targeting , leading to the proposal that association results from the additive binding energies of individual subclusters ( Heo et al . , 2006 ) . Similarly , the more abundant positive charges in SNAP25+10 ( C-to-G ) allow membrane association without palmitoylation . Why do SNAP25 and SNAP23 carry only a modest excess of positive charges instead of several charged clusters like those of the small GTPases ? It is tempting to speculate that the electrostatic force , based on three positive charges in wt-SNAP25 ( +3 ) , has been optimized for mediating anchoring that is just sufficient to increase the dwell time at the plasma membrane , but weak enough for dissociation after depalmitoylation . Otherwise , electrostatic anchoring would interfere with SNAP25 palmitoylation/depalmitoylation during SNAP25 recycling ( see 'Introduction' ) . Indeed , the SNAP25+10 ( C-to-G ) construct remains attached to the plasma membrane without being palmitoylated . Our data also show that hydrophobic forces ( Greaves et al . , 2009 ) are less crucial than electrostatic contacts in the plasma membrane binding of SNAP25 . Although the hydrophobic amino acids in the cysteine cluster remain unchanged , the removal of charges abolishes targeting , even though lysines are exchanged for more hydrophobic residues ( alanines in the constructs SNAP25-1distal , SNAP25-1proximal and SNAP25-5 , and leucines in SNAP25-5hydrophob ) . Hence , in a cascade of interactions , a modest modulation of the local charge regulates the efficiency of the cellular processes . It seems there are also other membrane contacts that could depend on charge , for instance regulation of the docking time of lipid transfer proteins . Here , the exchange of five lysines for glutamates in the protein loops that interact with the membrane leads to loss of Osh4-mediated sterol transfer ( Schulz et al . , 2009 ) . In conclusion , we have identified a mechanism of initial plasma membrane association of SNAP25 and SNAP23 , which precedes the stable membrane attachment mediated by palmitoylation of cysteines . The mechanism is based on a relatively small cluster of basic amino acid residues that anchor the protein by binding to acidic lipids , in particular to polyphosphorylated phosphoinositides , with PI ( 4 , 5 ) P2 being the most likely candidate . The data suggest that even small changes in protein electrostatics can have strong effects on a cellular mechanism , merely by transiently anchoring a protein to its site of destination .
Plasmids for the expression of GFP-SNAP25 ( Halemani et al . , 2010 ) and GFP-SNAP23 are based on the expression vector pEGFP-C1 ( GenBank accession No . U55763 , Clontech , Mountain View , CA ) , which contains a monomeric variant of mEGFP fused N-terminally to the sequence of full-length rat SNAP25B ( NP_112253 . 1 ) or SNAP23 ( NP_073180 ) . All fusion proteins contain a linker of five amino acids between mEGFP and the N-terminus of SNAP25B or SNAP23 . Mutations in the SNAP25 or SNAP23 coding sequence were introduced via fusion PCR with purchased oligonucleotides ( Eurofins Genomics ) , followed by insertion into the above-mentioned expression vector using the SacI and BamHI sites , or in the case of SNAP25 ( C-to-G ) , the XhoI and KpnI restriction sites . The construct for expression of C-terminally RFP-tagged rat Syntaxin 1A is based on the expression vector pEGFP-N1 ( GenBank accession No . U55762 , Clontech , Mountain View , CA ) , into which Syntaxin-RFP is inserted using the XhoI and the NotI restriction sites . A twelve amino acid linker connects Syntaxin 1A ( NP_446240 ) to a monomeric RFP ( Campbell et al . , 2002 ) lacking the first amino acid . For expression of GST-tagged constructs , the coding sequences for wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 were amplified with primers carrying restriction sites for BamHI ( forward primer ) and EcoRI ( reverse primer ) . The sequences were first subcloned into the pGEM-T easy vector system ( catalogue no . A1360 , Promega ) by TA cloning , and in a second step subcloned into the expression vector pGEX-6P1 ( GE Healthcare Life Sciences ) via the BamHI and EcoRI restriction sites . All constructs were verified by sequencing . PC12 cells ( a gift from Rolf Heumann , Bochum , Germany; similar to clone 251 ( Heumann et al . , 1983 ) ) were cultured in DMEM with high ( 4 . 5 g/l ) glucose ( PAN biotech ) supplemented with 10% horse serum ( Biochrom ) , 5% fetal calf serum ( Biochrom ) and 100 U/ml penicillin/100 ng/ml streptomycin ( PAN biotech ) . During the course of the project , characteristic features of the cell line — such as morphology , expression of neuronal proteins , and their responsiveness to NGF — were regularly confirmed . Cells were maintained at 37°C and 5% CO2 in a sterile incubator and tested negative for mycoplasmic infections ( GATC Biotech , Konstanz , Germany ) . Cells were transfected with the Neon Transfection System ( Thermo Fisher Scientific , Waltham , MA , USA ) . The tip ( 100 µl ) was loaded with 10 µg plasmid DNA of each construct . Cells were transfected by applying a pulse at 1410 V and 30 ms pulse width . Cells were plated onto poly-L-lysine ( PLL ) ( Sigma , Cat . No: P-1524 ) coated coverslips ( 25 mm diameter , Menzel Gläser , Braunschweig , Germany ) ) and maintained for at least 48 hr before imaging . For membrane sheet generation , cells were subjected to a brief ultrasound pulse in ice-cold sonication buffer ( 120 mM KGlu , 20 mM KAc , 20 mM HEPES-KOH , 10 mM EGTA; pH 7 . 2 ) . PC12 cells were imaged at 37°C in Ringer solution ( 130 mM NaCl , 4 mM KCl , 1 mM CaCl2 , 1 mM MgCl2 , 48 mM D ( + ) αGlucose , 10 mM HEPES; pH 7 . 4 ) using a FluoView1000 confocal laser scanning microscope ( Olympus ) and a UPLSAPO 60x oil objective ( NA 1 . 35 ) . Focussing on the glass–cell interface , all green cells were imaged provided they displayed a clear cell–glass contact area . This excludes dead or dying cells that are in the process of rounding up and detaching . In addition , we avoided large clumps of cells because for such cells it is difficult to identify a longer section of cell membrane required for analysis . For measuring membrane-to-cytosol ratios , PC12 cells expressing GFP-tagged constructs were scanned in a 256 pixel x 256 pixel field ( 12 bit image ) at a pixel size of 137 nm using for excitation 488 nm . The equatorial plane of the cell was imaged . To facilitate subsequent linescan analysis , the scanned field was rotated to allow for each cell placing a horizontal linescan ( length 100 pixel; five pixel width , averaged ) perpendicular to the plasma membrane that records the averaged fluorescence intensities along the 100 pixel linescan ( including background fluorescence ) . Fluorescence intensities were background corrected and normalized to the peak intensity at the plasma membrane . Traces from several cells imaged that day ( ranging from 14–40 per condition ) were aligned with reference to the peak intensity at the plasma membrane . The cytosolic fluorescence level was averaged over five pixels starting at a 10 pixel distance from the peak intensity at the plasma membrane , and the cell periphery/cytosol ratio was calculated subsequently . For mobility measurements by FRAP , the laser intensities of the 488 nm laser ( for GFP ) and the 543 nm ( for RFP ) were reduced to a minimum to prevent bleaching effects during the recording . Membrane sheets were analysed by scanning a 100 pixel x 100 pixel field with a pixel size of 0 . 414 µm . Recordings started with a pre-bleaching phase of three images , followed by a 500 ms bleaching step and the recording of the recovery phase . For bleaching , a region of interest ( ROI ) with a size of 7 pixels x 7 pixels ( 2 . 9 μm x 2 . 9 μm ) was bleached using a 488 nm laser in combination with a 405 nm laser ( both set to their maximum intensity ) . After bleaching , image sequences were taken at 1 . 2 Hz for 113 s with the scanning speed set to 40 µs per pixel . Recovery traces were background-subtracted and normalized to the average of the pre-bleach values . For one experiment , several normalized recovery traces were averaged ( 7–15 membrane sheets per condition ) . A hyperbolic curve y ( t ) = offset + maximal recovery x t/ ( t + t1/2 ) was fitted to the averaged recovery trace , yielding the half time ( t1/2 ) of recovery . To measure the association of SNAP25 constructs with the plasma membrane on plasma membrane sheets , a Zeiss Axio Observer D1 epifluorescence microscope equipped with a Plan-Apochromat 100x/NA 1 . 4 oil immersion objective and a 12 bit CCD camera ( 1376 × 1040 pixel ) was used , yielding a pixel size of 64 . 5 nm x 64 . 5 nm . Freshly prepared membrane sheets were imaged in sonication buffer supplemented with TMA-DPH ( 1- ( 4-tri-methyl-ammonium-phenyl ) −6-phenyl-1 , 3 , 5-hexatriene-p-toluenesulfonate; Thermo Fisher Scientific ) , up to ≈ 35 min after membrane-sheet generation . TMA-DPH staining was applied for visualization of the shape and integrity of the membrane sheets . Pictures were taken using filter sets F11-000 ( AHF Analysentechnik , Tübingen , Germany ) for TMA-DPH ( blue channel ) and F36-525 ( AHF Analysentechnik , Tübingen , Germany ) for GFP ( green channel ) . From individual membrane sheets , the fluorescence intensity was measured in 30 pixel x 30 pixel ROIs and background subtracted . For each experiment and condition , the values of 21–96 membrane sheets were averaged . About 48 hr after transfection , 9 × 106 PC12 cells were detached from the substrate by trypsin treatment for 2 min ( 0 . 05% Trypsin and 0 . 02% EDTA in PBS , PAN Biotech , Cat# P10-0231SP ) and trypsin was inactivated by adding medium . Cells were pelleted by centrifugation at 1000 x g at room temperature for 3 min and the pellet was washed with PBS . Cell pellets were resuspended in 750 µl ice-cold homogenization buffer ( 300 mM sucrose , 5 mM Tris-HCl , 0 . 1 mM EDTA , 1 mM PMSF freshly added; pH 7 . 4 ) . Using a Potter-Elvehjem homogenizer , cells were kept on ice and homogenized in a volume of 0 . 75 ml by applying 100 strokes . The homogenate was centrifuged for 8 min at 800 x g at 4°C , yielding pellet P1 ( containing non-homogenized cell debris ) and supernatant S1 . S1 was transferred into a new tube for a second centrifugation step for 120 min at 20 , 000 x g and 4°C , yielding pellet P2 ( containing the enriched membrane fraction , which was resuspended in 750 µl homogenization buffer ) and the supernatant S2 with the cytosolic fraction . Protein concentrations of P2/S2 were adjusted to the lowest concentration in the series , before adding respective amounts of 4x Laemmli buffer followed by incubation at 95°C for 10 min . Samples of 10 µg per lane were subjected to SDS-PAGE analysis using a 12% polyacrylamide gel . Proteins were then transferred onto a Roti-NC nitrocellulose membrane ( Carl Roth , Germany ) applying semi-dry blotting . Nitrocellulose membranes were blocked with PBS-T ( 0 . 05% Tween20 in PBS ) containing 5% milk powder for one hour , and incubated with primary antibody against GFP diluted in blocking solution overnight at 4°C . Membranes were washed for 20 min with PBS-T three times . For detection a second antibody tagged with HRP ( RRID:AB_631747 , Cat# sc-2030 , Santa Cruz Biotechnology , USA ) was applied for 1 hr and , after washing three times with PBS-T , chemiluminescence was detected with Luminol Reagent ( sc-2048 , Santa Cruz Biotechnology , USA ) using autoradiographic films . Films were scanned and band intensities were quantified from the digital images . Ten million PC12 cells were transfected with 15 µg GFP fusion constructs of wt-SNAP25 ( +3 ) , SNAP25-5 , SNAP25+10 or SNAP25 ( C-to-G ) . After one hour , the medium was replaced with DMEM containing 15% delipidized FCS ( PAN biotech ) and 100 µM palmitate-alkyne ( a kind gift from the Thiele lab , LIMES Institute , Bonn ) . After 15 hr of feeding , cells were harvested via trypsinization and scraping , washed once with PBS , and resuspended in lysis buffer ( 1% Triton , 1x cOmplete protease inhibitor cocktail , 150 mM NaCl , 5 mM MgCl2 , 25 mM HEPES; pH 7 . 2 ) . Lysis was promoted by vortexing and sonication . Samples were then centrifuged for 10 min at 14 , 000 x g , and the supernatant was bound to a GFP-trap ( Chromotek ) for 2 hr at 4°C to immunoprecipitate the GFP-SNAP25 fusion constructs . After several washing steps , incorporated palmitate alkyne was clicked to a Cy5-labelled azide ( Sigma , cfinal = 100 µM ) in 100 mM HEPES , pH 7 . 2 , containing 500 µM tetrakis ( acetonitrile ) copper ( I ) tetrafluoroborate ( Sigma ) for 1 hr at 37°C . The samples were then washed to remove non-bound Cy5 , and the GFP-constructs were eluted by boiling in Laemmli buffer , and loaded onto an SDS-PA gel . Proteins were then wet blotted to a nitrocellulose membrane . The membrane was blocked with a 1:1 mixture of Odyssey blocking buffer ( Li-Cor ) and PBS , and then incubated with a rabbit anti-GFP antibody ( RRID: AB_303395; abcam , catalog no . ab-290 , diluted 1:1000 in blocking solution containing 0 . 1% Tween-20 ) . After washing , the membrane was incubated with an IRDye 800CW-coupled goat-anti rabbit secondary antibody ( Li-Cor , catalog no . 9263221 , diluted 1:10 , 000 in blocking solution containing 0 . 1% Tween-20 ) . The Cy5 fluorescence of the palmitate and the IRDye 800CW fluorescence of the GFP were imaged with an Odyssey infrared imaging system ( Li-Cor ) at 700 nm and 800 nm , respectively . The fluorescent bands were quantified using ImageJ’s Gel Analyser . Atto647N-DPPE ( Att647N-1 , 2-dipalmitoyl-sn-glycero-3-phosphoethanolamine ) was purchased from Atto-Tec . All other lipids were from Avanti Polar Lipids . The complex lipid mixture ( 5 µmol total amount of lipid ) contains 34 . 5 mol % 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) , 15 mol % 1 , 2-dioleoyl-sn-glycero-3-phosphoserine ( DOPS ) , 20 mol % 1-hexadecanoyl-2-octadecenoyl-sn-glycero-3-phosphoethanolamine ( POPE ) , 25 mol % cholesterol ( from ovine wool ) , 5 mol % liver L-α-phosphatidylinositol ( PI , from liver ) and 0 . 5 mol % Atto647N-DPPE . For the lipid mixes containing brain L-α-phosphatidylinositol-4 , 5-bisphosphate ( PI ( 4 , 5 ) P2 ) or 1-stearoyl-2-arachidonoyl-sn-glycero-3-phospho- ( 1'-myo-inositol-3' , 4' , 5'-trisphosphate ) ( PI ( 3 , 4 , 5 ) P3 ) the amount of PI was reduced accordingly . The lipids were dissolved in chloroform or chloroform/methanol ( 3:1 ratio , for PI ( 4 , 5 ) P2 and PI ( 3 , 4 , 5 ) P3 ) , mixed and dried under a flow of nitrogen . The remaining chloroform was removed by vacuum for 4 hr . The lipids were dissolved in 1 ml reconstitution buffer ( 25 mM HEPES/KOH pH 7 . 4 , 200 mM KCl , 1% ( w/v ) OG ( n-Octyl-β-D glucopyranoside ) , 1 mM DTT ( 1 , 4-dithiothreitol ) ) by 30 min shaking . To form liposomes , OG was diluted below the critical micelle concentration by the addition of 2 ml buffer ( 25 mM HEPES/KOH; pH 7 . 4 , 200 mM KCl , 1 mM DTT ) . The residual OG was removed by flow dialyses with 4 L 25 mM HEPES/KOH pH 7 . 4 , 135 mM KCl , 1 mM DTT overnight . Subsequently , a Nycodenz gradient centrifugation was performed to isolate the liposomes . Therefore , the dialyzed samples were mixed with an equal volume of 80% ( w/v ) Nycodenz and transferred into two SW60-tubes ( Beckman Coulter ) . Layers of 750 µl 35% ( w/v ) Nycodenz , 150 µl 11 . 6% ( w/v ) Nycodenz and 100 µl fusion buffer were added on top of the 40% ( w/v ) Nycodenz/liposome solution . The gradient was spun at 55 , 000 rpm for 3 hr 40 min at 4°C . The liposomes were isolated , followed by a buffer exchange ( 25 mM HEPES/KOH pH 7 . 4 , 135 mM KCl , 1 mM DTT , 0 . 1 mM EGTA , 0 . 5 mM MgCl2 , 1 mM DDT ) using a PD MiniTrap G-25 ( GE Healthcare Life Sciences ) . The amounts of lipids were quantified by measurement of Atto647N fluorescence ( excitation 639 nm , emission 669 nm ) in a Fluoroskan Ascent FL Microplate Fluorometer ( Thermo Scientific ) . Wt-SNAP25 ( +3 ) , SNAP25-5 and SNAP25+10 pGEX-6P1 constructs were transformed into Escherichia coli ( Rosetta ( DE3 ) pLysS ) and expression of the GST fusion proteins was induced with 1 mM IPTG . After harvesting , cells were lysed for 60 min at 4°C in 150 mM NaCl , 50 mM Tris-HCl , pH 7 . 4 and 1 mM EDTA containing Roche cOmplete protease inhibitor , 1 mM DTT , 100 μg/ml lysozyme and two units/ml DNAse I . The suspension was sonicated , centrifuged for 30 min at 20 , 000 x g , frozen in liquid nitrogen and thawed for binding of the constructs directly to glutathione beads . To monitor SNAP25 interaction with liposomes , 42 µg GST-SNAP25 constructs or a equimolar amount of GST ( glutathione S-transferase ) were bound to 20 µl glutathione ( GSH ) sepharose four fast flow beads ( GE Healthcare Life Sciences ) prewashed 3 x with ddH2O and 3 x with fusion buffer ( 25 mM HEPES/KOH pH 7 . 4 , 135 mM KCl , 1 mM DTT , 0 . 1 mM EGTA , 0 . 5 mM MgCl2 , 1 mM DDT ) . 160 nmol liposomes in fusion buffer were added to the beads and incubated 1 hr at 4°C on a rotation wheel . The beads were washed once with 1 ml fusion buffer and resuspended in 80 µl fusion buffer . The bound liposomes were detected by measuring the Atto647N fluorescence . The amounts of liposomes specifically bound to the different GST-SNAP25 constructs were calculated by subtracting the values derived from the GST controls . | Cells often communicate with each other by releasing chemicals that normally are stored in small membrane-bound compartments called vesicles . For example , when a neuron is stimulated , vesicles merge with its cell membrane and release their content into a gap between itself and other neurons . This complicated process involves many steps and molecules , including proteins called SNAREs . Some SNARE proteins reside at the inner side of the cell membrane and help vesicles to fuse with this membrane . Two SNARE proteins called SNAP25 and SNAP23 are produced in the liquid inside the cell and initially float freely . Eventually , these proteins become directly anchored to the cell membrane , however , not much is known about what happens to these proteins in between these stages , or how they first attach to the membrane before anchoring to it . Electrostatic forces between oppositely charged molecules are known to be important for them to bind with each other . Here , electrostatic forces are less likely to occur because SNAP25 and SNAP23 are both mostly negatively charged , and should therefore be repelled from the cell membrane , which also typically has a negative charge . However , both SNAP25 and SNAP23 have a small cluster of positively charged amino acids ( the building blocks of proteins ) near the attachment site , and Weber et al . have now tested whether this charge is sufficient to overcome the predicted repulsion . The approach involved making mutant proteins with either more or less positively charged attachment regions . Mutant SNAP25 or SNAP23 proteins with more positive charges may stick more tightly but not necessarily more permanently to the membrane . However , when the number of positive charges was lowered , more of the proteins remained floating freely in the liquid inside the cell . These results suggest that even a small number of positively charged amino acids is sufficient to help a protein bind to a cell membrane for further processing . The findings of Weber et al . reveal an early step in the life cycle of SNAP25 and SNAP23 before they anchor to the cell membrane . They suggest that finely tuned protein electrostatics can regulate how long a protein spends at a specific site and thereby indirectly determine its fate . Such fine-tuned protein electrostatics are difficult to recognize and could represent an underestimated regulatory mechanism in all types of cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Electrostatic anchoring precedes stable membrane attachment of SNAP25/SNAP23 to the plasma membrane |
When choosing actions , we can act decisively , vacillate , or suffer momentary indecision . Studying how individual decisions unfold requires moment-by-moment readouts of brain state . Here we provide such a view from dorsal premotor and primary motor cortex . Two monkeys performed a novel decision task while we recorded from many neurons simultaneously . We found that a decoder trained using ‘forced choices’ ( one target viable ) was highly reliable when applied to ‘free choices’ . However , during free choices internal events formed three categories . Typically , neural activity was consistent with rapid , unwavering choices . Sometimes , though , we observed presumed ‘changes of mind’: the neural state initially reflected one choice before changing to reflect the final choice . Finally , we observed momentary ‘indecision’: delay forming any clear motor plan . Further , moments of neural indecision accompanied moments of behavioral indecision . Together , these results reveal the rich and diverse set of internal events long suspected to occur during free choice .
The study of decision making in the brain is often limited by the need to average over repeated trials . Yet if the decision process differs on different trials , we may miss the very events that would be most enlightening . To detect such events requires a moment-by-moment ( Briggman et al . , 2005; Yu et al . , 2009 ) readout of brain states on single trials ( Churchland et al . , 2007; Afshar et al . , 2011 ) . To achieve such a moment-by-moment view in the monkey , we applied state-space analysis methods ( Shenoy et al . , 2013 ) to simultaneous recordings from dorsal premotor cortex ( PMd ) and primary motor cortex ( M1 ) —brain areas centrally involved in preparing and producing the movements that effect decisions . When a monkey is instructed about an upcoming movement but not yet allowed to make it , neurons in these areas show changes in firing rate that reflect the properties of the upcoming reach , including direction ( Evarts , 1966; Tanji and Evarts , 1976; Weinrich and Wise , 1982; Godschalk et al . , 1985 ) , distance ( Riehle and Requin , 1989; Messier and Kalaska , 2000 ) , speed ( Churchland et al . , 2006a ) , and curvature ( Hocherman and Wise , 1991; Pearce and Moran , 2012 ) . PMd and M1 may ( Pastor-Bernier et al . , 2012; Thura and Cisek , 2014 ) or may not be involved in actually making decisions , but these areas are known to reflect the momentary decision state ( Thura et al . , 2012; Thura and Cisek , 2014 ) and therefore can be exploited as a window into the ongoing decision process . While studies of decision making typically focus on deliberative decisions based on noisy sensory evidence ( e . g . , Britten et al . , 1996; Shadlen and Newsome , 1996; Uchida and Mainen , 2003; Afraz et al . , 2006 ) or value optimization ( e . g . , Sugrue et al . , 2004 , 2005; Thura and Cisek , 2014 ) , we used a novel task in which the monkey was encouraged to decide quickly , yet often had time to change his mind if desired . Though recent studies have examined changes of mind driven by fluctuating sensory evidence ( Resulaj et al . , 2009; Bollimunta et al . , 2012; Insabato et al . , 2014; Kiani et al . , 2014 ) , the internal events that accompany truly free choices have largely remained hypothetical . Using single-trial methods , we were able to examine how free choices proceed despite the potential for this process to vary from trial to trial . In particular , we were able to test whether free choices generally proceed similarly to ‘forced choices’ in the motor system , determine how frequently the monkeys revised their initial motor plan , and provide insight into why some trials suffered slow reaction times ( RTs ) .
We employed a novel decision-making variant of the previously described maze task ( Churchland et al . , 2010; Kaufman et al . , 2013 , 2014 ) . Monkeys ( J and N ) touched and fixated a central spot , then were shown two targets and four virtual barriers ( Figure 1A ) . The monkeys were required to remain still and maintain eye fixation until a Go cue was given following a 0–1000 ms exponentially distributed delay ( Figure 1B ) . Two barrier configurations were used: ‘T-maze’ ( Figure 1C ) and ‘S-maze’ ( Figure 1D ) . In each configuration , two ‘key barriers’ appeared in one of three lengths making the nearby target easy ( shown as dark gray in Figure 1C , D ) , hard ( medium gray in Figure 1C , D ) , or inaccessible ( light gray in Figure 1C , D ) . Either key barrier could change difficulty at a random point in the trial ( ∼58% of trials ) . At most one change was made per trial . 10 . 7554/eLife . 04677 . 003Figure 1 . Decision-maze task . ( A ) Illustration of task setup . Two targets were presented along with four virtual barriers and a frame . The monkey performed the task with a cursor projected above his fingertip . Targets were rewarded equally . The cursor left a white trail on the screen . ( B ) Task timeline . ( C and D ) The two families of mazes used: ‘T-maze’ ( C ) and ‘S-maze’ ( D ) . Key barriers could take one of three positions , making each target easy , difficult , or blocked ( shown here as shades of gray ) . Reaches for trials with ≥300 ms delay shown . Faded colors , reach trajectories on forced choice trials; saturated colors , reach trajectories on free choice trials . ( E and F ) Overt changes of mind on free-choice trials with no barrier changes . Dataset J1 ( A–E ) ; dataset N3 ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 00310 . 7554/eLife . 04677 . 004Figure 1—figure supplement 1 . Overt changes of mind for the other five datasets ( labeled on panels ) . Displayed as in Figure 1E , F . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 004 This task presented a continuum of situations , ranging from complete experimenter control ( ‘forced choices’ , with only one target accessible ) , to moderate experimenter influence ( e . g . , a barrier that changed mid-trial might provoke a change of mind ) , to free choices with both targets accessible . This task is somewhat analogous to a driver choosing a lane . Often one attempts to pick a lane that will be most expeditious . The choice may be determined by the environment: for example , one lane is blocked by a large truck and the other is nearly empty . On other occasions the choice is more ambiguous . Both lanes may have some cars or both may be clear . We are all familiar with the experience of rapidly weighing these choices , and with the vacillation and hesitation that can result . Our task is roughly analogous: often the stimulus does not fully specify the right answer , leaving a choice to be made . In agreement with this idea , the monkeys' choices statistically reflected the relative difficulties of the two options ( Table 1 ) , similar to previous observations ( Cos et al . , 2011 ) . As expected , choices were also influenced by barrier changes during the delay period ( Table 1; dataset N2 showed imperfect responses to barrier changes , and was thus excluded from one analysis below ) . For example , if the leftward target became more accessible as the result of a barrier change then the probability of choosing left increased . While biases were common ( much as they would be for drivers choosing lanes ) both monkeys made reaches to both targets when given a choice , and reached more often to a target when it was easy than when it was hard . Finally , the monkeys occasionally ‘changed their mind’ mid-reach: they began reaching toward one target but performed a sharp turn and reached to the other target ( Figure 1E , F , Figure 1—figure supplement 1 ) . In most of these overt change-of-mind trials , both options were available when the monkey deviated from his original reach . Depending on the trial , these changes of mind were driven either by a barrier change that only altered the relative difficulties of the two options , or by some purely internal process . Thus , ‘freedom of choice’ was indeed exercised . 10 . 7554/eLife . 04677 . 005Table 1 . Choice probabilitiesDOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 005p ( left ) easy|easyp ( left ) hard|hardp ( left ) easy|hardp ( left ) hard|easyΔp for left-biasing changeΔp for right-biasing changeJ1-T0 . 660 . 020 . 950 . 0127%−38%J1-S0 . 160 . 570 . 660 . 1732%−19%J2-T0 . 920 . 000 . 920 . 0020%−44%J2-S0 . 230 . 991 . 000 . 0645%−38%N1-S0 . 981 . 001 . 000 . 2037%−37%N2-S0 . 790 . 830 . 880 . 70−4%*−15%N3-S0 . 790 . 540 . 790 . 4518%−16%The first four columns show the probability that the monkey chose the leftward target given the particular barrier configuration . ‘Easy|hard’ means that the barrier configuration was easy for the left and hard for the right . Trials included for these columns had no barrier changes . The Δp values show the change in p ( left ) when a trial presented a free choice throughout , but the difficulty of a key barrier changed during preparation ( from 100 ms after maze onset to 50 ms after Go ) . These are separated by whether the change favored the left target ( the left barrier became easier or the right harder ) or the right target . Because dataset N2-S did not have consistent behavioral effects on trials with a change in barrier difficulty ( starred entry ) , N2-S was not used for one analysis of biasing change trials . Trials were sorted post hoc to identify those relevant for the various analyses below . We first describe results from the ends of the choice spectrum ( fully forced choices and fully free choices ) , then consider cases where barrier changes encouraged changes of mind . Finally , we examine how the neural state related to RT . All analyses were , by necessity , performed on the data collected within a single day: the population of recorded neurons , as well as any monkey's internal events and behavior , likely change from day to day . To provide the necessary statistical power to observe internal events on a single-trial basis , neural data were recorded using two 96-electrode arrays ( in PMd and M1 ) per monkey . These recordings yielded 96–196 single- and multi-units per dataset . Since neurons were recorded simultaneously , we could neither optimize target locations for each neuron separately nor select for strongly responsive neurons . Instead , we selected target arrangements that evoked both possible choices and drove particularly robust population responses , which for some datasets yielded sufficient statistical power for good decoding of choice on single trials . This selection yielded seven datasets for analysis: two target arrangements ( T and S mazes ) for monkey J across 2 days ( J1 and J2 ) , and one target arrangement ( S maze ) for monkey N for 3 days ( N1 , N2 and N3 ) . Neural responses during forced-choice trials showed delay-period activity ( after maze onset but before the Go cue ) that rapidly ( in 100–200 ms ) reflected the available choice ( Figure 2A , B ) . In order for this to occur , the monkey needed to visually parse the arrangement of barriers and targets , determine which target was accessible , and form a motor plan that would reach the target and avoid the barriers . A key question is whether the neural activity during free choices reflects the eventual movement in a similar way and with a similar time course . This is not a given: during free choices , the preparatory activity might maintain its baseline state , achieve an intermediate state to make a change of mind ‘easier’ ( Cisek and Kalaska , 2005; Fleming et al . , 2009; Ifft et al . , 2012 ) , might develop more slowly , or might vacillate between the choices continually . 10 . 7554/eLife . 04677 . 006Figure 2 . Firing rates and decoding . Blue represents eventual leftward reaches , red indicates eventual rightward reaches . ( A and B ) Responses of example units , dataset J2-S . Thick traces , mean; thin traces , s . e . m , Maze , maze onset . Time in ms . Selectivity for left vs right movements was common , and responses were almost always similar for forced and free reaches ( A ) . Less commonly , forced and free evoked somewhat different responses ( B ) . ( C ) Schematic of decoder . Each dot represents neural state in a window of time on a single trial . ( D–F ) Decoded choice plots for forced-choice trials , generated by leave-one-out cross-validation . Percentages show fraction correct classification . Datasets J2-T ( D ) , J2-S ( E ) , N1-S ( F ) . Small dots are at last decoded time point . ( G–I ) Decoded choice plots for free-choice trials ( saturated colors ) with forced-choice trials shown for context ( faded colors ) . Datasets same as D–F . ( J–L ) Cross-validated decoded choice at final point for forced-choice trials . Datasets same as D–F . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 00610 . 7554/eLife . 04677 . 007Figure 2—figure supplement 1 . PSTHs for more example units . Both the preparatory epoch and the peri-movement epoch are shown . For display , we selected units that attained different firing rates for right vs left during preparation and which reflected the diversity of responses observed . ( A–C ) Units from J1-T . ( D–F ) Units from J1-S . ( G–I ) Units from J2-T . ( J–L ) Units from J2-S . ( M–O ) Units from N1-S . Units shown in panels J and K are the same units from Figure 2A , B . MAZE , target/maze onset; MOVE , movement onset . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 00710 . 7554/eLife . 04677 . 008Figure 2—figure supplement 2 . Decoding forced and free choices . These plots are like Figure 2D–I , for the remaining datasets . ( A ) Cross-validated decoded choices for forced-choice trials . Column headers indicate dataset . ( B ) Decoded choices for free choice trials using the decoder trained on forced-choice trials ( saturated colors ) . Forced-choice trials shown in faded colors for context . ( C ) Cross-validated decoded choices at final point for forced-choice trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 008 To answer this question , we segregated free choice trials by whether the monkey eventually reached left vs right ( Figure 2A , Figure 2—figure supplement 1; possible change-of-mind trials were excluded , see ‘Materials and methods’ ) . For most neurons , the average response only depended on which reach was made; responses were similar regardless of whether a choice was forced or free . Interestingly , for most neurons , the time course of the neural response was nearly identical for forced and free choices . This implies that the monkey formed a motor plan quickly when presented with a free choice: that is , the time to make an initial movement selection for free choices did not take appreciably longer than the time to determine which target was accessible for forced choices . For a smaller fraction of neurons , free-choice trials exhibited selectivity that was weaker than for forced-choice trials ( e . g . , Figure 2B ) . However , neurons consistently ‘preferred’ the same target in free-choice trials as in forced-choice trials ( 102 of 104 significantly tuned units maintained their preference ) , and neurons' trial-averaged responses during the delay strongly correlated for forced and free choices ( mean r = 0 . 76 , p < 0 . 001 for 6/7 datasets , p < 0 . 01 for N3-S; sign test; ‘Materials and methods’ ) . In sum , activity on free choice trials was consistent with the interpretation that initial choices were typically made rapidly . Below , we consider whether some of those choices might spontaneously waver or reverse . To analyze the population response on single trials , we trained a linear decoder using the forced-choice trials ( ‘Materials and methods’ ) . This decoder was simply a weighted sum of the neurons' smoothed spike counts ( Figure 2C ) . Empirically , the decoder chose to exploit the activity of most neurons , with the decoder's weights distributed over units approximately as a double-exponential distribution ( not shown ) . The same decoder was used across a set of delay-period time points: from 300 ms before maze onset to 200 ms before movement onset . Though it made decoding more challenging , the time invariance of the decoder was essential to ensure that any changes of mind we observed were solely due to changing neural activity . This decoder performed well on forced-choice trials , as verified by leave-one-out cross-validation . Each trial's decoded choice over time is represented by a single trace , with the horizontal coordinate representing the decoded choice and time unfolding upward ( Figure 2D–F; Figure 2—figure supplement 2A ) . Trials culminating in a leftward reach are shown in blue , while trials culminating in a rightward reach are shown in red . At the final time point ( 200 ms before movement onset ) , the decoder performed correctly on 99% of forced-choice trials for monkey J and 94% of trials for monkey N ( p < 10−6 for each dataset assessed individually , ‘Materials and methods’ ) . Conveniently , errors were all ‘small’ , lying just on the wrong side of the classifier boundary ( three datasets shown in Figure 2J–L; others in Figure 2—figure supplement 2C ) . This same decoder ( trained on forced-choice trials ) typically generalized well to free-choice trials . For the datasets containing at least 1000 trials , at the final time point ( 200 ms before movement onset ) the choice was decoded correctly on 96% of trials for monkey J and 88% of trials for monkey N . Free choice decoder performance was less accurate for the two datasets with fewer trials ( N2-S and N3-S ) , at 80% . Neural data from all the long-delay free choice trials ( including four barrier configurations ) , in which neither target was blocked and in which no barrier changes occurred , are shown in Figure 2G–I ( saturated traces , forced-choice trials shown as faded traces for context; see also Figure 2—figure supplement 2B; p < 10−6 for all datasets ) . As expected , decoding was unsuccessful before maze onset , for either forced or free choices . The slightly lower decoder performance on free choice trials was not due to any systematic shift in neurons' preferences: retraining the decoder using free-choice trials did not improve performance on free-choice trials ( 96% and 77% for monkeys J and N , assessed via leave-one-out cross-validation ) . Thus , in PMd and M1 the neural events during free choices closely resemble those during forced choices that result in the same movement . Rather , the lower performance of the decoder may reflect some additional complexity during free choices , such as last-moment changes of mind . The decoder is not given access to the data from shortly before movement onset , so any such last-moment changes of mind will necessarily be missed . It is immediately clear from these free-choice plots that there are some trials in which the decoder first indicated a plan toward one target , then switched to the other target ( traces crossing the midline/decoded value of 0 in Figure 2G–I ) . If the decoder is reliable at many time points , then these crossings would be evidence for covert changes of mind ( Horwitz and Newsome , 2001; Resulaj et al . , 2009; Bollimunta et al . , 2012; Kiani et al . , 2014; Thura and Cisek , 2014 ) . Below , we present evidence that these early decoder values are indeed reliable , then quantify these changes of mind . In order to search for fully covert processes such as vacillation ( unprompted changes of mind ) , it is important to determine the reliability of the decoder across time points . We did so in three ways: by exploiting our distribution of delays , by examining the decoded choice on trials where changes-of mind must occur a large fraction of the time , and by examining the RTs for trials with a particular set of events . For the first method , we considered trials with relatively short delays . Short-delay trials allow us to verify the accuracy of the decoder at various time points , since the monkey could not know in advance how long the delay would be for any given trial . If these early time points are decoded accurately , then we can rely on the short-latency decoded points for long-delay trials as well . Using these trials , we found that decoder accuracy plateaued ∼200–250 ms after maze onset ( Figure 3A ) . 10 . 7554/eLife . 04677 . 009Figure 3 . Validating the decoder . ( A ) Decoder performance vs time , forced-choice trials . ( B ) Maze icons illustrate free-to-forced trials . Saturated colors show the decoded choice for free-to-forced trials , which began as free but became forced when a barrier changed during the delay epoch . Faded colors show forced-choice trials for context . Large red dots indicate time of changes that made the rightward target the only available option; blue dots , leftward . Percentage indicates the fraction of trials for which the final decoded time point matched the monkey's choice . Dataset J2-S . ( C ) Same for dataset N3-S . ( D ) RT distributions for free-to-forced trials in which the barrier changed around the time of the Go cue . Black , trials where the monkey initially prepared a reach to the now-blocked side; gray , to the unchanging side . Data for monkey J pooled . Arrows , medians . ( E ) Same for monkey N . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 00910 . 7554/eLife . 04677 . 010Figure 3—figure supplement 1 . Decoded choices for free-to-forced trials . These plots are like Figure 3B , C , for the remaining datasets . ( A ) J1-T . ( B ) J1-S . ( C ) J2-T . ( D ) N1-S . ( E ) N2-S . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 010 Second , we wished to confirm that our readout could reflect the monkeys' choices at fast timescales . To do so , we examined ‘free-to-forced’ trials , in which one of the available choices became inaccessible due to a barrier change . Since we randomly chose which side to make inaccessible , on roughly half of such trials the monkey should have already fortuitously chosen the ‘lucky’ ( unchanging ) side . For the other ‘unlucky’ half of trials , we expect that the monkey must re-plan . This can be clearly seen in the decoded choice: when the barrier change occurred mid-delay ( at least 100 ms after maze onset but at least 50 ms before the Go cue ) , as expected there were large swings in the decoded choice on about half the trials following the change event ( Figure 3B , C , Figure 3—figure supplement 1; dots indicate barrier change ) . These large swings in the decoded choice contributed to strong decoder performance on free-to-forced trials: using the final decoded time , 91% of these trials were correctly decoded for monkey J , and 94% for monkey N . Thus , it was possible for the decoder to detect transient choices . Third , we hypothesized that our readout of the internal choice reflected behaviorally important aspects of the neural state . The free-to-forced trials considered above presented barrier changes during the delay , which gave the monkey time to re-plan . In contrast , when barrier changes occurred around the time of the Go cue , we would expect that the monkey must spend time re-planning and thus have a slower RT . This was the case in our data . We considered free-to-forced trials in which the barrier change occurred around the time of the Go cue ( no more than 50 ms before ) , and segregated them based on the decoded choice at the time of the barrier change . If the decoded choice indicated a plan to the unchanging side , the RT was faster than if it indicated a plan to the now-blocked side ( Figure 3D , E , median difference 68 ms for monkey J , p = 0 . 009; median difference 24 ms for monkey N , n . s . , Mann–Whitney U test ) . Do monkeys ever revise their choices ? We hypothesized that monkeys would quickly make an initial choice , then occasionally revise it when given more time to consider . For fully free choices , there is no reason this must happen , yet intuition suggests that a person would occasionally vacillate when presented with two viable options ( e . g . , a driver choosing a lane ) . Indeed , in both monkeys a modest number of trials exhibited such spontaneous vacillations . This was apparent above in Figure 2G–I ( and Figure 2—figure supplement 2B ) . Using a conservative algorithm ( ‘Materials and methods’ ) , we identified trials in which the decoder initially produced a value indicating one choice , then underwent a large swing to indicate the opposite choice . These trials are shown in Figure 4A , B ( saturated traces; see also Figure 4—figure supplement 1 for vacillation and non-vacillation trials plotted separately , and Video 1 for two examples ) . Using these same conservative criteria , 13% ( 3–24% range over datasets ) of free-choice trials exhibited such spontaneous vacillations , compared to only 2% ( 2–4% ) of forced-choice trials ( forced-choice assessed using leave-one-out cross-validated decoding ) . Vacillation trials were more common among free-choice trials than forced-choice trials in every one of the 7 datasets ( p < 10−9 pooled; χ2 2 × 2 contingency test; Figure 4I , left ) . This difference in frequency also reached statistical significance for 5/7 datasets when considered individually ( p < 0 . 05 , χ2 2 × 2 contingency test ) . To our knowledge , observations of such spontaneous vacillations when presented with a free choice—without changing external evidence—have not previously been described . 10 . 7554/eLife . 04677 . 011Figure 4 . Covert changes of mind . Maze icons indicate situation for adjacent decoded choice plots ( saturated colors ) , with faded colors showing forced-choice trials for context . ( A and B ) Apparent ‘vacillation’ on free choice trials . Datasets J1-S ( A ) and J2-S ( B ) . ( C and D ) Encouraged switch trials , which began as forced but became free when a barrier changed mid-delay . Red dots on traces indicate time of changes that made the rightward target more attractive; blue dots , leftward more attractive . J2-T . ( C ) Trials where the monkey chose the newly-available target . ( D ) Trials where the monkey chose the always-available target . ( E and F ) Biasing change trials . These trials were free throughout , but the difficulty of one side changed mid-delay . ( E ) Trials where the monkey reached to the target that was initially more difficult , and thus likely changed his mind . ( F ) Trials where the monkey reached to the target that was initially easier , and thus likely retained his initial decision . Dataset J1-S . ( G and H ) Distance from mean ‘baseline state’ ( −300 to −40 ms from maze onset ) for different trials and time epochs: during baseline ( gray ) , during free-choice vacillations around the time of the change in the decoded choice ( filled pink ) , during biasing-change trials around the time of the change in the decoded choice ( filled green ) , during all times for free ( hollow pink ) or biasing-change ( hollow green ) trials . J2-S ( G ) and N1-S ( H ) . ( I ) Probability that a trial of the given type exhibited a neural change of mind ( a large change in the decoded choice during the delay period ) . See ‘Results’ for details . Forced vs free and untaken vs taken switch , p < 10−9; unlikely vs likely change , p < 10−4 ( χ2 2 × 2 contingency test ) . Error bars show Wilson binomial confidence intervals equivalent to 1 s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 01110 . 7554/eLife . 04677 . 012Figure 4—figure supplement 1 . Free choice vacillation . All datasets contained free-choice trials exhibiting large , late changes in the decoded choice , presumably indicating spontaneous changes of mind . ( A ) Algorithmically-identified vacillation trials ( ‘Materials and methods’ ) shown in saturated colors; forced-choice trials shown in faded colors for context . Data from monkey J . Dataset identity indicated by column header . ( B ) All free-choice trials not included in ( A ) shown in saturated colors; faded colors same as in ( A ) . ( C ) Same as ( A ) for monkey N . ( D ) Same as ( B ) for monkey N . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 01210 . 7554/eLife . 04677 . 013Figure 4—figure supplement 2 . Example vacillation trials with choice decoded using PMd and M1 separately . Each row represents a free-choice trial ( saturated-color trace ) that was identified as containing a vacillation using both brain areas . Faded colors show forced-choice trials for context . The first column shows the decoded choice over time using both brain areas; the second column shows the decoded choice over time using PMd alone; the third column shows the decoded choice over time using M1 alone . ( A–C ) Trials in which vacillations appear similar using all three decoders . ( D ) Trial in which the M1 decoder disagreed with the combined and PMd decoders . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 01310 . 7554/eLife . 04677 . 014Figure 4—figure supplement 3 . Decoded choices for taken switch trials ( left column ) and untaken switch trials ( right column ) . These plots are like Figure 4C , D , for the remaining datasets . ( A and B ) J1-T . ( C and D ) J1-S . ( E and F ) J2-S . ( G and H ) N1-S . ( I and J ) N2-S . ( K and L ) N3-S . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 01410 . 7554/eLife . 04677 . 015Figure 4—figure supplement 4 . Biasing changes in the barriers encouraged changes of mind . These plots are like Figure 4E , F , for the remaining datasets . Left column , trials where the monkey reached toward the target that was initially more difficult , and thus changes of mind were likely; right column , trials where changes of mind were unlikely . ( A and B ) J1-T . ( C and D ) J2-T . ( E and F ) J2-S . ( G and H ) N1-S . ( I and J ) N3-S . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 01510 . 7554/eLife . 04677 . 016Figure 4—figure supplement 5 . Neural states during changes of mind typically did not resemble the baseline state . These plots are like Figure 4G , H , for the remaining datasets . ( A ) J1-T . ( B ) J1-S . ( C ) J2-T . ( D ) N2-S . ( E ) N3-S . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 01610 . 7554/eLife . 04677 . 017Video 1 . Two example free choice vacillation trials . The left panel shows the display that the monkey saw , plus descriptive text . The right panel shows the decoded choice trajectory for the same trial ( black ) , and the decoded choice trajectories for forced choice trials for context ( blue for left , red for right ) . Video is presented at 1/4 real-time speed . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 017 We performed two additional controls to aid in interpreting these apparent vacillations . First , we wished to verify that the activity in PMd and M1 agreed . For the 5 datasets that could be decoded most accurately ( all four J datasets and N1-S ) , we considered the data from PMd and M1 separately . Specifically , we asked whether the decoded choice from these two areas agreed at time points at least 160 ms into the delay period; this time point was chosen to be late enough that the decoded choice reflected a real choice and not just noise ( ‘Materials and methods’ ) . Despite the much higher noise levels involved when splitting our data into two ( not necessarily equal ) parts , the results from decoding PMd agreed with the results from decoding M1 with a similar reliability as expected from how well each decoder agreed with the behavior . This was true when considering all trials ( Table 2 ) or free choice trials alone ( Table 3 ) . Moreover , we could examine vacillation trials specifically . On the same trials , we could often see similar vacillations when decoding PMd and M1 separately ( Figure 4—figure supplement 2A–C ) , though this was not always the case ( Figure 4—figure supplement 2D ) . 10 . 7554/eLife . 04677 . 018Table 2 . Decoding using PMd and M1 separately , all trialsDOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 018Forced-choice decoder performance ( leave-one-out ) Decoder agreementCombinedPMdM1PMd-combinedM1-combinedPMd-M1J1-T0 . 990 . 990 . 900 . 920 . 850 . 80J1-S1 . 001 . 000 . 800 . 930 . 860 . 81J2-T0 . 980 . 970 . 800 . 960 . 870 . 86J2-S1 . 001 . 000 . 850 . 980 . 840 . 83N1-S0 . 900 . 840 . 950 . 930 . 720 . 67Statistics for decoding choice from PMd and M1 data separately . Decoder performance refers to the fraction of trials for which the final decoded point agreed with the target that the animal reached to . This was assessed on forced-choice trials with delays of at least 300 ms , using leave-one-out cross-validation . Decoder agreement refers to the mean fraction of time points per trial for which the decoded choice was the same using the two datasets indicated . For this statistic , all successful trials with delay periods of at least 300 ms were included . ‘Combined’ refers to the decoder trained using both PMd and M1 data . 10 . 7554/eLife . 04677 . 019Table 3 . Decoding using PMd and M1 separately , free choice trialsDOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 019Free-choice decoder performanceDecoder agreementCombinedPMdM1PMd-combinedM1-combinedPMd-M1J1-T0 . 940 . 860 . 770 . 900 . 830 . 77J1-S0 . 960 . 940 . 890 . 860 . 820 . 73J2-T0 . 970 . 960 . 820 . 960 . 820 . 80J2-S0 . 950 . 950 . 860 . 980 . 780 . 78N1-S0 . 880 . 820 . 790 . 910 . 720 . 67Same as Table 2 , but only considering free-choice trials with delay periods of at least 300 ms . The decoders were trained using forced-choice trials . Second , we addressed the concern that some of these vacillations might be related to the tendency for the baseline state ( before maze onset ) to produce a ‘leftward’ decoded choice in most datasets . To check for this possibility , we performed a simple control . For each vacillation trial , we considered the last time the decoded choice switched sign . We counted how many times it was left-to-right , and how many times it was right-to-left . In 3/7 datasets , left-to-right switches were more common; in 3/7 , right-to-left were more common; in 1 dataset they were tied . It is not entirely clear why at baseline the decode choice tended to register as ‘leftward’ , but it does not appear that this baseline bias affected our ability to identify vacillations . On some trials , we changed the difficulty of one option mid-trial to invite ( but not force ) the monkey to change his mind . We hypothesized that the monkey would sometimes change his mind on such trials , to be responsive to a changing environment . The most straightforward of these trials was the ‘encouraged switch’ ( forced-to-free ) case . These trials began as forced , then a barrier changed difficulty during the delay to provide a free choice . On some trials the monkey reached to the always-accessible option ( ‘untaken’ switch trials ) . On other trials ( ‘taken’ switch trials ) the monkey reached to the side that had changed , presumably reflecting a change of mind from the initial option to the newly-accessible target . Indeed , for taken-switch trials the decoded choice over time often reflected the change of mind well before movement onset ( saturated traces in Figure 4C; Figure 4—figure supplement 3; Video 2 ) . For these taken-switch trials , 40% exhibited a large change in the decoded choice ( 15–75%; assessed as for vacillations; ‘Materials and methods’ ) in the period from 160 ms after maze onset until 200 ms before movement onset . The remaining taken-switch trials presumably involved switches in the internal state that occurred after the analyzed period ( i . e . , just before or as the monkey began reaching ) , consistent with recent work showing that planning and re-planning can be rapid ( Ames et al . , 2014; Wong et al . , 2014 ) . On untaken-switch trials , the decoded choice was generally steady , as expected ( Figure 4D ) . Only 5% ( 0–17% ) of untaken-switch trials exhibited such a large change in the decoded choice . In all 7 datasets , large changes in the decoded choice were more common for taken-switch trials than untaken-switch trials ( p < 10−9 pooled; p < 0 . 02 for 6/7 datasets assessed individually , N3-S n . s . ; χ2 2 × 2 contingency test; Figure 4I , center ) . Thus , changes in the decoded choice were common for conditions where they were expected to be present , and uncommon for conditions where they were expected to be rare . This result provides further validation of the decoder . 10 . 7554/eLife . 04677 . 020Video 2 . Two example ‘taken-switch’ trials . These trials began as forced then became free , and the monkey reached to the newly-accessible target . A red or blue dot appears on the decoded choice trace at the time of the barrier change ( dot color indicates which side the change favored ) . Video is presented at 1/4 real-time speed . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 020 We then examined a case where decisions were more nearly free yet still under experimenter influence: ‘biasing change’ trials . These trials began with a free choice , with one side hard and the other easy . During the delay , the hard key barrier changed to become easy as well . Behavioral statistics indicate that these barrier changes successfully induced changes of mind on many trials ( Table 1 ) . In particular , there were many ‘likely change’ trials: for example , a trial where the rightward target had been hard , became easy , and was eventually chosen . Yet behavior alone cannot tell us which of those trials involved internal changes of mind: the monkey may have chosen the hard target from the start , as sometimes occurred during free choices . At the neural level , large changes in the decoded choice were frequently apparent: they comprised 41% ( 29–50% ) of such likely change trials ( saturated traces in Figure 4E; Figure 4—figure supplement 4; Video 3 ) . Conversely , on trials where the monkey eventually reached to a target that was easy throughout , changes of mind were presumably uncommon: most likely the monkey immediately chose the easy target and maintained this plan . Accordingly , for these ‘unlikely change’ trials changes in the decoded choice were observed less frequently: only 12% ( 4–23% ) of trials ( Figure 4F ) . Likely-change trials exhibited large changes in the decoded choice more frequently than unlikely-change trials in all 6 datasets ( N2-S excluded; see above discussion of Table 1 ) , and this difference was highly statistically significant ( p < 10−4 pooled; p < 0 . 05 for 3/6 individual datasets , χ2 2 × 2 contingency test; Figure 4I , right ) . 10 . 7554/eLife . 04677 . 021Video 3 . Two example ‘likely change’ trials . These trials began with one side hard and the other easy , then a change made the initially easy side hard or the initially hard side easy . The monkey reached to the target that was not initially easy . Video is presented at 1/4 real-time speed . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 021 We wished to ask in greater detail what occurred during changes of mind . First , we asked whether these changes of mind required a return to the baseline state . To do so , we examined the distance between the neural state during changes of mind and the neural state before the maze onset ( baseline ) . The changes of mind did not approach the baseline state for either induced changes ( on biasing-change trials ) or spontaneous vacillations ( Figure 4G , H , Figure 4—figure supplement 5 ) . That is , although we used a one-dimensional neural readout as a decoder , the neural state was multidimensional and in those other dimensions the neural state did not pass through baseline during changes of mind . This is consistent with previous findings about the neural response when a single target jumps from one location to another ( Archambault et al . , 2009 , 2011; Ames et al . , 2014 ) . Second , we asked whether there was a population signature during changes of mind . This is challenging because of the heterogeneity of event timing , but we could nonetheless ask whether individual units' firing rates were unusually high or low during these changes of mind . To do so , we compared firing rates during a time window around the change of mind with matched windows on similar non-change-of-mind trials ( ‘Materials and methods’ ) . For trials in which a change of mind was induced by a changing barrier , there was no systematic difference in firing rates relative to non-change-of-mind trials; units had higher-than-average and lower-than average firing rates with similar frequencies ( 30% were higher , 26% lower , 44% in the normal range; p = 0 . 16 pooled over datasets ) . However , during vacillation trials , firing rates tended to be slightly low relative to non-vacillation trials: 37% of units had lower-than-average firing rates during the vacillation , while only 27% were higher than average ( p = 0 . 04 for J; p = 0 . 009 for N; Z-test for proportions; remaining units were in the normal range ) . This suggests that changes of mind induced by outside events may represent a slightly different process from spontaneous vacillation . However , vacillations occurred earlier on average than shifts in choice due to barrier changes , so this firing rate difference may reflect the time course of motor preparation rather than a difference between spontaneous vacillation and induced changes of mind . Finally , we exploited the single-trial view to examine two potentially informative but rare and variable circumstances—cases in which decision and/or execution were slowed . Focusing first on execution , we returned to untaken-switch trials . These correspond to a situation we have all experienced: at the last minute an option becomes available , but we choose not to exercise it . This last-moment offer nonetheless produced behavioral consequences: the monkey occasionally exhibited slowed RTs even though his choice was unaffected ( Figure 5A , Figure 5—figure supplement 1 ) . We hypothesized that this slowing might be due to a transient re-planning event . To our surprise , the decoded choice on these trials never wavered: the same decoded choice was always observed throughout each trial in all datasets ( Figure 5B , Figure 5—figure supplement 1 ) . This suggests that the slow RTs on these trials were not due to re-planning/vacillation , but instead involved a ‘hesitation’ to execute . That is , when presented with a new option , the monkeys hesitated to execute their seemingly still-valid motor plan . 10 . 7554/eLife . 04677 . 022Figure 5 . Hesitation and indecision . ( A ) RT distribution for forced-choice trials ( gray ) and encouraged switch trials in which the monkey reached to the always-available target ( black ) . Arrows indicate medians; difference , 21 ms , p = 0 . 014 , Mann–Whitney U test . Dataset J1-T . ( B ) Decoded choice for the untaken switch trials from ( A ) with RTs >450 ms ( saturated colors ) , with forced-choice trials for context ( faded colors ) . Red dots , time of barrier change . ( C ) RT distributions for non-delayed forced ( gray ) and free ( black ) trials . Arrows indicate medians; difference , 56 ms , p < 0 . 001 , Mann–Whitney U test . Dataset J2-S . ( D ) Strength of decoded choice in direction of eventual reach ( positive toward reach ) at 100 ms after maze onset , vs RT for non-delayed free-choice trials . One point per trial; line indicates regression fit , dashed lines show 95% CI of fit . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 02210 . 7554/eLife . 04677 . 023Figure 5—figure supplement 1 . Decoded choice is undisturbed on slow-RT , untaken-switch trials . These plots are the same as Figure 5A , B , for the remaining datasets . ( A and B ) J1-S . ( C and D ) J2-T . ( E and F ) J2-S . ( G and H ) N1-S . ( I and J ) N2-S . ( K and L ) N3-S . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 02310 . 7554/eLife . 04677 . 024Figure 5—figure supplement 2 . Slow-RT free-choice trials were rare in other datasets ( J2-S shown in Figure 5C , D ) . All trials shown had no instructed delay . Left column , RT histograms; right column , RT vs strength of decoded choice in direction of eventual reach . ( A and B ) J1-T . ( C and D ) J1-S . Free choices for this dataset were significantly slower than forced choices ( p = 0 . 011 ) , but this difference was small . ( E and F ) J2-T . ( G and H ) N1-S . ( I and J ) N2-S . ( K and L ) N3-S . DOI: http://dx . doi . org/10 . 7554/eLife . 04677 . 024 Second , we examined an effect that corresponds to another intuitively common situation: the moment of indecision that can occur when faced with two good options . Such an effect is not inevitable ( there is little advantage to indecision when both choices are good ) and was not always observed . However , for one dataset the monkey did appear to undergo such moments of indecision: when a choice had to be made immediately upon maze presentation , the RT was often longer for two possible choices than for one choice ( Figure 5C , dataset J2-S; median difference 56 ms , p < 0 . 001 , Mann–Whitney U test ) . These indecisive moments were present for trials resulting in either left or right choices ( RT distributions were generally similar for left and right ) . Although this interesting behavior was only present in one dataset , the ability to analyze individual trials allows us to identify neural correlates . We hypothesized that these slow RTs might be due to a slowly-developing initial motor plan or quick vacillation . In accordance with this idea , there was a strong relationship between a trial initially exhibiting an ‘incorrect’ decoded choice ( relative to the eventual reach ) and that trial having a slow RT ( Figure 5D ) . Thus , this moment of indecision was distinct from hesitation , which did not have an obvious correlate in the decoded choice . As a control , we compared the datasets in which free-choice RTs were not slowed substantially ( Figure 5—figure supplement 2 ) ; the neural effect was absent in these datasets , as expected . Thus , neural ‘indecision’ was observed only when there was behavioral indecision .
While changes of mind have been hypothesized to occur during decision making , and have even been observed when sensory evidence changes ( Horwitz and Newsome , 2001; Bollimunta et al . , 2012; Kiani et al . , 2014; Thura and Cisek , 2014 ) , little has been known about how decisions proceed during truly free choices . Here , we trained monkeys to perform a novel task that often presented two viable options with either answer resulting in the same reward . We then examined the dynamics of decision making using single-trial , moment-by-moment decoding of preparatory activity from PMd and M1 . We found that motor plans for forced-choice and free-choice trials were typically similar and formed with similar speed . However , as suspected , these plans were sometimes revised later . These changes of mind were essentially never observed when the animal was presented with forced choices; this internal process was specific to free choices . Several pieces of evidence support the validity of the changes of mind we observed . First , the decoder performed very well on forced choices , with nearly all decoder errors lying just on the wrong side of the threshold and not strongly favoring the opposite choice . These ‘small’ errors could not readily give rise to strong apparent changes of mind . Second , trials with short delay lengths effectively acted as probes of the decoded choice at different time points . These probes permitted us to verify that the decoded choice matched the monkey's intention throughout the delay , since the monkeys did not know the delay duration in advance . Third , we observed many clear changes of mind under conditions where they were expected ( free-to-forced , taken-switch , and likely change trials ) , and much less frequently when they were expected to be uncommon ( forced choice , untaken-switch , and unlikely change trials ) . While neural responses to forced and free choices were generally similar , the decoder was somewhat less accurate for free choice trials than for forced choice trials . It is not entirely clear why this was the case . One candidate cause of this accuracy difference might have been systematic differences in tuning between forced and free . However , such differences do not seem to be at fault , because decoders trained using free choice trials performed similarly or less well . Other studies have presented two targets that were disambiguated only after the Go cue; in those circumstances , two competing motor plans have been observed both in the saccadic system ( Stanford et al . , 2010; Costello et al . , 2013 ) and for reaching movements ( Cisek and Kalaska , 2005; Cisek , 2006 ) . Here , however , competing motor plans seem unlikely to have interfered with the decoder on most trials: competing motor plans would tend to produce a decoded choice of intermediate magnitude , which was not typical in our data . Instead , we speculate that last-moment changes of mind , occurring after the epoch we decoded , were a more likely culprit . Two pieces of evidence support this hypothesis . First , it is known that responses can change quickly to follow a target that has jumped ( Archambault et al . , 2009 , 2011 ) . Second , our decoder's accuracy was lowest by far in cases where late changes of mind were frequent ( taken-switch and likely change trials ) . This suggests that some of these changes of mind either do not require re-planning ( Ames et al . , 2014 ) or that re-planning occurs at the very last moment ( Wong et al . , 2014 ) . This ability to change one's mind presumably reflects rich internal processing that can continually reevaluate both what act to make and whether to make it now . From this work , it is not clear where the decision itself is made . Other studies have argued that PMd may sometimes play a role in decision making ( Pastor-Bernier et al . , 2012; Thura and Cisek , 2014 ) . Our data do not bear on this question , and it is possible that the decision machinery for this task may be largely or entirely upstream . However , PMd and M1 apparently reflect the ongoing decision process at a fast timescale ( present ‘Results’ and Thura et al . , 2012; Thura and Cisek , 2014 ) , allowing us to use decoding as a tool for gaining insight into the decision making process . This ability to decode the motor plan at fast timescales allowed us to examine not only internally-driven changes of mind on single trials , but also features unique to individual datasets or occurring on only a relatively small fraction of trials . For example , indecision ( delay in making an initial motor plan when faced with a choice ) occurred in only one dataset , yet the decoder could provide insight into how a slowly-evolving initial motor plan might explain the slow RT . Such differences between datasets are likely to occur in rich cognitive behaviors , where individuals' strategies may reasonably differ . On hesitation trials ( slowed RT when a new option was presented but not exercised ) , the decoder revealed that the slow RT was not due to re-planning: no vacillations or weakening of the plan were observed on these trials . This permits comparison with saccadic inhibition ( Reingold and Stampe , 2002; Buonocore and McIntosh , 2012 ) , in which saccades are delayed when a distractor appears shortly after the Go cue . There is evidence that saccadic inhibition may be due to mutual inhibition of competing motor plans ( Bompas and Sumner , 2011 ) . Hesitation did not obviously involve the formation of a competing motor plan , but it remains possible that competing motor plans are formed in a trigger-related or upstream area , or during other behaviors ( such as indecision ) . As we learn more about each process , a more mechanistic comparison may become possible . The similarity we observed between activity for forced and free choices suggests that the forced-free distinction may lie largely upstream of the motor cortices . This similarity has the practical consequence that neural prosthetics trained using forced choices will likely generalize well to free choices ( Musallam et al . , 2004; Santhanam et al . , 2006 ) . In addition , this rapid-choice view challenges previous conclusions , based on trial-averaged data , which argue that humans make decisions unconsciously and only later become aware of them ( Libet et al . , 1983; Hallett , 2007 ) . In our data , motor plans were typically formed quickly , corresponding to preliminary choices . However , our data suggest that the moment when neural activity begins to change is not necessarily the moment a decision is finalized ( Schurger et al . , 2012 ) : the movement can be delayed ( hesitation ) or the choice revised ( vacillation ) .
Animal protocols were approved by the Stanford University Institutional Animal Care and Use Committee . Subjects were two adult male macaque monkeys ( Macaca mulatta ) , J and N , trained to perform a delayed reach task on a fronto-parallel screen for juice reward . Use of two monkeys is standard practice in the field . The surgical history of the monkeys has been described previously ( Churchland et al . , 2010 , 2012 ) ; briefly , they were implanted with a head restraint and two 96-electrode silicon arrays ( Blackrock Microsystems , Salt Lake City , UT ) in surface M1 and caudal PMd of both monkeys ( estimated from local anatomical landmarks ) . Both monkeys performed the task with their right arm , and the arrays were implanted in the left hemisphere . As described in the ‘Results’ , both monkeys performed a decision-making variant of the center-out delayed-reach ‘maze’ task . Both had previously performed other variants of the maze task , including thousands of different mazes ( Churchland et al . , 2010; Kaufman et al . , 2013 , 2014 ) . Typically , they completed novel mazes successfully on the first attempt . Errors for both novel and familiar mazes were almost always due to grazing the corner of a barrier along the correct path; other types of errors were much less common , and reaching toward a blocked target or into a dead end was extremely rare . These patterns of success and failure argue that the monkeys ‘understood’ the task and were not simply performing memorized stimulus-response associations . The task was performed with a cursor projected on the screen just above the monkey's fingertip . Initially , the two targets jittered slightly in place , indicating that the monkey was not yet allowed to move . The targets and barriers always appeared simultaneously . ‘Go’ was indicated by cessation of target jitter , the targets filling in , and the fixation point extinguishing . The monkey was then required to make a brisk movement , curving to avoid the barriers , and end at either target . The delay duration and latency to barrier change were drawn from independent , truncated exponential distributions ( τ = 500 ms ) . For the delay , the distribution was truncated at 1000 ms; for the barrier change , 1200 ms . To keep the task unpredictable , barrier locations and the occurrence and timing of barrier changes were randomized . All starting configurations and possible changes occurred with equal frequency . The two key barriers were not visibly different from the other barriers . Targets yielded equal rewards . In previous studies using the maze task , we have excluded reaches whose velocity profiles differed too much from typical ( correlation r < 0 . 9 ) . No such attempt was made here , since this could have introduced bias into decision-related analyses . However , to find overt changes of mind , rewarded trials with correlation coefficients less than 0 . 9 were screened by hand to identify those in which the monkey clearly began moving toward one target then switched to the other . All rewarded trials exhibiting a clear overt change of mind are plotted in Figure 1E , F and Figure 1—figure supplement 1 . Data from these same arrays has appeared in previous publications for both J ( Churchland et al . , 2010 , 2012; Kaufman et al . , 2014 ) and N ( Churchland et al . , 2012; Kaufman et al . , 2014 ) . Since the primary goal was to decode the monkeys' preparatory activity on single trials , all stable single- and multi-unit recordings with at least one spike during the relevant epoch were included ( 101 units for J1; 132 units for J2; 96 units for N1; 188 units for N2; 196 units for N3 ) . Spike sorting was performed offline using a custom software package ( available online as MKsort; https://github . com/ripple-neuro/mksort ) , and special attention was paid to ensuring the stability of isolations over the session . The total numbers of successful trials per dataset were: J1-T , 1302; J1-S , 1345; J2-T , 1108; J2-S , 1096; N1-S , 1282; N2-S , 869; N3-S , 739 . For any given trial subset of interest , the number of trials was necessarily much lower . Each monkey performed this task for approximately 2 weeks , resulting in eight possible datasets for monkey J and 10 for monkey N . For monkey J , on the first 4 days we did not yet have certain timing equipment in place; one of these datasets was used to pilot analyses but was not included for confirmatory analysis . On 2 days , we changed the mazes mid-session to achieve better behavior; these datasets were not analyzed . The remaining 2 days' data were included . For monkey N , on 6 days we changed the mazes mid-session or the monkey exhibited unstable preferences over the session . Those datasets were not analyzed further . For one dataset , the neural data could not be decoded . In the remaining three datasets , the neural data could be decoded for the S maze but not the T maze . These three S maze datasets were included . To determine whether neurons exhibited similar responses for forced and free choices , we performed a correlation analysis . For each neuron , we first collected a vector of trial-averaged firing rates over time for the forced left condition . Firing rates were smoothed with a Gaussian ( 30 ms SD ) . We concatenated this response vector with the smoothed response vector for the forced-right condition . We then correlated it ( Pearson correlation ) with the analogous response vector for the free-choice conditions . The resulting correlation coefficients were averaged over neurons . This analysis was performed separately for the T-maze and S-maze , and the epoch considered was from −200 to +300 ms from maze onset . We restricted the above analysis to units with a reasonable signal-to-noise ratio ( SNR ) , since ‘modulation’ for unresponsive units would simply be noise and produce correlation coefficients near zero . To compute signal , we concatenated the unit's forced and free response vectors then took the range of this vector . To compute the noise , we found the greatest SEM for any point in the concatenated response vector . The SNR was signal divided by noise . We included units with an SNR of at least four . This included a total of 219 units . In order to smooth the data in a principled way based on the data itself , and to reduce the possibility of overfitting the classifier , we first applied Gaussian Process Factor Analysis ( GPFA; Yu et al . , 2009 ) . Among other advantages , GPFA chooses smoothing kernels for each dimension to reflect both their natural spectral content and the signal-to-noise available in the data . For each trial , we took the portion of the data from 300 ms before maze onset to 200 ms before movement onset , and applied GPFA with a 20 ms bin width and 12 dimensions . Only successful trials were included . The dimensionality was chosen conservatively based both on previous analysis of PMd activity and cross-validation of the present data , which indicated that any dimensionality between ∼8–14 would retain most structure without increasing noise . Movement epoch data were excluded because firing rate changes tended to be much larger and thus dominated the resulting space , reducing the quality of the delay epoch trajectory estimation . For both monkeys , GPFA was performed on all successful trials from a single day together . The subsequent decoding did not appreciably change quality when performing the GPFA step on the T-maze and S-maze separately . As our classifier , we trained a Support Vector Machine ( SVM; Cortes and Vapnik , 1995 ) with a linear kernel ( using the svmtrain function in Matlab , Mathworks , Natick , MA ) . The linear SVM identifies a classifying hyperplane ( in this case , in the 12-D GPFA space ) based on a linear weighted sum of the dimensions . Since the GPFA step is linear as well , the resulting classifier is simply a weighted sum of the neurons' smoothed spike trains . We chose to use an SVM for two reasons . First , the SVM is a wide-margin classifier , which is both inherently regularizing and well suited to non-Gaussian data . Second , in practice it performed qualitatively and quantitatively better in cross-validation than other decoders tested ( such as logistic regression or a Naïve Bayes classifier ) . To obtain a graded decode of choice , we projected the data onto the vector normal to the hyperplane ( Figure 2C ) . This is equivalent to taking the signed distance from the classifying hyperplane . We trained the SVM using only forced-choice trials with delay durations of at least 300 ms . We used only these relatively-long-delay trials so that early delay activity was not overrepresented . To ensure that the decoder was trained on activity that at least partially reflected a choice , only time points from 80 ms after maze onset to 200 ms before movement onset were considered from these trials . For convenient scaling , the resulting decoded values were normalized by the 90th percentile of all decoded values for that dataset ( not just forced-choice trials ) . In order to test the decoder's performance with time ( Figure 3A ) , we divided the distribution of delay durations into 60 ms bins . For each dataset , we then identified all the trials for which the last decoded time point happened to fall in that bin . The value reported is the fraction of trials for which the final point was decoded correctly; that is , whether it agreed with the monkey's subsequent choice . We computed a p-value to verify that the cross-validated forced-choice decoder performed above chance . To do so , we performed a simulation in which we chose the ‘decoded choice’ for each trial to be left or right by chance ( according to their overall prevalence ) , then checked how often this simulation performed as well as the real decoder . 10 million runs were performed with different random seeds . For all decoder-based analyses except those noted , only trials with delay durations ≥300 ms were used . Where noted , trials with delay durations of zero were used ( the Go cue was presented simultaneously with maze onset ) . For producing PSTHs , trials with a delay duration of ≥350 ms were used . When forced or free choice trials are referenced without explicitly noting barrier changes , only trials with no barrier changes were included . For producing PSTHs , the presence of changes of mind represented a potential confound . We therefore used the moment-by-moment decoded choice to aggressively exclude all trials that might have contained such changes of mind . Specifically , we excluded all trials for which the eventual choice did not match the decoded choice at any point more than 160 ms after maze onset ( 150 ms is the approximate length of time it takes for movement preparation to complete , see Churchland et al . , 2006b; we rounded this value up to the next multiple of 20 ms to align with our binning ) . This criterion was applied to both forced choice and free choice trials . Change-of-mind trials were also excluded before determining whether selectivity was consistent for forced and free choices ( exclusion for free choices only ) . This exclusion was not performed before any other analyses . To identify trials with clear covert changes of mind , we used the following algorithm . For each trial with a delay of ≥300 ms , we considered time points ≥160 ms after maze onset ( see above for justification of timing ) . After this time , to be considered a change of mind the time course of the decoded choice had to meet three criteria: ( 1 ) at some time point , it needed to change from leftward to rightward or vice versa; ( 2 ) at some time point , it needed to be ≥10 times as likely to come from the ‘leftward’ forced-choice distribution of decoded choice as from the ‘rightward’ forced-choice distribution of decoded choice for the same time point; and ( 3 ) at some time point , it needed to be ≥10 times as likely to come from the ‘rightward’ forced-choice distribution of decoded choice as from the ‘leftward’ forced-choice distribution of decoded choice for the same time point . That is , the decoded choice needed to swing from ‘strongly left’ to ‘strongly right’ or vice versa . To determine likelihoods , Gaussian fits were used for the leftward and rightward distributions . For determining the forced-choice distributions of decoded choice at late time points , all time points ≥600 ms were pooled . To be conservative , when this algorithm was applied to leave-one-out cross-validation trials , the classifier and distributions from that cross-validation fold were used . All trials meeting these criteria are shown in the relevant plots . This algorithm was also used to identify trials for Figure 4G , H . We determined whether the neural state during a change of mind resembled the baseline state , in the 12-D GPFA space ( Figure 4G , H , Figure 4—figure supplement 5 ) . This 12-D state acts as a denoised summary of the firing rates of all the recorded neurons . We first determined the mean baseline neural state . To do so , we averaged the neural states of all successful trials from −300 to −40 ms from maze onset . We then found the distribution of the single-trial moment-by-moment baseline states around the mean . To do so , for each of the points from the previous step , we computed the Euclidean distance between the point and the mean baseline state . This distribution across trials and times is shown in gray . Next , we identified change of mind trials: subsets of free choice trials and biasing change trials ( which are free choices throughout , but in which one barrier changes difficulty during the delay ) . In each case , we identified these trials using the algorithm described in the section above . We then found the ‘crossings’ , the time points just before and just after the decoded choice changed sign . Each trial thus contributed two points . Finally , for each of these points , we found the Euclidean distance to the mean baseline state as above . The resulting distributions are plotted in filled pink and green ( Figure 4G , H , Figure 4—figure supplement 5 ) . Finally , for comparison , we considered all free choice trials and all biasing change trials ( not just the crossing trials ) . For each point at least 160 ms after maze onset , we again found the Euclidean distance to the mean baseline state . These distributions are plotted in hollow purple and green . To compare firing rates during changes of mind with firing rates on other trials , we had to choose a comparison population of trials and a matched window of time ( since firing rates were not perfectly stable over the delay ) . We therefore used the following procedure . For each unit , for each vacillation trial , we found the last time that the decoder switched from indicating one choice to the other . We then found the firing rate of that unit in the 200 ms window centered on this crossing time . Next , we found the mean firing rate in the same window for the same unit for all non-vacillation free left-choice trials , and for all non-vacillation free right-choice trials . After repeating this procedure for each vacillation trial , we found three average firing rates: for all the vacillation trials , for all non-vacillation free lefts , and for all non-vacillation free rights . Finally , we categorized the vacillation firing rate as lower than both free rates , between them , or higher than both free rates . The procedure for bias trials was similar: analogously , we compared bias trials on which the decoder did indicate a change of mind ( sign change in the decoded choice ) with those in which it did not . If instead we repeated the analysis using free choice non-vacillation trials for comparison , results were similar . This last point indicates that the observed difference between vacillations and induced changes of mind were not due to differences in the comparison trials . We wished to see whether a slowly resolving decoded choice might be responsible for the slow RTs sometimes observed on free choice trials ( Figure 5D , Figure 5—figure supplement 2 ) . To do so , we first identified free choice trials on which the Go cue was presented simultaneously with maze onset , and which evoked an RT ≥300 ms . For these trials , we considered the decoded choice at 100 ms after maze onset . For the regression , we ensured that a decoded choice agreeing with the eventual choice was positive while a decoded choice disagreeing with the eventual choice was negative . Thus , for trials on which the monkey chose the leftward target , we inverted the sign of the decoded choice . Linear regression was then performed . | Some decisions are easy to make . We know almost immediately what outcome we want to achieve and what actions are required to do so . But other decisions involve more deliberation: there may be more factors to consider or more at stake , or the best course of action may simply not be immediately apparent . Under such circumstances , we may hesitate , waver or even change our minds completely . To date , the majority of experiments that have explored the neural basis of decision making have been unable to detect the test subjects changing their mind as they made their decision . Instead , those experiments have measured the average brain response over multiple decisions because they lacked the power to detect what happened on each individual decision . Kaufman et al . have now addressed this issue by training two monkeys to perform a decision-making task that encourages wavering and changes of mind , and examining each decision one by one . The monkeys learned to use their finger to trace a path to one of two targets on a screen to earn a reward . As the monkeys performed this target-selection task , arrays of electrodes recorded the activity of two brain regions that are involved in the planning of movements . These signals reliably predicted which of the two targets the monkey was favouring , several hundred milliseconds before the monkey was instructed to start moving his finger . Moreover , these patterns of neural activity reflected whether the monkey responded immediately or hesitated , made a firm choice or wavered , or stuck to his original choice or changed his mind . While behaviours such as hesitation and wavering feel intuitively familiar , this is the first time that they have been observed at the neural level . However , it was not clear whether the two regions of the brain studied in the experiments were responsible for making the decisions about target selection , or if the activity in these areas reflected a decision that had been taken elsewhere in the brain . Nevertheless , the results indicate that the approach developed by Kaufman et al . allows researchers to follow aspects of the decision-making process as it happens , including those times when the monkey changes its mind . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Vacillation, indecision and hesitation in moment-by-moment decoding of monkey motor cortex |
Hox proteins are well-established developmental regulators that coordinate cell fate and morphogenesis throughout embryogenesis . In contrast , our knowledge of their specific molecular modes of action is limited to the interaction with few cofactors . Here , we show that Hox proteins are able to interact with a wide range of transcription factors in the live Drosophila embryo . In this context , specificity relies on a versatile usage of conserved short linear motifs ( SLiMs ) , which , surprisingly , often restrains the interaction potential of Hox proteins . This novel buffering activity of SLiMs was observed in different tissues and found in Hox proteins from cnidarian to mouse species . Although these interactions remain to be analysed in the context of endogenous Hox regulatory activities , our observations challenge the traditional role assigned to SLiMs and provide an alternative concept to explain how Hox interactome specificity could be achieved during the embryonic development .
There is mounting evidence that many protein–protein interactions ( PPIs ) are mediated by small peptide motifs called linear motifs ( LMs ) or eukaryotic/short linear motifs ( ELMs/SLiMs ) ( Neduva and Russell , 2005; Van Roey et al . , 2012 , 2014; Tompa et al . , 2014 ) . These compact interaction interfaces are typically less than 10 residues in length and are often located within intrinsically disordered regions of highly connected proteins . Due to their small size , SLiMs exhibit high evolutionary plasticity and mediate interactions with many different types of proteins . Moreover , SLiMs are known to be important for the rewiring of interaction networks , being the subject of tissue-specific regulatory mechanisms ( Buljan et al . , 2013 ) . The contribution of SLiMs to the functional diversification and specification of key regulatory TFs throughout development and evolution remains poorly understood . For example , very few SLiMs listed in the current databases relate to the regulatory interaction between transcription factors ( TFs ) ( Dinkel et al . , 2014 ) , suggesting that this particular type of functional interaction is more difficult to capture than others . More generally , classic large-scale screening methods based on affinity purification followed by mass spectrometry ( AP-MS ) or yeast two-hybrid are more efficient for detecting stable interactions between structured domains than for revealing transient interactions involving SLiMs ( Landry et al . , 2013 ) . Therefore , alternative approaches are needed to decipher SLiM-mediated interactions and functions within the context of developmental regulatory networks in vivo . Here , we tackle this issue by using Hox proteins as a case study . Hox proteins are homeodomain ( HD ) -containing TFs present in all cnidarian and bilaterian species ( Finnerty , 2003 ) . They are required throughout the embryogenesis for controlling specific cell fates and structures along different axes and in territories as different as the limb bud ( Zakany and Duboule , 2007 ) , cardiac outflow tract ( Bertrand et al . , 2011 ) , and female genital disc ( Foronda et al . , 2005 ) . The specific functions of Hox proteins in vivo contrast with their ability to recognize closely similar DNA-binding sites as monomers in vitro ( Berger et al . , 2008; Noyes et al . , 2008 ) . This so-called Hox paradox strongly suggests that additional cofactors are required for helping Hox proteins to elicit their diverse and specific transcriptional programs in vivo . To date , only one type of cofactors is described to specify Hox functions at the molecular level . These cofactors are collectively referred to as the PBC class of HD-containing TFs , and correspond to the Pbx1-4 and Extradenticle ( Exd ) proteins in mammals and Drosophila melanogaster , respectively ( Mukherjee and Bürglin , 2007 ) . Biochemical studies have shown that the interaction between Hox and PBC proteins relies on a highly conserved motif of Hox proteins , the hexapeptide ( HX ) ( Mann et al . , 2009 ) . The HX motif constitutes the generic signature of Hox proteins after key positions within the HD ( Merabet et al . , 2009 ) . It belongs to the LIG type of SLiM according to the ELM database ( http://elm . eu . org/ ) and contains a core YF/PWM sequence conserved in all but posterior Hox paralogs throughout animal evolution ( Merabet et al . , 2009 ) . More generally , the HX motif has been defined as corresponding to an invariant Tryptophan residue located in a hydrophobic environment and followed by basic residues from +2 to +5 ( In der Rieden et al . , 2004 ) . Crystal structures with truncated proteins emphasized the importance of the Tryptophan residue in establishing strong interactions with specific residues of the PBC partner ( Mann et al . , 2009 ) . However , in vivo analyses showed that mutations within the HX motif ( including that of the key Tryptophan residue ) did not systematically abolish PBC-dependent functions of Hox proteins ( Galant et al . , 2002; Merabet et al . , 2003 ) . In addition , Hox-PBC interactions are influenced by the promoter environment and can occur in absence of the HX motif in several cnidarian and bilaterian Hox proteins ( Hudry et al . , 2012 , 2014 ) . Reciprocally , the HX is also required for PBC-independent functions ( Merabet et al . , 2011 ) and for interacting with Bip2 , a TATA-binding protein associated factor in Drosophila ( Prince et al . , 2008 ) . Together these observations highlight that the HX motif is neither a unique nor an obligatory Hox protein interface for recruiting the PBC cofactor , suggesting that Hox-PBC interactions could rely on the presence of other specific SLiM ( s ) . Notably , another motif called UbdA and present in Ultrabithorax ( Ubx ) and AbdominalA ( AbdA ) proteins of protostome lineages was recently described to be important for the formation and activity of the Ubx/Exd complex in Drosophila ( Merabet et al . , 2007; Hudry et al . , 2012; Foos et al . , 2015 ) . In summary , our current knowledge on SLiM-mediated interactions in Hox proteins is limited to only two different types of TFs , the PBC and Bip2 proteins . Given the number of embryonic events controlled by Hox proteins , we hypothesize that Hox SLiMs such as the HX and UbdA motifs could interact with a higher number of TFs . Identifying these TFs represents a major challenge to understand part of the molecular cues underlying Hox transcriptional specificity and diversity in vivo . Here , we exploited the recently developed bimolecular fluorescence complementation ( BiFC ) ( Hudry et al . , 2011 ) to profile a wide range of Hox protein interactions in the Drosophila embryo and investigate whether SLiMs could influence their specificity in vivo . As a first step , we identified the respective sets of BiFC interactors of five Drosophila Hox paralogs , showing that each Hox interactome relies on a different combination of TFs . The role of the HX and UbdA motifs was then analysed in several Hox interactomes and in different tissues of the live embryo . Our data establish that the ablation of Hox SLiMs not only prevents several interactions but additionally leads to a number of ectopic interactions . These effects differ depending on the Hox protein and tissue considered , suggesting that SLiM activity could be strongly influenced by the protein environment . Furthermore , results obtained with mouse and cnidarian Hox proteins indicate that the inhibitory activity of SLiMs could be important for restricting the inherent binding potential of intrinsically disordered regions . Altogether , these findings provide new insights on how Hox transcriptional specificity could be reached in vivo and add to the functional repertoire of SLiMs .
BiFC relies on the property of monomeric fluorescent proteins to be reconstituted from two separate sub-fragments upon spatial rearrangement ( Ghosh et al . , 2000 ) . This property is used with different types of proteins in various cell and animal model systems to demonstrate the close proximity hence the existence of possible interactions between two putative protein partners ( Kerppola , 2008; Kodama and Hu , 2012 ) . We previously demonstrated that BiFC was sensitive and specific enough for analysing Hox-TF interactions in the live Drosophila embryo ( Hudry et al . , 2011 ) . Experimental parameters were established by using the partnership between AbdA and Exd as a case study . Interaction was visualized by fusing the two partners with complementary fragments of the Venus ( yellow: Figure 1A–A′ ) , mCherry ( red ) or Cerulean ( blue ) fluorescent protein . Among several controls , we showed that the simultaneous co-expression of a ‘cold’ AbdA protein ( i . e . , not fused with a fragment of the fluorescent protein ) with AbdA and Exd fusion proteins could induce a titration of the BiFC complex ( Hudry et al . , 2011 ) . Thus , cold interactions ( in this case AbdA–AbdA and AbdA–Exd interactions ) could compete against BiFC , leading to a significant decrease of fluorescent signals in the embryo ( Hudry et al . , 2011 ) . 10 . 7554/eLife . 06034 . 003Figure 1 . A BiFC competition screen identifies candidate transcription factors ( TFs ) as potential binding partners of the Hox protein AbdominalA ( AbdA ) . ( A–C ) Principle of the competition test . ( A–A′ ) Co-expression of Extradenticle ( Exd ) and AbdA proteins fused to the N- ( VN ) or C- ( VC ) terminal fragment of the Venus fluorescent protein leads to BiFC in the embryo . ( B–B′ ) Cases where no BiFC competition could be observed with a cold TF . B′ is an illustrative picture of non-competitive BiFC resulting from the simultaneous co-expression of the red fluorescent protein RFP ( see also Figure 1—figure supplement 2 ) . ( C–C′ ) Cases where BiFC competition could be observed with a cold TF . C′ is an illustrative picture of competitive BiFC resulting from the simultaneous co-expression of a nuclear-localized form of Exd ( see also Figure 1—figure supplement 1 ) . Note that AbdA-interacting partners do not obligatory lead to competitive BiFC . TFs that could be validated in the secondary step as AbdA-binding partners ( see Figure 2 ) are indicated ( dotted-red box ) . ( D ) Graph showing the repartition of competitive ( red bars ) and non-competitive ( green bars ) TFs with regard to their DNA-binding domain . See also Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 00310 . 7554/eLife . 06034 . 004Figure 1—figure supplement 1 . Illustrative pictures of competitive TFs . ( A ) Competitive control with the UAS-nls-Exd construct . ( B ) Competitive TFs . Each construct was co-expressed with AbdA and Exd fusion proteins and BiFC was analysed in the epidermis of stage 10–12 of live embryos . The competition was deduced when altered BiFC signals could be observed in the expected or nearly expected proportion of the embryo progeny ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 00410 . 7554/eLife . 06034 . 005Figure 1—figure supplement 2 . Illustrative pictures of non-competitive TFs . ( A ) Non-competitive control with the UAS-RFP construct . ( B ) Non-competitive TFs . Each construct was co-expressed with AbdA and Exd fusion proteins and BiFC was analysed in the epidermis of stage 10–12 of live embryos . Absence of competition was deduced when no clear altered BiFC signals could be observed in the embryo progeny ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 005 We reasoned that any protein capable of displacing the AbdA and/or Exd fusion protein from the BiFC complex could lead to a loss of the fluorescence . This readout could thus serve to rapidly identify putative interacting partners of AbdA . In this case , competitive BiFC could not be observed with proteins that are exclusively participating in the AbdA/Exd complex ( Figure 1B–B′ ) . In addition , the titration of BiFC signals could only occur when the cold interaction is strong enough to disrupt the assembly between AbdA and Exd ( Figure 1C–C′ ) . However , as competitive BiFC could not discriminate between AbdA- and Exd-specific interacting partners ( Figure 1C–C′ ) , a second experimental phase will be necessary to confirm the Hox interaction status . Despite these limitations , we decided to test our hypothesis with a reasonable number of candidate TFs and by using a fast genetic approach . To this end , we established a BiFC reporter fly line expressing AbdA and Exd fusion proteins under the control of the abdA-Gal4 driver ( ‘Materials and methods’ and [Hudry et al . , 2011] ) . This fly line can be crossed with individuals containing any UAS-driven cold candidate-binding partner and BiFC could directly be assessed in the embryo progeny . As a control , we verified that the co-expression of a nuclear-localized form of Exd ( Kammermeier et al . , 2004 ) could indeed affect BiFC fluorescent signals in the embryo ( Figure 1—figure supplement 1 ) . In comparison , co-expressing a red fluorescent protein ( RFP ) under the same condition did not lead to any changes in the BiFC profile ( Figure 1—figure supplement 2 ) . Here , we deliberately focused the screen on TFs that could participate in the transcriptional programs of Hox proteins in their various developmental contexts . We chose a starting set of 80 TFs covering different DNA-binding families and displaying distinct expression profiles in the three main germ layers of the embryo ( Supplementary file 1 ) . We observed that 33 of those TFs could compete against AbdA/Exd assembly ( Figure 1D and Figure 1—figure supplements 1 , 2 ) . Among them we found Biniou ( Bin ) and Mad , previously described to participate in the regulation of Hox target enhancers ( Grienenberger et al . , 2003; Walsh and Carroll , 2007 ) , and Teashirt ( Tsh ) , known to help Hox proteins in specifying the trunk segments of the embryo ( Fasano et al . , 1991 ) . Thus , these three TFs validate the competition screen . Interestingly , 65% of the tested HD and GATA TFs ( 9/14 and 4/6 , respectively ) were positive in the competition test ( Figure 1D ) , representing a strong tendency compared to the other tested TF classes . In total , none of the 33 competitors was previously described as a binding partner of AbdA or Exd , illustrating the efficiency of the competition screen for revealing new candidate cofactors in vivo . Our genetic competitive approach revealed proteins that could potentially bind to AbdA and/or Exd . We next analysed whether these positive competitors could more specifically interact with AbdA in a complementary BiFC-based approach . To this end , we generated fly lines carrying each corresponding TF as a UAS-driven fusion construct compatible for BiFC ( Figure 2A and ‘Materials and methods’ ) . Two additional TFs ( TFIIbeta and Knot [Kn] ) were added to the 33 positive competitors , reaching a total of 35 fusion TFs that could be used for BiFC in Drosophila ( Figure 2B and Supplementary file 1 ) . TFIIbeta , could not be tested by competition , as no corresponding UAS-driven fly line exists . It is however a good positive candidate as it is described to interact with Ubx in a yeast two-hybrid screen ( Bondos et al . , 2004 ) . Kn did not compete against BiFC in the first step and was used as a negative candidate interacting partner of the experiment . 10 . 7554/eLife . 06034 . 006Figure 2 . BiFC validates the AbdA-interaction status of competitive TFs . ( A ) Principle of the BiFC screen between AbdA and the 35 selected TFs . ( B ) Repartition of the 35 selected TFs with regard to their DNA-binding domain . ( C ) Illustrative pictures of BiFC signals obtained between VC-AbdA and the indicated VN-TF in the epidermis of stage 10–12 of live embryos . Fusion constructs are expressed with the abdA-Gal4 driver . Note that typical nuclear interaction profiles are observed between AbdA and different TFs ( white-dotted boxes ) . See also ‘Materials and methods’ , Supplementary files 2 , 3 and Figure 2—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 00610 . 7554/eLife . 06034 . 007Figure 2—figure supplement 1 . Co-immunoprecipitation ( co-ip ) between AbdA and TFs selected from the set used for BiFC in the Drosophila embryo . ( A ) Positive co-ip experiments . ( B ) Negative co-ip experiments . Co-ips were performed with an anti-HA antibody recognising the AbdA-HA variant . Presence of the TF was revealed with a polyclonal anti-GFP antibody recognising the VN fragment fused to each TF . The ip was performed with the fusion TF expressed alone ( lane 1 ) or together with AbdA ( lane 2 ) . Lane 3 corresponds to the input of cells expressing both proteins ( 20% of the total lysate ) before the ip . Gels on the bottom are western blots validating the efficiency of the ip with the anti-HA antibody . First gel corresponds to a control validating that the anti-GFP does not recognize AbdA-HA . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 00710 . 7554/eLife . 06034 . 008Figure 2—figure supplement 2 . BiFC between mesodermal TFs and AbdA in the mesoderm . Fusion proteins were expressed with the 24B-Gal4 driver recombined to UAS-mCherry ( red ) , which allows following the expression profile in the mesoderm . BiFC ( green ) was observed in stages 12–14 of live embryos . Note that weak BiFC signals could be observed with VN-Kn while VN-Lmd remained negative with VC-AbdA . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 008 BiFC was observed in the epidermis of stage 10 embryos , even for TFs that are not endogenously expressed in this tissue ( see ‘Materials and methods’ , Supplementary file 2 and [Hammonds et al . , 2013] ) . We anticipated that the epidermis was appropriate for the interaction since competition was observed in this tissue . In addition , the epidermis has been shown to tolerate the activation of mesodermal target genes upon the ectopic expression of mesoderm-specific TFs ( Cunha et al . , 2010 ) , suggesting it is a relatively neutral tissue . Finally , BiFC could also be increased in a heterologous tissue because of the absence of competition by the endogenous gene product , as previously described ( Hudry et al . , 2011 ) . Among the 35 TFs tested as fusion constructs , 31 led to BiFC signals with AbdA , including TFIIbeta ( Figure 2C ) . These fluorescent signals display homogenous or punctuate distribution within the nucleus , depending on the TF considered ( Figure 2C ) . BiFC was negative with Kn as expected , given that no competition was previously observed with this TF . BiFC was however also negative with Krüppel ( Kr ) , Lameduck ( Lmd ) , and Pangolin ( Pan ) , although these three TFs were positive competitors . This discrepancy suggests that the previously observed competition could result from the formation of specific cold complexes with the Exd and not with the AbdA fusion protein . Alternatively , the fusion topologies could forbid the interaction hence BiFC between the three TFs and AbdA . Indeed , the negative influence of fusion topologies on protein–protein interactions was previously described and is hardly predictable ( Hudry et al . , 2011 ) . To assess the potential influence of fusion topologies on AbdA-TF interactions , we performed co-immunoprecipitation ( co-ip ) experiments , using an anti-HA antibody recognising a HA-tagged form of AbdA . We reasoned that a small HA epitope ( 8 residues long ) should be more neutral than the Venus fragment ( 80 residues long ) for the interaction with the TF . Practically , the fusion TF was co-expressed with the AbdA-HA construct in S2 cells , and its presence was verified with an anti-GFP antibody recognizing the Venus fragment ( see ‘Materials and methods’ ) . Experimental parameters were established with the control Exd cofactor . All tested BiFC-positive TFs were found by co-ip ( Supplementary file 3 ) , highlighting that the S2 cell environment is appropriate for revealing interactions with tissue-specific TFs . Thus , observations from BiFC could be reproduced by co-ip , as previously noticed ( Lee et al . , 2011 ) . Co-ip was also performed with the three positive competitors that did not produce BiFC with AbdA ( Kr , Lmd , Pan ) . We observed that these three TFs could be immuno-precipitated with AbdA ( Supplementary file 3 ) . This was not the case for Kn , which was negative both in the competition and BiFC tests ( Figure 2—figure supplement 1 ) . We conclude that inappropriate fusion topologies are likely to be responsible for the absence of BiFC between AbdA and Lmd , Kr or Pan in the epidermis . Several of the tested TFs are not expressed in the epidermis ( Supplementary file 2 ) . We thus wondered whether the interaction with those TFs could also be reproduced in their endogenous expression tissue . To this end , we repeated BiFC analyses in the mesoderm , using mesodermal TFs that are not expressed in the epidermis , including Lmd and Kn ( Supplementary file 3 ) . We observed that TFs interacting with AbdA in the epidermis were also positive in the mesoderm ( Figure 2—figure supplement 2 ) . Fluorescent signals were however generally weaker than in the epidermis , probably due to the competition by the endogenous gene products . Surprisingly , weak BiFC signals could also be observed with Kn , while Lmd remained negative in the mesoderm ( Figure 2—figure supplement 2 ) . We concluded that Kn could interact with AbdA but that this interaction is more sensitive to the cell environment in order to occur . In summary , although our approach was performed with a limited set of TFs ( around 12% of all Drosophila TFs ) , it revealed an unexpectedly high number of new binding partners of AbdA . However , whether and how these binding partners could be used in the context of endogenous regulatory activities of AbdA remains to be investigated . This result still illustrates the strong propensity of AbdA to establish interactions with diverse TFs in vivo . In the following , we considered the set of 35 TFs as sufficiently representative for addressing the issue of the molecular mechanisms underlying Hox interactome specificity in vivo . Our experimental parameters allow quantifying subtle changes in fluorescent signal intensities in the whole Drosophila embryo ( ‘Materials and methods’ and [Hudry et al . , 2011] ) . These variations in the fluorescence intensity can be correlated to differences in interaction affinity . Indeed , high affinity partners lead to fast accumulation , hence strong BiFC signals , and vice versa ( Hudry et al . , 2011 ) . Here , levels of BiFC were used to measure the effects of AbdA mutations on the interaction potential with each of the 35 TFs . We first focused on the HD , which is responsible for the DNA-binding and the most conserved part of Hox proteins . The HD of Hox proteins is also described to interact with different types of cofactors ( Merabet et al . , 2009 ) , suggesting it could be involved in several of the observed interactions . More precisely , we asked ( i ) whether AbdA-TF interactions could depend on the DNA-binding activity of the HD , and ( ii ) whether the HD could be sufficient for AbdA-TF interactions . BiFC was performed with two corresponding mutant forms of AbdA: one carrying the N51A mutation in the HD , which abolishes the DNA-binding activity of full length AbdA ( AbdA51 ) ( Hudry et al . , 2011 ) , the other resulting in a truncated version that contains only the HD ( AbdAHD ) ( Boube et al . , 2014 ) . The two fusion constructs were inserted on the same genomic locus and expressed at similar levels in the embryo as the wild type AbdA fusion protein ( ‘Materials and methods’ and [Hudry et al . , 2011; Boube et al . , 2014] ) . BiFC was measured in the epidermis and considered as affected when the fluorescent signal was equal or lower than 50% of the fluorescent level normally obtained with wild type AbdA ( see ‘Materials and methods’ for quantification details ) . Among the 31 BiFC-positive interactions , 18 were strongly affected or lost with AbdA51 ( compared Figure 3A , B , and Figure 3—figure supplement 1 ) . Still , a significant proportion of the interactions ( for 13 TFs ) was retained , two of which were even stronger than with wild type AbdA ( leading to intensities higher than 120% of the wild type fluorescent signal ) . Thus , the two corresponding TFs ( Pannier and Slouch ) preferentially interact with a form of AbdA that is unable to bind DNA . Effects were more drastic with the minimal AbdAHD construct , which kept only eight interactions among the 31 positive TFs ( compare Figure 3A , C , and Figure 3—figure supplement 2 ) . One of these interactions ( with Serpent , Srp ) was also stronger compared to wild type AbdA . 10 . 7554/eLife . 06034 . 009Figure 3 . Role of the homeodomain ( HD ) in the AbdA interactome . ( A ) Interactome with wild type AbdA . ( B ) Interactome with the DNA-binding deficient form of AbdA ( mutated in the residue 51 of the HD , as illustrated with the red bar ) . See also Figure 3—figure supplement 1 . ( C ) Interactome with the HD of AbdA . See also Figure 3—figure supplement 2 . Each interactome is represented with the 35 TFs . The colour code for TFs corresponds to their type of DNA-binding domain , as shown in the Figure 2B . TFs that are not colour-filled correspond to TFs that do not interact with the wild type Hox protein . Those TFs are not connected to the Hox protein . Dotted lines indicate TFs that do interact with the wild type Hox protein . Interactions with Hox variants are depicted as the following: dotted lines indicate unaffected interactions ( in between 51% and 119% of the wild type interaction ) ; solid black lines indicate stronger ( equal or superior to 120% of the wild type interaction ) or novel interactions ( with a non-colour-filled TF ) ; absence of the dotted line with a colour-filled TF indicates a partial ( equal or below to 50% of the wild type interaction ) or complete loss of the interaction . Each Hox variant is schematized in the centre of the interactome . HD: Homeodomain . HX: Hexapeptide . See also ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 00910 . 7554/eLife . 06034 . 010Figure 3—figure supplement 1 . BiFC between the 35 TFs and the homeodomain ( HD ) -mutated form of AbdA . Illustrative pictures are provided for each TF as indicated . Quantifications were performed in the epidermis of stage 10 embryos and are depicted on the right of each picture as a boxplot representation ( see also ‘Materials and methods’ ) . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01010 . 7554/eLife . 06034 . 011Figure 3—figure supplement 2 . BiFC between the 35 TFs and the homeodomain ( HD ) of AbdA . Illustrative pictures are provided for each TF as indicated . Quantifications were performed as in Supplementary file 5 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No statistical quantification is provided in the case of a new interaction ( as seen with Krüppel ) . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 011 The observation of stronger interactions in some cases prompted us to analyse whether previously negative TFs with AbdA ( Kn , Kr , Lmd , and Pan ) could produce BiFC with either of the two AbdA mutant forms . We found that Kr could indeed lead to BiFC signals when using the AbdAHD version ( Figure 3C and Figure 3—figure supplement 2 ) . Thus , the interaction was strong enough in this particular case to be visualized despite unfavourable fusion topologies , illustrating that HD-surrounding region ( s ) could inhibit the interaction potential of the HD in vivo . Together these results show that the DNA-binding of AbdA is not systematically required for recruiting the TFs . The HD itself is also not sufficient in most cases , suggesting that surrounding protein region ( s ) are important for recruiting TFs . Interestingly , these HD-surrounding region ( s ) can also have an inhibitory role since their absence allows the formation of stronger interactions between the HD and two of the tested TFs . Given that HD-surrounding regions are less conserved in Hox proteins in general , we then asked whether some of the revealed interactions could also be found with other Drosophila Hox proteins . With the exception of Ubx , Drosophila Hox proteins have few redundant functions with AbdA , as reflected at the protein sequence and embryonic expression levels ( Figure 4A ) . We thus wondered whether common vs specific features between Hox proteins could be found in their respective interactomes . To this end , we repeated BiFC between the 35 TFs and four other Hox proteins ( see ‘Materials and methods’ ) , namely Sex combs reduced ( Scr ) , Antennapedia ( Antp ) , Ultrabithorax ( Ubx ) , and AbdominalB ( AbdB ) . 10 . 7554/eLife . 06034 . 012Figure 4 . Comparison between wild type and Hexapeptide ( HX ) -mutated Hox interactomes . ( A ) Embryonic expression profile ( grey ) and protein sequence identity of each of the five Drosophila Hox proteins under study . The percentage of sequence identity is given in comparison to AbdA . The conserved core sequence of the HX is also given for each Hox protein . ( B ) Heatmap showing the organisation of wild type Hox interactomes with the 35 TFs . See also Figure 4—figure supplements 1–4 . ( C ) Heatmap showing the organisation of HX-mutated Hox interactomes with the 35 TFs . See also Figure 4—figure supplements 5–9 . Interactions are symbolized by a colour code , as indicated . Note that the yellow colour , which corresponds to a gain of the interaction potential , appears with the HX mutation in all Hox proteins . Dendogram branches are coloured according to their bootstrap score: black 100% , grey 90–100% , blue 80–90% , green 70–80% , yellow 60–70% , orange 50–60% , pink 0 . 1–50% , red 0% support respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01210 . 7554/eLife . 06034 . 013Figure 4—figure supplement 1 . BiFC between the 35 TFs and Sex combs reduced ( Scr ) . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01310 . 7554/eLife . 06034 . 014Figure 4—figure supplement 2 . BiFC between the 35 TFs and Antennapedia ( Antp ) . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01410 . 7554/eLife . 06034 . 015Figure 4—figure supplement 3 . BiFC between the 35 TFs and Ultrabithorax ( Ubx ) . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01510 . 7554/eLife . 06034 . 016Figure 4—figure supplement 4 . BiFC between the 35 TFs and AbdominalB ( AbdB ) . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01610 . 7554/eLife . 06034 . 017Figure 4—figure supplement 5 . BiFC between the 35 TFs and hexapeptide ( HX ) -mutated Scr . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . No statistical quantification is provided in the case of a new interaction ( as seen for Distal-less , Even skipped , Eyes-absent , Knirps , Mef2 and Twist ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01710 . 7554/eLife . 06034 . 018Figure 4—figure supplement 6 . BiFC between the 35 TFs and hexapeptide ( HX ) -mutated Antp . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . No statistical quantification is provided in the case of a new interaction ( as seen for Bagpipe , Knot , Ladybird early , Midline , Nautilus and Pannier ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01810 . 7554/eLife . 06034 . 019Figure 4—figure supplement 7 . BiFC between the 35 TFs and hexapeptide ( HX ) -mutated Ubx . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . No statistical quantification is provided in the case of a new interaction ( as seen for Distal-less , Eyes-absent , Knot , Pannier , Pointed , Serpent , Seven-up and Zfh-1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 01910 . 7554/eLife . 06034 . 020Figure 4—figure supplement 8 . BiFC between the 35 TFs and hexapeptide ( HX ) -mutated AbdA . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . No statistical quantification is provided in the case of a new interaction ( as seen for Knot ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 02010 . 7554/eLife . 06034 . 021Figure 4—figure supplement 9 . BiFC between the 35 TFs and hexapeptide ( HX ) -mutated AbdB . Illustrative pictures of stage 10–12 embryos are provided for each TF as indicated . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . No statistical quantification is provided in the case of a new interaction ( as seen for Distal-less , Pangolin , Slouch and Spalt major ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 021 Overall , we observed an unexpected high proportion of positive interactions with the four additional Hox proteins: from 22 and 18 with Scr and Antp , to 21 and 26 with Ubx and AbdB , respectively ( Figure 4B and Figure 4—figure supplements 5–9 , Figure 5—figure supplement 1 ) . Because of this high proportion , the majority of interactions are common to several Hox proteins . Despite this number of common interactions , each Hox interactome contains a specific combination of binding partners . Interestingly , the hierarchical clustering of Hox interactomes ( see ‘Materials and methods’ ) does not reflect the protein sequence similarity between Hox proteins ( Figure 4A–B ) . For example , the interactomes of AbdA and AbdB appear closely similar although AbdB is much more divergent from AbdA than the other Hox proteins . Taken together these observations show that Hox proteins are able to bind to common and different types of TFs . Still , each Hox interactome contains a distinct combination of TFs , demonstrating a certain level of specificity . Moreover , the hierarchical clustering of Hox interactomes suggests that the recruitment of common cofactors would not obligatorily rely on the same Hox protein interface . BiFC analyses with AbdA showed that HD-surrounding regions are important for most of the revealed interactions . In addition to the HD , the HX motif represents the second generic signature of Hox proteins . We thus asked whether this common Hox SLiM could be important for recruiting common TFs in vivo . Its role was assessed within each Hox interactome by repeating BiFC between the 35 TFs and the corresponding HX-mutated Hox proteins ( see ‘Materials and methods’ and [Hudry et al . , 2012] ) . Heatmap representation shows that the HX mutation led to a complete reorganisation of Hox interactomes when compared to the wild type Hox proteins ( compare Figure 4B , C and Figure 4—figure supplements 5–9 ) . More precisely , the HX mutation affects the majority of Hox-TF interactions . Surprisingly , this mutation leads not only to a loss , but also to a gain of the Hox interaction potential , with the appearance of stronger or new interactions ( highlighted in yellow in Figure 4C ) . The balance between gain and loss was different depending on the Hox protein: Antp and AbdB were more sensitive to a loss while the reverse was observed with Ubx and AbdA . The HX of Scr was equally responsible for a gain or loss of interactions ( Figure 5 and Figure 5—figure supplement 1 ) . Overall , the HX mutation leads more often to a gain than a loss of the Hox interaction potential . Interestingly , the positive or negative influence of the HX on Hox-TF interactions is not identical for each TF . For example , the HX mutation has no effect on AbdA-Twist ( Twi ) interaction , while it leads to a loss of interaction between Antp and AbdB and the same TF , and to a stronger and novel interaction with Ubx and Scr , respectively ( Figure 5 and Figure 5—figure supplement 1 ) . Thus , the role of the HX appears dictated by the Hox protein to which it belongs and not by the interacting TF , therefore reinforcing the fact that HX neighbourhood is important for controlling its interaction properties ( Merabet and Hudry , 2011 ) . 10 . 7554/eLife . 06034 . 022Figure 5 . The HX mutation increases the interaction potential of Hox proteins with TFs in vivo: example in Ubx and AbdA . ( A–A′ ) Comparison between wild type and HX-mutated interactomes of Ubx . ( B–B′ ) Comparison between wild type and HX-mutated interactomes of AbdA . The HX mutation led more frequently to stronger or new interactions than to interaction loses in these two Hox proteins . Colour code and representation are as in Figure 3 . The HX mutation is indicated and highlighted in red . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 02210 . 7554/eLife . 06034 . 023Figure 5—figure supplement 1 . The HX mutation increases the interaction potential of Hox proteins with TFs in vivo: example in Scr , Antp , and AbdB . ( A–A′ ) Comparison between wild type and HX-mutated Scr . ( B–B′ ) Comparison between wild type and HX-mutated Antp . ( C–C′ ) Comparison between wild type and HX-mutated AbdB . Colour code and representation are as in Figure 3 . The HX mutation is indicated and highlighted in red . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 023 In summary , our results show that a short conserved motif , common to all Hox proteins , is specifically used both for promoting and limiting their interaction potential with TFs . Since we observed that the HX could inhibit the interaction potential of Hox proteins , we asked whether this property could also be found with other more specific Hox SLiMs . We focused on Ubx and AbdA , in which the HX mutation led to the highest number of gained interactions among the tested Hox proteins . Interestingly , Ubx and AbdA share the UbdA motif , which is conserved in most protostome lineages ( Balavoine et al . , 2002 ) . This motif was shown to be important for recruiting the Exd cofactor in Ubx ( Merabet et al . , 2007; Foos et al . , 2015 ) and for tissue-specific activities of AbdA in vivo ( Merabet et al . , 2011 ) . As previously done with the HX motif , the role of the UbdA motif was assessed by performing BiFC with UbdA-mutated forms of Ubx and AbdA ( Figure 6A and [Hudry et al . , 2012] ) . We focused on the 20 common binding partners of Ubx and AbdA to potentially reveal a common usage mode of the UbdA motif between the two Hox proteins . 10 . 7554/eLife . 06034 . 024Figure 6 . Usage mode of the HX and UbdA motifs in Ubx and AbdA proteins . ( A ) Schematic representation of wild type and HX- or UbdA-mutated Ubx and AbdA proteins . ( B ) Heatmap comparing interaction properties of HX- and UbdA-mutated Ubx proteins with a set of 20 TFs . These TFs are common to Ubx and AbdA for BiFC . See also Figure 6—figure supplement 1 . ( C ) Heatmap comparing interaction properties of HX- and UbdA-mutated AbdA proteins with the same set of TFs . See also Figure 6—figure supplement 2 . Note that the HX and UbdA mutations have distinct or opposite effects for most of the interactions in Ubx and AbdA . ( D ) Heatmap comparing interaction properties of HX-mutated Ubx and AbdA proteins with the 20 common TFs . ( E ) Heatmap comparing interaction properties of UbdA-mutated Ubx and AbdA proteins with the 20 common TFs . Note that a higher proportion of TFs is similarly affected by the UbdA mutation in Ubx and AbdA when compared to the HX mutation . Colour code is as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 02410 . 7554/eLife . 06034 . 025Figure 6—figure supplement 1 . BiFC with UbdA-mutated Ubx in the epidermis . BiFC was performed with the 20 TFs interacting both with Ubx and AbdA . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 02510 . 7554/eLife . 06034 . 026Figure 6—figure supplement 2 . BiFC with UbdA-mutated AbdA in the epidermis . BiFC was performed with the 20 TFs interacting both with Ubx and AbdA . Quantification was performed as in Figure 3—figure supplement 1 . Illustrative pictures of slightly older embryos are shown when signals were strongly affected . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 026 Results show that the UbdA mutation affects the majority of interactions in both Hox proteins , as previously noticed for the HX mutation ( Figure 6B–C and Figure 6—figure supplements 1 , 2 ) . Effects can also be categorized as a gain or a loss of the interaction potential . However , the UbdA motif has a more pronounced tendency to be required for the interaction rather than for inhibiting it ( Figure 6B–C ) . For example , the UbdA mutation leads to 12 loses and 5 gains among the 20 tested TFs with AbdA . In comparison , 3 losses and 12 gains were induced upon the HX mutation for the same set of interactions . In addition , the HX and UbdA mutations have distinct effects for the majority of interactions established by Ubx or AbdA ( Figure 6B–C ) . This result highlights that the two Hox proteins do not use the HX and UbdA motifs similarly . Interestingly , the UbdA mutation often leads to similar effects in Ubx and AbdA compared to the HX mutation ( Figure 6D–E ) . More precisely , the UbdA mutation affects the majority ( 13/20 ) of the tested TFs in a similar way , which is not the case for the HX mutation ( 8/20 ) . In conclusion , the UbdA motif displays preferential Ubx/AbdA-specific interaction properties compared to the HX motif . These interaction properties rely in part on inhibitory activities , highlighting that the negative influence on PPIs is not a specific property of the HX motif . To gain further insights into the molecular property of Hox SLiMs , we next examined whether the regulatory activity of the HX and UbdA motifs could change depending on the embryonic tissue considered . Previous work showed that the HX and UbdA motifs have tissue-specific functions in AbdA ( Merabet et al . , 2011 ) , suggesting that their interaction properties with TFs would not be identical in different embryonic tissues . To test this hypothesis , we analysed the interaction potential of HX- or UbdA-mutated AbdA in the mesoderm and nervous system , focusing on TFs that are normally expressed in one and/or both tissues . We also used a set of TFs that were all BiFC-positive with AbdA in the epidermis and all but one sensitive to the HX or UbdA mutation in this tissue ( Figure 7A ) . 10 . 7554/eLife . 06034 . 027Figure 7 . The HX and UbdA motifs of AbdA have different interaction properties in different embryonic tissues . ( A ) Interaction properties of wild type and HX- or UbdA-mutated AbdA in the epidermis . ( B ) Interaction properties of wild type and HX- or UbdA-mutated AbdA in the mesoderm . See also Figure 7—figure supplement 1 . ( C ) Interaction properties of wild type and HX- or UbdA-mutated AbdA in the nervous system . See also Figure 7—figure supplement 2 . Picture of an embryo making BiFC ( green ) and expressing the dsRed fluorescent protein under the control of the Gal4 driver illustrates the tissue of interest in each condition . Interactomes are represented as in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 02710 . 7554/eLife . 06034 . 028Figure 7—figure supplement 1 . BiFC with wild type , HX- or UbdA-mutated AbdA in the mesoderm . BiFC was performed with a set of 9 TFs that were all BiFC-positive with AbdA in the epidermis . Statistical quantification of BiFC signals with mutated AbdA proteins is shown as a boxplot representation . Quantification was performed in the mesoderm of stage 12 embryos ( see also ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 02810 . 7554/eLife . 06034 . 029Figure 7—figure supplement 2 . BiFC with wild type , HX- or UbdA-mutated AbdA in the nervous system . BiFC was performed with a set of seven TFs that were all BiFC-positive with AbdA in the epidermis . Statistical quantification of BiFC signals with mutated AbdA proteins is shown as a boxplot representation . Quantification was specifically performed in the ventral nerve cord of stage 14 of live embryos ( see also ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 029 We observed that the HX and UbdA motifs were less often required in the mesoderm since their mutation affected fewer interactions than in the epidermis ( compare Figure 7A , B , and Figure 7—figure supplement 1 ) . Still , affected TFs again correspond to a loss or a gain of the Hox interaction potential . The inhibitory activity of the HX and UbdA motifs was more pronounced in the nervous system , since their mutation led only to stronger interactions with TFs ( Figure 7C and Figure 7—figure supplement 2 ) . This effect was more obvious with the HX mutation , which led to a gain of interaction for all but one TF in this tissue ( Figure 7C ) . In total , all tested TFs were not similarly affected by the HX or UbdA mutation in the three different tissues . Thus , the two motifs are differently used depending on the Hox protein and tissue considered . This specific usage mode is based both on the positive and negative control of PPIs . Given the evolutionary conserved roles of Hox proteins in general , we then tested whether these novel facets of SLiM activity could also be found in Hox proteins from other animal species . Our work revealed that the HX could specify Drosophila Hox interactomes in part by limiting the interaction potential in a context-dependent manner . Here , we ask whether the inhibitory role of the HX on PPIs could constitute an evolutionary conserved property of Hox proteins . To this end , we used the mouse HoxB8 and Nematostella HoxE proteins as extreme representatives ( Figure 8A ) . HoxB8 is a central Hox protein containing a typical HX motif . HoxE was recently shown to display central-like molecular properties ( Hudry et al . , 2014 ) , although it contains a posterior-like derived HX motif ( Figure 8A ) . Overall , HoxB8 and HoxE show little sequence identity with the Drosophila AbdA protein outside the HX and HD ( Figure 8A ) . 10 . 7554/eLife . 06034 . 030Figure 8 . The Nematostella HoxE and mouse HoxB8 proteins interact with several Drosophila TFs . ( A ) Schematic representations of the Hox proteins and the corresponding animal phylogeny . The percentage of sequence identity is given in comparison to AbdA . ( B ) Interactome between mouse HoxB8 and the 35 Drosophila TFs . ( C ) Interactome between Nematostella HoxE and the 35 Drosophila TFs . Colour code and representation are as in Figure 3 . See also Figure 8—figure supplements 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 03010 . 7554/eLife . 06034 . 031Figure 8—figure supplement 1 . BiFC between Drosophila TFs and the mouse HoxB8 protein . Illustrative pictures are provided for each TF as indicated . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 03110 . 7554/eLife . 06034 . 032Figure 8—figure supplement 2 . BiFC between Drosophila TFs and the Nematostella HoxE protein . Illustrative pictures are provided for each TF as indicated . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 03210 . 7554/eLife . 06034 . 033Figure 8—figure supplement 3 . BiFC between Drosophila TFs and HX-mutated HoxB8 . Illustrative pictures are provided for each TF as indicated . BiFC was performed with TFs that were negative with wild type HoxB8 . The HX mutation leads to five new interactions among the eight tested TFs ( as seen for Apontic , Bagpipe , Biniou , Empty spiracles and Suppressor-of-Hairless ) . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 03310 . 7554/eLife . 06034 . 034Figure 8—figure supplement 4 . BiFC between Drosophila TFs and HX-mutated HoxE . Illustrative pictures are provided for each TF as indicated . BiFC was performed with TFs that were negative with wild type HoxE . The HX mutation leads to seven new interactions among the twelve tested TFs ( as seen for Bagpipe , Empty spiracles , Eyes absent , Knot , Pangolin , Seven up and Suppressor-of-Hairless ) . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 03410 . 7554/eLife . 06034 . 035Figure 8—figure supplement 5 . Drosophila AbdA , mouse HoxB8 and Nematostella HoxE interact with several common TFs in vivo . Colour code and interaction representation are as in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 035 We first addressed whether the HoxB8 and HoxE could interact with the 35 Drosophila TFs , as previously done with Drosophila Hox proteins ( see also ‘Materials and methods’ ) . Results show that the mouse and cnidarian Hox proteins interact with a surprisingly large number of TFs , respectively 27 and 22 ( Figure 8B–C and Figure 8—figure supplements 1 , 2 ) . Still , each Hox protein interacts with a different set of TFs , underlining the existence of preferential/specific interaction properties when considering the whole interactome . To directly assess whether the HX could act as an inhibitory motif , we analysed the interaction properties of HX-mutated HoxB8 and HoxE proteins , focusing on TFs that were negative with the corresponding wild type Hox proteins . We observed that more than half of the tested TFs became positive with HX-mutated proteins in both cases ( Figure 8—figure supplements 3 , 4 ) . Thus , the HX is also an inhibitory interaction motif in HoxB8 and HoxE . To further explore how the HX motif could inhibit PPIs , we considered the HoxE protein , which has a simple organisation in terms of secondary structures . Basically , this protein contains a long intrinsically disordered N-terminal region followed by the ordered HD ( Figure 9A ) . Still , this protein establishes a number of common interactions with HoxB8 and AbdA ( Figure 8—figure supplement 5 ) . Since the HD is unlikely to be sufficient for several of those interactions ( as deduced from AbdA: Figure 3C ) , we decided to test the N-terminal region of HoxE . We thus generated fly lines carrying a short HoxE variant , called Nter-HoxE , which corresponds to the residues 1 to 54 and does not contain the HX motif ( see ‘Materials and methods’ ) . BiFC with the 35 Drosophila TFs shows that only five interactions are lost when using this short variant of HoxE ( Figure 9B and Figure 9—figure supplement 1 ) . In contrast , this fragment interacts more strongly or establishes new interactions with seven TFs while 15 other interactions remained unaffected . In total , Nter-HoxE establishes as many interactions as the full length HoxE . 10 . 7554/eLife . 06034 . 036Figure 9 . The intrinsically disordered region of HoxE establishes a number of interactions with Drosophila TFs . ( A ) Scheme of full length Nematostella HoxE with its predicted SLiMs ( green bars ) , and disordered ( blue waves ) or ordered ( brown blocks ) regions . Adapted from iupred ( http://iupred . enzim . hu/ ) . The N-terminal disordered region used for BiFC is indicated ( Nter-HoxE ) . ( B ) Interactome between Nter-HoxE and the 35 Drosophila TFs . Colour code and representation are as in Figure 3 . See also Figure 9—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 03610 . 7554/eLife . 06034 . 037Figure 9—figure supplement 1 . BiFC between Drosophila TFs and the N-terminal ( 1-54 ) fragment of the Nematostella HoxE protein . BiFC was performed with the 35 TFs in the epidermis of stage 10–12 embryos , as indicated . Quantification was performed as in Figure 3—figure supplement 1 . No statistical quantification is provided in the case of a new interaction ( as seen for Apontic , Bagpipe , Knot , Mef2 and Zfh1 ) . No picture was taken in absence of BiFC . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 037 Together these results show that the long disordered region of HoxE is involved in a number of the heterologous interactions observed with Drosophila TFs . The observation of seven ectopic interactions also underlines that the interaction potential of the disordered region is tightly controlled in the context of the full-length protein . We propose that SLiMs such as the HX motif are important mediators of this control .
Apart from the PBC class , very little is known about the TFs that could help Hox proteins to elicit specific developmental programs in the embryo . As a consequence , the interactome underlying Hox embryonic functions remains largely elusive . This lack of knowledge is explained by the difficulty of identifying the transcriptional partners that could participate in each context-specific activity of Hox proteins . We hypothesize that many of those interactions are too weak and/or too dynamic to be efficiently trapped through classic high throughput approaches . In support of this , a yeast two-hybrid screen with the Drosophila Ubx protein led to the characterization of less than 15 TFs as interacting partners ( Bondos et al . , 2006 ) , strongly contrasting with the numerous functions ensured by this Hox protein during embryogenesis . A similar approach with the mouse HoxA1 protein also led to the characterization of less than 20 TFs ( Lambert et al . , 2012 ) . Our approach relied on the sequential analysis of tandem interactions between Hox proteins and individual TFs . The interaction screen was performed in two complementary steps with a set of TFs covering different DNA-binding families and displaying various expression profiles during Drosophila embryogenesis . This set of TFs is expected to be representative of the diverse transcriptional regulatory activities of Hox proteins in vivo . The competition experiment was performed in the epidermis , even for TFs that are not endogenously expressed in this tissue . Despite this limitation , we found a high proportion ( 33/80 ) of positive competitive events , which confirmed the sensitivity of BiFC for this type of approach . A similar strategy was reported in cell culture for identifying drug molecules that could affect the assembly or the localisation of a specific protein complex ( Morell et al . , 2008 ) . Thus , competitive BiFC could certainly be applied more generally in the future for selecting any kind of new interacting molecules upon the screening of subtle variations in fluorescent reporter signals . BiFC then showed that all but three competitive TFs could interact with AbdA , making a total of 31 TFs as new Hox interacting partners . In comparison , only seven TFs were so far described to interact with AbdA ( Merabet and Dard , 2014 ) . These results were reproduced in different tissues of the embryo or corroborated by co-ip experiments in S2 cells , illustrating that Hox-TF interactions could occur in different cell contexts upon co-expression . Importantly , the specificity of each interaction was supported by the loss of fluorescent signals when using mutated or truncated Hox variants , and by the observation of typical nuclear interaction profiles with different TFs . It is important to stress that all these results revealed an interaction potential between Hox proteins and TFs . Whether and how these interactions could be used in the context of the endogenous gene products is an open question . For example , some of the interactions revealed with the fusion Scr protein are unlikely to occur with the endogenous Scr product , which displays a quite restrained expression profile in the embryo compared to the other tested Hox proteins ( Supplementary file 3 and [Hammonds et al . , 2013] ) . In addition , the fact that TF-encoding genes were not expressed under the control of their endogenous promoter forbids assessing the role of tissue-specific expression levels in Hox interactome properties . In this context , the recent advent of genetic tools in Drosophila , including Mimic elements ( Gnerer et al . , 2015 ) and the CRISPR/Cas9 system ( Bassett and Liu , 2014 ) , could certainly add to the functional relevance of BiFC observations in the future . Nevertheless , the high proportion of positive events among a starting set of 80 TFs strongly suggests that AbdA , and Hox proteins in general , have a strong potential to interact with a number of different TFs in vivo . This assumption could be verified by using a library of 600 TFs compatible for BiFC in Drosophila ( Bischof et al . , 2013 ) . Hox proteins play numerous functions in all embryonic germ layers . These functions can be highly specific ( Brodu et al . , 2002; Li-kroeger et al . , 2008 ) or common to several ( Gebelein et al . , 2002; Coiffier et al . , 2008 ) Hox proteins , suggesting they could rely on the interaction with different types of cofactors . Here , we present the first interactomes of five Hox proteins with a set of 35 TFs . BiFC and co-ip experiments revealed that all tested TFs could interact with two or more Hox proteins . Although this result should be expanded to more TFs , it suggests that the interaction between Hox proteins and TFs is generally not exclusive . As a corollary , we hypothesize that Hox interactome specificity is unlikely to rely solely on the interaction with specific TFs . Despite a number of common interactions , each Hox interactome contains a different set of positive TFs . AbdA is the Hox protein establishing the highest number of interactions , which is consistent with the fact that it served as a bait protein in the starting competition screen . However , the observation that Hox proteins do not interact systematically with the same set of cofactors shows their specificity . Interestingly , this specificity is not only occurring at the DNA-binding level since the loss of AbdA DNA-binding activity did not affect all interactions ( 18 interactions of 31 were affected ) . Our results thus emphasize the need of considering post-DNA-binding mechanisms for understanding Hox functional specificity in vivo . The strong interaction potential of Hox proteins is consistent with their wide spectrum of regulatory activities in the embryo . It probably constitutes an inherent feature of other classes of developmental TFs that intervene in several regulatory processes throughout embryogenesis . Along this line , observations from genome-wide analyses showed that target cis-regulatory sequences allow the assembly of large multi-protein complexes ( Moorman et al . , 2006; Blaxter , 2010; Kvon et al . , 2012 ) . In addition , TFs can generally bind thousands of sites across the genome ( Li et al . , 2008 ) . Thus , the high number of possibilities in protein–protein and protein-DNA contacts likely reflects the propensity of developmental TFs to regulate target gene expression in many different cell contexts . Finally , clustering analyses showed that the similarity between the different Hox interactomes does not follow the level of sequence identity between Hox proteins . Thus , common interactions might rely on different Hox protein interfaces . Accordingly , we found that the HD , which contains the highest score of sequence identity between Hox proteins , was not sufficient to ensure the majority of interactions established by AbdA . In addition , the common HX motif is not similarly used in the different Hox interactomes . Interaction properties of the HX motif were also different depending on the tissue considered , highlighting the strong flexibility and adaptability of this motif to the surrounding protein environment . Restricting the analysis to the closely related Ubx and AbdA proteins did not reveal a higher level of similarity in the interaction properties of the HX motif . In contrast , the UbdA motif showed a more frequent similar usage mode between the two Hox proteins . Thus , SLiMs conserved at different evolutionary extents provide different levels of specificity to Hox interaction properties . A high interaction potential was not only observed with Drosophila , but also with mouse and cnidarian Hox proteins . This result is particularly striking with the cnidarian HoxE protein , which is capable of interacting with TFs that are specific of the Bilateria group , including Biniou , Midline , Pointed or Teashirt . Although these observations are not functionally informative , they indicate that the strong interaction potential of Hox proteins is an ancestral feature that was probably present before its full exploitation in bilaterian lineages . In addition , the observation that highly divergent Hox proteins could interact with the same set of cofactors questions the role of conserved and non-conserved regions in Hox functions . Interestingly , long intrinsic disordered regions characterize Hox proteins in general ( Merabet and Dard , 2014 ) , and a recent study showed that they could serve in Ubx to bind different partners in a competitive or cooperative way ( Hsiao et al . , 2014 ) . Results obtained with the cnidarian HoxE protein confirm the important role of a long disordered region in mediating interactions with different TFs . We suggest that the acquisition of long intrinsic disordered regions was a key for providing functional diversity to Hox proteins during animal evolution . Our work provides an original experimental strategy for analysing the role of SLiMs in the context of full-length proteins in vivo . Results show that the HX mutation affects a number of interactions in all tested Hox proteins . Surprisingly , the absence of the HX motif could lead to a stronger or new interaction potential with TFs . A gain of interaction was observed with Drosophila , mouse , and cnidarian Hox proteins , suggesting that this molecular property is evolutionary conserved in the animal kingdom . In total , the HX motif appears more often involved in limiting rather than in promoting interactions with TFs . The inhibitory effect of the HX motif on PPIs was most pronounced in Ubx and AbdA , which were also used for the analysis of the UbdA motif . We found that this motif was also required for limiting the interaction potential of the two Hox proteins , although to a lesser extent than the HX motif . Thus , the negative regulation of PPIs is not a specific property of the HX motif . Inhibitory activity of SLiMs on Hox protein function can be reconsidered in the light of previous functional data . For example , the HX mutation was shown to convert AbdA from a repressor to a strong activator of the decapentaplegic ( dpp ) target enhancer ( Merabet et al . , 2011 ) . This striking transcriptional conversion is difficult to assign to a simple loss of interaction with co-repressor ( s ) . Along the same line , the HX mutation increases the interaction potential of Ubx with Exd in vivo ( Hudry et al . , 2012 ) , and confers an AbdA-like activity to Ubx for segment specification in the epidermis ( Galant et al . , 2002 ) . Similarly , the mouse HoxB8 protein was shown to provoke dominant negative phenotypes in absence of its HX motif ( Medina-martinez & Ramı , 2003 ) , which is also difficult to reconcile with a simple loss of interactions . Finally , the interaction between HoxA11 and Foxo1a in placental mammal lineages is reported to result from the loss of a specific Foxo1a–inhibitory interaction domain and not from the gain of a new binding interface in HoxA11 ( Brayer et al . , 2011 ) . The gain of interaction between HoxA11 and Foxo1a comprises one of the major regulatory events that led to placentation in mammals , illustrating a so far unexpected role of a PPI inhibitory domain in interactome rewiring during evolution ( Lynch et al . , 2008 ) . More generally , protein autoinhibition is described for the regulation of other molecular events , including protein-DNA interactions and actin polymerisation ( Pufall , 2002; Lee et al . , 2005; Padrick and Rosen , 2010 ) . It relies on the presence of inhibitory modules that were recently found to be enriched in intrinsically disordered regions ( Trudeau et al . , 2013 ) . SLiMs can also be categorized as hiding motifs ( Van Roey et al . , 2013 ) , but this role has so far been described for few mammalian proteins in the context of their intracellular transport or post-translational modification . Here , we demonstrate that two different SLiMs could be used in a cell type-dependent manner for promoting or limiting the Hox interaction potential with TFs . We propose different models for explaining the underlying molecular mechanisms . In the classic situation , SLiMs are positively used as context-specific interaction modules , together with globular domains , for the recruitment of cell-specific cofactors ( cell context 1 in Figure 10 ) . Alternatively , SLiMs could also be important for restraining the interaction potential ( cell contexts 2 and 3 in Figure 10 ) . In one mechanism , the inhibitory activity would rely on the interaction with a particular partner that will mask or forbid the recruitment of the other SLiM-interacting cofactors ( cell context 2 in Figure 10 ) . This mechanism implies that the interaction between the SLiM and the hiding partner will be strong and stable enough to overcome the binding of the other cofactors . In a second mechanism , SLiMs could also directly act as a masking peptide , preventing recognition and/or binding of undesired cofactors ( cell context 3 in Figure 10 ) . SLiMs are classically required to make interactions with structured globular domains in trans ( Stein and Aloy , 2008 ) , but intramolecular contacts following post-translational modifications have also been reported in a few cases ( Pawson et al . , 2001 ) . Thus , SLiMs could also establish interactions in cis , potentially upon cell-specific modifications ( i . e . , phosphorylations ) , to eventually inhibit the recruitment of inappropriate cofactors . A similar role was previously described for the HX motif of Labial , which is able to prevent the binding of the HD on the DNA ( Chan et al . , 1996 ) . Interestingly , this inhibition is relieved upon interaction with Exd , highlighting the influence of the environment on the SLiM inhibitory activity . 10 . 7554/eLife . 06034 . 038Figure 10 . Molecular mechanisms underlying context-dependent activities of SLiMs in protein–protein interactions . The Hox protein is represented as containing a globular structured domain together with two different SLiMs embedded in a disordered region , as indicated . This protein will present different interaction properties with a set of cofactors that could vary depending on the cell context considered . Preferential interactions between cofactors and the protein domain and SLiMs are represented by a colour code . Black bars symbolize the various levels of interaction affinity . In the cell context 1 , cofactors are recruited through specific interactions with the globular domain and the two SLiMs . In the cell context 2 , there is a supplementary triangular cofactor that displays higher affinity with the red SLiM than the square cofactor . As a consequence , interaction will occur with this triangular ( hiding ) cofactor , which forbids the interaction with the other SLiM . In this context , the red SLiM behaves as an inhibitory interaction motif . In the cell context 3 , post-translational modifications in the disordered region ( yellow stars ) allow the inhibitory SLiM to establish interactions in cis with the globular domain . These intra-molecular contacts forbid the binding of the other cofactors . The last two mechanisms illustrate how the inhibitory activity of SLiMs could help in distinguishing/specifying interactomes with an identical set of cofactors . DOI: http://dx . doi . org/10 . 7554/eLife . 06034 . 038 In summary , we showed that highly conserved SLiMs are used in a context-dependent manner for constraining the interaction potential of Hox proteins with surrounding TFs . This molecular strategy has certainly been underestimated to date . We propose that the inhibiting interaction properties of SLiMs could apply more generally to the fine-tuning of highly connected interactomes . It is interesting to note that SLiMs can also be produced as individual molecules from short open reading frames ( Kondo et al . , 2007; Magny et al . , 2013 ) or even from long non coding RNAs ( Ruiz-Orera et al . , 2014 ) . We anticipate that short peptides could act as buffering molecules , helping hub proteins to discriminate their correct partners among hundreds of possible interactions within the ‘messy’ ( Tawfik , 2010 ) cell environment .
Several VC-Hox fusion constructs were previously generated ( Hudry et al . , 2012; Boube et al . , 2014 ) : these correspond to wild type and mutated/truncated variants of all Drosophila Hox proteins and wild type and HX-mutated HoxE . Other Hox fusion constructs ( HoxB8 , HoxB8HX and Nter-HoxE ) were generated by PCR and restriction-cloned in the pUAST or pUASTattB vector , in fusion with the C-terminal ( 155–238 ) fragment of Venus ( VC ) at 3′ end ( see Supplementary file 4 ) . Fusion TFs were also generated by PCR from full-length complementary DNAs and restriction-cloned in fusion in pUAST or pUASTattB vector with the N-terminal ( 1–173 ) fragment of Venus ( VN ) at the 5′ or 3′ end ( see Supplementary file 4 ) . Primers used for cloning of each TF are listed in Supplementary file 5 . For all fusion constructs , a linker of three to five amino acids was added to separate the Venus fragment from the protein . All constructs were sequence-verified before injection . Transgenic lines were established either by the PhiC31 integrase system ( with the pUASTattB vector [Venken et al . , 2006; Bischof et al . , 2007] ) or by classic P-element ( with PUAST vector ) mediated germ line transformation . Expression level of Hox fusion constructs was verified as previously described ( Hudry et al . , 2011 ) . Briefly , flies were crossed at different temperatures with the abdA-Gal4 driver and embryos were collected for immunostaining with a chicken anti-GFP antibody ( Abcam ab13970 , England ) recognizing the VN and VC fragments . Fluorescent revelation ( with a secondary anti-chicken antibody coupled to AlexaFluor488 , Abcam150169 ) was used to compare the expression level between the different conditions with wild type and mutated Hox fusion constructs . The temperature for each fly cross was adjusted accordingly , allowing comparing BiFC signals with Hox fusion proteins expressed at similar levels . The same anti-GFP antibody was used to verify the correct expression level of each generated VN-TF fly line . Gal4 drivers used are: Antp-Gal4 , Ubx-Gal4 , abdA-Gal4 , and AbdB-Gal4 ( de Navas et al . , 2006; Hudry et al . , 2012 ) . Fly lines generating in this work are listed in Supplementary file 4 . Competition tests were performed in one generation by crossing each candidate UAS-TF ( see Supplementary file 1 for the type of the UAS fly line ) with the BiFC reporter fly line containing the UAS-VC-abdA ( homozygous on the second chromosome ) and UAS-VN-exd ( homozygous on the fourth chromosome ) constructs , together with the abdA-Gal4 driver balanced over a TM6tubulineGal80 third balancer ( Hudry et al . , 2011 ) . Under these conditions , half of the embryo progeny ( with homozygous UAS-TF fly line ) could display affected BiFC signals in presence of a competitive TF . Complementation tests were performed by crossing en mass virgin females containing the VN-TF and VC-Hox constructs ( as non-established fly lines resulting from a previous cross between VN-TF and VC-Hox individuals ) with males carrying the corresponding Hox-Gal4 driver . Over night egg laying was performed at different temperatures , according to the expression level of the VC-Hox variant . Embryo collection , preparation , and immunodetections were performed according to standard procedures . The antibodies used were: chicken anti-GFP ( Abcam ab13970 , 1/500 ) , mouse anti-Scr ( 6H4 . 1 , 1/100 ) , mouse anti-Antp ( 4C3 , 1/100 ) , mouse anti-Ubx ( FP3 . 38 , 1/100 ) , rabbit anti-AbdA ( Dm . Abd-A . 1 , 1/100 ) , mouse anti-AbdB Co-ips were performed in S2 cells , which were transfected with a HA-tagged form of AbdA , together with an actin-Gal4 plasmid and the corresponding VN-cofactor . AbdA-HA construct was generated by PCR , using an oligonucleotide bearing the HA sequence and cloned into the pUASTattB vector . The construct was sequence verified . Transfection was realised using the X-tremeGENE HP DNA Transfection Reagent ( Roche ) . Cells were lysed 48 hr later and nuclear extracts were prepared as classically described . Ip was performed with a polyclonal rabbit anti-HA antibody ( Abcam ab9110 ) . Presence of AbdA-HA was verified by western blot using a monoclonal anti-HA antibody ( HA . 11 from Covance ) . Presence of the associated VN-cofactor was revealed with a chicken anti-GFP antibody recognising the VN fragment ( Abcam ab13970 ) . Experimental parameters allowing a comparable expression level between wild type and mutated VC-Hox proteins were previously established for several constructs ( Hudry et al . , 2012; Boube et al . , 2014 ) or deduced from additional immunostaining experiments for new constructs ( HoxB8 , HoxB8HX , and Nter-HoxE ) . Fly crosses for BiFC analyses were set up at the defined temperature over night . After the removal of the flies , the embryos were kept at 4°C for 24 hr before live imaging . Live embryos were dechorionated and mounted in the halocarbon oil 10S ( commercialized by VWR , Pennsylvania , USA ) . Quantification of the BiFC signals was realised by taking unsaturated images at the same desired stage , depending on the tissue considered ( epidermis: stage 10 , mesoderm: stage 12 , nervous system: stage 14 ) . For BiFC analysis in the CNS , embryos were manually aligned on the dorsal face . Observations were performed at least twice ( from two different over night egg laying periods ) for a same genotype . A minimum of 5 embryos of the correct developmental stage was considered in each case . Pictures were acquired using a LSM780 confocal microscope ( Zeiss , Jena , Germany ) . For Venus fluorescence , filters were adjusted at 500 nm for excitation and 535 nm for emission . Identical parameters of acquisition were applied between the different genotypes . The number and intensity of the all pixels ( for each embryo ) were measured in the tissue of interest using the histogram function of the ImageJ Software . The quantification of fluorescence complementation is shown for each condition by boxplot representation using R-Software . Boxplot depicts: the smallest value , lower quartile , median ( black line ) , upper quartile , and largest value for each condition . Networks were represented using Cytoscape 3 . 0 ( Shannon et al . , 2003 ) . A hierarchical clustering algorithm ( with Euclidian distance and average linking ) was applied to the matrix using the MeV software suite ( Saeed et al . , 2006 ) . The bootstrap method was used for resampling the data and provides a statistical support for each tree node . | In all animals , it is important that cells are correctly organised into tissues and organs . This organisation starts in the embryo , and cells are instructed to perform different roles depending on their position within the body . A family of proteins called the Hox proteins coordinates the organisation of the cells in the animal embryo by binding to and controlling the expression of specific genes . To properly control their target genes , Hox proteins need to interact with other proteins called transcription factors that can also bind to the genes . However , only a few of these transcription factors have been identified so far , and it is not clear how Hox proteins are able to interact with them . Here , Baëza , Viala , Heim et al . identified several more transcription factors that can bind to the Hox proteins in fruit fly embryos . The experiments show that Hox proteins are able to bind to many transcription factors that are very different from each other . Baëza , Viala , Heim et al . also show that two short sections within the Hox proteins known as short linear motifs are important for controlling these interactions . A fly Hox protein that was missing these motifs was able to interact with new transcription factors . This inhibitory role was found in Hox proteins from mice and sea anemones , suggesting that these motifs may play the same role in all animals . Baëza , Viala , Heim et al . 's findings challenge the traditional view of the role of the short linear motifs in interactions between proteins . Also , the findings provide an alternative explanation for how the Hox proteins are only able to interact with particular transcription factors in animal embryos . The next step will be to find out whether the inhibitory role of short linear motifs could more generally apply to many other protein families . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2015 | Inhibitory activities of short linear motifs underlie Hox interactome specificity in vivo |
Animals collect sensory information from the world and make adaptive choices about how to respond to it . Here , we reveal a network motif in the brain for one of the most fundamental behavioral choices made by bilaterally symmetric animals: whether to respond to a sensory stimulus by moving to the left or to the right . We define network connectivity in the hindbrain important for the lateralized escape behavior of zebrafish and then test the role of neurons by using laser ablations and behavioral studies . Key inhibitory neurons in the circuit lie in a column of morphologically similar cells that is one of a series of such columns that form a developmental and functional ground plan for building hindbrain networks . Repetition within the columns of the network motif we defined may therefore lie at the foundation of other lateralized behavioral choices .
All animals have to collect sensory information from the environment and use it to produce adaptive behavioral choices . While theory , modeling , and experimentation has led to substantive ideas about circuit wiring that might effectively implement behavioral choice by vertebrate brains ( van Ravenzwaaij et al . , 2012; Shadlen and Newsome , 2001; Drugowitsch et al . , 2012; Doya and Shadlen , 2012; Bogacz et al . , 2006; Faumont et al . , 2012; Gaudry et al . , 2013; Gaudry and Kristan , 2009; Shaw and Kristan , 1997; Song et al . , 2015; Svoboda and Fetcho , 1996 ) , defining the cellular and synaptic organization of neurons involved in such choices has been more challenging . The solution to this problem might be more accessible in an evolutionarily old , conserved region of the brain , the hindbrain , which mediates what are some of the most fundamental behavioral choices made by vertebrate nervous systems: moving adaptively to avoid predators or to capture prey . The hindbrain is not only at the core of many sensory/motor responses , but also has a strikingly conserved ground plan of neuronal classes arranged early in life in columns ordered by structure and function ( Gray , 2013; Kinkhabwala et al . , 2011; Koyama et al . , 2011 ) . We set out to reveal a circuit motif for a simple , primitive behavioral choice mediated by hindbrain neurons within this vertebrate ground plan . One of the most fundamental behavioral choices of all bilateral animals is whether to respond to the left or to the right by movements of the body , head , limbs and eyes . We studied a relatively simple , primitive , but critical lateralized behavior: the escape response of larval zebrafish , which involves a rapid turn away from a potential threat ( Faber et al . , 1989 ) . This escape , like that in most aquatic vertebrates , begins with the firing of a single action potential in one of a bilateral pair of hindbrain neurons , called Mauthner cells , whose axons cross in the brain and extend into the spinal cord . Activation of a Mauthner cell on one side initiates a turn away from an attacking predator that is the shortest latency and fastest motor response in vertebrates . Whether an escape is to the left versus the right is determined by which of the two cells fire , leading others to propose that this could be a simple model of decision making ( Korn and Faber , 2005 ) . A correct behavioral choice here is vital- one wrong turn in a lifetime can lead to death . Earlier work showed that auditory inputs entering one side of the brain excite both the Mauthner cell and inhibitory neurons that were thought to be important for the laterality of the escape response ( Koyama et al . , 2011; Faber et al . , 1989 , 1978; Zottoli and Faber , 1980; Takahashi et al . , 2002 ) ; see diagram in Figure 1A . The inhibitory neurons lie at the bottom of a column of morphologically similar neurons within the hindbrain ( Figure 1B1–4 ) that is one of a series of columns forming the ground plan from which the many circuits in vertebrate hindbrains arise ( Kinkhabwala et al . , 2011; Koyama et al . , 2011 ) . The repetitive structure of the hindbrain suggested that if we could determine the wiring and test the functional roles of neurons important for mediating the left/right choice in the Mauthner network , we would not only reveal a network motif with a role in a behavioral choice , but also one that is likely repeated within the columnar patterning to mediate other lateralized sensory motor responses of vertebrates . 10 . 7554/eLife . 16808 . 003Figure 1 . Connectivity of inhibitory neurons implicated in the laterality of the escape response . ( A ) The blue region marked on the drawing of the fish contains the two Mauthner cells and neuronal candidates for their control , which may determine whether the animal escapes to the left or right . Right side shows some known contacts in this network , with sensory inputs from the eighth nerve exciting the ipsilateral M-cell as well as inhibitory interneurons that inhibit both M-cells . ( B1 ) A cross section through the hindbrain in the region of the Mauthner cell from a transgenic line with glycinergic neurons labeled with GFP ( green ) . The two Mauthner cells ( blue ) and a filled feedforward ( FF ) inhibitory interneuron ( red ) were labeled via patch pipettes . The FF neuron lies at the bottom of the most lateral of three columns ( arrowheads ) of glycinergic neurons , near the lateral dendrite of the M-cell . ( B2 ) A horizontal view of the region shows that red processes of the FF neuron are located in the vicinity of both the ipsilateral and contralateral M-cells . Optical sections through the region of the contralateral ( B3 ) and ipsilateral M-cells ( B4 ) show swellings ( arrowheads ) of the processes from the inhibitory FF cell adjacent to both M-cells . ( B5 ) A plot of the number of boutons adjacent to ipsilateral versus contralateral M-cells from 14 fish ( 4 days old ) in which individual FF cells and both M-cells were labeled . In every case , the number of boutons apposed to the contralateral cell exceeded the number apposed to the ipsilateral one , with a highly significant difference between the two sides ( p<0 . 0001 ) . ( C1 ) Triple patch recordings from an FF neuron and the two M-cells show that firing the FF cell ( asterisk ) by current injection produces larger IPSPs in the contralateral M-cell than the ipsilateral one ( arrowheads ) . IPSPs from four sweeps are shown at potentials above and below resting potential by the injection of plus or minus 1 nA of current into the M-cells . This electrophysiology is from the neurons whose morphology is shown in B . ( C2 ) A plot of the amplitude of the IPSPs ( for negative 1 nA injection ) in the ipsilateral versus contralateral M-cell from 19 triple patch experiments like the one in C1 . Seven of these were from 6 days old nacre fish and 12 from 4 days old relaxed fish . In 18 of 19 experiments , the IPSPs were larger in the contralateral M-cell , with a highly significant difference between the two sides ( p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 003
We first determined the pattern and strength of the connections of the inhibitory neurons thought to play a role in the laterality of the escape because they are driven by auditory input and connect to the Mauthner cells . To reveal how they distributed their inhibition to the two Mauthner cells , we performed 19 triple patch recordings from individual feedforward inhibitory neurons and the two Mauthner cells , followed by high-quality dye filling of the three cells in 10 of the experiments . The neurons were targeted based on their location at the bottom of a lateral column of glycinergic neurons and the knowledge that at least some of the population straddled the lateral dendrite of the Mauthner cell ( Kinkhabwala et al . , 2011; Koyama et al . , 2011 ) . They could thus be targeted in a line with GFP labeled glycinergic neurons by recording from GFP positive neurons located in that column of glycinergic cells and lying adjacent to the M-cell lateral dendrite , which was visible in DIC optics . Individual inhibitory neurons had commissural axons that branched on both sides of the body and had putative synaptic boutons apposed to both Mauthner cells as in Figure 1B1–4 . In 14 cases where we filled the three neurons ( 10 with prior physiology and 4 only labeled anatomically to minimize cellular damage ) and reconstructed them in 3D from confocal optical sections , every feedforward neuron had more boutons contacting the contralateral M-cell than the ipsilateral one , with a highly significant difference between the two sides ( Figure 1B5; p<0 . 0001 ) . This clear morphological asymmetry was associated with a functional one revealed by firing the inhibitory neurons while recording the postsynaptic responses in both M-cells . Figure 1C1 shows an example of one experiment in which the feedforward inhibitory neuron produced a larger IPSP in the contralateral M-cell than the ipsilateral one ( morphology of this set is in Figure 1B1–4 ) . This experiment was repeated at two different ages ( 4 and 6 days ) in two lines of fish paralyzed either by bungarotoxin ( in nacre line with reduced pigmentation ) , or by a mutation blocking calcium release in muscle ( relaxed line; see Materials and Methods for details ) . The main result was the same independent of age , genetic line , and approach to paralysis . In 18 of 19 triple patch experiments , the strength of the contralateral IPSP was greater than the ipsilateral one and the overall difference was significant ( mean contra/ipsi ratio was 2 . 44 , range 0 . 97–6 . 61; p<0 . 001 ) . We conclude that individual inhibitory neurons are driven by ipsilateral sensory input and inhibit both M-cells , but at different strengths . The weaker inhibition of the ipsilateral M-cell , along with its direct excitation by excitatory sensory afferents might be expected to make that M-cell more likely to fire to an ipsilateral stimulus than the contralateral M-cell , resulting in an escape bend away from the stimulus source . A sensory stimulus , however , will often activate sensory inputs on both sides of the body to differing extents , so the typical natural situation would be one in which the left and right populations of inhibitory neurons compete at the level of the two M-cells to influence which reaches threshold first . The problem of escaping a threat requires turning away from the side that receives the strongest sensory stimulus ( typically the side of the attack ) over a broad range of stimulus strengths above the minimum that signals a potential predatory attack . If the inhibitory neurons only competed at the level of the M-cells , asymmetric , but very large stimuli on the two sides might lead to massive inhibition of both M-cells , possibly delaying or blocking an escape . The intuition that strong bilateral inputs might pose a problem with the known connectivity was examined more formally in a model incorporating our data from the inhibitory neurons ( Figure 2 A1–3 and Figure 2—figure supplement 1: Table 1 ) . Here we focus on the versions of the model most closely tied to the experimental evidence , although all of the variations tested and their implications are presented in Figure 2—figure supplement 1 . We initially modeled a circuit containing inhibitory neurons driven by sensory inputs , but with connections only to the ipsilateral M-cell . As expected , because the two M-cells are controlled independently , this network led to non-adaptive bilateral M-cell responses even when there were large differences in the input strengths on opposite sides of the body ( Figure 2B1 ) . Adding commissural inhibitory connections to the contralateral M-cell that were stronger than those to the ipsilateral one , in the proportions revealed by our data , allowed for some unilateral M-cell responses to strongly asymmetric bilateral inputs ( Figure 2B2 ) . The model performance was still flawed , however , because the strong crossed inhibition of the contralateral M-cell led to a broad range of strengths of bilateral sensory input over which neither M-cell fired , in line with our prior intuition that the strong commissural inhibition might not facilitate rapid , adaptive responses away from the side receiving the strongest input when inputs to both sides are substantial ( Figure 2B2 ) . 10 . 7554/eLife . 16808 . 004Figure 2 . Output of a computational model of the studied neurons in the Mauthner network . ( 2A1 ) Connectivity implemented in the model . Gray arrows indicate mixed electrical and excitatory glutamatergic connections . White arrows indicate pure electrical connections . Black arrows indicate inhibitory glycinergic connections . ( 2A2 ) Mauthner response to eighth nerve stimulation . The eighth nerve on one side was stimulated ( left panel ) and the EPSP in the lateral dendrite of Mauthner cell was recorded experimentally ( middle panel ) . The EPSP in the model was simulated by synchronous activation of eighth ganglion neurons ( right panel ) . ( 2A3 ) Inhibitory input from an FF neuron to a Mauthner cell . A single actual FF neuron is activated by current injection to fire a single spike while monitoring IPSPs in the ipsilateral Mauthner cell ( left panel ) . Four traces are shown at potentials above and below reversal potential by injection of plus or minus 1 nA of current into the M-cells ( middle panel ) . The modeled IPSP in the ipsilateral Mauthner cell during injection of plus or minus 1 nA of current ( right panel ) . ( 2B1 ) Output of the computational model of Mauthner circuit with only ipsilateral feedforward inhibitory connections ( upper panel ) . Various combinations of left ( x-axis ) and right stimuli ( y-axis ) are presented to the model circuit and each stimulus condition is color-coded based on the activation of Mauthner cells ( blue: only right Mauthner cell fires , red: only left Mauthner cell fires , white: both Mauthner cells fire ) ( middle panel ) and also based on the latency of the activation ( lower panel ) . ( 2B2 ) Output as in B1 , but with the addition of asymmetric bilateral feedforward inhibitory connections ( upper panel ) . Middle and lower panels were formatted as in B1 . ( 2B3 ) Output as in B2 , but with the addition of putative reciprocal inhibitory connections between the FF neurons ( upper panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 00410 . 7554/eLife . 16808 . 005Figure 2—figure supplement 1 . Outcomes of all of the circuit configurations modeled . The layout is the same as in Figure 2 , with panels A , C and E duplicated from Figure 2 to allow comparison with other variants of the circuit . ( B ) A circuit like that in C with ipsilateral and contralateral inhibition of the M-cells , but with symmetrical crossed inhibition of the M-cells , instead of the biased crossed inhibition in C that we actually observed experimentally . The symmetrical inhibition of the M-cells in network B leads to a smaller , though still substantial , zone of no escape responses than a similar model with biased inhibition ( C ) . ( D–E ) Models like those in B and C , but including mutual inhibition between the inhibitory interneurons on opposite sides . Adding mutual inhibition between the inhibitory neurons in each of these recovers escape responses to strong bilateral inputs . This , however , leads to a seemingly more adaptive outcome when the M-cell inhibition is contralaterally biased ( E ) , as this narrows the zone in which both M-cells are simultaneously activated . This configuration matches what we observed experimentally . F and G show model performance when there is only contralateral M-cell inhibition at symmetrical weak or strong levels ( the levels are those used for the weak or strong biased connections in other panels , which are based on experiment ) . Both perform poorly , with large zones of simultaneous or no responses . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 00510 . 7554/eLife . 16808 . 006Table 1 . Experimentally measured properties and settings used in the Modeling . Top: Experimentally derived basic properties of Mauthner and Feedforward glycinergic neurons at 4 dpf . ( Mauthner: n=24 , FF: n=28 ) . Rm , input resistance; Erest , resting membrane potential; ECl , reversal potential of IPSP; Espike , spiking threshold; Tau , membrane time constant . Corrected for liquid junction potential . Bottom: Parameters used for conductance-based models of the Mauthner circuit . Rm , input resistance; Erest , resting membrane potential; ECl , reversal potential of IPSP; Espike , spiking threshold; Tau , time constant; El , leak reversal potential; EK , reversal potential of potassium; ENa , reversal potential of sodium; gl , leak conductance per area; gNa , sodium conductance per area; gK , potassium conductance per area . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 0061 . Measured parametersMauthnerFFRm [Mohm]10 . 3 ± 0 . 9411 . 7 ± 41 . 16Erest [mV]−78 . 5 ± 0 . 9−77 . 3 ± 0 . 7ECl [mV]−75 . 4 ± 0 . 94 NAEspike [mV]−61 . 3 ± 1 . 1−60 . 9 ± 1 . 0Tau [ms]22 . 6 ± 7 . 49 . 9 ± 1 . 22 . Model parameters LIF model parameters Mauthner FFRm [Mohm]10400Erest [mV]−79−77Ecl [mV]−75−75Espike [mV]−61−61Tau [ms]2310No of Cells/side130HH model parameters Auditory El [mV]−79EK [mV]−90ENa [mV]50gl [msiemns/cm2]0 . 05gNa [msiemens/cm2]100gK[msiemens/cm2]200No of Cells/side30Surface area [um2]2000–20000Connectivity Parameters Synapse type Conductance [nS] Delay [ms] Connectivity pattern FF -> ipsilateral MGlycine 250 . 3all to oneFF-> contralteral MGlycine2 . 5 × FF->ipsiM0 . 3all to oneAuditory-> MGap15NAall to oneAuditory-> MGlutamate120 . 7all to oneAuditory->FFGap10NAone to oneFF -> contra FFGlycine120 . 3all to allSynapse time constants Glutamate Glycine Tau [ms]22 One solution to this problem is for the inhibitory neurons to reciprocally inhibit one another , which might serve to reduce the overall level of inhibition during bilateral inputs and provide inhibition in proportion to the difference between the strengths of sensory inputs on the two sides of the body ( Mysore and Knudsen , 2012 ) . Our modeling shows that such a reciprocal connection could solve the problem of a lack of escapes to strong bilateral inputs ( Figure 2B3 ) . Such reciprocal connections between the inhibitory neurons were unknown , so we tested the prediction of their existence by using pairwise patch recording followed by intracellular labeling . We recorded from 17 bilateral pairs of feedforward inhibitory neurons ( 10 in relaxed fish at 4 dpf , 7 in nacre fish at 6 dpf ) and filled them with dye to confirm their identity . 14 of the 17 pairs were connected . Of these 14 , 10 pairs were connected in just one direction ( left cell inhibiting right , or right inhibiting left ) , as in the neurons shown in Figure 3A–B . The other 4 reciprocally inhibited each other at the single cell level as in Figure 3C–D . The connections were blocked by strychnine , consistent with the glycinergic phenotype of the feedforward neurons ( Figure 3C–D ) . The distribution of patterns of connectivity is summarized in Figure 3E1 for the two groups of fish . The amplitudes of the PSPs measured at rest ranged from 0 . 34 to 12 . 08 mV and averaged 3 . 85 mV ( Figure 3E2 ) , with no statistical difference in the distributions of strength in connections from right to left versus from left to right . In connected pairs , the processes of the presynaptic neuron were apposed to the postsynaptic cell ( Figure 3B1–2; cell filled in the physiology experiment in 3A ) , as expected for a monosynaptic connection . The neurons , like those reported in Figure 1 , also had processes adjacent to the contralateral M-cell , indicating that individual neurons might inhibit both that M-cell and contralateral feedforward inhibitory cells . 10 . 7554/eLife . 16808 . 007Figure 3 . Connections between FF inhibitory neurons on the two sides . ( A ) Patching of pairs of FF neurons on opposite sides of the brain . In this case , firing the right cell ( Right FF ) led to an IPSP in the FF on the left side ( Left FF ) ( middle panel ) , but the left cell did not produce a response in the right one ( right panel ) . B1 and B2 show horizontal and cross sections respectively , of the neurons recorded in A after filling them with dye . Green processes from the right neuron give rise to swellings apposed to the left neuron ( arrowheads on insets in B1 , B2 , which show the soma at higher magnification ) , consistent with the physiological connection . ( C ) Paired patch recordings of two reciprocally connected FF neurons . Firing the right neuron led to a depolarizing response in the left cell and vice versa ( dark traces ) . Both responses were blocked by strychnine ( 1 µM , gray traces ) , consistent with the PSP being a glycinergic inhibition with a reversal potential just above resting potential in our recording conditions . ( D1–2 ) The locations of the neurons recorded in C in confocal images of the dye filled neurons after recordings . ( E1 ) Patching of 17 bilateral pairs of FF neurons from either 4 day ( relaxed ) or 6 day ( nacre ) old fish revealed that most were connected to each other in one or both directions . Numbers in the histograms indicate the number of pairs with connections in neither direction ( none ) , one direction ( 1-way ) , or reciprocal connections ( 2-way ) . ( E2 ) Box plot of the strengths of the connections from left to right and right to left ( only measured in cases where there was a connection in a given direction ) . The connection strengths were not significantly different in the two directions . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 007 The pattern of connections in the network revealed by our experiments , along with the modeling , led to two predictions about the behavioral role of the inhibitory neurons in the initiation of the escape turn . The first , suggested by others before us , is that these neurons may contribute to determining the time at which an M-cell fires , which is when the decision to escape is made ( Faber et al . , 1989 , 1978 ) . Sensory inputs directly excite the ipsilateral M- cell , but also inhibit it by exciting the feedforward inhibitory neurons that project to that M-cell . This leads to the expectation that the timing of the escape is set by a balance between direct excitation and feedforward inhibition , which would determine when an M-cell reaches threshold to a unilateral stimulus and thus when a choice to escape is made . The model incorporating known connection strengths also shows this is possible , as lowering the level of inhibition shortens the response latency to a unilateral stimulus because the excitation dominates and drives the cell to threshold sooner ( All Ablated , Half Ablated , Figure 4A ) . 10 . 7554/eLife . 16808 . 008Figure 4 . Output predicted by the model after ablation of FF neurons . ( A ) Modeled latency of the Mauthner cell spike as a function of the simulated stimulus strength in control ( black ) and in cases where the entire pool of FF-neurons was removed from the model on the stimulated side ( red ) or half the pool was removed ( orange ) . The response latency becomes shorter after removal of FF neurons in the model . ( B ) Biased left-right decision after one-sided FF-removal in the model . A series of left and right stimuli is presented to the control model ( left panel ) and the FF-removed models ( right panels ) . The stimulus conditions that lead to the activation of one of the Mauthner cells are color-coded based on the activation pattern ( blue: only right Mauthner cell fires , red: only left Mauthner cell fires , white: both Mauthner cells fire ) . The likelihood of left Mauthner activation increases after the removal of right FF neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 008 The second prediction is that the inhibitory neurons are playing a central role in determining which of the two M-cells fires , and thus the laterality ( initial left or right bend ) of the escape behavior . This contribution of the inhibitory neurons to laterality would be mediated both by their stronger inhibition of the contralateral M-cell and by the inhibition of their contralateral inhibitory counterparts . The model leads to the prediction that removing inhibitory neurons on one side will lead on average to more escapes initiated by the Mauthner cell on the opposite side when the balance of sensory inputs on the two sides is varied ( Figure 4B ) . Thus , the experimentally derived network structure and the modeling suggest roles for the inhibitory neurons both in determining when a choice is made and what choice is made . We set out to test these two predictions by assessing behavioral performance following laser ablation of the feedforward inhibitory neurons . The first prediction was that removal of the inhibitory neurons on one side would reduce the sensory driven inhibition of the ipsilateral M-cell , while leaving the excitation unchanged , thus shifting the balance to excitation and leading to both a quicker depolarization to threshold and a shorter latency escape response to a stimulus on that side . To test this , we used a high-energy , long wavelength femtosecond laser to ablate inhibitory neurons on one side in a transgenic line in which glycinergic neurons were GFP labeled and the M-cell was labeled by single cell electroporation with Texas red dextran . As a control for the specificity of the lesioning and possible damage to the M-cell , we targeted the laser to other glycinergic neurons near the Mauthner cell , but medial to the ones in the escape network . We also directly targeted the M-cell’s lateral dendrite to either sever the dendrite from the cell body , or kill the M-cell . The approximate laser powers for killing the M-cells or severing their lateral dendrite were established by a systematic exploration of the effects of pulses at different energy levels ( Figure 5 ) . These data led us to use levels of 70–90 nanojoules ( nJ ) of energy to kill or cut the lateral dendrite of the M-cell because these outcomes occurred with high frequency at those levels . We used lower levels of 50-55 nJ for ablation of the smaller glycinergic neurons , and still lower energy pulses in control , non-ablation laser exposure ( 0-20 nJ for glycinergic neurons . The high energy pulsed laser we used ablates by production of a plasma rather than thermal damage . As a consequence , collateral damage is minimized and it produces very specific lesions of single cell precision in XY and Z at the power levels we used , as shown for an inhibitory neuron in Figure 6 . The effectiveness of all of the perturbations and the lack of obvious damage to adjacent neurons was confirmed in confocal or two photon 3D reconstructions from the live fish at the end of all of the experiments , as illustrated in the examples in Figure 7 . 10 . 7554/eLife . 16808 . 009Figure 5 . Assessment of impact of laser intensity on the Mauthner cell . Plot shows the results of targeting the lateral dendrite of 72 Mauthner cells with laser pulses of different intensities , showing the percentage of cases in which the cell died , the lateral dendrite was cut off with the neuron surviving , or no obvious change in the cell . Numbers on bars indicate the number of fish in each category . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 00910 . 7554/eLife . 16808 . 010Figure 6 . Specificity of laser targeting of inhibitory neurons . ( A ) FF neurons were backfilled with Texas red in a transgenic line with neurons labeled with nuclear targeted GCaMP6s . ( B1 ) Image prior to laser targeting showing the cell to be targeted marked by the white arrow head . ( B2 ) GCaMP6s fluorescence from the targeted cell following laser illumination for ablation shows a massive increase in fluorescence , also visible in images of the targeted neuron in horizontal ( C1 ) and cross section ( C2 ) images taken after illumination . Note the bright green signal is evident only in the targeted cell and not in surrounding neurons next to and above it , showing the specificity of the targeting . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 01010 . 7554/eLife . 16808 . 011Figure 7 . Examples of Laser ablations . ( A ) Diagram showing the approach to ablations , with initial filling of the Mauthner cell with red dye by single cell electroporation in a line with GFP labeled glycinergic neurons , followed by imaging and ablations from above with a femtosecond laser . ( B1–B3 ) Control fish exposed to laser power below threshold for lesions shown pre irradiation ( B1 ) , immediately post irradiation ( B2 ) and between 1 and 2 days later ( B3 ) . Glycinergic neurons are green , with arrows pointing to the cluster of feedforward cells on either side of the red Mauthner cell . ( C1–C3 ) As in B , but with laser ablation directed at the visible feedforward neurons ( arrows in C1 ) , most of which were removed ( C3 ) , while preserving the adjacent red dendrite of the M-cell . ( D1–D3 ) Laser targeting of the lateral dendrite of the Mauthner cell ( arrow ) sometimes severs the lateral dendrite , while leaving the adjacent inhibitory neurons and the soma of the M-cell intact ( D3 ) . ( E1–E3 ) Laser targeting of the lateral dendrite that led to death of the M-cell ( soma marked by arrowhead in E1 is absent in E3 ) , but preservation of adjacent inhibitory neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 011 The experiments involved six comparison groups , all with unilateral targeting of the laser: control sham lesions directed at the feedforward neurons , but at laser energy below levels producing damage; ablation of feedforward glycinergic neurons; ablation of glycinergic neurons not implicated in escapes , that are located adjacent to the M-cell soma , but in the most medial glycinergic column ( see Figure 1B1 , 2 ) about 40 µm medial to the feedforward cell type; cutting off the lateral dendrite of the M-cell which lies adjacent to the somata of the inhibitory neurons; and ablation of the M-cell . We targeted 4 to 8 inhibitory neurons for ablation per animal . The number of cells that lost fluorescence in the post-ablation stack averaged 5 . 2 ( s . d . =1 . 3 ) . We do not have a firm way of establishing the overall size of the feedforward inhibitory population , but we estimated it based on backfills from the Mauthner cell on one side . We found most filled cells on the contralateral side were located in the ventral half of the lateral glycinergic stripe ( Koyama et al . , 2011 ) . Because backfilling is very likely to miss some cells , to estimate the overall size of the inhibitory population we counted the number of glycinergic cells in the lateral stripe in rhombomere 4 located ventral to the most dorsal backfilled neurons , whether the cells in that region were backfilled or not . This led to an average of 31 . 3 cells ( se=0 . 7 , n=8 fish ) . We suspect this is an overestimate ( possibly a large one ) of the number that actually connect to the M-cell because the assumption here is that all the glycinergic cells in that region connect to the M-cell . If we take this value as an upper limit of the size of the population , we estimate that we will have removed minimally 13–25% of the population . The neurons we targeted with the laser were reliably identifiable based on their position relative to the lateral dendrite , and we were confident based on our many pairwise recordings that neurons in this location were members of the feedforward population . The day after the ablations , we tested the escape latency to a highly directional stimulus – a pressure-ejected pulse of water directed at either the lesioned or unlesioned sides of the caudal body/tail of the fish . This gave precise temporal control of the stimulus as well as a predominantly unilateral stimulus . A tail stimulus was chosen because it is known to reliably engage the M-cell ( Liu and Fetcho , 1999; O'Malley et al . , 1996 ) . In addition , we showed that this stimulus , though tail directed , excites auditory afferents based upon recordings from the eighth nerve during such a stimulus ( Figure 8 ) . The stimulus is , however , very likely multimodal , as we expect it to excite somatosensory and lateral line inputs , along with auditory ones . 10 . 7554/eLife . 16808 . 012Figure 8 . Tail directed water pulse leads to activation of axons in the eighth cranial nerve . ( A ) Diagram of the experiment in which extracellular recordings from the eighth cranial nerve in a paralyzed larvae were performed while applying a pulse of water directed at the tail . ( B ) Image of the glass recording electrode on the eighth nerve . ( C ) The water pulse led to extracellularly recorded activity in fibers of the eighth nerve , showing that a water pulse like that we used on freely swimming fish can produce an auditory response , even though tail directed . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 012 The results of these experiments are shown in Figure 9 . In control animals , with inhibitory neurons targeted with an attenuated laser , the median latency for the initial turns following a stimulus was 11 ms ( mean=16 . 4 s . e . = 2 . 2 , n=38 ) on one side and 12 ms on the other ( mean=13 . 6 ms s . e . =1 . 4 , n=31 ) , with no significant difference between the two sides . Control lesions of medial inhibitory neurons near the M-cell , but not implicated in escape , led to median latencies of 10 ms with no difference between the two sides ( mean=11 . 3 ms s . e . = 0 . 8 , n=73; mean=11 . 5 ms s . e . =0 . 7 , n=86 ) . Killing the M-cell produced a large and significant increase in latency to a median of 42 ms ( mean= 36 . 8 s . e . =4 . 3 , n=17; significance values in Figure 9 caption ) to a stimulus on the lesioned side and a significant difference in latency between the lesioned and unlesioned sides of the fish , consistent with earlier work ( Liu and Fetcho , 1999 ) . Severing the lateral dendrite of the M-cell led to a trend toward increased latencies on the lesioned side to a stimulus on that side ( median = 13 . 0 ms , mean=16 . 0 , s . e . =1 . 6 , n=51 ) , but the lesioned and unlesioned sides were not significantly different . 10 . 7554/eLife . 16808 . 013Figure 9 . Escape latency of control and lesioned fish in response to a unilaterally directed squirt of water . ( A ) Example trial showing the escape elicited by a squirt of colored water from a needle on the left side of the frame . The bend begins in the second frame ( * ) and the fish performs a rapid escape bend away from the stimulus in subsequent frames , with the peak initial bend at 10 ms ( ** ) . ( B . ) All of the data for the latency of escape response from lesioned ( red ) and unlesioned ( blue ) sides in controls and in the different ablation conditions . Box and whisker plots represent median as well as first and third quartiles . Outlier points are solid . Asterisks mark significant differences between responses on intact and lesioned sides ( *p<0 . 05 , ***p<0 . 001 , corrected for multiple comparisons ) . Sham ablations and ablation of medial glycinergic cells near the M-cell , but not implicated in escapes , did not affect the latency . Killing the Mauthner cell led to a significant increase in latency . Cutting dendrite of the M-cell also led to a trend toward a longer latency , although it was not significantly different from the intact side . Ablation of the lateral FF glycinergic neurons led to a significant reduction in the latency to respond . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 013 Most importantly , however , removing the feedforward inhibitory neurons caused an effect opposite to that seen by directly targeting the adjacent M-cell and one not seen after removal of other nearby glycinergic cells or sham ablations . Ablation of the feedforward neurons ( Lateral glycine ablation in Fig . 9B ) significantly shortened the latency of the response to a stimulus on the lesioned side ( median=6 . 0 ms , mean=8 . 3 ms , s . e . =1 . 7 , n=26 ) , so that it was about half of the latency of the escape response of control fish and significantly different from the latency to stimuli on the unlesioned side ( median=10 . 0 ms , mean=11 . 9 ms , s . e . =1 . 9 , n =27 ) . This difference in the effect of removing the feedforward inhibitory cells versus damaging the immediately adjacent Mauthner cell or other inhibitory neurons near it supports the specificity of the lesions and argues against the observations arising from a general nonspecific effect of damage in the region . Lesions of inhibitory neurons thus 'improved”' the fish’s response time by shortening the latency to respond . This is consistent with the prediction that the inhibitory neurons help to determine when the decision to escape is normally made by elevating the sensory stimulus required for the M-cell to reach threshold . A relatively high response threshold may be important to avoid production of escapes to weak , non-threatening stimuli . A unilaterally directed stimulus was useful for revealing how the balance of excitation and inhibition affects the time to the response , but many normal stimuli will activate sensory inputs bilaterally to varying extents , with the left/right direction of the escape determined by which of the two M-cells fires first . The circuit we defined suggests that each side competes with the other to control the choice of escape direction by inhibiting the M-cell and the feedforward inhibitory neurons on the opposite side . We infer , and our modeling supports , the possibility that the inhibition might act to directly suppress the contralateral M-cell and also indirectly promote activation of the ipsilateral one by blocking the inhibition coming from the other side . The circuit configuration , synaptic strengths , and modeling lead to the prediction that if escapes were elicited by stimuli over a range of directions and strengths , then the removal of FF inhibitory cells on one side would result in more escapes initiated by the opposite , intact side . We tested this prediction by eliciting escapes in a situation in which the fish was freely swimming in a dish on the surface of a vibration plate that produced a sudden , brief vertical displacement of 9 g acceleration ( Figure 10A , B; see materials and methods ) . Because the location and orientation of the fish varied in the dish from trial to trial we expected that the balance of sensory input to the two sides ( and thus the perceived location of its source ) would also vary from trial to trial leading , over a series of trials , to an equal number of escape bends to the left and to the right in an intact fish . Figure 10C shows that unlesioned control fish did on average produce half of their responses to the left and half to the right . 10 . 7554/eLife . 16808 . 014Figure 10 . Response laterality of control and lesioned fish in response to an omnidirectional stimulus . ( A ) Diagram of the experiment in which a dish containing a freely swimming fish is vibrated and the left/right direction of the initial escape turn is monitored . ( B ) An example of the high speed video recordings of a trial in which the escape response occurred to the fish’s right side . Asterisks mark the beginning ( * ) and end ( ** ) of the initial bend . ( C ) All of the data for response laterality for the control and lesion conditions , showing the percentage of responses initiated by the ablated side , which led to a turn away from that side and toward the intact side . Every lesion condition , except the ablation of medial glycinergic cells not implicated in escape , was significantly different from sham controls , with the magnitude of the change differing depending upon the type of lesion ( *p<0 . 05 , ***p<0 . 001 , corrected for multiple comparisons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 014 We decided a priori to focus on laterality and not on response latency in these experiments because the dish displacement stimulus did not offer as narrow a time estimate as the squirting for when the fish sensed the stimulus . Nonetheless , the large majority of the responses were short latency , rapid C-bends consistent with those produced by involvement of the M-cell . For example , in 134 responses from control and feedforward lesioned fish , the mean latency was 10 . 1 ms ( s . e . =0 . 82; only 6 were longer than 20 ms ) measured from the onset of the dish displacement . Responses with the kinematics of an escape ( Liu and Fetcho , 1999 ) of all latencies were included in the analysis . We asked how different laser perturbations changed the ratio of left/right responses to the omnidirectional stimuli . The outcome of the removal of a control set of medial glycinergic neurons adjacent to the M-cell was like that of sham ablations , with no resulting imbalance between the two sides . Killing the M-cell on one side led to the strongest effect with a median of zero percent of short latency escapes produced by the lesioned side ( mean=3 . 5 , s . e . =1 . 9 , n=9 ) , consistent with short latency turns being driven largely by the intact M-cell when the other M-cell is eliminated . Cutting off the lateral dendrite of the M-cell also led to a clear bias in the laterality of responses , with more short latency escape bends initiated by the side opposite to the lesion . The median percentage of responses initiated by the lesioned side was only 12 . 9 percent ( mean=18 . 5 , s . e . =6 . 6 n=6 ) . This result might be expected because substantial excitatory input onto the cut dendrite is removed by the cut , thereby reducing the ability of the lesioned cell to respond to the stimulus , thus biasing responses to the intact side . The most important result , however , in the context of the role of the FF inhibitory neurons in behavioral choice is the effect of their ablation on response laterality . After unilateral removal of a portion of the FF inhibitory population , we found a statistically significant drop in the fraction of escapes produced by the lesioned side ( Lateral glycine ablation , Figure 10 C; median=25 , mean=29 . 6 , s . e . =8 . 3 , n=10 p<0 . 05 ) . The majority ( 70–75% ) of escapes were produced by the intact side , as predicted . In summary , the changes in both behavioral paradigms were those predicted based upon the patterns and strengths of the neuronal connectivity we revealed , supporting the conclusion that we have defined a network motif important for a simple two alternative behavioral choice .
Even simple nervous systems have to select alternative motor responses based on sensory evidence . In bilateral animals like vertebrates , the laterality of a response is one of the most fundamental and primitive behavioral choices as it is critical for avoiding predators and navigating through the environment , however big a brain the animal might have . Our work was directed toward revealing a network motif that contributes to this basic behavioral choice in vertebrate nervous systems . One advantage of Mauthner system we studied is that these two cells are the locus of a left right decision . If one of the Mauthner cells fires , the fish generates a strong bend ( Nissanov et al . , 1990 ) . The firing of one of the two cells thus represents a choice of laterality , potentially offering insight into simple networks for behavioral choice – here considered in the broadest sense as collecting sensory evidence and using it to select an adaptive motor response from possible alternatives ( Korn and Faber , 2005 ) . In escapes , like most behaviors , correct behavioral choices require producing the appropriate behavior at the right time . The connectivity motif of the excitatory , and inhibitory interneurons defined here ( Figure 11A ) contributes to solving both of these problems through its influence on both the laterality of an escape bend and its timing . The key aspects of the motif are excitatory inputs conveying sensory evidence ( here , auditory inputs ) for the alternatives to the neurons driving the output of the choice ( here , the right or left M-cell ) . The choice depends on a competition between the two alternatives , as sensory evidence for one choice inhibits the other through inhibitory neurons that suppress the alternative outcome ( the contralateral M-cell ) . Reciprocal connections between those inhibitory neurons likely assure that the response is fast and lateralized even for strong bilateral inputs by providing inhibition to choice neurons ( M-cells ) that scales with the difference in the level of sensory input on the two sides rather than the magnitude of the sensory drive . Finally , the timing of the response is determined by a balance between excitation and inhibition at the choice point ( the M-cell ) , including a feedforward inhibition that likely prevents escape responses to weak inputs . The circuit motif we describe has a pattern of connections similar to network structures that theoretical work shows can allow for optimal choices between two alternative responses based on sensory evidence , also supporting its likely importance for behavioral choice ( Bogacz et al . , 2006; Mysore and Knudsen , 2012 ) . 10 . 7554/eLife . 16808 . 015Figure 11 . Summary of the circuit and its relationship to hindbrain structure . ( A ) Diagram of the circuit motif for the left right choice revealed and tested behaviorally in this paper . Relative synaptic strengths are indicated by contact size . Dotted line shows ipsilateral output of the FF . M: Mauthner cell , FF ( red ) : Feedfoward inhibitory neuron . ( B ) Image of the disposition of the neurons in the hindbrain . The inhibitory neurons ( red , FF ) lie at the bottom the most lateral of three columns of glycinergic inhibitory neurons ( green ) marked by arrows . This lateral column contains commissural inhibitory neurons with outputs on both sides of the hindbrain , while more medial columns contain neurons with contralateral ( middle column ) or ipsilateral outputs ( medial column ) . Scale bar=20 µm . ( C ) The column containing the FF neurons is one of a series of interleaved glutamatergic and glycinergic columns in the hindbrain . The variety of sensory inputs to the lateral glycinergic column and the shared morphology of the neurons in it raise the possibility that other neurons in the column might also contribute to the laterality of other behavioral responses to sensory inputs by implementing the same circuit motif . DOI: http://dx . doi . org/10 . 7554/eLife . 16808 . 015 The roles of the motif in the timing and lateralization of the escapes are supported by lesion experiments that produced outcomes predicted by the connectivity pattern we revealed , including changes in the laterality of the response and the ability to shorten an already very short latency response . Extensive controls support the specificity of the perturbations . Our simple model incorporating the known connections and strengths could produce adaptive responses away from the side of the strongest input over a very broad range of bilateral sensory strengths , showing the possibility that such a network , even with a small number of cell types , could account for the behavior – a proof of principle at least . Importantly , removal of neurons in the model led to results consistent with lesion experiments . In combination with the outcomes of perturbations in the model that were not possible experimentally , the model points to the importance of the pattern of connections for producing escape responses away from the side of the strongest stimulus at relatively short latency ( but not to weak stimuli ) over a range of bilateral stimulus strengths . Together , the experimental evidence and the modeling support a key role for the circuit motif we revealed , shown in Figure 11A , in determining the timing and laterality of a behavioral choice . Here , the timing and laterality of the escape bend depend on inhibitory as well as excitatory inputs , in contrast to recent work suggesting that the side of the initiation of swimming in tadpoles is determined by an imbalance in excitation reaching reticulospinal neurons on opposite sides ( Buhl et al . , 2015 ) . That work detected no inhibition , even though the possibility for it was left open . A purely excitatory mechanism might falter if a precise choice of the laterality of the response is required , as in the escape behavior . In the initiation of swimming , the choice of the side on which the propagating wave starts may not be as critical as the direction of a turn away from a predator . Importantly , the feedforward inhibitory neurons are not the only neurons that might contribute to the features of the escape response in different contexts . The Mauthner cell network also has excitatory interneurons that can influence its firing ( Lacoste et al . , 2015 ) , and there are very likely other unknown inputs as well that could also have an effect on which cell fires when . In addition , while the timing and laterality of the response is determined by when one of the two M-cell fires , the magnitude of the response depends on patterns of activation of other neurons in the hindbrain and spinal cord ( Liu and Fetcho , 1999; O'Malley et al . , 1996; Kohashi and Oda , 2008 ) . Given the many lateralized behaviors ( eye movements , head movements , limb movements ) produced by animals and their variations in magnitude , we might expect that a circuit motif underlying lateralized responses in the brain would be repeated to control different lateralized behaviors . While on the face of it , the escape network seems very specialized , we now know that it arises in an orderly way from a columnar ground plan in the hindbrain that increasing evidence suggests is shared by vertebrates broadly ( Figure 11A–C ) ( Gray , 2013; Kinkhabwala et al . , 2011; Koyama et al . , 2011 ) . The feedforward inhibitory neurons that we have shown are key to determining the threshold and directionality of escapes are a small subset of one of the hindbrain columns that contains other bilaterally projecting glycinergic inhibitory neurons . This raises the possibility that the other morphologically similar neurons in the column might also receive sensory inputs and have similar patterns of connectivity , but with different output neurons than the M-cell , to implement left right choices in other networks that move the eyes , head , body and limbs . Prior studies of both the spinal cord and hindbrain in zebrafish suggest how repetition of such circuit motifs might be organized ( Kinkhabwala et al . , 2011; McLean et al . , 2007; McLean and Fetcho , 2009; McLean et al . , 2008 ) . Early in life , hindbrain neurons are ordered by age and function , which roughly map onto their dorsoventral position in larval zebrafish . In the hindbrain columns prior to neuronal migration ( much of which occurs after larvae have hatched and already have the ability to swim , escape and feed to survive ) , older , more ventral neurons like the M-cell , the ventral feedforward inhibitory neurons , and ventral descending excitatory neurons in a more medial column ( positive for the Chx10/Alx transcription factor ) control powerful movements such as escape and fast swimming ( Kinkhabwala et al . , 2011; Koyama et al . , 2011 ) . Increasingly younger , more dorsal neurons are recruited for successively slower , weaker swimming responses . Lateralized motor responses can also vary widely in strength as well as latency to respond; for example , escape turns produced by old neurons are at the most powerful end of a range of turning movements that also include much slower and longer latency turns ( Budick and O'Malley , 2000; McElligott and O'malley , 2005; Burgess and Granato , 2007 ) . We therefore predict that younger inhibitory neurons located more dorsally in the column containing the escape feedforward cells might share patterns of connectivity defined here , but will collect sensory information over increasingly longer time scales to drive appropriately lateralized behaviors that occur at longer latencies and reduced strengths than the escape . Longer latency responses might be produced by neurons with longer integration times ( there is already evidence for a gradient of integration times in zebrafish oculomotor circuits that maps onto location in the hindbrain [Miri et al . , 2011] ) or by the accumulation of evidence via a gradual recruitment within a larger population of neurons than is engaged in the very fast escape behavior . If so , then repeated network motifs in the hindbrain may contain sensory motor circuits that accumulate sensory information over different time scales to drive behavioral choices to turn left or right over a range of different latencies , speeds and magnitudes .
All experiments were performed on 4 to 7 day post-fertilization ( dpf ) zebrafish obtained from a laboratory stock of transgenic and mutant adults . All procedures conform to the US National Institute of Health guidelines regarding animal experimentation and were approved by Cornell University’s Institutional Animal Care and Use Committee . The transgenic lines , TgBAC ( slc6a5:GFP ) ( McLean et al . , 2007 ) TgBAC ( slc17a6b:loxP-DsRed-loxP-GFP ) ( Koyama et al . , 2011 ) in the nacre background ( [Lister et al . , 1999] mitfab692/b692 ) were used . Tg ( glyt2:GFP ) in relaxed background ( [Ono et al . , 2001] cacnb1ts25/ts25 ) was used in some electrophysiology experiments ( Koyama et al . , 2011 ) . Tg ( elavl3:H2B-GCaMP6s ) in the casper background ( Vladimirov et al . , 2014 ) was used in some control ablation experiments . Patch-clamp recordings were performed as described previously ( Koyama et al . , 2011 ) . Tg ( glyt2:GFP ) was used to target glycinergic neurons . In initial experiments larvae at 4 dpf , in a relaxed background ( Ono et al . , 2001 ) , in which a mutation prevents the release of calcium in internal stores in muscle , were used to immobilize the fish without a cholinergic blocker . The Tg ( glyt2:GFP ) line in a nacre background at 6 dpf was used with α-bungarotoxin paralysis ( Koyama et al . , 2011; Lister et al . , 1999 ) in later experiments to confirm the physiological observations in the cancb1 background also apply to the lines used for the behavioral experiments . Feedforward glycinergic cells and Mauthner cells were targeted based on previous anatomical and physiological characterization ( Koyama et al . , 2011 ) . Whole-cell current clamp recordings were made with a Multiclamp 700A and 700B ( Molecular Devices , Sunnyvale CA ) at a gain of 50 with a 500 MOhm feedback resistor ( 50 MOhm for the Mauthner cell ) filtered at 10 kHz and digitized at 50 kHz with a Digidata 1440A ( Molecular Devices ) using Clampex 10 . 2 ( Molecular Devices ) . The reversal potential of IPSPs was checked by injecting a series of depolarizing and hyperpolarizing currents in the postsynaptic cell , with 5 traces for each current amplitude . A glycinergic blocker , strychnine ( 1 µM; Sigma-Aldrich ) , was applied to the bath to confirm the neurotransmitter responsible for the IPSP in some experiments . Data were analyzed with MATLAB ( MathWorks ) . The peak of each IPSP was detected and the mean was calculated for each postsynaptic cell . The morphology of the recorded cells was visualized under a confocal microscope ( LSM510 META , Zeiss ) after filling them with dye ( 0 . 02% , Alexa Fluor 568 or 647 hydroazide , Invitrogen ) dissolved in intracellular solution . Z stacks were volume rendered using the program Imaris ( Bitplane ) . We made certain in triple recordings of an inhibitory neuron and two M-cells that the recordings of the cells were of high quality , with low access resistance , healthy resting potentials in the M-cells ( approximately −68 mV without correction for junction potential [12 mV] ) , and resting potentials that were well matched between the two M-cells . Because the reversal potential for the inhibitory synapses is near resting potential , we measured the synaptic responses while injecting 1 nA of positive or negative current into the M-cells on different trials . Putative presynaptic terminals of feedforward neurons were identified based on their characteristic puncta-like structure , and the number of puncta apposed to the cell body of Mauthner cells on each side was counted and compared with a paired Wilcoxon signed rank test with ties using the 'coin' package in R ( http://cran . r-project . org/web/packages/coin/index . html ) . To statistically examine whether the inhibitory strength of feedforward neurons differs between the Mauthner cells on each side , the data set for the mean peak of the IPSPs was fitted with the following generalized linear mixed model with a random effect for cell using 'lme4' package in R ( http://cran . r-project . org/web/packages/lme4/index . html ) . The log link function for the Gaussian distribution was used to satisfy the assumptions of the linear model ( normality , homogeneity of variance , independence and linearity ) . ISPS ∼ Side + Age + ( 1∣CellID ) The significance of the effects was examined by a likelihood ratio test of nested models . Tg ( glyt2:EGFP ) in the nacre background at 4 dpf were anesthetized and embedded in 1 . 7% low melting-point agar dissolved in 10% Hanks solution . Twenty percent Texas Red Dextran , 10 , 000 MW , anionic , lysine fixable ( Invitrogen ) was electroporated into the Mauthner cell using the protocol described previously ( Bhatt et al . , 2004 ) . Fish were removed from the agar and immediately transferred to a small petri dish of 10% Hank’s buffer after the electroporation . Fish were re-embedded again the following day ( 5 dpf ) and imaged with a custom-made two-photon microscope ( Farrar et al . , 2011 ) controlled by MPScope or ScanImage ( Nguyen et al . , 2011; Pologruto et al . , 2003 ) . A z stack of images was acquired before the ablation using a 20× water-immersion objective lens ( numerical aperture ( NA ) =1 . 0; Zeiss ) . 920-nm , 87-MHz , 100-fs pulses from a Ti:sapphire laser oscillator ( MIRA HP; Coherent ) and 1043-nm wavelength , 1-MHz , 300-fs pulses from a fiber laser ( FCPA μJewel D-400; IMRA ) were used to excite GFP ( inhibitory cells ) and Texas Red dextran ( Mauthner cells ) . We used emission filters at 645/65 nm ( center wavelength/bandwidth ) and 517/65 nm ( Chroma Technology ) to isolate fluorescence from Texas Red dextran and GFP , respectively . In order to ablate a cell body or a dendrite , we delivered one or two 100-fs pulses from a regenerative amplifier ( 800 nm wavelength , 1 kHz repetition rate; Legend , Coherent ) for each target while continuously imaging the target and the surrounding region . Because the damage was mediated by an electron-ion plasma formed by nonlinear optical absorption and there was very little thermal energy deposited , the damage was largely confined to the focal volume ( Nishimura et al . , 2006 ) . In order to ablate feedforward glycinergic neurons , we used 50 to 55 nJ pulses to target four to eight GFP positive glycinergic neurons on one side of the hindbrain located at the ventral-most part of the lateral glycinergic stripe , close to the lateral dendrite of the Mauthner cell . These were previously shown to be feedforward glycinergic neurons by retrograde labeling and paired recordings ( Koyama et al . , 2011 ) , as confirmed by many additional recordings in this paper . In control experiments , we targeted a similar number of glycinergic neurons not implicated in escapes that were near the M-cell , but medial to the feedforward cells . In other experiments , we targeted the lateral dendrite of the Mauthner cell on one side of the hindbrain at a site next to the feedforward glycinergic neurons to examine the effects of direct damage to the M-cell . We used energy ranging from 70–90 nJ , which either successfully cut off the lateral dendrite or killed the cell , probably because the membrane failed to reseal in some cases . In each round of the experiments , we used approximately half of the fish for the ablation and the other half for controls , in which we deposited smaller energy pulses that would not damage the target ( 0–20 nJ for glycinergic neuron ablation and 30 – 50 nJ for lateral dendrite ablation ) . A z stack was acquired immediately after the ablations to examine the lesions . Fish with successful lesions were then transferred from agar into a small petri dish containing 10% Hank’s buffer and allowed to recover overnight before the subsequent behavioral assay . In order to demonstrate the spatial specificity of our ablation protocol , we used a transgenic line that expresses a genetic calcium indicator pan-neuronally to assess the damage in the cells near the target cell based on their calcium level ( Lister et al . , 1999 ) . Feedforward glycinergic neurons were backfilled with 20 percent Texas Red Dextran , 10 , 000 MW , anionic , lysine fixable ( Invitrogen ) from the ipsilateral Mauthner cell using the procedure described previously ( Kinkhabwala et al . , 2011 ) in Tg ( elavl3:H2B-GCaMP6s ) at 3 dpf . Then the backfilled feedforward glycinergic neurons were ablated using the same protocol used above at 5 dpf while monitoring the signals from Texas Red Dextran and H2B-GCaMP6s in the target cell . Z stacks were acquired before and after the ablation . Escape behavior of individual larvae was examined 1 day after the ablation ( 6 dpf ) in a petri dish ( 3 . 5 cm ) filled with 10% Hank’s buffer to a depth of 3–4 mm . We delivered two types of stimuli described below that elicit an escape response and filmed the fish’s behavior at 1000 Hz with a high-speed video system ( FASTCAM PCI , Photron ) . To rule out any bias in the experimental procedure and subsequent analysis , fish were coded by an individual other than the experimenter to blind the experimenter from the ablation outcomes until the escape responses were analyzed . Neuronal ablations were re-confirmed the following day ( 7 dpf ) either with the 2-photon microscope or with a confocal microscope ( LSM510 Meta , Zeiss ) . In one series of experiments , escape responses to a unidirectional stimulus were examined by delivering water pulses to the tail as described previously ( Liu and Fetcho , 1999 ) . A pulse of colored water ( 0 . 05 mg/ml , fast green ) was delivered to the caudal body/tail by a picospritzer ( ~17 psi , 5 ms ) through a syringe and 27 gauge needle cut blunt and bent to about 100 degrees . These settings were determined to be the lowest level that can reliably produce an escape response without the water stream disturbing the animal’s movements . Stimuli were delivered to the right and left sides of the tail alternately with a minimum inter-trial interval of 2 min . At least 10 trials were collected per fish ( 5 trials for each side ) . The escape latency was calculated for each trial as the time from the contact of the pulse of water ( visible because of the green dye ) with the fish to the beginning of the bending movement . The trials in which fish did not show the characteristic kinematics of the escape movement ( Liu and Fetcho , 1999 ) were excluded from analysis . To examine the effects of ablations on the escape latency , fish were categorized into the following groups based on the post-hoc imaging: 1 ) lateral feedforward inhibitory cells ablated; 2 ) Medial inhibitory cells ablated; 3 ) Mauthner’s lateral dendrite cut; 4 ) Mauthner cell ablated; and 5 ) control . For statistical testing , the data for escape latency were fitted with the following linear mixed model with a random effect for subject using the 'nlme' package in R ( http://cran . r-project . org/web/packages/nlme/index . html ) . Latency was log-transformed to satisfy the assumptions of the linear model . log ( Latency ) ∼ Ablation ∗ Side + ( 1∣SubjectID ) The effects of ablations were tested by comparing the ablated and un-ablated side for each ablation group with correction for multiple comparisons using the 'multcomp' package in R ( http://cran . r-project . org/web/packages/multcomp/index . html ) . In another series of experiments , escape responses to an omnidirectional stimulus were examined by shaking the petri dish with a vibrating transducer ( Taparia Magnetics , Mumbai , India ) placed underneath it . The vibration was produced by a single 100–150 ms voltage pulse applied to the transducer , which produced a peak acceleration at pulse onset of 9 g , as measured in the vertical direction with a calibrated accelerometer ( ACC-103 Omega engineering , Stamford CT ) . A relatively long duration pulse ( as compared to the 10–20 ms duration of the initial escape bend ) was used to delay any potential response to the vibration caused by the offset of the pulse . A minimum of 10 trials was collected per fish with an inter-trial interval of at least 2 min . The stimulus strength elicited escape responses reliably without fatigue at this inter-trial interval . The data were analyzed statistically by fitting the number of left versus right escapes to the following generalized linear mixed model with a random effect for subject , using a logit link function for a binomial distribution and the 'lme4' package in R ( http://cran . r-project . org/web/packages/nlme/index . html ) . Direction ∼Ablation + ( 1∣SubjectID ) The effects of ablations were examined with Dunnett’s multiple comparisons with a control by using the 'multcomp' package in R ( http://cran . r-project . org/web/packages/multcomp/index . html ) . Conductance-based models of the Mauthner circuit were built using the Brian spiking neural network simulator ( http://briansimulator . org/ ) ( Goodman and Brette , 2008 ) . Basic membrane properties of the Mauthner cell and feedforward inhibitory neurons were derived from the dataset from 4 dpf fish acquired in this study ( Mauthner cell: n=24; feedforward glycinergic neurons: n=28 ) ( Table 1; corrected for liquid junction potential of 12 mV ) . Input resistance was calculated based on the current amplitude required to hyperpolarize the cell to 10 mV below the resting membrane potential . The relaxing phase of voltage traces after hyperpolarizing current injections were fitted with double exponentials using fmincon in MATLAB’s optimization toolbox and the slowest time constant was used to calculate the capacitance of the cell ( White and Hooper , 2013 ) . Putative spiking threshold was estimated from phase plots ( dVm/dt vs Vm ) of intracellular voltage ( Vm ) ( Bean , 2007 ) . The reversal potential for chloride in the Mauthner cell was estimated from the linear fit of the IPSP amplitude and the holding potential of Mauthner cell . Based on these parameters , the Mauthner cell and feedforward glycinergic neurons were modeled as conductance-based leaky integrate and fire neurons ( Supp . Table ) . To represent the spike-like waveform from gap junctional inputs from the VIIIth nerve , we modeled VIII ganglion neurons using a Hodgkin-Huxley model with adjustments in the reversal potentials to set the resting membrane potentials identical to that of the Mauthner cell . The membrane area of the VIII ganglion neurons was varied so that the number of recruited neurons increased in response to stronger stimulation . We connected one VIII ganglion neuron to only one feedforward neuron and treated them as one functional unit that recruited systematically as stimulus strength increased ( Figure 2A1–3 ) . The connection between them was specified as a purely electrical connection based on previous literature ( Zottoli and Faber , 1980 ) and its strength was set just enough to initiate a spike in feedforward neurons in the absence of inhibitory inputs to account for the presence of feedforward inhibition even at a low stimulus strength ( Oda et al . , 1995 ) . The relative conductance of chemical and electrical connections between the VIIIth nerve and Mauthner cell was set heuristically based on the experimentally-derived VIII nerve response in the Mauthner cell in 5 dpf zebrafish ( Figure 2A2 ) . We used the membrane time constant of 0 . 4 ms to model the eighth nerve response in the lateral dendrite based on previous literature ( Pereda et al . , 2011 ) . The inhibitory connection from a feedforward neuron to the ipsilateral Mauthner cell was tuned based on the dataset from this study ( Figure 2A3 ) . The relative balance of the excitatory and ipsilateral inhibitory connections was determined so that the latency range of the Mauthner spike in response to various strengths of unilateral activation of VIII ganglion neurons matched that of the Mauthner-mediated escape to unilateral stimuli ( Liu and Fetcho , 1999 ) . The relative strength of contralateral and ipsilateral inhibitory inputs from feedforward neurons was determined based on the experimentally derived ratio ( Figure 1 ) . The strength of the putative mutual inhibition between feedforward neurons was set to recover the Mauthner spikes in response to equally strong left and right stimuli . The ablation of feedforward neurons was modeled by removing feedforward neurons on one side and their associated connections . | Humans and other vertebrate animals constantly make choices about whether to respond to the left or to the right . Do they look left or right; turn left or right; reach left or right ? In humans , the distinction between left and right is so fundamental that it has entered our collective thinking . Many societies define their political positions , for example , in terms of leaning to the left or to the right . However , we know little about the wiring of the brain that accomplishes the task of making physical left-right choices . Koyama et al . therefore set out to identify the neural circuit responsible for the decision to turn either left or right . Zebrafish larvae were chosen as subjects because they execute rapid left or right turns to escape predators . Given that one wrong turn can result in the death of the zebrafish , a correct choice matters more than in most of the other decisions that animals make . Experiments revealed that a process of competition between neurons on the left and right sides of the brain underlies this decision-making . Neurons on the right collect evidence that an attack is coming from the right , and drive turns to the left , away from the threat . These neurons also attempt to silence competing neurons on the left that act to produce turns to the right . By weighing up the evidence from left and right sides , the circuit as a whole comes to a decision about the best direction in which to turn . The region of the brain that controls the left versus right escape response in zebrafish is present in all vertebrates . Moreover , it appears to have a similar structure across species , consisting of repeating columns of neurons . This raises the possibility that other left-right choices in fish and other animals occur in a similar way – a principle that can be tested in future work . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | A circuit motif in the zebrafish hindbrain for a two alternative behavioral choice to turn left or right |
When mammalian cells detect a viral infection , they initiate a type I interferon ( IFNs ) response as part of their innate immune system . This antiviral mechanism is conserved in virtually all cell types , except for embryonic stem cells ( ESCs ) and oocytes which are intrinsically incapable of producing IFNs . Despite the importance of the IFN response to fight viral infections , the mechanisms regulating this pathway during pluripotency are still unknown . Here we show that , in the absence of miRNAs , ESCs acquire an active IFN response . Proteomic analysis identified MAVS , a central component of the IFN pathway , to be actively silenced by miRNAs and responsible for suppressing IFN expression in ESCs . Furthermore , we show that knocking out a single miRNA , miR-673 , restores the antiviral response in ESCs through MAVS regulation . Our findings suggest that the interaction between miR-673 and MAVS acts as a switch to suppress the antiviral IFN during pluripotency and present genetic approaches to enhance their antiviral immunity .
Type I interferons ( IFN ) are crucial cytokines of the innate antiviral response . Although showing great variation , most mammalian cell types are capable of synthesizing type I IFNs in response to invading viruses and other pathogens . Once type I IFNs are secreted , they activate the JAK-STAT pathway and production of interferon-stimulated genes ( ISGs ) in both the infected and neighbouring cells to induce an antiviral state ( Ivashkiv and Donlin , 2015 ) . Two major signalling pathways are involved in IFN production in the context of viral infections . The dsRNA sensors RIG-I and MDA5 initiate a signalling cascade that signals through the central mitochondrial-associated factor MAVS , ultimately activating Ifnb1 transcription . The cGAS/STING pathway is activated upon detection of viral or other foreign DNA molecules and uses a distinct signalling pathway involving the endoplasmic reticulum associated STING protein ( Chan and Gack , 2016 ) . Despite its crucial function in fighting pathogens , pluripotent mammalian cells do not exhibit an IFN response . Both mouse and human embryonic stem cells ( ESCs ) ( Wang et al . , 2013; Chen et al . , 2010 ) as well as embryonic carcinoma cells ( Burke et al . , 1978 ) fail to produce IFNs , suggesting that this function is acquired during differentiation . The rationale for silencing this response is not fully understood but it has been proposed that in their natural setting , ESCs are protected from viral infections by the trophoblast , which forms the outer layer of the blastocyst ( Delorme-Axford et al . , 2014 ) . ESCs exhibit a mild response to exogenous IFNs , suggesting that during embryonic development , maternal IFN could have protective properties ( Hong and Carmichael , 2013; Wang et al . , 2014 ) . In mouse ESCs , a Dicer-dependent RNA interference ( RNAi ) mechanism , reminiscent to that of plants and insects , is suggested to function as an alternative antiviral mechanism ( Maillard et al . , 2013 ) . And in humans , ESCs intrinsically express high levels of a subgroup of ISGs in the absence of infection , bypassing the need for an antiviral IFN response ( Wu et al . , 2018; Wu et al . , 2012 ) . All these suggest that different antiviral pathways are employed depending on the differentiation status of the cell . Silencing of the IFN response during pluripotency may also be essential to avoid aberrant IFN production in response to retrotransposons and endogenous retroviral derived dsRNA , which are highly expressed during the early stages of embryonic development and oocytes ( Ahmad et al . , 2018; Grow et al . , 2015; Macia et al . , 2015; Peaston et al . , 2004; Macfarlan et al . , 2012 ) . Furthermore , exposing cells to exogenous IFN induces differentiation and an anti-proliferative state , which would have catastrophic consequences during very early embryonic development ( Borden et al . , 1982; Hertzog et al . , 1994 ) . All these observations support a model in which cells gain the ability to produce IFNs during differentiation . One particular class of regulatory factors that are essential for the successful differentiation of ESCs are miRNAs ( Greve et al . , 2013 ) . These type of small RNAs originate from long precursor RNA molecules , which undergo two consecutive processing steps , one in the nucleus by the Microprocessor complex , followed by a DICER-mediated processing in the cytoplasm ( Treiber et al . , 2018 ) . The Microprocessor complex is composed of the dsRNA binding protein DGCR8 and the RNase III DROSHA which are both essential for mature miRNA production ( Gregory et al . , 2004; Lee et al . , 2003 ) . In addition , mammalian DICER is also essential for production of siRNAs ( Bernstein et al . , 2001 ) . The genetic ablation of Dgcr8 or Dicer in mice blocks ESCs differentiation suggesting that miRNAs are an essential factor for this , as these are the common substrates for the two RNA processing factors ( Wang et al . , 2007; Kanellopoulou et al . , 2005 ) . In this study , we show that miRNAs are responsible for suppressing the IFN response during pluripotency , specifically to immunostimulatory RNAs . We found that miRNA-deficient ESCs acquire an IFN-proficient state , are able to synthesize IFN-β and mount a functional antiviral response . Our results show that miRNAs specifically downregulate MAVS ( mitochondrial antiviral signalling protein ) , an essential and central protein in the IFN response pathway . In agreement , ESCs with increased MAVS expression or knock-out of the MAVS-regulating miRNA miR-673 , resulted in an increased IFN production and antiviral response . Our results support a model where the MAVS-miR-673 interaction acts as a switch to suppress the IFN response and consequently virus susceptibility during pluripotency .
There are two major pathways for sensing intracellular viral infections and consequent activation of the IFN response in cells . One senses dsRNA , usually originating from RNA viruses , with MAVS as a central factor , and the second senses dsDNA , from DNA- and retroviruses signalling through STING ( McFadden et al . , 2017 ) . It has been shown that mouse ESCs do not produce type I IFNs in response to poly ( I:C ) transfection , a synthetic analogue of dsRNA classically used to mimic viral RNA replication intermediates ( Wang et al . , 2013 ) . In contrast , it is still unknown how mouse ESCs respond to immunostimulatory DNA . To study this , two different mouse ESC cell lines ( ESC1 and ESC2 ) were transfected with poly ( I:C ) and G3-YSD , an HIV-derived DNA that stimulates the cGAS/STING pathway ( Herzner et al . , 2015 ) . As controls , NIH3T3 fibroblasts and BV-2 microglial cells were included . As expected , the transfection of poly ( I:C ) did not result in Ifnb1 expression in both ESC lines ( Figure 1A ) . ESCs also failed to activate Ifnb1 expression upon G3-YSD transfection , suggesting that the cGAS/STING pathway was also inactive ( Figure 1B ) . Similarly , NIH3T3 cells , which have also been previously shown to have a defect in this specific pathway ( Cheng et al . , 2018 ) , did not express Ifnb1 in response to G3-YSD ( Figure 1B ) . These same cell lines were infected with the ( + ) ssRNA virus TMEV ( Theiler's Murine Encephalomyelitis Virus ) and showed that ESCs are at least 30 times more sensitive than NIH3T3 and BV-2 cells , which correlates with the ability of these cell lines to induce Ifnb1mRNA expression ( Figure 1C ) . The ability of cells to express IFN in response to viruses or immunogenic nucleic acids is assumed to be acquired during differentiation . To test this model , we in vitro differentiated both ESC lines with retinoic acid and determined their ability to respond to poly ( I:C ) . Briefly , embryoid bodies were generated by a hanging droplet method for 48 hr before being cultured in the presence of retinoic acid for 2 or 10 days . Samples from each of these time points were analysed for expression of pluripotency and differentiation markers . The pluripotency markers Nanog and Pou5f1 ( Oct4 ) showed a rapid decrease in mRNA expression during differentiation in both the cell lines ( Figure 1—figure supplement 1A ) , whereas differentiation markers Neurog2 , Gata6 and Gata4 showed a gradual increase ( Figure 1—figure supplement 1B ) confirming successful differentiation of the ESCs . Next , we compared the ability of ESCs ( day 0 ) and retinoic-acid differentiated cells after 10 days ( day 10 ) to express Ifnb1 mRNA in response to poly ( I:C ) , and confirmed that differentiated cells acquired the ability to synthesize Ifnb1 to similar levels to the positive control cell line , BV-2 ( Figure 1D ) . Given the relevance of RNAi as an antiviral mechanism in mouse ESCs ( Maillard et al . , 2013 ) , we next asked if ESCs , in the absence of the central factor for RNAi , ICER , would be more susceptible to RNA viruses . Unexpectedly , Dicer-/- ESCs were more resistant to viruses compared to their wild-type counterparts ( previously named ESC2 ) ( Figure 2A , left ) . Similar results were obtained using the ( - ) ssRNA virus , Influenza A ( IAV ) ( Figure 2A , right ) . Importantly , mammalian Dicer has a dual function , being essential for both siRNA and miRNA biogenesis . To determine whether these differences in viral susceptibility were due to the activity of Dicer on siRNA or miRNA production , we compared Dicer-/- cells with ESCs lacking the essential nuclear factor for miRNA biogenesis , Dgcr8 . The absence of Dgcr8 also decreased TMEV and IAV viral susceptibility , suggesting that miRNAs are responsible for suppressing the antiviral response in ESCs ( Figure 2A ) . Interestingly , Dgcr8-/- cells were more resistant to virus infection than Dicer-/- cells , which supports a dual function for DICER by also acting as a direct antiviral factor targeting viral transcripts for degradation by RNAi . To rule out the possibility of morphological differences influencing viral susceptibility , we performed a virus binding and entry assay which showed no differences ( Figure 2—figure supplement 1 ) . Even though ESCs lack an IFN response , we wondered whether the differential resistance to viral infections were the result of abnormal IFN activation due to the absence of miRNAs . To test this hypothesis , we transfected the dsRNA analogue , poly ( I:C ) and the immunogenic G3-YSD DNA in Dgcr8 or Dicer deficient mESCs , and quantified Ifnb1 expression by RT-qPCR and ELISA . ESCs lacking miRNAs ( Dgcr8-/- or Dicer-/- ) were able to respond to the dsRNA analogue , poly ( I:C ) and express Ifnb1 mRNA and protein in a dose dependent manner ( Figure 2B and Figure 2—figure supplement 2A–C ) , whereas no significant response was observed with immunostimulatory DNA ( Figure 2B ) . These results show there is a correlation between viral susceptibility and the ability of miRNA-deficient ESCs to express IFN-β , and suggest that miRNAs are responsible for silencing the IFN response to dsRNA . To further establish the involvement of IFN expression in these observations , we blocked IFN signalling using the JAK1/2 inhibitor Ruxolitinib before infecting cells with TMEV . As a result we observed no , or a very mild increase in TMEV viral replication in wild-type ESCs , but a significant increase in viral replication in miRNA-deficient ESCs ( Figure 2C and Figure 2—figure supplement 2D ) . ESCs were also stimulated with exogenous IFN-β and confirmed that mouse ESCs retain the ability to respond to external IFNs , and , importantly , that miRNA deficiency did not alter ISG expression levels , supporting the hypothesis that the miRNA-mediated silencing of the IFN pathway in ESCs occurs upstream of IFN production ( Figure 2D ) . To verify that the observed results are solely due to the absence of miRNAs , we rescued the knock-out cell lines by reintroducing Dgcr8 and Dicer and observed that these reverted to wild-type viral replication and susceptibility levels ( Figure 2E , F and Figure 2—figure supplement 2E ) . As a control , we confirmed rescue of miRNA production by Northern blot ( Figure 2E , F ) . To understand where the IFN pathway is silenced in ESCs we blocked the interferon response at defined points in the pathway and measured viral susceptibility . The inhibitor BX795 blocks TBK1/IKKε phosphorylation and consequently IRF3 transcriptional activity , whereas BMS345541 is an inhibitor of the catalytic subunits of IKK and thus blocks Nf-κB-driven transcription . Both transcription factors are essential for the expression of Ifnb1 and other pro-inflammatory cytokines and initiation of an antiviral response ( Lawrence , 2009; Schafer et al . , 1998 ) . Both inhibitors increased viral susceptibility in wild-type cells lines , however , the effect was far greater in the knock-out cell lines ( Figure 3A , B and Figure 3—figure supplement 1A–C ) , suggesting that miRNAs regulate the interferon pathway upstream Ifnb1 transcription . We next aimed to identify the mechanism by which miRNAs silence IFN expression in ESCs , and analysed the proteomes of Dgcr8-/- and the rescued cell line by mass spectrometry . STRING analyses of the expression profiles revealed significant differences in a number of pathways , including ribosome structure/function , mitochondrial activity and the oxidative phosphorylation pathway , which were downregulated in the absence of miRNAs ( Figure 3C , for complete list see Figure 3—source data 1 ) . Measurement of Rhodamine 123 uptake in mitochondria , as an indirect measure for oxidative phosphorylation activity ( Scaduto and Grotyohann , 1999 ) , confirmed lower oxidative phosphorylation activity in the absence of miRNAs ( Dgcr8-/- and Dicer-/- ) ( Figure 3—figure supplement 1D ) . A search for differentially expressed proteins involved in the IFN response did not reveal any significant changes except for the Mitochondrial antiviral-signalling protein ( MAVS ) , which in contrast to many other mitochondria-related proteins , was upregulated in the absence of miRNAs . This protein has a central role in the RLR-induced ( Rig-I-like receptors ) IFN pathway , where activated MDA5 and RIG-I receptors translocate to the mitochondria and bind MAVS to ultimately induce Ifnb1 expression ( Kawai et al . , 2005 ) . Western blot and qRT-PCR analysis confirmed that MAVS was the only factor consistently expressed to higher levels in both miRNA-deficient cell lines , Dgcr8-/- and Dicer-/- ( Figure 3D , lanes 2 and 5 , and Figure 3—figure supplement 1E ) , compared to a panel of other components of the same innate immune response pathway ( Figure 3—figure supplement 1F ) . To confirm the involvement of miRNAs on MAVS expression , a dual luciferase assay system was used where the 3’UTRs of Mavs , Mda5 and Rig-I were fused to a luciferase reporter gene to compare luciferase activity in wild-type and knock-out ESCs . Only the Mavs 3’UTR showed relatively higher luciferase expression levels in the knock-out lines when compared to the empty plasmid , suggesting that the 3’UTR of Mavs is strongly regulated by miRNAs in ESCs ( Figure 4A ) . For this reason , a miRNA-resistant form of Mavs , lacking its natural 3'UTR , was overexpressed in wild-type ESCs and infected with TMEV to test if cells regain viral resistance similar to miRNA deficient ESCs ( Figure 4B ) . A 15-fold decrease in TCID50 and significant reduction in vRNA levels were found compared to wild-type ESCs ( Figure 4C ) . MAVS overexpressing cells also regained the ability to produce Ifnb1 after stimulation with poly ( I:C ) ( Figure 4D ) . All these experiments show that MAVS is a crucial target for the absence of the IFN response in ESCs . We next aimed to identify the miRNA ( s ) responsible for the regulation of MAVS in ESCs and selected a number of miRNA candidates based on literature , prediction software and public miRNA expression databases for further investigations . Previous experimental evidence has shown that human MAVS is regulated by miR-125a , miR-125b and miR-22 ( Hsu et al . , 2017; Wan et al . , 2016 ) . However , only miR-125a-5p and miR-125b-5p have conserved binding sites in mouse MAVS . Two additional miRNAs , miR-185–5p and miR-673–5p , were selected based on their DICER and DGCR8-dependent biosynthesis pathway , their high expression levels in mouse ESCs and number of predicted binding sites in the Mavs 3’UTR ( Tang et al . , 2006; Babiarz et al . , 2008 ) . We transfected Dgcr8-/- cells with mimics of these miRNAs and measured Mavs mRNA and protein levels by RT-qPCR and western blot , respectively . Results showed reductions in MAVS protein and mRNA levels for all tested miRNAs ( Figure 5A and Figure 5—figure supplement 1A ) . The infection of miRNA-transfected Dgcr8-/- cells with TMEV resulted in an increase in both susceptibility and viral replication for miR-125a-5p , miR-125b-5p and miR-673–5p , which correlated with the ability of these miRNAs to downregulate MAVS protein levels ( Figure 5B and Figure 5—figure supplement 1B ) . As an alternative approach , Dgcr8+/+ cells were transfected with inhibitors to miRNAs miR-125a-5p , miR-125b-5p and miR-673–5p . Western blot analysis showed a clear increase in MAVS protein expression , especially for anti-miR-673–5p ( Figure 5C ) . Because miR-673–5p showed the largest effect on MAVS protein expression both when depleted and overexpressed , we hypothesize that miR-673 is a crucial miRNA involved on MAVS regulation . We further investigated the role of miR-673–5p in ESCs by creating stable knock-out cell lines for this miRNA by CRISPR/Cas9 . Three cell lines were selected based on the genomic deletion and confirmed undetectable expression of miR-673–5p ( Figure 5—figure supplement 2A , B ) . The absence of miR-673–5p was enough to observe an increase in MAVS expression both at the mRNA and protein levels ( Figure 5D and Figure 5—figure supplement 2C ) . In addition , we measured miR-673 and MAVS expression levels in the mouse fibroblasts cell line , NIH3T3 , which is proficient in producing IFN in response to dsRNA . Mouse fibroblasts had no detectable miR-673–5p , and MAVS protein expression was comparable to miRNA-deficient ESC ( Figure 5D and Figure 5—figure supplement 2B ) , highlighting the correlation of MAVS expression with the ability of cells to activate Ifnb1 expression in response to immunogenic RNA . Next , miR-673-deficient cell lines were tested for TMEV susceptibility , which showed a consistent decrease in virus replication , similar to that observed in the absence of all miRNAs ( Dgcr8-/- ) , suggesting this miRNA is essential in regulating the innate antiviral response in ESCs ( Figure 5E ) . To test the relevance of IFNs on the increased antiviral resistance of miR-673-/- cell lines , we compared their sensitivity to TMEV infections in the presence of the JAK1/2 inhibitor , Ruxolitinib . Whereas inhibition of IFN signalling did not significantly increase the accumulation of viral RNA in wild-type ESCs ( Dgcr8+/+ ) , both miR-673-deficient and Dgcr8-/- ESCs showed a significant increase in viral RNA upon treatment , confirming the role of miR-673 and IFNs in viral susceptibility ( Figure 5F and Figure 5—figure supplement 3A ) . As controls , we confirmed that knocking out miR-673–5p expression did not affect the ability of ESCs to proliferate or induced spontaneous differentiation , which could cause differences in viral susceptibility ( Figure 5—figure supplement 3B–E ) . Interestingly , during ESC differentiation with retinoic acid , expression of miR-673–5p became silenced , confirming previous results obtained with alternative differentiation protocols ( Knelangen et al . , 2011; Zhao et al . , 2014; Hadjimichael et al . , 2016; Yang et al . , 2016 ) , and suggesting that the expression levels of this miRNA negatively correlate with the ability of cells to activate the IFN response ( Figure 5G ) . Collectively , these data show that the IFN response in mouse ESCs is silenced by the post-transcriptional control of Mavs expression by miR-673–5p .
Several studies suggest that the pluripotent state of a cell is incompatible with an active IFN response ( Guo et al . , 2015 ) . Both mouse and human stem cells fail to synthesize IFNs in response to dsRNA ( Wang et al . , 2013; Chen et al . , 2010 ) , implying that this characteristic is acquired during differentiation ( D'Angelo et al . , 2016 ) . Embryonic carcinoma cells , which are still pluripotent , also fail to produce IFNs in response to viral RNA mimics ( Burke et al . , 1978 ) . In agreement , reprogramming of somatic cells to iPSCs ( induced pluripotent stem cells ) leads to a loss of IFN response , suggesting the presence of regulatory mechanisms able to switch this antiviral pathway on or off between the differentiated and pluripotent states ( Chen et al . , 2012 ) . Another feature of pluripotent cells is their attenuated response to exogenous type I IFNs . Mammalian pluripotent stem cells , iPSCs and embryonic carcinoma cells exhibit an attenuated production of ISGs upon type I IFN stimulation ( Hong and Carmichael , 2013; Irudayam et al . , 2015; Wang et al . , 2014; Burke et al . , 1978 ) . Why these activities are supressed is still not understood , but it has been hypothesized that type I IFN stimulation could impair their self-renewal capacity , since these compounds are well-known antiproliferative agents and inducers of cell death ( Bekisz et al . , 2010 ) . Indeed , type I IFNs are capable of inhibiting tumour cell division in vitro and are currently employed as an adjuvant to treat several types of cancers , acting as stimulants of the innate immune cellular response ( Bracci et al . , 2017 ) . Mouse ESCs express low levels of the RNA sensors TLR3 , MDA5 and RIG-I , which could explain their inability to respond to dsRNA although no functional studies support this model so far ( Wang et al . , 2013 ) . Our data shows an alternative scenario in which MAVS is the key factor for controlling the IFN response . The overexpression of a miRNA-resistant form of MAVS in wild-type ESCs is enough to enable dsRNA-mediated IFN activation , suggesting that dsRNA sensing is not a limiting step in the IFN pathway in ESCs . Regulation of MAVS alone proves to be an efficient mechanism to block dsRNA induced IFN expression compared to suppressing individual dsRNA sensors . The observation that miRNAs only suppress RNA-mediated IFN activation , but not the DNA-mediated pathway , leads us to speculate about the reasons for silencing this specific response during pluripotency . Embryonic stem cells , and also earlier stages of embryonic development are characterized by high expression levels of specific retrotransposons ( non-LTR ) and endogenous retroviruses ( LTR ) , which are a hallmark of their pluripotent state . This is in contrast to most somatic cell types that silence their expression ( Yin et al . , 2018 ) . These repetitive elements produce cytoplasmic RNA molecules as an intermediate for mobilisation , which can be accidentally recognised as immunogenic or non-self RNAs , as it has been previously shown for the human non-LTR retroelement Alu in the context of Aicardi-Goutires syndrome or for endogenous retroviruses ( Ahmad et al . , 2018; Chiappinelli et al . , 2015; Roulois et al . , 2015 ) . Therefore , silencing the RNA-mediated IFN response during pluripotency would act as a protective mechanism for aberrant IFN activation by transposon-derived transcripts . Cells that are incapable of activating the RNA-mediated IFN response have developed alternative antiviral defence pathways . The endonuclease DICER can act as an antiviral factor in mouse ESCs by generating antiviral siRNAs ( Maillard et al . , 2013 ) . Detection of antiviral DICER activity is facilitated in the absence of a competent IFN response , such as in the case of pluripotent cells , but also in somatic cells where the type I IFN response has been genetically impaired ( Maillard et al . , 2016 ) . These findings are supported by the observation that in IFN-competent cells , the RNA sensor LGP2 acts as an inhibitor of DICER cleavage activity on dsRNA ( van der Veen et al . , 2018 ) . However , DICER activity has also been reported in other cell lines , independently of their IFN-proficiency capacity ( Li et al . , 2016 ) . Interestingly , when we disrupt Dicer in ESCs , which inherently lack an IFN response and would theoretically render these cells highly sensitive to viral infections , they become more resistant by acquiring an active IFN response . All these results support the presence of extensive cross-talk between the different antiviral strategies , and suggests that cells have developed mechanisms to compensate for the loss of a specific antiviral pathway . Our model shows that MAVS and miR-673 levels are the key factors regulating the IFN response to dsRNAs during pluripotency . Accordingly , overexpressing MAVS or knocking-out this single miRNA in ESCs is enough to enhance their antiviral response . Interestingly , this miRNA is only conserved in rodents , despite human ESCs also suppressing type I IFNs expression ( Hong and Carmichael , 2013 ) . This suggests that either other miRNAs regulate MAVS expression in human ESCs , or alternative mechanisms operate to silence IFN . Interestingly , human and mouse ESCs have been suggested to constitutively express a subset of ISGs to protect them from viruses ( Wu et al . , 2018 ) . Our proteomics data suggest that , from all the ISGs detected , miRNAs did not significantly affect production of these antiviral factors , such as IFITM1 , IFTIM2 , IFITM3 amongst others . We have shown that engineering ESCs to acquire a functional IFN response significantly increases their antiviral immunity , highlighting the powerful antiviral effects of IFNs even during pluripotency . Previous findings also support a general role for DICER and miRNAs acting as negative regulators of the IFN response in human and mouse models outside pluripotency ( Papadopoulou et al . , 2012; Witteveldt et al . , 2018 ) . In agreement , an indirect approach to deplete cellular miRNAs , by overexpressing the viral protein VP55 from Vaccinia virus , showed that miRNAs are also relevant to control the expression of pro-inflammatory cytokines during viral chronic infections , but not in the acute antiviral response ( Aguado et al . , 2015 ) . However , the concept of miRNAs acting as direct antiviral factors is still controversial . It is relevant to mention that some of the results leading to this conclusion have been primarily generated in DICER1-/- HEK293T human cell line ( Bogerd et al . , 2014; Tsai et al . , 2018 ) , which has an attenuated IFN response due to low PRRs expression ( Rice et al . , 2014; Witteveldt et al . , 2018 ) . We have shown that overexpression of MAVS or silencing specific miRNAs in a transient or stable manner improves the antiviral response of ESCs . These findings are the basis to further study the conservation of the miRNA-mediated regulation of the IFN response in somatic cells and in the context of human pluripotency . All these investigations will provide a deeper understanding and tool set on how to enhance the innate immunity of ESCs and their differentiated progeny , an especially relevant aspect in clinical applications .
Dgcr8 knockout ( Dgcr8-/- ) mouse ESCs were purchased from Novus Biologicals ( NBA1-19349 ) and the parental strain , v6 . 5 ( Dgcr8+/+ , also named in the text ESC1 ) from ThermoFisher ( MES1402 ) . Dicer flox/flox ( Dicer+/+ , also named ESC2 ) and Dicer knockout ( Dicer-/- ) mouse ESCs were provided by R . Blelloch lab ( University of California , San Francisco ) . All mESC cells were cultured in Dulbecco’s modified Eagle Medium ( DMEM , ThermoFisher ) supplemented with 15% heat-inactivated foetal calf serum ( ThermoFisher ) , 100 U/ml penicillin , 100 µg/ml streptomycin ( ThermoFisher ) , 1X Minimal essential amino acids ( ThermoFisher ) , 2 mM L-glutamine , 103 U/ml of LIF ( Stemcell Technologies ) and 50 µM 2-mercaptoethanol ( ThermoFisher ) . Cells were grown on plates coated with 0 . 1% Gelatine ( Embryomax , Millipore ) , detached using 0 . 05% Trypsin ( ThermoFisher ) and incubated at 5% CO2 at 37°C . MDCK , BHK-21 , BV-2 and RAW264 . 7 cells were cultured in Dulbecco’s modified Eagle Medium ( DMEM , ThermoFisher ) supplemented with 10% heat-inactivated foetal calf serum ( ThermoFisher ) , 100 U/ml penicillin , 100 µg/ml streptomycin ( ThermoFisher ) , 2 mM L-glutamine and incubated at 5% CO2 at 37°C . NIH3T3 cell line was provided by A . Buck lab , and grown in DMEM supplemented with 10% FCS . To determine cell proliferation kinetics , cells were seeded in 6-well plates at a density of 2*105 cells/well . Cell numbers were determined in triplicates after 12 , 24 , 36 and 48 hr using a CASY cell counter . Stocks of TMEV strain GDVII were grown on BHK-21 cells and frozen in aliquots at −80°C . Stocks of Influenza A virus strain PR8 ( kindly provided by P . Digard , University of Edinburgh ) were grown on MDCK cells in the absence of serum and in the presence of 2 µg/ml TPCK-treated trypsin and frozen in aliquots at −80°C . For TMEV infections , cells were infected for 1 hr with the required dilution , followed by replacement with fresh medium and incubation for the desired time . For the 50% Tissue Culture Infective dose ( TCID50 ) assays , seven serial dilutions of TMEV were prepared and at least six wells ( in 96-well format ) per dilution were infected and incubated for at least 24 hr before counting infected wells . TCID50 values were calculated using the Spearman and Kärber algorithm . Influenza A virus infections were performed by infecting cells in the absence of serum for 45 min with the addition of 2 µg/ml TPCK-treated trypsin . After replacement of the inoculum with fresh serum containing medium the cells were incubated for the desired period . To differentiate mESCs , they were first cultured as hanging droplets to induce embryoid body formation . For this , a single-cell suspension of 5 × 105 cells/ml was prepared in medium without LIF and 20 µl drops are pipetted on the inside of the lid of a 10 cm petri dish and hung upside-down . The petri-dish was filled with PBS to prevent drying of the hanging drops and incubated at 37°C , 5% CO2 for 48 hr . The embryoid bodies were consequently washed from the lids and transferred to petri dishes to further differentiate , all in the absence of LIF . After another incubation time of 48 hr , medium was removed and replaced with fresh medium containing 250 nM of retinoic acid ( Sigma-Aldrich ) and incubated for 7 days while replacing the medium every 48 hr . After this incubation time , the embryoid bodies were collected and plated on normal gelatine-coated cell culture plates which allowed the embryoid bodies to adhere to the plastic and the cells to migrate from the embryoid bodies . Again , the medium was refreshed every 48 hr for the cells to further differentiate . Total RNA ( 15 µg ) was loaded on a 10% TBE-UREA gel . After electrophoresis , gel was stained with SYBR gold for visualization of equal loading . Gel was transferred onto a positively charged Nylon membrane for 1 hr at 250 mA . After UV-crosslinking , the membrane was pre-hybridized for 4 hr at 40°C in 1xSSC , 1%SDS ( w/v ) and 100 mg/ml single-stranded DNA ( Sigma ) . Radioactively labelled probes corresponding to the highly expressed ESCs miRNAs miR-130–3 p , miR-293–3 p , and miR-294–3 p were synthesized using the mirVana miRNA Probe Construction Kit ( Ambion ) and hybridized overnight in 1xSSC , 1%SDS ( w/v ) and 100 mg/ml ssDNA . After hybridization , membranes were washed four times at 40°C in 0 . 2xSSC and 0 . 2%SDS ( w/v ) for 30 min each . Blots were analysed using a PhosphorImager ( Molecular Dynamics ) and ImageQuant TL software for quantification . Oligonucleotides used are listed in Supplementary file 1 . To activate the IFN response , cells were transfected with either the dsRNA analogue poly ( I:C ) ( Invivogen ) or the Y-shaped-DNA cGAS agonist ( G3-YSD , Invivogen ) using Lipofectamine 2000 ( ThermoFisher ) . Transfections were performed in 24-well format , with cells approximately 80% confluent , using different concentrations of poly ( I:C ) , from 0 , 5 to 2 , 5 µg per well ( as indicated in the figures ) or 0 . 5 µg of G3-YSD . Cells were incubated for approximately 16 hr for poly ( I:C ) - and 8 hr for DNA-transfections before harvest and further processing . IFN-β expression was measured using a quantitative ELISA kit ( Mouse IFN-β , Quantikine , R and D systems ) according to manufacturer’s instructions . Cells were transfected with 2 . 5 µg/ml poly ( I:C ) , incubated for 16 hr after which supernatant was collected and assayed for IFN-β . To activate ESCs with exogenous IFN-β ( R and D systems ) , cells were incubated with 10 . 000 U/ml of IFN-β for 4 hr , followed by RNA extraction and quantitative RT-PCR . For the miRNA mimics ( miScript , Qiagen ) a final concentration of 1 μM was transfected into cells using Dharmafect ( Dharmacon ) , incubated for the desired period and further processed . The same procedure was followed for the antagomirs ( Dharmacon ) , but at a concentration of 100 nM . All experiments were performed in 24-well format , with cells at approximately 80% confluency . Total RNA from cells was isolated using Tri reagent ( Sigma-Aldrich ) according to the manufacturer’s instructions . 0 . 5–1 µg RNA was subsequently reverse transcribed using M-MLV ( Promega ) and random hexamers , and used for quantitative PCR in a StepOnePlus real-time PCR machine ( ThermoFisher ) using GoTaq master mix ( Promega ) . Data was analysed using the StepOne software package . Oligonucleotides used are listed in Supplementary file 1 . Cells used for Western blot analysis were lysed in RIPA buffer ( 50 mM TRIS-HCl , pH 7 . 4 , 1% triton X-100 , 0 . 5% Na-deoxycholate , 0 . 1% SDS , 150 mM NaCl , protease inhibitor cocktail ( Roche ) , 5 mM NaF , 0 . 2 mM Sodium orthovanadate ) . Lysates were spun and protein concentrations measured using a BCA protein assay kit ( BioVision ) . After adjusting protein concentrations , lysates were mixed with reducing agent ( Novex , ThermoFisher ) and LDS sample buffer ( Novex , ThermoFisher ) and boiled at 70°C for 10 min before loading on pre-made gels ( NuPAGE , ThermoFisher ) . Proteins were transferred to nitrocellulose membrane using semi-dry transfer ( iBlot2 , ThermoFisher ) . Membranes were blocked for 1 hr at room temperature in PBS-T ( 0 . 1% Tween-20 ) and 5% milk powder before overnight incubation at 4°C with primary antibody . Antibodies used were: Anti-rabbit HRP ( Cell Signaling Technology ) , Anti-mouse HRP ( Bio-Rad ) , MAVS ( E-6 , Santa Cruz Biotechnology ) , PKR ( ab45427 , Abcam ) , MDA5 ( D74E4 , Cell Signaling Technology ) , RIG-I ( D12G6 , Cell Signaling Technology ) , phospho-IRF-3 ( D601M , Cell Signaling Technology ) and α-tubulin ( CP06 , Merck ) . Proteins bands were visualised using ECL ( Pierce ) on a Bio-Rad ChemiDoc imaging system . Protein bands were quantified using ImageJ ( v1 . 51p ) software and expression levels calculated normalized to α-tubulin . The 3’UTRs from Mda5 , Rig-I and Mavs were amplified from genomic DNA based on the annotation from UTRdb ( utrdb . ba . itb . cnr . it ) using primers containing restriction sites . The fragments were cloned in the psiCHECK-2 vector ( Promega ) at the 3’ end of the hRluc gene . Cells in 24-well format were transfected with 250 ng plasmid using Lipofectamine 2000 and incubated for 24 hr . Cells were subsequently lysed and assayed using the Dual-Glo Luciferase assay system ( Promega ) . Luminescence was measured in a Varioskan flash ( ThermoFisher ) platereader . For the total proteome comparison , 6 replicates of the Dgcr8-/- and Dgcr8resc cell lines were prepared by lysing cells in Lysis buffer ( 50 mM TRIS-HCl , pH 7 . 4 , 1% triton X-100 , 0 . 5% Na-deoxycholate , 0 . 1% SDS , 150 mM NaCl , protease inhibitor cocktail ( Roche ) , 5 mM NaF and 0 . 2 mM Sodium orthovanadate ) at 4°C . Samples were subsequently sonicated 4 × 10 s , at 2μ amplitude , reduced by boiling with 10 mM DTT and centrifuged . The samples were further processed by Filter-aided sample preparation ( FASP ) by mixing each sample with 200 µl UA ( 8M Urea , 0 . 1 M Tris/HCl pH 8 . 5 ) in a Vivacon 500 filter column ( 30 kDa cut off , Sartorius VN01H22 ) , centrifuged at 14 . 000 x g and washed twice with 200 µl UA . To alkylate the sample , 100 µl 50 mM iodoacetamide in UA was applied to the columns and incubated in the dark for 30 min , spun , followed by two washes with UA and another two washes with 50 mM ammonium bicarbonate . The samples were trypsinized on the column by the addition of 4 µg trypsin ( ThermoFisher ) in 40 µl 50 mM ammonium bicarbonate to the filter . Samples were incubated overnight in a wet chamber at 37°C and acidified by the addition of 5 µl 10% trifluoroacetic acid ( TFA ) . The pH was checked by spotting onto pH paper , and peptide concentration estimated using a NanoDrop . C18 Stage tips were activated using 20 µl of methanol , equilibrated with 100 µl 0 . 1% TFA ) and loaded with 10 µg peptide solution . After washing with 100 uL 0 . 1% TFA , the bound peptides were eluted into a Protein LoBind 1 . 5 mL tube ( Eppendorf ) with 20 µl 80% acetonitrile , 0 . 1% TFA and concentrated to less than 4 µl in a vacuum concentrator . The final volume was adjusted to 6 µl with 0 . 1% TFA . Five µg of peptides were injected onto a C18 packed emitter and eluted over a gradient of 2–80% ACN in 120 min , with 0 . 1% TFA throughout on a Dionex RSLnano . Eluting peptides were ionised at +2 kV before data-dependent analysis on a Thermo Q-Exactive Plus . MS1 was acquired with mz range 300–1650 and resolution 70 , 000 , and top 12 ions were selected for fragmentation with normalised collision energy of 26 , and an exclusion window of 30 s . MS2 were collected with resolution 17 , 500 . The AGC targets for MS1 and MS2 were 3e6 and 5e4 respectively , and all spectra were acquired with one microscan and without lockmass . Finally , the data were analysed using MaxQuant ( v 1 . 5 . 7 . 4 ) in conjunction with uniprot fasta database 2017_02 , with match between runs ( MS/MS not required ) , LFQ with one peptide required . Average expression levels were calculated for each protein and significant differences identified using a two tailed t-test assuming equal variance ( homoscedasticity ) with a p-value lower than 0 . 05 . Plasmids containing the sequence of mouse DICER ( pCAGEN-SBP-DICER1 , Addgene ) , MAVS ( GE-healthcare , MMM1013-202764911 ) and DGCR8 ( Macias et al . , 2012 ) were used to amplify the open reading frame using specific primers containing restriction sites ( Supplementary file 1 ) . The amplified and digested fragments were ligated in pLenti-GIII-EF1α for MAVS and pEF1α-IRES-dsRED-Express2 for DGCR8 and DICER . Verified plasmids containing the genes of interest were transfected in mESCs using Lipofectamine 2000 and selected with the appropriate antibiotic . After several weeks of selection , colonies were isolated , expanded and tested for expression by qRT-PCR and Western blot . The mitochondria specific dye Rhodamine 123 ( Sigma-Aldrich ) was used to measure mitochondrial activity . Suspended cells were incubated with Rhodamine 123 at 37°C and samples were taken at various intervals , washed three times with PBS at 4°C and the fluorescence measured in a VarioSkan flash ( ThermoFisher ) plate reader ( excitation 508 , emission 535 ) . Cells were pre-incubated with the inhibitors BX795 , which blocks the phosphorylation of the kinases TBK1 and IKKε , and consequently IRF3 activation and IFN-β production ( 10 µM , Synkinase ) and the inhibitor BMS345541 , which targets IKβα , IKKα and IKKβ and consequently NF-κβ signalling ( 10 µM , Cayman Chemical ) for 45 min before infection with TMEV . 24 hr post-infection in the presence of the inhibitor , infected wells were scored and the TCID50 calculated . For Ruxolitinib ( Cell Guidance Systems ) , cells were pre-incubated for 45 min with 50 µM Ruxotlitinib , infected with TMEV and incubated for 16 hr followed by extensive washing with PBS , RNA extraction and analysis by quantitative RT-PCR . To create a cell line lacking mmu-mmiR-673–5p , the Alt-R CRISPR-Cas9 System ( IDT ) was used . Two different crRNAs were designed to target sequences within the pri-miRNA sequence hairpin to induce structural changes disrupting processing by the Microprocessor and DICER . Cas9 protein and tracrRNAs were transfected with the Neon Transfection System followed by cell sorting to create single cell clones . Genomic DNA was purified and screened by PCR followed by restriction site disruption analyses for the pri-miRNA sequence . Genomic DNA of the pri-miRNA sequence of candidates was amplified using primers in Supplementary file 1 , and cloned into pGEMt-easy vector for sequencing . Total RNA ( 100 ng ) was used to quantify mmu-mmiR-673–5p levels . RNA was first converted to cDNA using miRCURY LNA RT kit ( Qiagen ) . cDNA was diluted 1/25 for RT-qPCR using miRCURY LNA SYBR Green kit and amplified using mmu-mmiR-673–5p specific primers ( Qiagen ) and U6 as a loading control . Quantitative PCR was carried out on a Roche LC480 light cycler and analysed using the second derivative method . All processed Mass spectrometry data is provided as Figure 3—source data 1 , including LFQ intensity values for each protein detected in each of the samples . All raw data are available from corresponding author upon request . | Living cells are under constant attack from disease-causing agents , such as viruses and bacteria . As a result , they have evolved various protective mechanisms to fight off these agents . One of the most important ways that an animal cell protects itself from infection is through the interferon response , which warns the cell of approaching viruses , prompting it to prepare to defend itself . Virtually all healthy cells have an active interferon response , except for stem cells , which have switched off this defensive mechanism , for unknown reasons . This makes stem cells more susceptible to infections . Stem cells are specialized cells that play an essential role in developing the early embryo . The two defining characteristics of these cells – their ability to divide indefinitely , and develop into all cell types – offers great therapeutic potential , as they can be used to ‘replace’ damaged cells and tissues . However , without an interferon response , stem cells are likely to become infected when moved into a new environment , counteracting their therapeutic benefits . Now , Witteveldt et al . investigate how stem cells turn off this viral defence mechanism , and whether turning it back on will affect their ability to divide and form new tissues . Using stem cells taken from the embryos of mice , Witteveldt et al . found that the interferon response is turned off by specific small molecules of RNA . These small RNA molecules block a protein in the pathway that recognizes viruses and activates a defence . Genetically engineering stem cells to be deficient in these small RNA molecules led to an increased resistance to viral infections . Importantly , modifying stem cells in this manner had no obvious impact on the characteristic traits that give stem cells their therapeutic potential . Temporarily increasing the interferon response of stem cells as they are moved into a new environment could potentially make stem cell treatments more effective . However , more work is needed to investigate whether the same approach can be applied to human cells , and determine what negative effects may be associated with turning on the interferon response . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2019 | MicroRNA-deficient mouse embryonic stem cells acquire a functional interferon response |
Sodium/proton antiporters are essential for sodium and pH homeostasis and play a major role in human health and disease . We determined the structures of the archaeal sodium/proton antiporter MjNhaP1 in two complementary states . The inward-open state was obtained by x-ray crystallography in the presence of sodium at pH 8 , where the transporter is highly active . The outward-open state was obtained by electron crystallography without sodium at pH 4 , where MjNhaP1 is inactive . Comparison of both structures reveals a 7° tilt of the 6 helix bundle . 22Na+ uptake measurements indicate non-cooperative transport with an activity maximum at pH 7 . 5 . We conclude that binding of a Na+ ion from the outside induces helix movements that close the extracellular cavity , open the cytoplasmic funnel , and result in a ∼5 Å vertical relocation of the ion binding site to release the substrate ion into the cytoplasm .
Na+/H+ antiporters are essential secondary-active transporters of the cation-proton antiporter ( CPA ) family ( Brett et al . , 2005 ) . CPA antiporters are conserved across all biological kingdoms and play crucial roles in pH , ion and volume homeostasis ( Padan , 2013 ) . The CPA1 branch of the family includes the archaeal NhaP antiporters from Methanocaldococcus jannaschii ( MjNhaP1 ) and Pyrococcus abyssii ( PaNhaP ) and the medically important human NHE sodium proton exchangers ( Donowitz et al . , 2013; Fuster and Alexander , 2014 ) . CPA1 antiporters are electroneutral and exchange one Na+ against one H+ ( Calinescu et al . , 2014; Wöhlert et al . , 2014 ) . CPA2 antiporters , including EcNhaA from E . coli and TtNapA from Thermus thermophilus are electrogenic , exchanging one Na+ against two H+ ( Lee et al . , 2013a; Taglicht et al . , 1993 ) . Previous electron crystallographic studies have shown the structure of the MjNhaP1 dimer in the membrane at 7 Å resolution ( Vinothkumar et al . , 2005; Goswami et al . , 2011 ) and revealed substrate-induced conformational changes within the range of physiological Na+ concentrations and pH . MjNhaP1 shares significant sequence homology of functionally important regions with the mammalian NHEs ( Goswami et al . , 2011 ) . MjNhaP1 and NHE1 are both thought to use a sodium gradient to maintain the intracellular pH by expelling protons from the cell ( Lee et al . , 2013b; Paulino and Kühlbrandt , 2014 ) , but the mechanism by which this happens has remained unknown .
MjNhaP1 has 13 transmembrane helices ( TMH ) , referred to as H1-13 . The N-terminal H1 is essential for transport activity ( Goswami et al . , 2011 ) , but its orientation in the membrane has not been determined experimentally . Comparison to the EcNhaA structure predicts the cytoplasmic location of the MjNhaP1 C-terminus . We performed GFP/PhoA activity assays with MjNhaP1 expressed in E . coli , which indicated that the C-terminus is indeed on the cytoplasmic side ( Figure 1 ) . Because there are 13 membrane spans , the N-terminus is on the extracellular side . 10 . 7554/eLife . 03583 . 003Figure 1 . Membrane topology of MjNhaP1 . ( A ) PhoA activity assay . Enzymatic activity of a C-terminal MjNhaP1 fusion construct with alkaline phosphatase expressed in the PhoA-deficient E . coli strain CC118 was measured spectroscopically . PhoA is active only in the periplasm . ( + ) PhoA fused to the periplasmic C-terminus of YiaD or to the cytoplasmic C-terminus of YedZ were used as positive ( + ) or negative ( − ) controls . Western blot analysis ( inset ) shows that constructs were expressed at comparable levels . ( B ) GFP assay . Normalized whole-cell fluorescence of C-terminal MjNhaP1-GFP-fusion constructs expressed in BL21 ( DE3 ) cells . GFP is fluorescent only in the cytoplasm . EcNhaA , which has a cytoplasmic C-terminus , was used as positive control and untransformed BL21 ( DE3 ) cells as negative control . In-gel florescence of the constructs is shown in the inset . The activity of the positive control was set to 100% . The low activity of the PhoA construct , together with the fluorescence of the GFP fusion construct indicates that the C-terminus of MjNhaP1 expressed in E . coli is on the cytoplasmic side . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 003 The structure of MjNhaP1 was determined by molecular replacement with the related PaNhaP ( Wöhlert et al . , 2014 ) ( pdb 4cz8 ) , using crystals grown with 100 mM NaCl at pH 8 . The two dimers in the asymmetric unit were refined to 3 . 5 Å resolution ( Table 1 ) . The 13 TMHs in the monomer are arranged into a 6-helix bundle and a row of seven helices at the dimer interface , as in PaNhaP , but the two structures differ in important details . Seen from above or below the membrane , the MjNhaP1 dimer is oval ( Figure 2A , Figure 2—figure supplement 1A ) rather than rectangular . Seen from the side , it is about 10 Å shorter than PaNhaP ( Figure 2B , Figure 2—figure supplement 1B ) . Its cytoplasmic surface is flat and does not extend more than 10 Å above the membrane . The cytoplasmic cavity protrudes only 13 Å into the interior of the protomer . On the extracellular side , a negatively charged funnel , which is considerably wider and deeper than in PaNhaP , extends 15 Å towards the center of the protomer ( Figure 2—figure supplement 2 ) . A short loop on the extracellular side of MjNhaP1 connects H6 to H7 and a short amphipathic helix connects H12 to H13 above the center of the helix bundle , whereas in PaNhaP , H6 and H7 are connected by a helix and H12 and H13 by a short loop ( Video 1 ) . 10 . 7554/eLife . 03583 . 004Table 1 . X-ray crystallographic dataDOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 004NativeData collection Wavelength0 . 976 Space groupP21 Cell dimensions a , b , c ( Å ) 98 . 4 , 102 . 5 , 132 . 1 α , β , γ ( ° ) 90 . 0 , 105 . 6 , 90 . 0 Resolution ( Å ) 31 . 93–3 . 5 ( 3 . 72–3 . 5 ) Rpim0 . 086 ( 0 . 573 ) I / σI6 . 0 ( 1 . 7 ) CC*0 . 999 ( 0 . 918 ) Completeness ( % ) 99 . 3 ( 99 . 2 ) Multiplicity7 . 0 ( 7 . 2 ) Refinement Resolution ( Å ) 31 . 93–3 . 5 ( 3 . 72–3 . 5 ) Unique reflections58 , 249 Reflections in test set3115 Rwork/Rfree ( % ) 25 . 2/30 . 2 ( 34 . 4/39 . 2 ) CC ( work ) /CC ( free ) 0 . 905/0 . 930 ( 0 . 799/0 . 645 ) Wilson B-Factor ( Å2 ) 141 Atoms in asymmetric unit12 , 548 Protein12 , 535 Ligands13 r . m . s . deviations: Bond lengths ( Å ) 0 . 003 Bond angles ( ° ) 0 . 78510 . 7554/eLife . 03583 . 005Figure 2 . X-ray structure of MjNhaP1 . ( A ) The MjNhaP1 dimer seen from the cytoplasmic side . Helices H1 to 13 are color-coded and numbered . In one protomer , only the partly unwound helices H5 and 12 are colored . ( B ) Side view with the N-terminus of H1 on the extracellular side . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 00510 . 7554/eLife . 03583 . 006Figure 2—figure supplement 1 . X-ray structure of MjNhaP1 . Cartoon representation of MjNhaP1 with helices shown as cylinders . ( A ) Cytoplasmic and ( B ) side view of the dimer , colour-coded as in the main figure . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 00610 . 7554/eLife . 03583 . 007Figure 2—figure supplement 2 . Cavities in MjNhaP1 at pH 8 . Side view of the dimer with protomer A shown in grey and protomer B in colour . The extracellular and cytoplasmic funnel and the enclosed cavity are colored by surface potential . The hydrophobic cavity between both protomers is shown in olive . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 00710 . 7554/eLife . 03583 . 008Figure 2—figure supplement 3 . Na+/H+ antiport activity of H1 truncation mutants . Activity depends strongly on the first 15 residues in the MjNhaP1 sequence and is lost completely when H1 is deleted . Constructs were expressed in the pTrcHis2TOPO plasmid in KNabc cells , and measurements were performed with everted vesicles at pH 6 . Activities are expressed as percentages of fluorescence relative to the mean of at least four independent experiments . The apparently higher activity of the first two mutants is explained by differences in expression levels ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 00810 . 7554/eLife . 03583 . 009Figure 2—figure supplement 4 . Sequence alignment of CPA antiporters . Comparison of 13 archaeal NhaP antiporters ( blue ) , one fungal NHX plus six mammalian NHE exchangers ( purple ) , and six bacterial CPA2 antiporters ( green ) . The two unwound regions in H5 and H12 , and the ND/DD motif in H6 are boxed . Black diamonds mark residues in the PaNhaP ion-binding site . The partly ( ▽ ) or strictly ( ▼ ) conserved CPA1 residues D93 , R320 and R347 affect transport but do not participate directly in ion binding . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 00910 . 7554/eLife . 03583 . 010Video 1 . Comparison of the x-ray structures of MjNhaP1 and PaNhaP . One protomer of MjNhaP1 ( colored ) is superposed on one protomer of PaNhaP ( grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 010 While the 6-helix bundle of MjNhaP1 closely resembles that of PaNhaP , there are significant differences at the dimer interface . H10 of MjNhaP1 is much shorter and does not protrude into the cytoplasm . MjNhaP1 has no equivalent for His292 in H10 , a key residue for transport in PaNhaP ( Wöhlert et al . , 2014 ) . On the cytoplasmic side of the MjNhaP1 dimer interface , a line of ionic and polar residues parallel to the membrane surface maintains tight interactions between protomers . On the extracellular side , dimer interactions are mediated largely by the hydrophobic H1 , although deletion of this helix , which is necessary for function , did not abrogate dimer formation ( Goswami et al . , 2011 ) . Mutational analysis indicates that the first 15 N-terminal residues of H1 are dispensable for activity ( Figure 2—figure supplement 3 ) . On the dimer interface of MjNhaP1 there is a deep , hydrophobic cavity , which , unlike that in PaNhaP , spans nearly the entire thickness of the membrane and is covered on the extracellular side ( Figure 2—figure supplement 2 ) . As in PaNhaP , this cavity contains density indicative of co-purified , flexible lipids , which was however not sufficiently well-defined for building an atomic model . Structural homology to PaNhaP indicates that the substrate ion , which is not resolved in MjNhaP1 , is coordinated by Asp132 and the backbone of Thr131 in the unwound stretch of H5 . Other key residues in the ion-binding site are Ser157 and Asp161 in H6 ( Ser155 and Asp159 in PaNhaP; Figure 3 ) . MjNhaP1 lacks an equivalent of the ion-coordinating Glu73 in H3 of PaNhaP , which is however not essential for transport ( Wöhlert et al . , 2014 ) . Instead , Thr76 in H3 of MjNhaP1 interacts with Glu154 in H6 , which may guide the substrate ion from the binding site to the cytoplasm . Asn160 in H6 is part of the characteristic ND motif in the CPA1 antiporters ( Figure 2—figure supplement 4 , Figure 3—figure supplement 1 , Figure 3—figure supplement 2 ) . Comparison with PaNhaP suggests that Asn160 is unlikely to coordinate the substrate ion directly . Rather , it forms a hydrogen bond to the hydroxyl group of Thr131 , which in PaNhaP does participate in substrate-ion coordination via its main-chain carbonyl . Nevertheless , changing N160 to alanine renders MjNhaP1 inactive ( Figure 4A ) . A mutant in which N160 is exchanged against aspartate has reduced activity but is not electrogenic ( Figure 4B , C ) . 10 . 7554/eLife . 03583 . 011Figure 3 . Detailed views of the MjNhaP1 x-ray structure . ( A ) Cytoplasmic view of the 6-helix bundle with residues involved in substrate-ion binding and transport . The essential N160 in the ND motif points away from the ion-binding site and interacts with Asp93 . Glu154 in H6 interacts with Thr76 in the interface helix H3 . Thr76 replaces the ion-coordinating Glu73 of PaNhaP . The 6-helix bundle interacts via Thr135 in H5C with Arg285 in H10 and Asp31 in H2 . Arg320 forms an ion bridge with Glu156 . ( B ) Stereo view of the 6-helix bundle . The red circle marks the ion-binding site in PaNhaP ( Wöhlert et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 01110 . 7554/eLife . 03583 . 012Figure 3—figure supplement 1 . 6-helix bundle in the CPA1 antiporters MjNhaP1 and PaNhaP . Stereo views of the conserved substrate-binding site in MjNhaP1 , compared to the known binding site in the homologous PaNhaP ( Wöhlert et al . , 2014 ) . The substrate-binding site is located between the 6-helix bundle and the dimer interface helix 3 . In the PaNhaP structure , the substrate ion with a bound water molecule and another water close to the ND-motif ( Asn158 and Asp159 ) were resolved . The arginines and glutamates forming an ion bridge in the 6-helix bundle ( Arg320 and Glu156 in MjNhaP1 , Arg337 and Glu154 in PaNhaP ) are conserved in CPA1 antiporters . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 01210 . 7554/eLife . 03583 . 013Figure 3—figure supplement 2 . 6-helix bundle in the CPA2 antiporters EcNhaA and TtNapA . Stereo views of the putative ion-binding site in the CPA2 antiporters EcNhaA and TtNapA . In the structure of monomeric EcNhaA ( pdb 1ZCD ) , Lys300 points towards helix XII while in the structure of the EcNhaA dimer ( pdb 4AU5 ) , Lys300 forms an ion bridge with Asp163 in the DD motif in helix V . In TtNapA the CPA2-conserved Lys305 forms an ion bridge with Asp156 in the DD motif . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 01310 . 7554/eLife . 03583 . 014Figure 4 . Activity of Asn160 mutants . ( A ) Na+/H+ antiporter activity of wildtype ( WT ) and the N160A mutant measured under asymmetrical conditions in everted vesicles at pH 6 . Constructs were expressed in Na+/H+ antiporter-deficient KNabc cells at comparable levels ( inset ) . Everted vesicles were preloaded with protons by addition of 2 . 5 mM Tris/DL-lactate ( ↓ ) . Transport was initiated by addition of NaCl to 25 mM ( ▲ ) , and proton efflux was monitored by acridine orange fluorescence dequenching . ( B ) Activity of purified wildtype MjNhaP1 ( WT ) and the N160D mutant reconstituted into proteoliposomes under symmetrical pH . ( C ) Activity at pH 6 was unaltered in the presence of valinomycin , demonstrating that N160D is not electrogenic . Asterisks in ( B ) and ( C ) mark the addition of proteoliposomes . The pH gradient was dissipated with 25 mM NH4Cl ( ▽ ) . ( D ) Transport by MjNhaP1 is inhibited by the Myc-His-Tag at pH 8 but not at pH 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 014 Measurements of 22Na+ uptake by wildtype MjNhaP1 reconstituted into proteoliposomes indicate an activity maximum at pH 7 . 5 ( Figure 5A ) . Transport increases by a factor of approximately two from 0 . 94 ions per second per protomer at pH 6 ( Figure 5B ) to 1 . 68 ions per second at pH 8 ( Figure 5C ) at a KmNa+ of 0 . 84 mM . Unlike PaNhaP , MjNhaP1 is not cooperative at any pH tested . Linear plots ( Figure 5—figure supplement 1 ) show that near-saturation is reached around 5 mM NaCl at either pH . 10 . 7554/eLife . 03583 . 015Figure 5 . Transport activity of MjNhaP1 . ( A ) pH profile measured by 22Na uptake with acidic-inside proteoliposomes at 1 . 5 mM NaCl . MjNhaP1 activity is highest between pH 7 and pH 8 and drops to background level at pH 4 and pH 10 . ( B ) 22Na+ transport at pH 6 follows Michaelis–Menten kinetics , with a Km of 1 . 11 mM ± 0 . 20 mM and a vmax of 1226 ± 77 nmol · min−1 · mg−1 . ( C ) At pH 8 , Km drops to 0 . 84 ± 0 . 11 mM and vmax increases to 2187 ± 88 nmol · min−1 · mg−1 ( 1 . 68 s−1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 01510 . 7554/eLife . 03583 . 016Figure 5—figure supplement 1 . Na+-dependent transport activity on a linear scale . The linear plots show that near-saturation is reached around 5 mM NaCl at both pH 6 ( A ) and pH 8 ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 016 3D crystals of MjNhaP1 without sodium or at low pH either failed to grow or were poorly ordered . We therefore determined the structure of the sodium-free state at low pH by electron cryo-crystallography of 2D crystals grown at pH 4 without NaCl ( Paulino and Kühlbrandt , 2014 ) . Amplitudes and phases obtained from 128 images were merged to yield a 3D map ( Table 2 , Figure 6 ) with an in-plane resolution of 6 Å ( Figure 6—figure supplement 1 ) . The unit cell contained two dimers , with one protomer in the asymmetric unit . The MjNhaP1 x-ray structure was fitted manually to the EM map to obtain an atomic model ( Figure 7 ) . A 3D difference map calculated between the experimental 6 Å EM density and the x-ray map truncated to this resolution indicated clear changes in the orientation of key helices ( Figure 8 ) . Of the interface helices , only H10 required a tilt of ∼6° about the helix center . Within the 6-helix bundle , H6 was tilted by ∼14° and H12E by ∼15° . H5C and H5E changed direction by ∼15° and ∼7° , respectively . The 6-helix bundle as a whole tilted by ∼7° towards the dimer interface on the cytoplasmic side and away from the dimer interface on the extracellular side , about an axis in the membrane plane roughly parallel to the dimer interface . On the cytoplasmic side of the EM model , H5C and H6 are closer to the interface than in the x-ray structure and obstruct access to the substrate-binding site . On the extracellular side , the tilt of the 6-helix bundle , especially of H6 and 12E , widens and deepens the exterior funnel . Whereas the x-ray structures of MjNhaP1 and PaNhaP ( Wöhlert et al . , 2014 ) both show the inward-open conformation , the EM structure of MjNhaP1 is closed on the cytoplasmic and open on the extracellular side ( Figure 9 ) . We refer to this conformation as an ‘outward-open’ state of MjNhaP1 . In the transition from inward-open to outward-open , the ion-binding site moves towards the extracellular side by about 5 Å ( Figure 9 ) . 10 . 7554/eLife . 03583 . 017Table 2 . Electron crystallographic dataDOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 017pH 4 , 0 mM NaClUnit cell dimensionsa = 81 . 5 Å , b = 103 . 3 Å , c = 200 Å , γ = 90°Two-sided plane groupp22121Number of images ( tilt angles in brackets ) 15 ( 0° ) , 23 ( 20° ) , 44 ( 30° ) , 46 ( 45° and above ) In-plane resolution6 ÅResolution in z directiona14 ÅDefocus range0 , 12–1 , 8 μmTilt range0–54°Total number of observed reflectionsb47 , 064Observed unique reflections15 , 509Unique reflections in asymmetric unit2686Overall weighted phase residualb12 . 1°Overall weighted R-factorb24 . 8%acalculated from the point spread function of the experimental data . bcalculated with program LATLINEK . a , b Reflections with IQ ≤ 6 Å were included . 10 . 7554/eLife . 03583 . 018Figure 6 . 3D EM map of MjNhaP1 at pH 4 in the absence of Na+ . The map was contoured at 1 . 7σ and a B-factor of −200 Å2 was applied . ( A ) Cytoplasmic view of one dimer . The 6-helix bundle is shaded yellow and the dimer interface green . ( B ) The side view indicates low noise level outside the membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 01810 . 7554/eLife . 03583 . 019Figure 6—figure supplement 1 . Electron crystallographic amplitudes and phases . ( A ) Lattice lines for four representative reflections . The variation of phases ( upper panel ) and amplitudes ( lower panel ) along z* were fitted by weighted least squares . Each blue cross represents one measured reflection . ( B ) IQ plots of single images recorded at the tilt angles indicated . Circles indicate 12 Å , 9 Å , 7 Å and 6 Å resolution . h , k vectors and the tilt axis ( TAXA ) is shown . ( C ) Tilt angle distribution according to ( Cheng and Yeager , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 01910 . 7554/eLife . 03583 . 020Figure 7 . 3D EM structure of MjNhaP1 at pH 4 without sodium . Cytoplasmic ( A ) and side view ( B ) of 3D EM map with one MjNhaP1 protomer fitted . The map was sharpened with a B factor of −200 Å2 and contoured at 1 . 7σ . Connecting helices and loops without EM density were omitted for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 02010 . 7554/eLife . 03583 . 021Figure 8 . 3D Difference maps . Difference densities were calculated between the unsharpened experimental EM map ( blue mesh ) and the x-ray map truncated to 6 Å resolution . Difference maps are shown for H6 ( green ) and the 5C/12E pair of half helices ( yellow/pink ) . For clarity , only negative difference densities are shown ( red mesh ) . The x-ray structure of the inward-open state is grey . Maps were plotted at 2σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 02110 . 7554/eLife . 03583 . 022Figure 9 . Sections through the MjNhaP1 x-ray and EM structures . Sections through the inward-open x-ray structure and the outward-open EM structure of MjNhaP1 . In the x-ray structure ( left ) the ion-binding site ( red circle ) is accessible from the cytoplasm . In the EM structure ( right ) , the ion-binding site has moved upwards by ∼5 Å and is accessible through the extracellular funnel . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 02210 . 7554/eLife . 03583 . 023Figure 9—figure supplement 1 . Comparison of MjNhaP1 and TtNapA . ( A ) Sections through protomer volumes of the MjNhaP1 EM structure ( left ) and the outward-open TtNapA structure ( Lee et al . , 2013 ) ( right ) . ( B ) Superposition of helices H5 , H6 , H12 and H13 in the MjNhaP1 EM structure ( blue ) and corresponding helices in TtNapA ( green ) . Helix positions suggest that TtNapA is more open on the extracellular side than MjNhaP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 023
A projection difference map between the x-ray structure and the EM model calculated at 6 Å resolution ( Figure 10A–C ) indicates significant lateral changes in helix position and orientation in the 6-helix bundle , whereas the dimer interface changes only minimally . It is instructive to compare this difference map to Figure 5 of an earlier paper that describes substrate-ion induced conformational changes in MjNhaP1 ( Paulino and Kühlbrandt , 2014 ) . Figure 10 shows that the positions and relative strength of difference peaks between the inward-open and outward-open state are nearly identical to those that are observed when the 2D crystals of MjNhaP1 are taken from 0 mM NaCl to 500 mM NaCl , either at pH 8 or at pH 4 . Together with the 3D data presented here , this allows us to conclude that an increase in NaCl concentration converts the antiporter from the outward-open conformation in the absence of salt to the inward-open conformation in the presence of salt . 10 . 7554/eLife . 03583 . 024Figure 10 . Projection difference maps . ( A ) Superposition of the three-dimensional MjNhaP1 outward-open structure at pH 4 without sodium ( blue ) and the inward-open structure at pH 8 with sodium ( red ) . Helices are shown as cylinders as seen from the cytoplasmic side . ( B ) 6 Å projection difference map calculated between the structures shown in ( A ) . Major difference peaks are observed in the 6-helix bundle , whereas difference peaks at the dimer interface are weak . ( C ) Superposition of the inward-open and outward-open MjNhaP1 structures on the projection difference map shown in ( B ) . ( D ) 6 Å projection difference map between MjNhaP1 2D crystals at pH 8 with and without sodium ( adapted from Figure 5 in ( Paulino and Kühlbrandt , 2014 ) top row , 500 mM NaCl brought to the same phase origin ) . The projection difference map calculated from the 3D structures closely resembles the projection map that shows sodium-induced changes in Paulino and Kühlbrandt ( 2014 ) . The conformational change from the inward-open to outward-open state of MjNhaP1 is therefore induced by sodium ions . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 024 Comparison of the MjNhaP1 inward-open x-ray structure to the 3 . 45 Å x-ray structure of EcNhaA reveals similarities in the 6-helix bundle ( Goswami et al . , 2011 ) , but clear differences at the dimer interface and its position relative to the 6-helix bundle . The outward-open state of the MjNhaP1 EM structure is confirmed by comparison with the outward-open structure of the CPA2 antiporter TtNapA ( Lee et al . , 2013a ) , which looks strikingly similar , especially with respect to the extracellular funnel ( Figure 9—figure supplement 1A ) , the conformation of H6 and the H5/12 pairs of half helices ( Figure 9—figure supplement 1B ) . The structure of the apical sodium-dependent bile acid symporter ASBT , which surprisingly has the same fold as the sodium/proton antiporters , has been solved in both the inward-open and the outward-open state ( Hu et al . , 2011; Zhou et al . , 2014 ) . Comparison reinforces our conclusion that the x-ray structure of MjNhaP1 shows the inward-open and the EM structure the outward-open state ( Figure 9 ) . In both MjNhaP1 and ASBT , the 6-helix bundle performs a rigid-body rotation around the same axis in the membrane parallel to the dimer interface ( Video 2 ) . The resulting up-and-down movement of the substrate-binding site is more pronounced in ASBT than in MjNhaP1 , as might be necessary to facilitate translocation of the larger bile acid substrate . 10 . 7554/eLife . 03583 . 025Video 2 . Conformational changes in MjNhaP1 and ASBT . Morphing the transition from the outward-open to the inward-open states in MjNhaP1 ( green ) and ASBTYf ( Zhou et al . , 2014 ) ( purple ) reveals a very similar rigid-body movement of the 6-helix bundle relative to the dimer interface in both proteins . Cytoplasmic view ( left ) and side view ( right ) . Structures were superimposed on the dimer interface and intermediate states were calculated using the program LSQMAN . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 025 A different and much larger conformational change has been postulated for TtNapA on the basis of its outward-open x-ray structure and an inward-open state modeled on the dissimilar EcNhaA structure ( Lee et al . , 2013a ) . The inward-open model of TtNapA implied that the 6-helix bundle moves up and down by 10 Å and rotates by 21° about an axis roughly perpendicular to that in MjNhaP1 and ASBT . The similarity of the MjNhaP1 EM structure to the TtNapA x-ray structure suggests however that the inward-open state of TtNapA closely resembles the x-ray structure of MjNhaP1 rather than that of EcNhaA . We conclude that all sodium/proton antiporters undergo essentially the same conformational changes in the course of their transport cycles , as represented here by the two states of MjNhaP1 . The electroneutral CPA1 and electrogenic CPA2 antiporters have different ion transport stoichiometries . CPA2 antiporters , such as EcNhaA and TtNapA , exchange two protons against one Na+ . One of the predicted motifs for the electrogenic transport is the DD motif in helix V in place of the ND motif in H6 of CPA1 antiporters , such as MjNhaP1 and PaNhaP ( Figure 2—figure supplement 4 , Figure 3—figure supplement 1 , Figure 3—figure supplement 2 ) . In CPA2 antiporters the two conserved aspartates have been proposed to each bind one of the translocated protons ( Taglicht et al . , 1991; Hunte et al . , 2005; Arkin et al . , 2007 ) . However , replacing N160 in the ND motif does not render MjNhaP1 electrogenic ( Figure 4C ) . The inactive N160A mutant ( Figure 4A ) shows that this sidechain is important for ion translocation , even though it does not participate in ion coordination directly . The reduced activity of the N160D mutant ( Figure 4B , C ) suggests a possible role in stabilizing the proton or substrate-bound state , which can also be fulfilled by an aspartate . Note that an asparagine in this position renders EcNhaA and TtNapA inactive ( Inoue et al . , 1995; Lee et al . , 2013a ) , probably because it cannot form an ion bridge , as observed for Lys305 and Asp156 in TtNapA ( Lee et al . , 2013a ) , which may be important for protein stability ( Figure 3—figure supplement 2 ) . In MjNhaP1 and PaNhaP the arginine replacing this lysine does not interact with the ND motif but forms an ion bridge to the neighboring conserved glutamate in H6 ( Figure 3A and Figure 3—figure supplement 1 ) , which would stabilize the 6-helix bundle . In terms of its overall structure , TtNapA is more similar to MjNhaP1 and PaNhaP than to EcNhaA ( Hunte et al . , 2005 ) , especially with respect to the relative position of the dimer interface with the seven helices . The tertiary structure of CPA antiporters thus does not correlate with their transport stoichiometry . There appear to be two types , one of which , represented by MjNhaP1 , PaNhaP and TtNapA , is more common than the other type , that seems to be confined to EcNhaA and its close relatives . 22Na+ uptake measurements with MjNhaP1 proteoliposomes acidified by an ammonium sulfate gradient indicated a bell-shaped pH profile with a pH maximum at around pH 7 . 5 . Activity dropped to background levels below pH 4 or above pH 9 . Earlier studies ( Hellmer et al . , 2003; Vinothkumar et al . , 2005; Goswami et al . , 2011 ) had found that MjNhaP1 is active at pH 6 but inactive at pH 7 . 5 or above . This discrepancy is due to the C-terminal affinity tag on the construct that was used in previous transport measurements ( Hellmer et al . , 2003; Goswami et al . , 2011 ) . We repeated the measurements with this tagged construct under symmetrical pH conditions and found that it was indeed inactive at pH 8 but active at pH 6 ( Figure 4D ) . Apparently the affinity tag at the C-terminus of H13 , which is part of the 6-helix bundle , impairs the mobility of the bundle at elevated pH . This movement is an integral feature of the transport mechanism . For reliable functional measurements on flexible , conformationally active membrane transporters it is therefore advisable to use untagged proteins . 22Na+ uptake by untagged MjNhaP1 reconstituted at a high lipid/protein ratio into proteoliposomes acidified by an ammonium gradient also avoids other potential problems associated with the limited pH range of fluorescent dyes , everted vesicles energized by a process that is itself pH-dependent ( Reenstra et al . , 1980 ) , or with leaky proteoliposomes produced at low lipid/protein ratio ( Tsai et al . , 2013 ) . The bell-shaped pH profile of MjNhaP1 and PaNhaP ( Wöhlert et al . , 2014 ) confirms an earlier conclusion that the antiporter has to shut down at acidic or basic pH for physiological reasons ( Vinothkumar et al . , 2005 ) . Our finding that proton-driven Na+ uptake drops at decreasing external pH ( Figure 5A ) is in good agreement with the inverse experiment , which showed that Na+ efflux in MjNhaP1 proteoliposomes increases under these conditions ( Calinescu et al . , 2014 ) . In the Na+ uptake experiment , substrate ions bind less well from the outside at low external pH due to proton competition . Conversely , in the efflux experiment ( Calinescu et al . , 2014 ) , a decrease in external pH has no effect on Na+ binding from the inside . The turnover number of MjNhaP1 was derived from vmax = 2187 ± 88 nmol · min−1 · mg−1 at pH 8 measured at 0°C . At any higher temperature , transport was too fast to be reliably recorded . The measured rate of 1 . 68 ions per second per protomer is more than 400 times higher than the transport rate of PaNhaP extrapolated to 0°C . The difference is most likely due to an extra acidic sidechain ( Glu73 ) in the ion-binding site of PaNhaP , which has no equivalent in MjNhaP1 . Removal of this sidechain in PaNhaP increases the transport rate , because the substrate ion is released more readily ( Wöhlert et al . , 2014 ) . Although compared to EcNhaA ( Taglicht et al . , 1991 ) , neither MjNhaP1 nor PaNhaP are particularly fast at ambient conditions , the activity of PaNhaP increases exponentially with temperature to an extrapolated turnover number of 5000 ions per second at 100°C ( Wöhlert et al . , 2014 ) . Given its similarity to PaNhaP , MjNhaP1 will be very much faster at its physiological temperature of 85°C than at room temperature , and most likely also considerably faster than PaNhaP under physiological conditions . The x-ray structure of MjNhaP1 was determined at pH 8 in the presence of substrate ions , where the antiporter is highly active . The x-ray structures of PaNhaP were determined at pH 4 or pH 8 also in the presence of Na+ . Under both conditions PaNhaP is inactive ( Wöhlert et al . , 2014 ) . The close resemblance of the MjNhaP1 and PaNhaP x-ray structures proves that there is no pH-induced conformational switch to regulate either antiporter . This is in excellent agreement with a recent study ( Paulino and Kühlbrandt , 2014 ) , which indicated that a change in pH in the absence of substrate ions has only minimal effects on the conformation of MjNhaP1 , whereas Na+-binding induces helix movements that are similar in the entire activity range and consistent with the changes in helix orientation described here ( Figure 10 ) . We propose that these considerations hold true for all CPA1 antiporters . Like the mammalian NHE exchangers , the electroneutral Na+/H+ antiporters MjNhaP1 and PaNhaP are thought to work as Na+-driven proton transporters . They maintain an intracellular neutral pH by utilizing the inward Na+ gradient that is always present in their native saline habitat . In PaNhaP , Na+ coordination stabilizes the inward-open state , whereas the apo or proton-bound state of MjNhaP1 adopts the outward-open conformation ( Paulino and Kühlbrandt , 2014 ) . Thus , their default resting state is inward-open with a sodium ion in the binding site . As suggested for EcNhaA ( Mager et al . , 2011 ) and recently shown for MjNhaP1 ( Calinescu et al . , 2014; Paulino and Kühlbrandt , 2014 ) , protons and Na+ compete for a single binding site in the protomer . When the intracellular pH drops , the binding site becomes protonated and the Na+ ion is released into the cytoplasm . The conformational changes we observe are consistent with a rocking bundle mechanism , as proposed for other secondary transporters ( Forrest and Rudnick , 2009 ) . The rocking movement of the 6-helix bundle controls alternating access to the ion-binding site from the outside medium or from the cell interior ( Video 3 ) . 10 . 7554/eLife . 03583 . 026Video 3 . Transport cycle of MjNhaP1 . Substrate-induced conformational changes in MjNhaP1 from the outward-open sodium-free state at pH 4 to the inward-open state at pH 8 in the presence of sodium . The 6-helix bundle rotates by ∼7° with respect to the dimer interface , and the ion-binding site moves by ∼5 Å , resulting in alternating access to the ion-binding site from the cytoplasm or from the extracellular side . Cytoplasmic view ( left ) and side view ( right ) . Structures were superimposed on the dimer interface and intermediate states were calculated using the program LSQMAN . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 026 In summary , the simplest mechanism for MjNhaP1 and other CPA1 antiporters entails the following four steps ( Figure 11 ) . ( i ) In the default resting state , a proton from within the cell replaces the bound Na+ , which is released to the cytoplasm; ( ii ) upon protonation of Asp161 , the antiporter switches to the outward-open state , making the ion-binding site accessible to extracellular Na+; ( iii ) a Na+ ion diffusing to the binding site from the extracellular medium displaces the proton at Asp161 or its equivalent; the deprotonated sidechain engages in Na+ coordination; ( iv ) deprotonation and Na+ binding triggers the switch from the outward-open back to the inward-open state . Na+ is released into the cytoplasm , a proton binds , and the cycle repeats . 10 . 7554/eLife . 03583 . 027Figure 11 . The MjNhaP1 transport cycle . In the outward-open state ( Ce ) , Na+ from the exterior medium gains access to the ion-binding site through the open extracellular funnel , where it replaces a bound proton . Na+ binding triggers the transition from the outward-open to the inward-open , substrate-bound conformation ( CSi ) via a substrate-occluded state ( CSec ) . In the inward-open state , the cytoplasmic funnel widens and the extracellular funnel closes . A cytoplasmic proton releases the bound Na+ to the cell interior . Na+ release triggers the conformational change to the outward-open state via an occluded inward-open proton-bound conformation ( CSic ) , and the cycle repeats . H5 , 6 and 12 are color-coded . The 6-helix bundle ( blue ) performs a ∼7° rocking motion around an axis parallel to the dimer interface ( light brown ) , which remains fixed in the membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 03583 . 027 The most important remaining questions concern the exact molecular events during the transition from the inward-open to the outward-open state , and the underlying energetics . These questions are best addressed by molecular dynamics simulations . Given that its structures are now available in both the inward-open and the outward-open state and its transport cycle may take only 2 µs , MjNhaP1 would be an ideal target .
The relative orientation of the protein in the membrane was determined by GFP/PhoA fusion ( Daley et al . , 2005; Drew et al . , 2002; ter Horst and Lolkema , 2012 ) . For the GFP assay , MjNhaP1 or EcNhaA genes were cloned into the pWaldo plasmid carrying a C-terminal GFP-tag , using the restriction sites XhoI and EcoRI and the following primers: MjNhaP_XhoI_s: 5′-CCGCCGCTCGAGATGGAACTGATGATGGCGATCG-3′ MjNha1_EcoRI_as: 5′-CGGCGGGAATTCATGGTGGCTTTCTTCTTTATATTTCG-3′ EcNhaA_XhoI_s: 5′-CCGCCGCTCGAGGTGAAACATCTGCATCGATTC-3′ EcNhaA_EcoRI_as: 5′-CGGCGGGAATTCAACTGATGGACGCAAACGAAC-3′ BL21 ( DE3 ) cells were transformed and grown at 37°C in 10 ml LB medium with 50 μg/ml kanamycin . Protein expression was induced at an OD600 of ∼0 . 4 by addition of 1 mM IPTG and the cells were harvested at an OD600 of 0 . 5–0 . 6 . The cell pellet was washed 50 mM Tris/HCl pH 8 , 200 mM NaCl and 15 mM EDTA and resuspended in 300 μl of the same buffer . Samples were analyzed by in-gel fluorescence and Western blot analysis using α-His or α-GFP antibodies . 180 μl of whole cell suspension were transferred into 96-well Nunc plates and incubated for 1 . 5 hr . GFP fluorescence was measured at 512 nm with excitation at 485 nm . The mean fluorescence was calculated from three independent measurements and normalized to respective OD600 . Untransformed BL21 ( DE3 ) cells were used as negative control and the EcNhaA constructs as positive controls . For the PhoA assay , MjNhaP1 was cloned into the pHA-1 plasmid carrying a C-terminal PhoA-tag , using the restriction sites XhoI and BsiWI/Acc651 and the following primers: MjNhaP1_XhoI_s: 5′-CCGCCGCTCGAGATGGAACTGATGATGGCGATCG-3′ MjNhaP1_BsiWI_as: 5′-GGCGGCGTACGGCATGGTGGCTTTCTTCTTTATATTT-3′ Plasmids carrying PhoA-fusions constructs of known topology ( pos: YiaD; neg: YedZ ) were kindly provided by Gunnar von Heijne and used as controls . The PhoA deficient E . coli strain CC118 was transformed and grown in 10 ml LB medium with 100 μg/ml ampicillin at 37°C . Protein expression was induced with 0 . 2% arabinose at OD600 = 0 . 15 , cultures were harvested after 2–2 . 5 hr and 1 mM iodoacetamide was added . The optical density for all cultures was measured and normalized . Cell pellets were washed and resuspended in 10 mM Tris/HCl pH 8 and 1 mM iodoacetamide . For the PhoA assay , 0 . 2 ml of the cell suspension were added to 0 . 8 ml 1 M Tris/HCl pH 8 , 0 . 1 mM ZnCl2 . To permeabilize the cells , 50 μl 0 . 1% SDS and 50 μl chloroform were added to the reaction mixture . The cells were incubated 5 min at 37°C and 5 min on ice . The reaction was started by addition of 0 . 1 ml p-nitrophenyl phosphate . After incubation at 37°C for 90 min , the reaction was stopped by addition of 120 μl 1:5 0 . 5 M EDTA pH 8 , 0 . 1 M KH2PO4 . For each sample the absorption at 420 nm and 550 nm was recorded and the mean activity was obtained from four independent measurements . For Western blot analysis , 50 μl sample was centrifuged and cell pellet was resuspended in 20 μl SDS sample buffer . A PhoA antibody was used for fusion-protein detection . For 3D crystallization and 22Na+ uptake measurements MjNhaP1 was cloned into a pET-21a vector with a C-terminal Cysteine Protease Domain ( CPD ) fusion ( Shen et al . , 2009 ) that produces untagged protein . E . coli BL21- ( DE3 ) cells were transformed with the resulting plasmid and MjNhaP1 was expressed at 37°C in ZYM-5052 autoinduction medium ( Studier , 2005 ) . Cultures were harvested and cells were broken with a microfluidizer ( M-110L , Microfluidics Corp . , Westwood , MA ) . Membranes were isolated by centrifugation at 100 , 000×g at 4°C for 1 hr , resuspended in 50 mM Tris/HCl pH 7 . 5 , 140 mM choline chloride , 250 mM sucrose and diluted 1:2 in 150 mM MOPS/KOH pH 7 . 0 , 45% glycerol and ∼2 . 0% Foscholine-12 . After incubation at 4°C for 2 hr the solution was clarified by centrifugation at 125 , 000×g for 1 hr . The supernatant was supplemented with 5 mM imidazole and 150 mM NaCl , incubated for 2 hr with TALON resin ( Clontech , Mountain View , CA ) at 4°C and loaded on a Biorad column . Unspecifically bound protein was eluted with 15 column volumes of 20 mM Bis-Tris pH 7 . 0 , 300 mM NaCl , 10 mM imidazole , 0 . 1% Foscholine-12 , 0 . 24% Cymal-5 and 20 column volumes of 20 mM Bis-Tris pH 6 . 5 , 0 . 2% Cymal-5 , 300 mM NaCl . MjNhaP1 was eluted from the column with 10 column volumes of 20 mM Bis-Tris pH 6 . 5 , 300 mM NaCl , 0 . 2% cymal-5 , 10 µM inositol-hexaphosphate , concentrated to 8 mg/ml using a concentrator with 50 kDa cut-off and dialysed against 25 mM Sodium-Acetate pH 4 . 0 , 100 mM NaCl , 0 . 2% Cymal-5 . For 2D crystallization MjNhaP1 was expressed in the pET26b vector with a C-terminal hexa-histidine tag and purified by Ni-NTA affinity chromatography as described ( Paulino and Kühlbrandt , 2014 ) . To ensure Na+-free conditions the column was washed with 10 column volumes of 15 mM Tris/HCl pH 7 . 5 , 500 mM NaCl , 15 mM imidazole and 0 . 03% dodecyl maltoside ( DDM ) , followed by 8 column volumes of sodium-free buffer ( 15 mM Tris/HCl pH 7 . 5 , 200 mM KCl and 0 . 03% DDM ) . MjNhaP1 was eluted with 50 mM potassium acetate pH 4 , 100 mM KCl , 5 mM MgCl2 and 0 . 03% DDM , concentrated and stored at −80°C . Prior to 3D crystallization , MjNhaP1 was incubated at 85°C for 15 min and centrifuged for 1 hr at 125 , 000×g . The supernatant was passed through a 0 . 1 µm filter ( Ultrafree-MC , Millipore , Billerica , MA ) , supplemented with 5 mM K2Pt ( CN ) 4 and mixed 1:1 with 100 mM Tris/Cl pH 8 . 2 , 24% PEG 1000 . MjNhaP1 crystals grew in hanging drops within 10–14 days to a maximal size of 350 µm and were vitrified directly in liquid nitrogen . Data were collected at the ESRF beamline id23 . 1 , processed with XDS ( Kabsch , 1993 ) and scaled with AIMLESS in the CCP4 package ( Collaborative Computational Project 4 , 1994 ) . Resolution cut-offs were set on the basis of cross correlation between half datasets , completeness and I/σ ( I ) -values in high resolution shells ( Karplus and Diederichs , 2012 ) . COOT ( Emsley and Cowtan , 2004 ) was used for model building and the PHENIX package ( Adams et al . , 2010 ) for refinement . Phases were obtained by molecular replacement with PHASER ( McCoy , 2007 ) using a polyalanine dimer model of the PaNhaP x-ray structure ( 4cz8 ) as search template . After density modification with Parrot ( Zhang et al . , 1997 ) the protein backbone was rebuild into the density-modified map and a run of molecular replacement was started with the new template . This process was repeated several times . After correcting the backbone geometry , side chains were fitted to the electron density , starting with the residues that are conserved between PaNhaP and MjNhaP1 in several rounds of iterative model building and refinement using phenix . refine ( Adams et al . , 2010 ) . 2D crystals of MjNhaP1 were grown with E . coli polar lipids ( Avanti Polar Lipids , Inc . , Alabaster , AL ) at a final protein concentration of 1 mg/ml , a final decyl maltoside concentration of 0 . 15% and a lipid-to-protein ratio ( LPR ) of 0 . 5 ( Paulino and Kühlbrandt , 2014 ) . 2D crystals were grown at 37°C in sodium-free 25 mM K+ acetate pH 4 , 200 mM KCl , 5% glycerol and 5% 2-4-methylpentanediol . EM grids were prepared by the back-injection method ( Wang and Kühlbrandt , 1991 ) in 4% trehalose and rapidly frozen in liquid nitrogen . Images were recorded with an electron dose of 20–30 e−/Å2 on Kodak SO-163 film with a JEOL 3000 SFF electron microscope at a nominal temperature of 4K , an acceleration voltage of 300 kV , a magnification of 53 , 000 at a defocus range of 0 . 1–1 . 8 μm in spot scan mode . Images of tilted crystals were recorded with fixed tilt angle cryo holders . Lattice images were screened by optical diffraction , and well-ordered areas of 4k × 4k or 6k × 6k pixels were digitized at 7 μm step size on a Zeiss SCAI scanner . Images were processed by the 2dx software package ( Gipson et al . , 2007 ) . A total of 128 image areas ( 15 at 0° , 23 at 20° , 44 at 30° and 46 at 45° or above ) were used for 3D reconstruction . Data quality was improved by synthetic unbending ( Arheit et al . , 2013 ) . To compensate for the resolution-dependent degradation of image amplitudes a negative temperature factor of B = −200 Å2 was used . A molecular model was build by fitting the MjNhaP1 x-ray structure into the 3D EM density in COOT ( Emsley and Cowtan , 2004 ) . 3D difference maps were calculated with scripts from the CCP4 ( Winn et al . , 2011 ) and the PHENIX ( Adams et al . , 2010 ) software packages ( as indicated below ) . The EM and x-ray density maps were expanded , scaled and the resolution cut to 6 Å ( sftools ) . Maps were superimposed in PHENIX ( Adams et al . , 2010 ) and placed into a cell with identical units ( mapmask , maprot ) . The density of one protomer was masked with the help of the pdb models ( pdbset , ncsmask , maprot , mapmask ) , and both maps were subtracted from one another ( overlapmap ) . Figures and movies were prepared with PyMOL ( DeLano and Lam , 2005 ) . Superimpositions were performed using Secondary Structure Superimposition within COOT ( Emsley and Cowtan , 2004; Krissinel and Henrick , 2004 ) . For morphing LSQMAN ( Kleywegt , 1996 ) from the Uppsala Software Factory was used to generate a series of intermediates between structures . Sequence alignments were performed using ClustalX ( Larkin et al . , 2007 ) and adjusted in JalView ( Waterhouse et al . , 2009 ) . The potential surface was calculated with pdb2pqr ( Dolinsky et al . , 2007 ) and APBS ( Baker et al . , 2001 ) . Analysis of transport pathways , channels and cavities was performed with Hollow ( Ho and Gruswitz , 2008 ) and visualized within PyMOL . For sodium efflux under symmetrical pH , E . coli polar lipids were dried under nitrogen and resuspended in 10 mM choline citrate/Tris/glycine pH 6–9 , 5 mM KCl , 200 mM sodium chloride , 10 mM β-mercaptoethanol . Liposomes were preformed using polycarbonate filters with a pore size of 400 nm and destabilized by addition of octyl glucoside to a final concentration of 1% . MjNhaP1 was added at an LPR of 200:1 and the suspension was incubated for 1 hr . Detergent was removed by dialysis ( 14 kDa cut-off ) overnight against detergent-free reconstitution buffer . To ensure complete detergent removal , 1 g of Biobeads were added to the dialysis buffer per 4 ml of lipids . Proteoliposomes were centrifuged at 300 , 000×g for 20 min and washed once with reconstitution buffer . Liposomes were centrifuged again and resuspended at a lipid concentration of 60 mg/ml in reconstitution buffer . 2 µl of proteoliposomes were diluted in 2 ml of reaction buffer ( 10 mM choline citrate/Tris/glycine at the same pH , 5 mM KCl , 2 µM acridine orange ) to start the reaction . Antiport activity of MjNhaP1 establishes a ΔpH across the membrane , observed as acridine orange quenching . As a control , ( NH4 ) 2SO4 was added to a final concentration of 25 mM at the end of the reaction to dissipate the pH gradient . Measurements were performed at 25°C and acridine orange fluorescence was monitored at 530 nm ( excitation: 495 nm ) in a Hitachi fluorimeter . For 22Na uptake measurements , reconstitution was performed as described for sodium efflux under symmetrical pH conditions , with the following changes . Lipids were resuspended in 20 mM choline citrate/Tris/glycine pH 4–10 , 10 mM ( NH4 ) 2SO4 , 10 mM β-mercaptoethanol and the LPR was 400:1 . Transport was initiated by diluting 2 µl of proteoliposomes into 200 µl of reaction buffer ( 20 mM choline citrate/Tris/glycine at the same pH , 10 mM choline chloride , 2 mM MgSO4 ) containing NaCl at final concentrations between 25 µM and 6 . 5 mM and 1 µCi/ml 22Na . Dilution of proteoliposomes in ( NH4 ) 2SO4-free reaction buffer results in NH3 efflux , acidifying the interior of the liposomes ( Dibrov and Taglicht , 1993 ) . Transport was stopped by filtering the proteoliposomes on 0 . 2 µm nitrocellulose filters and washing with 3 ml ice-cold 22Na-free reaction buffer . Before counting , filters were transferred to counting tubes and 4 ml scintillation cocktail ( Rotiszint , Roth , Germany ) was added . All measurements were performed on ice and repeated at least three times . The activities of N-terminal truncated MjNhaP1 constructs and the Asn160 mutant were determined under asymmetrical pH by fluorescence in everted vesicles or proteoliposomes ( Goswami et al . , 2011 ) . MjNhaP1 constructs were cloned into the pTrcHis2-Topo plasmid , carrying a C-terminal Myc-His tag , via NcoI and EcoRI restriction sites and the following primers: MjNhaP_6s_NcoI: 5′-CCGCCGCCATGGCCCTTGCTATTGGTTACCTTGGATAGC-3′ MjNhaP_10s_NcoI: 5′-CCGCCGCCATGGCCCTTGGATTAGCTTTAGTTCTTGGTTC-3′ MjNhaP_16s_NcoI: 5′-CCGCCGCCATGGCCCTTCTTGGTTCGTTAGTGGCAAAAATTG-3′ MjNhaP_426as_EcoRI: 5′-CAAAGTATAAAGAAGAATCCCACCATAAGGGCGAATTCGCCGCC-3′ Point mutations were generated using the QuickChange II Site-Directed Mutagenesis Kit ( Agilent Technologies ) and the following primers: MjNhaP1_N160A_s: 5′-GTTAGAGGCGGAGAGTATCTTTGCCGACCCATTGGGAATAGTTTC-3′ MjNhaP1_N160A_as: 5′-GAAACTATTCCCAATGGGTCGGCAAAGATACTCTCCGCCTCTAAC-3′ MjNhaP1_N160D_s: 5′-GTTAGAGGCGGAGAGTATCTTTGACGACCCATTGGGAATAGTTTC-3′ MjNhaP1_N160D_as: 5′-GAAACTATTCCCAATGGGTCGTCAAAGATACTCTCCGCCTCTAAC-3′ Coordinates and structure factors for the pH 8 X-ray structure and the pH 4 EM structure were deposited in the PDB with the accession code 4czb and 4d0a , respectively . The 3D EM map was deposited in the EM data bank with the accession code EMD-2636 . | Although the membrane that surrounds a cell is effective at separating the inside of a cell from the outside environment , certain molecules and ions must enter or leave the cell for it to work correctly . Proteins embedded in the cell membrane , called transporters , ensure this occurs . Transporters that are found in all organisms include the sodium/proton antiporters , which exchange protons from inside the cell with sodium ions from outside . However , exactly how these antiporters work was unknown . Paulino , Wöhlert et al . have now examined the structure of a sodium/proton antiporter from a single-celled organism called Methanocaldococcus jannaschii , a species of archaea that thrives at high temperature . Using X-ray crystallography , Paulino , Wöhlert et al . uncovered the structure of the antiporter in the presence of sodium ions and in alkaline conditions . Under these conditions the sodium/proton antiporter adopts an ‘inward-open’ state , where the substrate-binding site—the region where the ions bind to be transported—of the transporter is open towards the cell interior . Paulino , Wöhlert et al . also used electron cryo-microscopy to investigate the antiporter's structure under acidic conditions in the absence of salt . This revealed an ‘outward-open’ state , where the substrate-binding site of the transporter is open towards the space outside of the cell . The main difference between this and the inward-open state is the movement of a bundle of six helices within the antiporter . This activating structural change occurs when a sodium ion binds to the antiporter rather than by a change in acidity . Paulino , Wöhlert et al . found that the structure of the M . jannaschii antiporter is very similar to the structures of an antiporter from another archaea species , which was studied in separate work . The acidity range under which the two transporters are most active is different , indicating that minor changes in the amino acid sequence that make up their structure can have a substantial effect on the activity of these antiporters . The next step will be to use computer simulations to calculate how sodium/proton antiporters change from an inward-open to an outward-open state . The M . jannaschii antiporter will be particularly suitable for such simulations , as Paulino , Wöhlert et al . found that it transports ions more rapidly than any previously known transporter . Understanding how these transporters work is also medically relevant , as defects in related sodium/proton antiporters in humans are implicated in serious and life-threatening diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Structure and transport mechanism of the sodium/proton antiporter MjNhaP1 |
Mutation and natural selection shape the genetic variation in natural populations . Here , we directly estimated the spontaneous mutation rate by sequencing new Drosophila mutation accumulation lines maintained with minimal natural selection . We inferred strong stabilizing natural selection on quantitative traits because genetic variation among wild-derived inbred lines was much lower than predicted from a neutral model and the mutational effects were much larger than allelic effects of standing polymorphisms . Stabilizing selection could act directly on the traits , or indirectly from pleiotropic effects on fitness . However , our data are not consistent with simple models of mutation-stabilizing selection balance; therefore , further empirical work is needed to assess the balance of evolutionary forces responsible for quantitative genetic variation .
Mutation is the ultimate source of genetic variation . In natural populations of finite size , however , mutations and the genetic variation they introduce are constantly removed by genetic drift and by purifying and stabilizing natural selection . Therefore , to understand how genetic variation is generated and maintained , one must know the rate by which spontaneous mutations occur and their effects on fitness and quantitative traits , and the forms and consequences of natural selection on genetic variation . To answer these fundamental questions , we take advantage of mutation accumulation ( MA ) lines of Drosophila melanogaster derived from a single inbred genome and independently maintained under conditions of minimal natural selection , as well as a panel of wild-derived inbred lines representing a balanced state of mutations , drift and natural selection . The expectation is that MA lines are experiencing severe genetic bottlenecks thus have minimal natural selection . Therefore , unless a mutation is highly deleterious or lethal , its fate is determined primarily by genetic drift . This property of MA lines is distinctly different from natural populations and allows us to derive the expectation of genetic variation under neutrality using the estimated mutational variance among the MA lines . By comparing the expectation under neutrality to the observed level of genetic variation in a natural population , one can infer the form and consequences of natural selection on genetic variation ( Figure 1 ) . 10 . 7554/eLife . 14625 . 003Figure 1 . Experimental design of the DGRP and mutation accumulation ( MA ) lines . ( a ) The DGRP is a collection of inbred strains derived from the wild population in Raleigh , NC . The spectrum of mutations and genetic variation in the DGRP is a reflection of the combined effects of mutations , drift , and selection . In this study , we used the DGRP genotype data ( Huang et al . , 2014 ) to estimate the parameter ∑i=1GVar ( gi ) ( see Materials and methods ) , the microarray expression data collected by ( Ayroles et al . , 2009 ) and organismal phenotypes ( sleep traits from Harbison et al . , 2009 ) to estimate genetic variance among the DGRP lines ( Vg ) . ( b ) We derived 25 MA lines from the inbred line DGRP_360 . These lines were kept with small population sizes of 10 females and 10 males such that there is minimal natural selection . At generation 60 , we sequenced the MA lines to estimate mutation rate ( μ ) and ∑i=1MVar ( mi ) , and obtained microarray gene expression and organismal phenotypic data to estimate mutational variance ( Vm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 003 We use whole genome sequencing to determine the rate and characteristics of spontaneous mutations , and quantitative measurements of gene expression and organismal traits in the MA and wild-derived lines to understand the origin and maintenance of quantitative genetic variation . Previous MA studies have successfully used similar strategies to identify mutations and estimate mutation rates in yeast ( Lynch et al . , 2008 ) , algae ( Ness et al . , 2012 ) , nematodes ( Denver et al . , 2004 ) , flies ( Haag-Liautard et al . , 2007; Keightley et al . , 2009; Schrider et al . , 2013 ) , and plants ( Ossowski et al . , 2010 ) ; and to characterize the nature of natural selection on gene expression ( Denver et al . , 2005; Rifkin et al . , 2005 ) . However , compared to the present study , they were small in scale , did not simultaneously identify mutations at the DNA level and estimate mutational variance for gene expression and organismal quantitative traits , or did not compare mutational variation to standing genetic variation in the same equilibrium population from which the MA lines were derived .
To accumulate spontaneous mutations , we split one sequenced inbred line from the Drosophila melanogaster Genetic Reference Panel ( DGRP ) ( Mackay et al . , 2012 ) into 25 MA lines , and maintained them in small populations ( 10 females and 10 males ) for many generations ( Figure 1b ) . The small population size will minimize natural selection and allow non-lethal mutations to drift to high frequency or fixation and accumulate over time . We sequenced all 25 MA lines at generation 60 and obtained deep sequence data for 23 lines ( Supplementary file 1A ) . We detected a total of 1 , 456 mutations that were either fixed or segregating at high frequency ( >0 . 20 ) on the euchromatic nuclear genome in the MA lines . The number of mutations per MA line ranged from 35 to 193 , with a mean of 63 ( Supplementary file 1B ) . To validate mutations detected by Illumina’s sequence-by-synthesis chemistry , we sequenced a randomly selected set of 51 mutations in five MA lines at generation 130 by capillary sequencing ( Sanger ) . We reasoned that the 70 generations between mutation detection and validation allowed sufficient time for mutations that were segregating at generation 60 to drift to fixation , and hence be more reliably detected by Sanger sequencing . The observed fixation probabilities of putative mutations agreed well with the expected probabilities given their initial frequencies at G60 ( Figure 2 ) , suggesting that both the mutations and their frequency estimates were accurate . 10 . 7554/eLife . 14625 . 004Figure 2 . Validation of mutations . 51 Mutations detected at G60 ( frequency > 0 . 2 ) in MA lines 2 , 11 , 13 , 17 , or 24 were randomly selected and sequenced by Sanger sequencing at G130 and classified as Fixed ( cross ) or Lost ( circle ) by manually inspecting chromatograms . The fixation status is plotted against the initial mutant frequency at G60 . A LOESS smooth line was fitted to the data ( point estimates = solid line , 95% confidence interval = broken line ) to estimate the fixation probability . Expectation of fixation probability ( = initial mutant allele frequency ) is indicated by the grey diagonal line . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 004 Among the 1456 mutations , there were 1203 single base substitutions ( SBS ) , 17 multiple base substitutions ( MBS ) , 141 deletions , 47 insertions , and 48 complex mutational events that involved a combination of base substitutions and indels ( Supplementary file 1B ) . The numbers of different types of mutations are proportional to their mutation rates; thus the indel mutation rate was approximately 1/6 that of SBS , and deletions occurred at a frequency three times higher than insertions . Of the 1203 SBS , 595 were transitions ( Ti ) and 608 were transversions ( Tv ) , corresponding to a Ti/Tv ratio of 1 . 96 ( Figure 3 ) . There were twice as many base substitution mutations at G or C sites ( n = 828 ) than at A or T sites ( n = 401 ) , which , given the 42 . 48% genomic GC content , indicated a strong bias ( ~ 2 . 80 fold increase ) towards mutations at G or C bases . These numbers agreed well with a previous study using the same sequencing strategy of three D . melanogaster MA lines , which reported a Ti/Tv ratio of 1 . 95 and an approximately two-fold increase in mutations at G/C bases ( Keightley et al . , 2009 ) . We then asked if genomic context affected whether or not a mutation occurred . While there was no appreciable difference in local sequence ( 20 bp up and downstream ) GC content between mutations and randomly sampled sites , indels appeared to occur more often in low complexity regions , where homopolymers and short tandem repeats can often be found and are prone to replication slippage ( Figure 4 ) . 10 . 7554/eLife . 14625 . 005Figure 3 . Classification of single base substitutions ( SBS ) . Single base substitutions were classified according to their ancestral alleles ( top and bottom box ) and mutant alleles ( middle boxes ) . The size of each of the middle boxes indicates number of mutations in each class . Blue and red lines indicate transitions and transversions , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 00510 . 7554/eLife . 14625 . 006Figure 4 . Genomic context of mutations . Sequence composition ( GC content , a ) and complexity ( b ) for local sequences ( 20 bp up and downstream of mutations ) are plotted as box plots and compared between different types of mutations . The 'Random genomic' class contains 1000 randomly chosen sites in the genome . Sequence complexity is measured as –lnS , where S is calculated using the algorithm in NCBI’s DUST program and measures sequence complexity . P values were computed by Wilcoxon’s rank sum tests comparing data in each category to the 'Random genomic' category . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 006 To characterize the functional effects of spontaneous mutations , we annotated their genomic locations ( exonic , intronic or intergenic ) and compared the distribution to that of standing variation in molecular polymorphisms in the DGRP , and to the fraction of total genomic sites in each of these categories . Standing variation in the DGRP reflects the demographic history of the natural Raleigh , NC population and is depleted of exonic variants ( Figure 5a ) . However , the extent of depletion of exonic spontaneous mutations was much weaker – the spectrum of spontaneous mutations detected in the MA lines more closely reflected the proportion of exonic , intronic and intergenic bases in the genome ( Figure 5a ) . We further classified coding variants and mutations according to their functional impacts on polypeptide sequences . Remarkably , the proportions of frame-shift , stop gained or lost , and nonsynonymous spontaneous mutations were significantly greater than the proportion of segregating polymorphisms in the DGRP for these categories ( Figure 5b ) . While the fitness effects of these protein sequence mutations may differ under laboratory and natural environments , these results clearly indicate that deleterious mutations that would otherwise be lost have accumulated in the MA lines . In keeping with the inference that spontaneous mutations accumulated under minimal natural selection , the numbers of mutations per gene was primarily a function of gene length ( Figure 6 ) ; and there was no gene ontology ( GO ) category enrichment for genes harboring new mutations ( Supplementary file 1C ) . 10 . 7554/eLife . 14625 . 007Figure 5 . Mutation accumulation lines accumulate deleterious mutations . Annotation of genomic bases , standing variation , and mutations in MA lines according to their ( a ) genomic locations and ( b ) functional impact on protein sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 00710 . 7554/eLife . 14625 . 008Figure 6 . Relationship between number of mutations and gene length . Number of mutations detected in MA lines for each gene is plotted against the total number of bases covered for mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 008 To estimate the spontaneous mutation rate , we first inferred the effective population size in the MA lines based on the observed mutation frequency spectrum of presumably neutral mutations . This was achieved by a maximum likelihood procedure where the probability density was obtained by simulation . The parameter Ne was estimated by taking the value that maximizes the likelihood of observing the data given the probability density . Although the census population size of each MA line was 2Ne=40 haploid genomes , the effective population size was estimated to be 2Ne=19 for the X chromosome and 2Ne=23 for autosomes ( Figure 7a , b ) . The frequency spectrum of the observed effective population size visually agreed with the expectation ( Figure 7a , b ) . At this population size , for selection coefficients s ranging between -0 . 01 and 0 . 01 and assuming additivity , the fixation probability u=1−e−2s1−e−4sNe ( Kimura , 1957 ) is between 0 . 034 and 0 . 054 , which closely centers around the fixation probability for neutral sites . The large difference between census and effective population sizes could be due to the large variance in number of offspring commonly observed in flies ( Crow and Morton , 1955 ) . Although the X to autosome Ne ratio was slightly higher than the expected 0 . 75 , this difference was not statistically significant ( P = 0 . 40 ) . The estimates of Ne did not differ whether or not presumably non-neutral mutations ( mutations that changed amino acids ) were included , further suggesting that drift was the primary force driving mutation frequency dynamics in these MA lines . Given the estimated effective population size , the marginal ( integrated over 60 generations ) probability of a mutation to attain the frequency cutoff ( 0 . 2 ) was 7 . 20% on the X chromosome and 6 . 18% on autosomes . We used these estimates of effective population size to estimate the spontaneous mutation rate on the X chromosome and autosomes , given the number of observed mutations in each line , taking into account variable sequence coverage in each line . 10 . 7554/eLife . 14625 . 009Figure 7 . Inference of effective population size and mutation rate . Expected and observed distributions of mutant allele frequency on autosomes ( a ) and the X chromosome ( b ) . The expected distribution was generated based on estimates of effective population size ( 2Ne ) . Mutation rate estimates of autosomes and X chromosomes are compared for different types of mutations ( c ) and for different lines ( d ) . In ( d ) , the Pearson’s correlation coefficient ( r ) is calculated with ( orange line ) or without ( green line ) MA19 ( orange circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 009 The median spontaneous mutation rate was 5 . 21 × 10–9 per base on autosomes and 5 . 07 × 10–9 on the X chromosome for single base substitutions and 0 . 79 × 10–9 on autosomes and 0 . 65 × 10–9 on the X chromosome for indels . The overall spontaneous mutation rate for all types of mutations combined was 6 . 25 × 10–9 on autosomes and 6 . 96 × 10–9 on the X chromosome ( Figure 7c ) , similar to recent mutation rate estimates from MA studies using high throughput sequencing ( Keightley et al . , 2009; Schrider et al . , 2013 ) . There was substantial variation in spontaneous mutation rates among the MA lines ( Figure 7c ) . The mutation rate in MA19 was nearly a magnitude greater than the lines with the smallest mutation rates ( MA08 for autosomes and MA15 for the X chromosome ) . Such a large difference cannot be solely explained by variability in 2Ne among the MA lines , as varying 2Ne from 10 to 40 can only account for a 28% difference in mutation rate ( difference in 2Ne*p , see Materials and methods ) . The mutation rates on autosomes and the X chromosome showed no systematic difference ( Figure 7c , paired Wilcoxon rank sum test P = 0 . 64 ) and were positively correlated ( Figure 7d ) , further suggesting the random distribution of mutations . To assess the effects of mutations on quantitative traits , specifically the rate at which mutations introduce genetic variation , we analyzed the genetic variation of genome-wide gene expression profiles , two bristle number traits , and five sleep and activity traits . We partitioned the phenotypic variation observed among the collective sample of individuals from the MA lines into the between-line genetic variation due to mutations ( VMA ) and the within-line variation due to environmental or technical noise ( Ve ) . Although any pair of MA lines only differ on average by 126 sites , all organismal phenotypes and the expression of a large fraction of genes in both sexes ( 1526/7566 = 20 . 2% of expressed genes in females and 3 , 872/8 , 136 = 47 . 6% in males ) accumulated significant between-line variance ( Supplementary file 1D–E , Figure 8 ) . Mutational variance ( Vm ) , the amount of genetic variation introduced by mutations in each generation , scaled by environmental variance and expressed as the mutational heritability ( hm2= Vm/Ve ) , had a median of 0 . 55 x 10–3 for gene expression traits in females and 0 . 75 x 10–3 in males ( Figure 8 ) . hm2 for organismal traits ranged between 0 . 30 x 10–3 and 2 . 88 x 10–3 , of the same order of magnitude as observed in earlier studies ( Houle et al . , 1996 ) . These values were near or among the upper quartile of that for gene expression traits with the exception of abdominal bristle number in males ( Supplementary file 1D ) . The difference in mutational heritabilities between gene expression traits and organismal phenotypes may be due to a larger number of QTLs influencing organismal traits and thus a larger mutational target size . 10 . 7554/eLife . 14625 . 010Figure 8 . High rate of mutational variance . Histogram of mutational heritability ( hm2=Vm/Ve ) for females ( a ) and males ( b ) are plotted on the log scale . The placements of organismal traits in the hm2 bins are indicated by lines connecting the bars and the trait names . AB = abdominal bristle and SB = sternopleural bristle . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 010 We next assessed whether genes exhibiting mutational variation for gene expression were associated with GO categories . For each GO category , we compared the mutational variance of genes in the category to those not in the category . For significant GO categories , we further polarized the difference in rate of accumulation of mutational variance as faster or slower within the category . Although mutations occurred randomly across the genome and were not associated with GO categories , the rate of accumulation of genetic variation was significantly different for genes within many GO terms than the rest of the transcriptome ( Supplementary file 1F ) . For example , in both sexes , mutational variance accumulated faster for genes involved in chitin metabolism , iron binding , and sensory perception of chemical stimulus , but slower for genes involved in protein translation , mRNA splicing , and mitotic spindle organization . Finally , we compared mutational variation in gene expression to plasticity of gene expression across a wide range of environments ( Zhou et al . , 2012 ) . We observed that genes that are more plastic to macro-environmental perturbations accumulated genetic variation at a faster rate ( Figure 9 ) , suggesting a shared control of gene expression variation by mutations and environmental perturbations ( Landry et al . , 2007 ) . 10 . 7554/eLife . 14625 . 011Figure 9 . Correlation between mutational variance and environmental plasticity . Mutational variance ( Vm ) is plotted against variance due to environmental plasticity ( VENV ) for females ( a ) and males ( b ) . The dashed line is a LOESS fit to the data . Spearman’s correlation ( ρ ) and the P value of a test for its significance are also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 011 Under the neutral model of polygenic phenotypic evolution , the among-line variance ( Vg ) in the DGRP inbred lines is 4NVm , where N is the effective size of the wild population from which DGRP was derived ( Lynch and Hill , 1986 ) . We estimated the genome-wide nucleotide diversity in the DGRP ( π ) to be π = 4 . 92 x 10–3 in the DGRP ( Huang et al . , 2014 ) , from which we estimated N to be 186 , 363 ( π4μ ) . For all organismal and the majority of gene expression traits , the magnitude of standing genetic variation in the DGRP was much smaller than that predicted under the assumption of neutrality ( Figure 10 ) . Therefore , there must be strong stabilizing selection that acts either directly on the traits ( direct stabilizing selection ) or through pleiotropic fitness effects of new mutations ( apparent stabilizing selection ) , which constrains the accumulation of genetic variation by mutations for gene expression and organismal traits ( Denver et al . , 2005; Rifkin et al . , 2005 ) . To understand the consequence of the strong apparent stabilizing selection , we estimated the expectation of allelic effects based on the observed sequence and quantitative trait variation among the MA and DGRP lines . The amount of genetic variation is proportional to sequence variation by a factor of E ( a2 ) ( see Materials and methods ) , where a is the allelic effect of a mutation on quantitative traits . For the majority of traits , the ratio of mutational to standing genetic variance , Vm/Vg , far exceeded the expectation given the observed sequence variation ( Figure 10 ) ; thus the allelic effects of spontaneous mutations , E ( am2 ) , were several orders of magnitude larger than that of standing DNA variation , E ( ag2 ) . This result suggests that the apparent stabilizing selection , directly for fitness and indirectly for traits correlated with fitness , had either eliminated mutations with large effects on quantitative traits or modified their effects . The former appears to be at least partly true given the obvious difference in functional categorization of spontaneous mutations and standing DNA variation ( Figure 5 ) . 10 . 7554/eLife . 14625 . 012Figure 10 . Strong apparent stabilizing selection of quantitative trait variation . Distributions of Vm/Vg in females ( a ) and males ( b ) are plotted on the log scale . The blue , red , and purple bars indicate genes with significant ( FDR = 0 . 05 ) among-line variance in DGRP only , MA lines only , and both DGRP and MA lines respectively . Placements of organismal traits are indicated by lines connecting the bars and the trait names . AB = abdominal bristles and SB = sternopleural bristles . Neutral expectations 1/4N and ∑Var ( mi ) /k∑Var ( gi ) are also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 012 Using Vm/Vg as an indicator of the strength of the apparent stabilizing selection ( higher values = stronger selection ) , we examined the properties of genes associated with variation in Vm/Vg for gene expression . First , genes expressed in both sexes are under slightly stronger selection than those expressed in only one sex , and the strength of selection has a modest but highly significant positive correlation between the two sexes ( Figure 11 ) . Second , there is stronger selection for genes on the X chromosome than autosomal genes , and this effect is more pronounced in males than females ( Figure 12 ) . Third , we partitioned the genome with respect to GO categories and assessed the significance of the difference of Vm/Vg for genes associated with each GO category and those not in the category . Many genes in GO categories associated with essential cellular functions related to transcription , translation , cell cycle , and energy metabolism , among others , appeared to be under stronger selection in both sexes ( Supplementary file 1G , Figure 13 ) . 10 . 7554/eLife . 14625 . 013Figure 11 . Strength of apparent stabilizing selection in females and males . ( a ) Vm/Vg for gene expression in females are plotted against that in males . ( b ) Boxplots of Vm/Vg for sex-specific ( S ) and non-specific ( NS ) genes . Within each sex , Vm/Vg for sex-specific and non-specific genes are compared using Wilcoxon’s rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 01310 . 7554/eLife . 14625 . 014Figure 12 . Strength of apparent stabilizing selection on autosomes and X chromosome . Boxplots of Vm/Vg are plotted for each chromosome in females and males . Wilcoxon’s rank sum test is used to compare Vm/Vg on autosomes ( A ) and on X chromosome ( X ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 01410 . 7554/eLife . 14625 . 015Figure 13 . Strength of apparent stabilizing selection differs for genes in different functional categories . Boxplots of Vm/Vg for genes in selected Gene Ontology ( GO ) categories as compared to all genes . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 015
The origin and maintenance of quantitative trait variation are fundamental problems in evolutionary biology and have profound implications in agriculture and medicine , where most economically and medically relevant traits are quantitative in nature . Mutations are the ultimate source of genetic variation , but they are rare and constantly being removed by natural selection and genetic drift . The purifying property of both selection and drift on mutations makes it especially difficult to study the characteristics of spontaneous mutations in natural populations because it is difficult to attribute a mutational pattern to either one of these evolutionary forces . In this study , we combined the classical mutation accumulation design , which subjects inbred lines of Drosophila melanogaster to genetic bottlenecks for many generations , with modern high throughput technologies . This allowed us to largely separate the effects of genetic drift from the combined effects of both selection and drift , a key advantage that is not possible from observations only on natural populations . Our estimates of mutation rates could be biased either upwards or downwards , depending on the initial fitness of the inbred line . On the one hand , lethal and highly deleterious mutations would not accumulate in these lines . On the other hand , beneficial mutations fix at a higher probability than the neutral probability assumed . Although spontaneous mutations occurred at a relatively low rate of 6 . 60 × 10–9 averaged over autosomes and the X chromosome , they replenished genetic variation at a significant rate for gene expression ( sex averaged rate = 0 . 65 × 10–3Ve per generation ) and organismal traits ( sex averaged rate = 1 . 36 × 10–3Ve per generation ) . Our estimates of hm2 for gene expression traits were higher than observed in previous studies ( Rifkin et al . , 2005 ) , possibly due to smaller technical variation and thus smaller Ve in this study . Although the MA lines were derived from the same progenitor line and were raised under homogeneous conditions , mutation rates among the MA lines varied significantly . This implies either that non-genetic factors affect mutagenesis and/or that some spontaneous mutations themselves affect mutation rate; the latter is less plausible because such mutations must occur early in many lines to have a pronounced effect . The MA lines accumulated a broad functional spectrum of spontaneous mutations , including a significant fraction of high impact mutations that would otherwise likely be removed by natural selection . The relatively small number of mutations and the large number of genes with significant mutational variance in expression implies pervasive pleiotropic effects by new mutations . Gene expression traits often form co-expression modules ( Ayroles et al . , 2009 ) , and therefore mutations that directly influence expression of a small number of loci can cause secondary trans effects at a much larger number of genes . Taking this networked view of gene expression traits , any selection will not act on individual traits , but rather on the combined effects of all traits . Consistent with this notion , we found stronger selection on genes that would cause trans pleiotropic effects such as transcription factors ( Supplementary file 1G ) . Because of the scale of experiment and the large number of MA lines needed to accurately estimate mutational variance , we chose an experimental design that focused on a single inbred strain to derive a relatively large number of MA lines . This design , however , does not allow us to infer the effects of genetic background on the rates of mutation and introduction of genetic variation . There is limited evidence for genuine genetic variation for mutation rate . However , this is primarily due to technical limitations because mutation rate is sensitive to environmental and physiological factors that cannot be easily controlled and it is prohibitive to collect genetic data on mutation rates in natural populations . The effects of genetic background can be studied with a modified design using the DGRP by deriving MA lines from multiple genetically diverse inbred strains and controlling environments . Because of the close proximity of mutation rates in this study and earlier studies of the same species , the conclusions drawn in this study are likely to hold across genetic backgrounds , especially given the magnitude of apparent stabilizing selection , which is unlikely to be attributable to genetic factors unique to the ancestral line . It is entirely possible that mutation rates may be different in inbred conditions than in natural populations , which has significant theoretical consequences ( Agrawal , 2002 ) . While we could not formally exclude the possibility , the mutational variance we observed in the MA lines appears to be too large ( by a few orders of magnitude ) to be solely explained by elevated mutation rates under stressful conditions . The issue of why genetic variation for quantitative traits segregates at appreciable levels in natural populations despite the tendency of genetic drift and directional and stabilizing selection to erode it remains an unresolved puzzle . Our data unequivocally reject the neutral mutation – random drift balance model for maintenance of quantitative genetic variation ( Lynch and Hill , 1986 ) . The observed magnitude of segregating variation is much less than predicted under this model given our estimates of population size in nature and mutational variation . Therefore , we inferred that a form of stabilizing selection on a large fraction of gene expression as well as organismal traits constrains naturally occurring genetic variation , consistent with earlier studies using a similar analysis in C . elegans ( Denver et al . , 2005 ) and inter-specific variation in Drosophila ( Rifkin et al . , 2005 ) . Theoretical models of maintenance of quantitative genetic variation by mutation – stabilizing selection balance assume either direct stabilizing selection on each trait or stabilizing selection as a deleterious pleiotropic side-effect of new mutations on fitness ( Johnson and Barton , 2005 ) . The former class of model has different quantitative predictions depending on the relative magnitude of mutation and selection . The house-of-cards approximation holds when mutation rates are low , mutational effects are large , and selection is strong ( Turelli , 1984 ) ; while the Gaussian approximation holds under conditions of weak stabilizing selection , high mutation rates and small mutational effects ( Lande , 1975 ) . Our observations of low mutation rates and mutational effects that are much larger than standing DNA polymorphisms favor the former parameterization . Under this model , Vg=4nμVs , where n is the number of loci potentially affecting the trait , μ is the mutation rate , and Vs represents the strength of stabilizing selection ( Turelli , 1984 ) . Assuming strong direct stabilizing selection ( e . g . Vs=20Ve ) , high heritabilities ( e . g . Vg=Ve ) and our estimated mutation rate ( μ=6 . 60 × 10−9 ) , then n=1 . 9 × 106 , which seems implausibly large given the total size of the D . melanogaster euchromatic genome of approximately 1 . 2 × 108 . Under simple pleiotropic models of apparent stabilizing selection ( all mutations are equally deleterious and have a reduction of heterozygous fitness of s , and the strength of selection is strong ( ~10–2 ) against new mutations ) the equilibrium genetic variance is Vg=Vm/s and s=Vm/Vg . On average , our estimate of Vm/Vg had a median 1 . 94 × 10–3 for organismal traits and 1 . 82 × 10–3 for gene expression traits . Thus , while we detect apparent stabilizing selection on quantitative traits , it is too weak by an order of magnitude for the majority of genes and quantitative traits to be consistent with observed selection against heterozygous effects of new mutations ( Mukai et al . , 1972; Mackay et al . , 1992 ) . Alternatively , the rate of generation of new mutations is too weak a force to counter strong apparent stabilizing selection , which removes variation faster than it is generated . Our estimates of μ , Vm and Vg do not alter the conclusion that simple mutation – stabilizing selection models for either direct or apparent stabilizing selection cannot maintain the observed amounts of segregating genetic variation with the observed mutational input and strong selection: the mutational variance is too low and/or the standing genetic variance is too high ( Turelli , 1985; Hill and Keightley , 1988; Barton and Turelli , 1989; Zhang et al . , 2002; Turelli and Barton , 2004 ) . It is possible that stabilizing selection in nature is weaker than assumed ( Kingsolver et al . , 2001; Kingsolver and Diamond , 2011 ) , and other mechanisms such as balancing selection , fluctuating allelic effects in the face of temporally and spatially varying environments , canalization of mutational effects , and a combination of direct and apparent stabilizing selection all contribute ( Barton , 1990; Zhang et al . , 2002; Turelli and Barton , 2004; Zhang and Hill , 2005 ) . Ultimately , artificially introducing individual mutations or combinations of mutations and assessing their effects , which is now feasible , will be needed to understand the balance between mutations and selection in maintaining segregating genetic variation for quantitative traits .
The D . melanogaster strain DGRP_360 was generated by 20 generations of strict full-sib mating from an isofemale line derived from the Raleigh , NC USA population ( Mackay et al . , 2012 ) . We divided DGRP_360 into 25 replicate sublines , each maintained at a census population size of 10 virgin females and 10 males per generation . All lines were maintained in shell vials with 10 ml cornmeal-agar-molasses medium at 25°C , 70% humidity and 12 hr/12 hr light/dark cycle . Genomic DNA was extracted from 100 female flies per MA line at generation 60 using the QIAGEN Genomic-tip 100/G kit ( Qiagen , Valencia , CA ) . The flies were homogenized with a mortar and pestle to a fine white powder using liquid nitrogen and lysed for 2 hr ( 50°C ) with Buffer G2 supplemented with RNAse A ( 1 . 5 mg ) and Proteinase K ( 12 mg ) . The samples were centrifuged at 7000 x g for 30 min at 4°C and the clear lysates applied to a Genomic-tip 100/G that had been equilibrated with Buffer QBT . Once the lysates had passed through the columns , the columns were washed twice with Buffer QC . Genomic DNA was eluted from the column with Buffer QF , precipitated with 100% isopropanol and the DNA pellets washed with 70% ethanol . The genomic DNA pellets were re-suspended in 130 μL of nuclease-free water . Purified genomic DNA ( 1 μg ) was fragmented to an average size of 300–400 bp using Covaris shearing ( Covaris , Woburn , MA ) . DNA libraries were prepared from the fragmented DNA using the Illumina TruSeq DNA Sample Preparation Kit ( Illumina , San Diego , CA ) by following the manufacturer’s procedure . The fragmented DNA was subjected to end-repair , adenylation of 3’-ends , ligation of indexed paired-end adapters and PCR-enrichment of the barcoded DNA . The libraries were quantified by qPCR using the KAPA SYBR FAST Master Mix Universal 2X qPCR Master Mix ( Kapa Biosystems , Wilmington , MA ) . The sizes of the PCR-enriched libraries were verified by Bioanalyzer using the high sensitivity DNA chip ( Agilent , Santa Clara , CA ) . We multiplexed and sequenced 5 libraries per lane on the HiSeq 2000 ( Illumina ) . Sequence data are deposited to the NCBI SRA database with the accession number for the BioProject SRP068116 . Sequence reads from the parental line ( DGRP_360 ) and each of the 25 MA lines were aligned to the Drosophila melanogaster reference genome ( BDGP5 ) using BWA-MEM with default parameters ( Li , 2013 ) . Alignments were locally realigned around known indels in the DGRP and around target regions identified across all samples using GATK ( DePristo et al . , 2011 ) . After PCR duplicate removal and base quality recalibration using GATK , overlapping bases from paired end reads were clipped using bamUtil ( http://genome . sph . umich . edu/wiki/BamUtil ) . Finally , alleles ( mapping quality ≥13 , base quality ≥13 ) were piled up using freebayes ( Garrison and Marth , 2012 ) . Only lines whose median filtered coverage was above 15 were considered for mutation detection . We considered 109 , 260 , 235 sites where between 15 and 250 reads were observed in the parental line and in at least 10 MA lines , and no more than 10 possible alleles were observed in all lines combined . A mutation was called if: ( 1 ) no read supported the mutant allele in the parental line; ( 2 ) the p value for a Fisher’s exact test assessing strand bias of alleles was conservatively >0 . 001; ( 3 ) no more than two possible alleles were observed in the mutant line; ( 4 ) the mutant allele frequency was greater than 20% in the mutant line; ( 5 ) no line other than the mutant line contained the mutant allele at frequency greater than 5%; ( 6 ) no more than two other lines contained any reads supporting the mutant allele . To validate mutations , we sequenced PCR fragments flanking 51 randomly selected mutations in pooled DNA from five MA lines ( MA02 , MA11 , MA13 , MA17 , and MA24 ) in G130 using Sanger sequencing . We sequenced DNA in G130 to allow sufficient time for mutations to fix , because Sanger sequencing cannot readily distinguish low frequency polymorphisms from background noise and is subject to bias in PCR amplification of alleles . On average , mutations fix at a probability equal to their initial frequencies after 4Ne generations . Genomic DNA was extracted from 20 females from the MA lines using the Gentra Puregene Tissue Kit ( Qiagen ) . The flies were homogenized with 2 spherical ceramic beads ( MP Biomedical ) using the TissueLyser ( Qiagen ) and lysed for 1 hr ( 56°C ) with Cell Lysis Buffer supplemented with RNAse A ( 1 . 5 mg ) and Proteinase K ( 12 mg ) . Proteins were removed from the clear lysates with protein precipitation solution followed by centrifugation . Genomic DNA was precipitated with 100% isopropanol and the DNA pellets washed with 70% ethanol . The genomic DNA pellets were re-suspended in 100 ul of nuclease-free water . Each sample was diluted to 5 ng/ul and subjected to PCR using 95 primer pairs ( Supplementary file 1B ) with the following cycling parameters: 95°C for 2 min followed by 30 cycles of 95°C / 30 s + 56°C / 30 s + 72°C / 30 s followed by a final extension step at 72°C for 4 min . The PCR products were purified using the PureLink Pro 96 PCR Purification Kit ( Life Technologies , Carlsbad , CA ) and sequenced with each corresponding forward primer using the BigDye Terminator Cycle Sequencing Kit ( Life Technologies ) . To estimate mutation rate in the MA lines , we first inferred the effective population size using a maximum likelihood approach , assuming that synonymous and non-exonic SNPs are neutral and effective population size and mutation rate stays constant over time and among MA lines . For each 2Ne value ranging from 10 to 40 , we simulated 1 , 000 , 000 MA lines where one mutation occurred independently per line per generation . At G60 , frequencies of all 60 , 000 , 000 unlinked mutations were summarized ( based on samples of 200 chromosomes in the last generation ) to obtain the expected frequency distribution of mutations under a given 2Ne at G60 , which allowed us to calculate the likelihood of observing the frequency distribution of neutral sites . We calculated the multinomial likelihood L= ( Mm1 , m2 , … , m16 ) ∏i=116Pimi of the observed mutant frequency distribution , where i indexed one of the 16 equally sized allele frequency bins between 0 . 2 and 1 ( ( 0 . 2 , 0 . 25] , ( 0 . 25 , 0 . 30 ) , … , ( 0 . 95 , 1] ) . Pi is the probability of observing a mutant allele in the ith bin given the expected mutant frequency distribution based on simulation , and m1 , m2 , ⋯ , m16 are the number of mutations in each bin and summed to M . We inferred effective population size for autosomes and X chromosomes separately . Because of the small number of mutations in each line , the mutant frequency distribution was summarized across all 23 lines to provide an overall estimate of 2Ne across all lines . Mutation rate in each line was then estimated as μ=mt*2Ne*p*B , where m is the number of mutations , t=60 is the number of generations , 2Ne is the effective population size , p is the estimated marginal probability of a mutation attaining 0 . 2 frequency given 2Ne , and B is the number of bases considered for mutation calling in that line . At G60 we assessed whole genome transcript profiles of the 25 MA lines for 3–5 day old males and females , with two biological replicates per sex and line , using Affymetrix Drosophila 2 . 0 arrays . All samples were harvested between 9–11 am . Whole bodies of 10 flies per sample were homogenized with 1 mL of QIAzol lysis reagent ( Qiagen ) and two ¼ inch ceramic beads ( MP Biomedical ) using the TissueLyser ( Qiagen ) adjusted to a frequency of 15 Hz for 1 min . Total RNA was extracted using the miRNeasy 96 kit ( Qiagen ) with on-column DNAse I digestion and following the spin technology protocol as outlined in the manufacturer’s manual . The RNA was eluted with 45 μL of RNAse-free water . Total RNA was quantified using a NanoDrop 8000 spectrophotometer ( Thermo Scientific , Carlsbad , CA ) . The 100 RNA samples were processed at all stages in a strict randomized design . Fragmented biotin-labeled aRNA were prepared for hybridization to GeneChip Drosophila Genome 2 . 0 arrays as described in the GeneChip 3’ IVT Express Kit user manual ( Affymetrix P/N 702646 Rev . 5 ) . Briefly , 200 ng of total RNA was reverse-transcribed to synthesize first-strand cDNA . The cDNA was converted to double-stranded DNA and used as a template for in vitro transcription to synthesize biotin-labeled aRNA . The aRNA was purified using magnetic RNA binding beads and quantified using a NanoDrop 8000 spectrophotometer ( Thermo Scientific ) . 12 μg of purified biotin-labeled aRNA were fragmented and hybridized to GeneChip Drosophila Genome 2 . 0 arrays ( Affymetrix ) . All microarray data have been deposited to ArrayExpress with accession number E-MTAB-4117 . In addition to the 3’ IVT array data that were generated in this study , we downloaded raw microarray expression data from a previous study profiling the transcriptomes of 40 DGRP lines ( Ayroles et al . , 2009 ) ( ArrayExpress E-MEXP-1594 ) , another profiling the transcriptomes of a synthetic outbred population derived from the same 40 DGRP lines under different environmental conditions ( Zhou et al . , 2012 ) ( ArrayExpress E-MTAB-639 ) . Arrays for DGRP_514 , MA-21 and MA-23 were not considered because we lacked sequence information for these lines . Probe intensities were corrected for background hybridization using GCRMA ( Wu et al . , 2004 ) and quantile normalized within each sex . We lifted the probe target alignment to FlyBase annotation ( Release 5 . 57 ) and retained only those that mapped entirely ( can span exon-exon junctions ) to constitutive non-overlapping exons and contained no DNA variation among the MA lines or the 40 DGRP lines . The normalized probe intensities were log 2 transformed and estimates of gene expression were obtained using median polish , which adjusted probe effect and removed potential outliers . We did a preliminary normalization and stringently removed arrays that appeared to be contaminated by flies of the opposite sex or contained more than 1% of genes whose expression was more than five standard deviations higher or lower than the mean expression of the same genes across all arrays . A total of 23 out of the 368 arrays were removed before a final normalization was performed on the remaining 345 arrays . We analyzed gene expression for each sex within each experiment separately . For each gene , we first transformed the data to be normally distributed by taking the quantiles of a normal distribution with the mean equal to the median expression and the variance equal to the square of the median absolute deviation ( by a scaling factor of 1 . 4824 to account for the expected difference of median absolute deviation and variance , which has no effect on the statistical inference ) . We defined a gene as expressed if its expression level was above 3 . 71 in females or above 4 . 26 in males . These cutoffs were chosen such that by modeling the expression profile as a mixture of two normal distributions , for a gene with expression higher than the cutoff , its probability of belonging to the high expression group was at least twice that of belonging to the low expression group ( Figure 14 ) . Partition of variance into between-line and within-line variances was performed using the lme4 package in R . It is important to note that the microarray data were collected in three studies such that data sources were completely confounded with the experiments ( MA , DGRP , and environmental exposure ) . Importantly , all estimation of parameters was performed within each experiment such that any between-experiment batch effects that shifted genome-wide gene expression profiles would not affect our results . Batch effects that increased or decreased variances in gene expression , however , may affect the results . Nonetheless , as line or treatment was randomized within each experiment , any such batch effect on variance would be much more likely to affect the within-line variance than the among-line variances that were used for comparison . More importantly , the magnitude of differences in variances , which was also consistent with other studies that assessed the differences within a single batch ( Denver et al . , 2005; Rifkin et al . , 2005 ) , was too large to be explained by a batch effect . 10 . 7554/eLife . 14625 . 016Figure 14 . Classification of genes as expressed or not expressed in each sex . Within each sex ( females in a , males in b ) , the distribution of median expression for each gene across the MA lines is subject to a mixture model analysis with two normal components . The mean and variance of each component distribution is estimated using the mixtools package in R . A gene is called expressed if its posterior probability of belonging to the normal distribution with the larger mean is higher than 2/3 . The histograms show the observed distribution and the estimated normal distributions and their mixture . DOI: http://dx . doi . org/10 . 7554/eLife . 14625 . 016 All organismal traits were scored at generation 61 , one generation after the DNA was sequenced and RNA was analyzed . Abdominal bristle number is the sum of the numbers of abdominal chaetae on the two most posterior abdominal sternites , and sternopleural bristle number is the sum of the total number of macrochaetae and microchaetae on the left and right sternopleural plates . Bristle numbers were scored on 10 males and 10 females from each of two replicate vials per line ( total N=1 , 000 ) . Five sleep traits ( total sleep duration during the night and day , numbers of sleep bouts during the night and day , and total waking activity ) ( Harbison et al . , 2009 ) were measured on 3–5 day old flies , for 16 virgin males and 16 virgin females per line ( N=800 ) . Prior to sleep measurement , all flies were maintained at a constant density of 30 flies per same-sex vial to mitigate the effects of both social exposure and mating on sleep . We recorded seven continuous days of sleep and activity using the Drosophila Activity Monitoring System ( Trikinetics , Waltham , MA ) , which measures the numbers of times each fly crosses an infrared beam . Data from flies that did not survive the entire recording period were not used in the sleep calculations . Sleep duration was calculated as any period of inactivity lasting at least five minutes . Waking activity was calculated as the number of times the fly crossed the infrared beam divided by the total time awake . We also measured bristle numbers in the 39 DGRP lines for which we had expression data or obtained the sleep phenotypic data for the same lines from a previous study ( Harbison et al . , 2009 ) . The same analysis as the gene expression data was performed to partition variance for the organismal quantitative traits for both MA and DGRP lines , except for bristle numbers , for which an additional between-replicate term was also included . The between-line variance for MA lines VMA under a neutral and polygenic model is approximately , VMA=kVm , where k=2[ t− ( 2Ne−5 ) ( 1−e−t2Ne ) ] , t=60 ( for gene expression traits ) or 61 ( for organismal traits ) is the number of generations since the lines were established , Ne is the effective population size ( 2Ne=21 ) by taking the average of autosomes and X chromosomes ) and Vm is the mutational variance ( Lynch and Hill , 1986; Mackay et al . , 1992 ) . We scaled Vm by the environmental variance Ve to obtain the mutational heritability hm2=Vm/Ve . For gene expression traits , the within-line variance is Ven+Vr , where n=10 is the number of flies per sample and Vr is the technical variation or measurement error of microarrays . Vr is believed to be small thus we multiplied the within-line variance by n to obtain an upwardly biased approximation of Ve , which underestimates hm2 . For organismal traits , Ve was assumed to be the within-line variance ( sleep traits ) or the sum of within-line and between-replicate variances ( bristle numbers ) . At equilibrium , the among-line variance between a set of inbred lines such as the DGRP ( Vg ) is approximately 4NVm , where N is the effective population size ( Lynch and Hill , 1986 ) of the wild population from which the DGRP was derived . Therefore VmVg=14N when quantitative traits evolve neutrally . We used π=4Nμ=4 . 92 × 10−3 in the DGRP to estimate N . Because whole genome sequences are available , we can also derive the amount of quantitative trait variance due to sequence divergence , without the assumption of neutrality . The expectation for among-line variance for a trait is Vg= ∑i=1GE[Var ( aigi ) ]=∑i=1GE ( ai ) Var ( gi ) =E ( ag2 ) ∑i=1GVar ( gi ) , where ai is the allelic effect expressed as a deviation from the ancestral allele at each locus and has an expectation of E ( ag2 ) , gi is the number of copies of the mutant allele in the ith line and can take a value in the range of 0 and 2 , with numbers between 0 and 2 ( twice of the segregating frequency in that line ) used to represent lines where the alleles are still segregating , and Var ( gi ) is simply the sample variance of gi . This formulation is insensitive to the sign of a and therefore does not require polarization of alleles in the DGRP . For MA lines , mutations below a frequency of 0 . 2 were randomly drawn from the expected distribution based on the inferred effective population size . The expectation of among-line variance follows the same form: Vm= ∑i=1ME[Var ( aimi ) ]/k=∑i=1ME ( ai ) Var ( mi ) /k=E ( am2 ) ∑i=1MVar ( mi ) /k , where ai is the allelic effect of the mutation and has an expectation of E ( am2 ) , mi measures the number of mutant alleles in the MA line , and k is as defined above . Therefore VmVg=E ( am2 ) E ( ag2 ) ∑Var ( mi ) /k∑Var ( gi ) , and the ratio of Vm to Vg measures the difference between E ( am2 ) and E ( ag2 ) . If the effect size distributions were equal , VmVg=∑Var ( mi ) /k∑Var ( gi ) . Finally , we estimate variance in gene expression traits due to environmental variation by the among-treatment variance ( VENV ) in a study of DGRP derived flies subject to 20 diverse and potentially harsh environments ( Zhou et al . , 2012 ) . | A key challenge in evolutionary biology is to understand how genetic variation – differences in the DNA of individuals in a population – is generated and maintained to create the enormous diversity that exists in nature . Mutations to the DNA introduce new variation , but these are constantly removed from populations by two other evolutionary forces: natural selection and genetic drift . Natural selection removes harmful genetic mutations that affect an organism’s fitness and reproduction , and genetic drift is the random increase in , or loss of , a genetic variant from a population over time . However , disentangling the effects of these evolutionary forces is challenging because the genetic variation we observe is often the final product of a long history of interaction between them . Huang et al . have now investigated genetic variation by breeding fruit flies in the laboratory . Natural selection was minimized for these flies; genetic drift was therefore the main force that removed variation . Huang et al . then sequenced the DNA of the flies to estimate the rate at which genetic mutations spontaneously occur . The sequences contained many more “high-impact” mutations ( which directly affect how proteins in the fly’s cells work ) than seen in sequences taken from a natural fly population . Traits that are produced by the cumulative actions of many genes and the environment are known as quantitative traits . By examining how much variation genetic mutations introduced into the quantitative traits of each generation of the laboratory-grown flies , Huang et al . estimated how much variation should occur in a natural population whose quantitative traits evolved without natural selection . This estimate was much higher than the levels of genetic variation seen in nature , suggesting that natural selection acts to eliminate mutations that significantly affect quantitative traits . Simple theoretical models cannot explain the relatively high spontaneous mutation rate and low genetic variation seen in the quantitative traits of natural populations . Therefore , further work is now required to understand more about the balance of evolutionary forces that maintain quantitative genetic variation . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2016 | Spontaneous mutations and the origin and maintenance of quantitative genetic variation |
Emerging evidence suggests that the nervous system is involved in tumor development in the periphery , however , the role of the central nervous system remains largely unknown . Here , by combining genetic , chemogenetic , pharmacological , and electrophysiological approaches , we show that hypothalamic oxytocin ( Oxt ) -producing neurons modulate colitis-associated cancer ( CAC ) progression in mice . Depletion or activation of Oxt neurons could augment or suppress CAC progression . Importantly , brain treatment with celastrol , a pentacyclic triterpenoid , excites Oxt neurons and inhibits CAC progression , and this anti-tumor effect was significantly attenuated in Oxt neuron-lesioned mice . Furthermore , brain treatment with celastrol suppresses sympathetic neuronal activity in the celiac-superior mesenteric ganglion ( CG-SMG ) , and activation of β2 adrenergic receptor abolishes the anti-tumor effect of Oxt neuron activation or centrally administered celastrol . Taken together , these findings demonstrate that hypothalamic Oxt neurons regulate CAC progression by modulating the neuronal activity in the CG-SMG . Stimulation of Oxt neurons using chemicals , for example , celastrol , might be a novel strategy for colorectal cancer treatment .
Colorectal cancer ( CRC ) is the third most commonly diagnosed malignant tumor and the second leading cause of cancer death globally . There were 1 . 8 million new cases , and 900 , 000 patients died of CRC annually worldwide ( Bray et al . , 2018 ) . It is estimated that there were more than 1 . 5 million people living with a previous CRC diagnosis in the United States in 2019 ( Miller et al . , 2019 ) , and approximately 147 , 950 new cases will be diagnosed and 53 , 200 individuals will die of CRC in 2020 ( Siegel et al . , 2020 ) . Besides , prevalence of CRC is rapidly rising in developing countries . For instance , incidence and mortality of CRC rank third and fifth in both men and women among all cancers in China ( Cao et al . , 2020 ) . Thus , it is imperative to understand the mechanism ( s ) of CRC development . Negative moods , including anxiety , stress , and depression , are frequently associated with the occurrences of cancers ( Antoni et al . , 2006; Lillberg et al . , 2003 ) . Anxiety is linked to a greater damage of adaptive immunity ( Lutgendorf et al . , 2008 ) and impaired quality of life among cancer patients ( Delgado-Guay et al . , 2009 ) . Stress is related to the incidence or mortality of CRC in women ( Kikuchi et al . , 2017; Kojima et al . , 2005; Nielsen et al . , 2008 ) . Although negative mood is associated with the development of cancer , the underlying neural mechanism remains poorly understood . The hypothalamus is a key brain region in mood regulation ( Price and Drevets , 2010; Schindler et al . , 2012 ) . Oxytocin ( Oxt ) neuropeptide-producing neurons in the paraventricular nucleus ( PVN ) of the hypothalamus are critical in the regulation of anxiety , stress , and depression ( Neumann , 2008; Neumann and Landgraf , 2012 ) . Previous work demonstrated that Oxt was anxiolytic when administered to humans ( Heinrichs et al . , 2003 ) and rodents ( Blume et al . , 2008; Ring et al . , 2006; Windle et al . , 1997 ) , whereas disruption of Oxt gene elevated anxiety level in mice ( Amico et al . , 2004; Mantella et al . , 2003 ) . Hence , Oxt plays a crucial role in mood control . Recent work indicated that nerve fibers of the autonomous nervous system are critically involved in the progressions of prostate ( Magnon et al . , 2013 ) , stomach ( Hayakawa et al . , 2017 ) , and breast cancers ( Kamiya et al . , 2019 ) . Furthermore , the central nervous system ( CNS ) , in particular the hypothalamus , was shown to regulate peripheral tumor progression ( Cao et al . , 2010 ) . However , the neuronal population ( s ) involved in this process remain unclear . In this work , by combining genetic , chemogenetic , pharmacological , and electrophysiological approaches , we show that Oxt neurons in the PVN regulate tumor progression in a CRC mouse model .
Dysregulation of mood is frequently associated with the occurrences of cancer ( Antoni et al . , 2006; Lillberg et al . , 2003 ) , while Oxt produced in the hypothalamus has an anxiolytic effect ( Neumann , 2008; Neumann and Landgraf , 2012 ) , suggesting that modulation of Oxt neurons may impact tumor progression in the periphery . To address this possibility , we crossed the OxtCre ( Wu et al . , 2012 ) with the Rosa26DTA176 knockin ( Wu et al . , 2006 ) mice ( Figure 1A ) . By doing so , we obtained OxtCre and the littermate OxtCre;Rosa26DTA176 ( OxtCre;DTA ) mice , in which the Oxt-producing neurons in the brain had been depleted ( Figure 1B and C ) . To confirm the importance of Oxt neurons in anxiety modulation , we analyzed the anxiety-like behavior of OxtCre and OxtCre;DTA mice . In the open field test , OxtCre;DTA mice spent less time in the central region than that of the OxtCre mice ( Figure 1—figure supplement 1A ) . In the elevated plus maze test , lesion of Oxt neurons decreased the time spent in the open arms ( Figure 1—figure supplement 1B ) . Moreover , in the light/dark box test , depletion of Oxt neurons significantly shortened the time spent in the light box ( Figure 1—figure supplement 1C ) . Thus , lesion of Oxt neurons elevates anxiety level in mice . Next , we administered azoxymethane ( AOM ) and dextran sodium sulfate ( DSS ) into the adult male OxtCre and OxtCre;DTA mice to induce colitis-associated cancer ( CAC ) in the colon and rectum ( Figure 1—figure supplement 1D ) . Depletion of Oxt neurons did not significantly impact the body weight or food intake in mice fed a normal chow diet ( Figure 1—figure supplement 1E , F ) . After the treatment , colorectal tissues and plasma samples were collected . Indeed , the plasma Oxt levels in OxtCre;DTA mice were barely detectable ( Figure 1—figure supplement 1G ) , suggesting the disruption of Oxt-producing neurons . Notably , the number and diameter of CAC were both increased in the OxtCre;DTA mice ( Figure 1D and E; Figure 1—figure supplement 1H ) , while colorectal length was not significantly affected ( Figure 1—figure supplement 1I ) . Depletion of Oxt neurons promoted cell proliferation in the CAC , as demonstrated by the increased number of cells positive for proliferating cell nuclear antigen ( PCNA ) , a marker for proliferating cell ( Figure 1F; Figure 1—figure supplement 1J ) . Moreover , lesion of Oxt neurons inhibited cell apoptosis in the tumors as revealed by the reduced number of cells positive for terminal deoxynucleotidyl transferase dUTP nick end labeling ( TUNEL ) ( Figure 1G; Figure 1—figure supplement 1K ) . Together , these data indicate that depletion of Oxt neurons promotes CAC development in mice . Given that depletion of Oxt neurons elevated anxiety level in mice , and that the dysregulation of hypothalamic-pituitary-adrenal ( HPA ) axis can elicit stress , next , we assessed the circulating adrenocorticotropin ( ACTH ) and corticosterone levels in OxtCre and OxtCre;DTA mice with AOM/DSS-induced CAC . Plasma ACTH and corticosterone levels were evidently increased in the OxtCre;DTA mice comparing with the OxtCre mice ( Figure 1—figure supplement 1L , M ) . Thus , the dysregulation of the HPA axis may contribute to the CAC development in the OxtCre;DTA mice . Next , we asked whether stimulation of Oxt neurons in the PVN ( OxtPVN ) inhibits CAC progression . To do so , we employed the designer receptor exclusively activated by designer drug ( DREADD ) ( Roth , 2016; Smith et al . , 2016 ) approach to manipulate these neurons . Specifically , OxtCre mice were bilaterally injected with adeno-associated virus ( AAV ) carrying GFP ( AAV-hSyn-GFP ) , or Cre-dependent hM3Dq-mCherry into the PVN ( Figure 1H ) . To validate the DREADD system , CAC was induced in virus-injected mice . These animals were then intraperitoneally ( i . p . ) administered with a synthetic ligand , clozapine-N-oxide ( CNO ) every other day for 3 weeks . Two hours after the final dose of CNO , the mice were perfused with 4% paraformaldehyde ( PFA ) , and then brain tissues were harvested . Immunofluorescent staining showed that treatment with CNO elicited a robust c-Fos expression in the OxtPVN neurons of hM3Dq AAV-injected mice compared with the controls ( Figure 1I and J ) , suggesting the activation of these neurons . Mirrored with the results of Oxt neuron depletion , activation of OxtPVN neurons significantly relieved anxiety-like behavior in mice ( Figure 1—figure supplement 2A-C ) . Thereafter , control and hM3Dq-mCherry AAVs were injected into the PVN of OxtCre mice . CAC was induced in these mice using AOM and DSS , and then CNO was i . p . administered every other day for 3 weeks ( Figure 1—figure supplement 2D ) . After the treatment , plasma Oxt level was elevated , whereas body weight and food intake had not been significantly affected in hM3Dq AAV-infected mice ( Figure 1—figure supplement 2E , F ) . Notably , the elevation of plasma Oxt level following chemogenetic excitation of Oxt neurons has been observed previously ( Grund et al . , 2019 ) . Both tumor number and tumor diameter were reduced in mice whose OxtPVN neurons had been excited ( Figure 1K and L; Figure 1—figure supplement 2G ) , whereas colorectal length was not impacted ( Figure 1—figure supplement 2H ) . In agreement with the reduction in tumor size , the number of proliferating cells , revealed by the immunostaining for PCNA , was significantly decreased in hM3Dq AAV-injected mice compared with the controls ( Figure 1M; Figure 1—figure supplement 2I ) . Besides , the TUNEL assay showed that the number of apoptotic cells was evidently increased ( Figure 1N; Figure 1—figure supplement 2J ) . Thus , activation of OxtPVN neurons inhibits CAC progression by suppressing cell proliferation and promoting cell apoptosis . Our assays indicated that plasma ACTH and corticosterone levels were markedly decreased in the hM3Dq AAV-injected mice ( Figure 1—figure supplement 2K , L ) , implying that the reduced activity of HPA axis may contribute to the tumor suppression effect of OxtPVN neuron activation . The activation of the anti-tumor immunity is crucial for cancer treatment , hence , we asked whether any of the immune cells contributes to the anti-tumor effect of OxtPVN neuron activation . To address this question , we assessed these cells in the tumor tissues . Indeed , the number of CD8+ T cells was markedly increased in hM3Dq AAV-injected mice compared with controls ( Figure 1—figure supplement 3A , F ) , and there was no significant change in CD4+ T cells , B cells , NK cells , or macrophages ( Figure 1—figure supplement 3B-E , and G-J ) . Hence , activation of OxtPVN neurons may enhance the anti-tumor immunity by increasing the number of CD8+ T cells . Oxt neurons regulate peripheral physiology via both the neural and the endocrinal pathways ( Zhang et al . , 2021 ) . Next , we asked whether the central action is important for OxtPVN neuron activation to suppress CAC progression . To this end , we elected to centrally block Oxt receptor using L-368 , 899 , an Oxt receptor ( OTR ) antagonist . Specifically , adult male OxtCre mice were bilaterally injected with control or hM3Dq AAV into the PVN , and then CAC was induced using AOM and DSS . Subsequently , these mice were i . p . administered with CNO and i . c . v . injected with aCSF ( artificial cerebrospinal fluid ) or L-368 , 899 every other day for 3 weeks ( Figure 2—figure supplement 1A ) . After the treatment , these mice were perfused with 4% PFA , and then brain tissues were sectioned . Immunofluorescent staining showed that treatment with CNO elicited a dramatic c-Fos expression in the OxtPVN neurons of hM3Dq AAV-injected mice compared with the controls ( Figure 2A and B ) , suggesting the excitation of OxtPVN neurons . Treatment with CNO and L-368 , 899 did not significantly impact the body weight or food intake in mice ( Figure 2—figure supplement 1B , C ) . As anticipated , activation of OxtPVN neurons inhibited CAC progression in mice ( Figure 2C–E ) . Notably , brain treatment with L-368 , 899 significantly abrogated this effect ( Figure 2C–E ) . Colorectal length remained not impacted in the mice administered with CNO and L-368 , 899 ( Figure 2F ) . Moreover , the immunostaining for PCNA revealed that excitation of OxtPVN neurons inhibited cell proliferation , however , this effect was markedly attenuated when the mice were administered with L-368 , 899 ( Figure 2G and H ) . Furthermore , the TUNEL assay showed that the effect of activation of OxtPVN neurons on cell apoptosis was diminished when the mice were administered with L-368 , 899 ( Figure 2I and J ) . Collectively , these data suggest that the tumor suppressive effect of OxtPVN neuron activation is dependent on its action in the CNS . The CNS is known to control peripheral physiology via both the sympathetic nervous system ( SNS ) and the parasympathetic nervous system ( PNS ) . Besides , the sympathetic celiac-superior mesenteric ganglion ( CG-SMG ) predominantly innervates colon and rectum . Hence , we examined the effect of OxtPVN neuron activation on CG-SMG neuronal activity . To do this , adult male OxtCre mice were injected with control and hM3Dq AAV into the PVN . After recovery , these mice were i . p . administered with CNO . Two hours later , CG-SMG was dissected and fixed in 4% PFA . Double immunofluorescence staining for c-Fos and tyrosine hydroxylase ( TH ) , a marker of catecholamine neuron , revealed that the activities of the sympathetic neurons in CG-SMG were significantly inhibited following the activation of OxtPVN neurons ( Figure 3A and B ) . To confirm this OxtPVN neuron -> THCG-SMG neuron pathway , we cut the preganglionic nerve fiber of CG-SMG , and then assessed the neuronal activity in this ganglion using in vivo single-unit recordings . Specifically , adult male OxtCre mice were injected with control and hM3Dq AAV into the PVN , and were also implanted with infusion cannula directed to the third ventricle . After recovery , these animals were performed sham operations , or the transection of the preganglionic fiber of CG-SMG ( Figure 3—figure supplement 1A-C ) . Subsequently , the 6 min control ( 1% DMSO in aCSF ) spiking activity was acquired before CNO ( 1 µg per mouse ) application through the pre-implanted cannula . Single-unit spikes from 30 ( sham ) and 34 ( transection ) CG-SMG neurons were isolated , and the firing rates were compared before and after CNO infusion ( Figure 3—figure supplement 1D ) . Group data showed that i . c . v . administration of CNO significantly reduced the firing frequency of CG-SMG neurons , however , transection of preganglionic fiber significantly abolished this effect ( Figure 3—figure supplement 1E ) . Scatterplot of mean firing frequency of individual CG-SMG neuron revealed a mixed modulation following OxtPVN neurons activation ( Figure 3—figure supplement 1F , G ) . The majority of CG-SMG neurons ( 67% ) displayed a decreased firing frequency after CNO infusion . Only a small proportion of neurons ( 16% ) showed an increased firing frequency . The remainder ( 17% ) maintained their activity level after CNO infusion . Yet , after the transection of the preganglionic fiber , the majority of CG-SMG neurons ( 65% ) maintained their activity level after CNO infusion . Hence , following OxtPVN neuron activation , the signal that leads to the suppression of CG-SMG neurons is transmitted through the preganglionic fiber . Next , we assessed the OxtPVN neuron -> THCG-SMG neuron connection using the CAC mouse model . To this end , CAC was induced in the adult OxtCre and OxtCre;DTA mice using AOM and DSS . After the first cycle of DSS treatment , CG-SMG resection and sham surgeries were performed in mice ( Figure 3C and D ) . These manipulations did not significantly impact body weight or food intake in mice ( Figure 3—figure supplement 2A , B ) . While depletion of Oxt neurons led to the increasing of CAC number and diameter , CG-SMG resection markedly attenuated these effects ( Figure 3E–G ) . We noted that colorectal length was not affected in these mice ( Figure 3—figure supplement 2C ) . In agreement with the data of tumor number and size , the effects on cell proliferation and cell apoptosis were both attenuated when CG-SMG were removed from these mice ( Figure 3H–K ) . Taken together , the promotion of CAC development owing to Oxt neuron deficiency is mediated by the sympathetic CG-SMG . Celastrol is a pentacyclic triterpenoid initially extracted from the root of thunder god vine . A recent study showed that treatment with celastrol decreased the body weight in obese mice , but not mice with normal weight ( Ma et al . , 2015 ) . A following study suggested that hypothalamus is critical for celastrol to regulate energy balance ( Liu et al . , 2015 ) . Therefore , we assessed the effect of i . c . v . administered celastrol on hypothalamic neuronal activity . The data showed that the number of c-Fos-positive cells was increased in the PVN , but not other nuclei ( Figure 4—figure supplement 1A , B ) , suggesting that brain treatment with celastrol stimulates neurons in the PVN . Oxt neurons in the PVN play a critical role in energy balance control , therefore , we asked whether its activity is modulated by celastrol . To answer this question , we analyzed Oxt neuron excitability after bath application of celastrol via slice electrophysiology . The hypothalamic slices were obtained from OxtCre;Rosa26-LSL-EYFP ( OxtCre;EYFP ) mice , in which enhanced yellow fluorescent protein ( EYFP ) was expressed in Oxt neurons ( Figure 4A ) . In response to 500 ms current steps , Oxt neurons fired more action potentials ( AP ) across increasing current injections in celastrol condition , suggesting an enhanced neuronal excitability ( Figure 4B and C ) . We also analyzed the AP waveforms , and found that celastrol increased the size of afterhyperpolarization ( Figure 4D and E ) , but did not impact AP threshold , AP amplitude , AP half-width , or AP area ( Figure 4D and F; Figure 4—figure supplement 1C-E ) . Moreover , celastrol increased input resistance of Oxt neurons , which might increase neuronal excitability ( Figure 4G ) . These data implicate that celastrol enhances Oxt neuron firing . Besides , the above data suggested that celastrol might promote Oxt release from the OxtPVN neurons . To address this possibility , we carried out an ex vivo Oxt release assay . The PVN slices were dissected from the male adult C57 BL/6 mice . These tissue slices were balanced in normal Locke’s solution , and then in the same solution supplemented with celastrol . The data showed that treatment with celastrol enhanced the rate of Oxt releasing ( Figure 4—figure supplement 1F ) . Altogether , these data demonstrate that celastrol could excite OxtPVN neurons . Next , we assessed the effect of brain administered celastrol on CAC progression . To this end , CAC was induced in adult male C57 BL/6 mice using AOM and DSS ( Figure 4—figure supplement 2A ) . These mice were then implanted with a guide cannula directed to the third ventricle . After surgical recovery , vehicle and celastrol were administered into the third ventricle via the pre-implanted cannula every other day for 3 weeks ( Figure 4—figure supplement 2A ) . Mice receiving celastrol treatment exhibited higher plasma Oxt level than that of the controls ( Figure 4—figure supplement 2B ) , suggesting that this chronic treatment stimulated OxtPVN neurons . Consistent with the previous study ( Liu et al . , 2015 ) , treatment with celastrol did not impact energy balance in CAC mice with normal body weights ( Figure 4—figure supplement 2C , D ) . This treatment significantly reduced tumor number and diameter ( Figure 4H , I; Figure 4—figure supplement 2E ) , while it did not affect colorectal length ( Figure 4J ) . Besides , cell proliferation was suppressed , and cell apoptosis was enhanced in the tumor tissue of mice treated with celastrol ( Figure 4K and L; Figure 4—figure supplement 2F , G ) . Collectively , these data indicate that brain treatment with celastrol suppresses CAC progression in mice . The above data suggested that hypothalamic Oxt neurons are important for celastrol to suppress CAC progression in mice . To address this question , the CAC was induced in the OxtCre and OxtCre;DTA mice ( Figure 5A ) . These mice were then i . p . injected with vehicle versus celastrol every other day for 3 weeks ( Figure 5A ) . Treatment with celastrol did not significantly impact the body weight or food intake in mice ( Figure 5B and C ) . While celastrol inhibited CAC progression in mice , lesion of Oxt neurons could markedly abrogate this effect ( Figure 5D–F ) . Lesion of Oxt neuron or celastrol treatment did not have noticeable effect on colorectal length ( Figure 5G ) . Notably , the effects of celastrol on cell proliferation and cell apoptosis in CAC were both attenuated in the mice deficient for Oxt neurons ( Figure 5H–K ) . Thus , hypothalamic Oxt neurons are required for celastrol to suppress CAC progression . Next , we interrogated whether activation of SNS target , that is , β2 adrenergic receptor ( β2AR ) , would attenuate the anti-tumor effect of OxtPVN neuron excitation . Our data showed that isoprenaline , an agonist for β2AR , did not affect the activity of CG-SMG neurons ( Figure 6—figure supplement 1A , B ) , suggesting that it is proper to use this drug to target CAC cells . Thereafter , adult male OxtCre mice were bilaterally injected with control and hM3Dq AAV into the PVN , and then CAC was induced . These mice were i . p . administered with CNO every other day , and were also i . p . injected with saline or isoprenaline on a daily basis . These treatments were continued for 3 weeks ( Figure 6—figure supplement 1C ) . Subsequently , these mice were perfused with 4% PFA , and then brain tissues were sectioned . Immunofluorescent staining showed that treatment with CNO elicited a robust c-Fos expression in the OxtPVN neurons of hM3Dq AAV-injected mice compared with the controls ( Figure 6A and B ) , suggesting the activation of these neurons . Treatment with CNO and/or isoprenaline did not impact the body weight or food intake in control and hM3Dq AAV-injected mice ( Figure 6—figure supplement 1D , E ) . Excitation of OxtPVN neurons suppressed CAC progression in mice , however , activation of β2AR with isoprenaline significantly abolished this effect ( Figure 6C–E ) . Colorectal length was not significantly impacted in the mice administered with isoprenaline ( Figure 6F ) . The histological data revealed that the effects of OxtPVN excitation on cell proliferation and cell apoptosis were dramatically attenuated when isoprenaline was administered ( Figure 6G–J ) . Hence , activation of β2AR can significantly abrogate the anti-tumor effect of OxtPVN neuron activation . Our data indicated that Oxt neurons are important for celastrol to restrict CAC development in mice ( Figure 5 ) . Next , we asked whether i . c . v . administered celastrol could similarly regulate CG-SMG neuronal activity . To address this question , adult male C57 BL/6 mice were implanted with a guide cannula , and were then allowed to recover from surgeries . Subsequently , the preganglionic fiber of CG-SMG was transected , or left intact ( sham ) . These mice were i . c . v . administered with vehicle versus celastrol . Two hours later , CG-SMG was dissected and fixed in 4% PFA . Double immunofluorescence staining for c-Fos and TH revealed that administration of celastrol suppressed the activity of sympathetic neurons in the CG-SMG . Notably , this effect was markedly diminished when the preganglionic nerve fiber of CG-SMG was transected ( Figure 7—figure supplement 1A-C ) . Thereafter , we asked whether brain OTR is crucial for centrally administered celastrol to suppress the CG-SMG neuronal activity . To this end , adult male C57 BL/6 mice were implanted with a guide cannula directed to the third ventricle . After surgical recovery , these mice were i . c . v . administered with vehicle control or L-368 , 899 , the OTR antagonist , an hour before in vivo single-unit recordings . Subsequently , the 6 min control spiking activity was acquired before celastrol application through the guide cannula ( Figure 7A and B ) . Single-unit spikes from 68 CG-SMG neurons ( vehicle ) and 44 CG-SMG neurons ( OTR antagonist ) were isolated , and the firing rates were compared before and after celastrol infusion ( Figure 7C and D ) . Group data showed that treatment with celastrol significantly reduced the firing frequency of CG-SMG neurons , however , blockade of OTR abrogated this effect ( Figure 7E ) . Scatterplot of mean firing frequency of individual CG-SMG neuron revealed a mixed modulation by celastrol ( Figure 7F ) . The majority of CG-SMG neurons ( 63% ) displayed a decreased firing frequency after celastrol infusion . Only a small proportion of neurons ( 18% ) showed an increased firing frequency . The remainder ( 19% ) maintained their activity level during celastrol infusion . However , when L-368 , 899 was applied , the majority of CG-SMG neurons ( 57% ) maintained their activity level during celastrol infusion ( Figure 7G ) , suggesting that blockade of OTR could attenuate the inhibitory effect of celastrol on neuronal firing rate in CG-SMG . Together , these data suggest that brain OTR is crucial for centrally administered celastrol to suppress the neuronal activity in the CG-SMG . Lastly , we interrogated whether the activation of β2AR could attenuate the anti-tumor effect of celastrol . To do so , the AOM/DSS-induced CAC mice were implanted with a guide cannula directed to the third ventricle . After recovery , these animals were i . c . v . administered with vehicle versus celastrol every other day for 3 weeks . Besides , these mice received daily saline or isoprenaline treatment ( Figure 7—figure supplement 2A ) . Treatment with celastrol and/or isoprenaline did not impact the body weight or food intake in mice ( Figure 7—figure supplement 2B , C ) . As anticipated , brain treatment with celastrol suppressed CAC progression in mice . Yet , treatment with isoprenaline significantly abrogated this effect ( Figure 7H , I; Figure 7—figure supplement 2D ) . Treatment with celastrol and/or isoprenaline did not impact colorectal length ( Figure 7—figure supplement 2E ) . The immunohistochemistry data revealed that treatment with celastrol inhibited cell proliferation , however , this effect was markedly attenuated when the mice were administered with isoprenaline ( Figure 7J; Figure 7—figure supplement 2F ) . Besides , the TUNEL assay showed that the effect of brain treatment with celastrol on cell apoptosis was diminished when the mice were treated with isoprenaline ( Figure 7K; Figure 7—figure supplement 2G ) . Overall , these data suggest that activation of β2AR can significantly abolish the anti-tumor effect of centrally administered celastrol .
Negative mood is associated with the occurrences of cancers , however , the underlying mechanisms remain less well understood . In this study , we show that excitation of OxtPVN neurons remarkably ameliorated CAC progression in mice , and that this effect was mediated by inhibiting the neuronal activities in the CG-SMG . Also , brain treatment with celastrol suppressed the progression of CAC , and this effect required hypothalamic Oxt neurons . Moreover , we show that β2AR was involved in these processes . Together , our current work demonstrates that modulating hypothalamic Oxt neurons can impact the CAC progression in mice . Negative moods , such as anxiety , depression , and stress , are implicated in tumor progression . As for CRC , a recent study has revealed a significant association of perceived stress with the incidences of rectal cancer ( Kikuchi et al . , 2017 ) . Perceived stress at work and stressful life events elevated the risk of CRC ( Azizi and Esmaeili , 2015; Blanc-Lapierre et al . , 2017 ) . Besides , stress is one of the key contributing factors to the onset and development of spontaneous colitis in humans ( Mitchell and Drossman , 1987; Salem and Shubair , 1967 ) . This association , in particular the one between chronic stress and colitis , was further confirmed in murine models ( Gao et al . , 2018; Reber et al . , 2006; Reber et al . , 2008 ) . Moreover , chronic psychosocial stress was shown to result in the deterioration of CAC progression in mice ( Peters et al . , 2012 ) . Hence , these findings suggest that stress is critical for CRC progression . Previous studies showed that Oxt has an anxiolytic effect in both humans ( Heinrichs et al . , 2003 ) and rodents ( Blume et al . , 2008; Ring et al . , 2006; Windle et al . , 1997 ) . Conversely , our current and others’ previous studies ( Amico et al . , 2004; Mantella et al . , 2003 ) demonstrated that disruption of Oxt neuron or Oxt gene increased anxiety level in mice . Importantly , we show that depletion of Oxt neuron promoted tumor progression in CAC mice , which agrees with the previous findings showing that increased stress level could promote colorectal tumor progression . Remarkably , our data indicated that chronic excitation of OxtPVN neurons or treatment with celastrol could significantly inhibit CAC progression in mice . These results are consistent with previous reports displaying that social support reduced the risk of colon cancer ( Ikeda et al . , 2013; Kinney et al . , 2003 ) . Besides , recent work demonstrated that Oxt has a prosocial role in humans ( Kosfeld et al . , 2005 ) and rodents ( Lukas et al . , 2011; Teng et al . , 2013 ) . Altogether , these findings suggest that the anxiolytic property of Oxt is critically important in its anti-tumor effect . Previous studies unveiled a crucial role for nerve fiber in the tumorigenesis of various organs and tissues . For instance , both the densities of SNS and PNS nerve fibers were correlated with the aggressiveness of human prostate cancer ( Magnon et al . , 2013 ) . Intriguingly , blockade of SNS activity suppressed the development of prostate cancer , whereas blockade of PNS activity inhibited the invasion and metastasis of prostate cancer in mice ( Magnon et al . , 2013 ) . A further study indicated that norepinephrine released from SNS nerves drove angiogenesis in prostate cancer ( Zahalka et al . , 2017 ) . Besides , a recent study showed that vagal innervation contributed to the development of stomach cancer via muscarinic acetylcholine M3 receptor ( Zhao et al . , 2014 ) . Infiltration of nerve fibers was associated with the aggressiveness of breast cancer ( Pundavela et al . , 2015 ) . The sensory neurons were able to facilitate the initiation and progression of pancreatic ductal adenocarcinoma in mice ( Saloman et al . , 2016 ) . Together , these findings underscore an important role for nerve fiber of the autonomous nervous system in the initiation , invasion , or metastasis of cancers in peripheral organs , and hence the term ‘cancer neuroscience’ was coined ( Demir et al . , 2020; Monje et al . , 2020 ) . However , whether the CNS is similarly important remains largely unknown . In this work , we show that stimulation of OxtPVN neurons could suppress CAC progression in mice . Thus , in concert with other evidence ( Cao et al . , 2010; Liu et al . , 2014 ) , our current study implicates a critical role for the CNS , in particular the hypothalamus , in peripheral tumor development . In summary , our current study indicates that chemogenetic stimulation of OxtPVN neurons or brain treatment with celastrol can suppress CAC progression in mice . The anti-tumor effect of celastrol requires hypothalamic Oxt neurons . Overall , these results suggest that modulating Oxt neuronal activity might be a relevant strategy for the treatment of CRC .
The OxtCre ( Wu et al . , 2012 ) mouse line was purchased from the Jackson Laboratory ( Bar Harbor , ME ) . Rosa26DTA176 ( Wu et al . , 2006 ) and Rosa26-LSL-EYFP ( Srinivas et al . , 2001 ) mice have been described previously . We generated the OxtCre;Rosa26DTA176 mice by crossing the OxtCre with the Rosa26DTA176 mice , and the OxtCre;Rosa26-LSL-EYFP ( OxtCre;EYFP ) mice by crossing the OxtCre with the Rosa26-LSL-EYFP mice . C57 BL/6 mice were purchased from the Vital River Laboratory Animal Technology ( Beijing , China ) . Rodent chow diet was purchased from HFK Bioscience ( Beijing , China ) . All mice were housed in a 12-hr light/12-hr dark cycle in a temperature-controlled room ( 22–24°C ) . Rabbit and goat anti-c-Fos , mouse anti-TH , anti-CD4 , and anti-CD11b antibodies were purchased from Santa Cruz Biotechnology ( Santa Cruz , CA ) . Rabbit anti-c-Fos antibody was purchased from Abcam ( Cambridge , UK ) . Rabbit anti-Oxt antibody was obtained from Immunostar ( Hudson , WI ) . Mouse anti-PCNA antibody was purchased from Boster Biological ( Wuhan , China ) . Rabbit anti-CD8α antibody was purchased from Bioss ( Woburn , MA ) . Rat anti-B220 and mouse anti-NK1 . 1 antibodies were obtained from BD Biosciences ( San Diego , CA ) . Alexa Fluor ( AF ) 488 goat anti-rabbit , AF 555 donkey anti-rabbit , AF 633 donkey anti-goat , and AF 488 donkey anti-mouse secondary antibodies were purchased from Thermo Fisher ( Waltham , MA ) . Azoxymethane , isoprenaline , and Avertin were purchased from Sigma-Aldrich ( St Louis , MO ) . Dextran sulfate sodium was obtained from TdB Labs ( Uppsala , Sweden ) . CNO was purchased from MedChemExpress ( Monmouth Junction , NJ ) . Celastrol was obtained from Mengry Bio-Technology ( Shanghai , China ) . L-368 , 899 was purchased from Santa Cruz Biotechnology . Male mice were i . p . injected with the azoxymethane ( 12 . 5 mg kg–1 ) . A week later , mice were administrated with two cycles of 5-day oral exposure to DSS ( 2 . 5% in drinking water ) and then 16-day normal drinking water ( Neufert et al . , 2007 ) . Third ventricle cannulation: The procedures have been described before ( Wu et al . , 2017; Zhang et al . , 2008 ) . Briefly , mice were anesthetized with Avertin ( 300 mg kg–1 ) and were then placed on an ultra-precise stereotaxic instrument ( David Kopf , Tujunga , CA ) . Next , a guide cannula ( RWD Life Science , Shenzhen , China ) was placed directed to third ventricle ( coordinates: A/P –2 . 0 mm posterior to bregma , D/V –5 . 0 mm ) . Mice were allowed to fully recover from surgeries . For AAV injection , mice were anesthetized and placed on the stereotaxic instrument . With the help of a guide cannula , viral solution was injected bilaterally into the PVN ( coordinates: A/P , –0 . 85 mm posterior to bregma , M/L , ± 0 . 2 mm , D/V , –4 . 8 mm ) . AAVs carrying GFP ( AAV-hSyn-GFP ) or Cre-dependent hM3Dq-mCherry ( AAV-hSyn-DIO-hM3Dq-mCherry ) were purchased from Obio Technology ( Shanghai , China ) . Adult male OxtCre mice were bilaterally injected with AAVs into the PVN , and were then allowed to recover from surgeries . After the induction of CAC , mice were i . p . administered with CNO ( 3 mg kg–1 , every other day for 3 weeks ) to activate the hM3Dq-expressing Oxt neurons . Treatment with CNO and L-368 , 899: The control and hM3Dq AAVs were injected into the PVN of adult OxtCre mice . CAC was induced using AOM and DSS . These mice were i . p . injected with CNO and i . c . v . administered with vehicle or L-368 , 899 ( 2 µg per mouse ) every other day for 3 weeks . Body weight and food intake in mice were recorded throughout the experiment . Celastrol: Adult male C57 BL/6 mice bearing AOM and DSS-induced CAC were implanted with a guide cannula directed to the third ventricle , and were then allowed to recover from surgeries . aCSF and celastrol ( 0 . 5 µg per mouse ) was i . c . v . administered every other day for 3 weeks . In a separate experiment , adult male and female OxtCre and OxtCre;DTA mice were administered with AOM and DSS to induce CAC , and were then i . p . injected with vehicle ( 1% DMSO in saline ) or celastrol ( 1 mg kg–1 ) every other day for 3 weeks . Body weight and food intake were regularly assessed throughout the experiment . Treatment with CNO and isoprenaline: The control and hM3Dq AAVs were injected into the PVN of male OxtCre mice , in which CAC was then induced . These mice were i . p . administered with CNO ( 3 mg kg–1 ) every other day for 3 weeks . During this period , saline and isoprenaline ( 10 mg kg–1 ) were i . p . administered on a daily basis . Body weight and food intake in mice were assessed . Treatment with celastrol and isoprenaline: Adult male C57 BL/6 mice bearing CAC were i . c . v . administered with vehicle or celastrol ( 0 . 5 µg per mouse ) every other day for 3 weeks . In the meanwhile , these mice were i . p . injected with saline or isoprenaline ( 10 mg kg–1 ) on a daily basis . Body weight and food intake in mice were measured . Mice were anesthetized using Avertin , and then the abdomen was cut open . Abdominal viscera were gently pulled out and held in warm sterile saline-soaked gauze . The intersection of the descending aorta and the left renal artery was identified , where the superior mesenteric artery was located . The CG-SMG is wrapped around the superior mesenteric artery and associated lymphatic vessels . Fine forceps and microdissection scissor were used to remove CG-SMG or transect its preganglionic nerve fiber . The OxtCre;EYFP mice ( 4 months of age ) were euthanized with an overdose of sodium pentobarbital ( 40 mg kg–1 , i . p . ) . Coronal PVN slices ( 300 μm in thickness ) were cut in a solution containing ( in mM ) : 228 sucrose , 26 NaHCO3 , 11 glucose , 2 . 5 KCl , 1 NaH2PO4 , 7 MgSO4 , and 0 . 5 CaCl2 , and recovered in aCSF containing ( in mM ) : 119 NaCl , 26 NaHCO3 , 11 glucose , 2 . 5 KCl , 1 NaH2PO4 , 1 . 3 MgSO4 , and 2 . 5 CaCl2 . Recordings were performed in a submerged-style chamber mounted under an infrared-differential interference contrast microscope ( BX-51 WI , Olympus , Tokyo , Japan ) . Slices were constantly perfused with heated aCSF ( 35°C ) and bubbled continuously with 95% O2 and 5% CO2 . Oxt neurons were identified by EYFP epifluorescence . Whole-cell recordings were achieved using a Multiclamp 700B amplifier ( Molecular Devices , San Jose , CA ) . Signals were filtered at 10 kHz , and then sampled by Digidata 1550B4 ( Molecular Devices ) at 20 kHz using Clampex 10 acquisition software . The pipette resistance was about 4–6 MΩ with an internal solution containing ( in mM ) : 135 K-gluconate , 8 KCl , 10 HEPES , 0 . 25 EGTA , 2 MgATP , 0 . 3 Na3GTP , 0 . 1 spermine , 7 phospho-creatine ( pH 7 . 25–7 . 3; osmolarity 294–298 ) . For celastrol condition , celastrol ( 5 μM ) was added to the incubation chamber 20 min prior to recording and was added in bath aCSF throughout recording . Liquid junction potential ( 16 mV ) has been corrected in the text and figures . Male mice ( 8 weeks of age ) were implanted with a guide cannula directed to the third ventricle . Two weeks later , in vivo single-unit recordings were performed and analyzed as described previously ( Tseng et al . , 2011 ) . The guide tubes housed 16-channel electrodes using 25 . 4 μm formvar-insulated nichrome wire ( 761500 , A-M System , Sequim , WA ) . The final impedance of the electrodes was 700–800 kΩ . On the recording day , the CG-SMG located at the intersection of the descending aorta and left renal artery was identified , and the 16-channel electrodes were manually placed into CG-SMG . A sterile cotton swab was dipped in saline solution , and was then placed by the CG-SMG to maintain tissue humidity during recording . Spiking activities were digitized at 40 kHz , bandpass-filtered from 250 to 8000 Hz , and stored on a PC for further offline analysis . For administration of celastrol and L-368 , 899 , the C57 BL/6 mice were implanted with an infusion cannula directed to third ventricle and were then singly housed to allow recovery from surgeries . On the recording day , aCSF and L-368 , 899 were applied through the pre-implanted cannula 1 hr before recordings . The 6 min control ( 5% DMSO in aCSF ) spiking activity was acquired before celastrol ( 0 . 5 µg per mouse ) application through the infusion cannula . In the CG-SMG preganglionic nerve fiber transection experiment , adult OxtCre mice were injected with control or hM3Dq AAV into the PVN . These mice were also implanted with an infusion cannula directed to third ventricle . After recovery , the preganglionic nerve fiber of CG-SMG was transected before recording . In the control group , sham operations were carried out before recording . Subsequently , the 6 min control ( 1% DMSO in aCSF ) spiking activity was acquired before CNO ( 1 µg per mouse ) application through the infusion cannula . The single-unit spike sorting was performed with Offline Sorter V4 . 0 ( Plexon , Dallas , TX ) . Spikes were detected when a minimum waveform reached an amplitude threshold of –4 . 50 standard deviation greater than the noise amplitude . Principal component analysis and automatic scan were employed to separate single-unit waveforms into individual clusters . Manual checking was then performed to ensure that the cluster boundaries were clearly separated . All isolated single units exhibited recognizable refractory periods ( >1 ms ) in the inter-spike interval histograms . Only well-isolated units ( L ratio <0 . 2 , isolation distance >15 ) were included in the data analysis . The response of single unit was analyzed with Neuroexplorer V5 . 0 ( Plexon ) . Well-separated units were used to analyze the responses before ( baseline ) and after celastrol or CNO infusion . Firing rates of neurons during baseline , 10 and 20 min after celastrol or CNO infusion were compared to determine the significance of difference in firing rates ( paired Student’s t-test , 95% confidence interval ) . For heatmap analysis , z-score of each bin ( 10 s ) was calculated by the following equation: z = ( x-μ ) /σ , in which x is the raw firing rate , μ is the mean firing rate during the baseline period , and σ is the corresponding standard deviation . Further normalization was utilized for better presentation . All of the single-unit z-scores were plotted using Matlab R2019b ( Natick , MA ) . The detailed procedures have been described previously ( Shen et al . , 2020 ) . Mice were anesthetized using Avertin , and were then transcardially perfused with 4% PFA . Mouse brains were removed , post-fixed in 4% PFA , and infiltrated with 20–30% sucrose solutions . Brain tissues were sectioned using a cryostat . Tissue sections were washed with phosphate buffered saline ( PBS ) , blocked with 5% serum/0 . 3% Triton X-100/PBS for 30 min , incubated with primary antibodies at 4 °C overnight , and fluorophore-conjugated secondary antibodies at room temperature for 1 hr . Cell nuclei were counterstained with DAPI . Immunofluorescence staining of CG-SMG: Mice were euthanized , and then the CG-SMG were dissected , fixed in 4% PFA for 10 min . The tissues were infiltrated with 75–100% ethanol , and were then embedded in paraffin and sectioned ( thickness: 3 μm ) . The tissue sections were deparaffinized and rehydrated using graded ethanol . Antigen retrieval was then performed . Tissue sections were washed with 1× PBS , blocked with 5% serum/0 . 3% Triton X-100/PBS for 30 min , incubated with primary antibodies at 4°C overnight , and fluorophore-conjugated secondary antibodies at room temperature for 1 hr . Cell nuclei were counterstained with DAPI . Images were acquired with the LSM 780 confocal microscope ( Carl Zeiss , Jena , Germany ) . Cells were manually counted in one representative image collected for each mouse . Paraffin-embedded tissue sections were deparaffinized , rehydrated , and antigen-recovered . Sections were then blocked with 5% serum/0 . 3% Triton X-100/PBS for 30 min , incubated with primary antibodies at 4°C overnight and followed by a reaction using a SABC-POD kit ( Boster Biological ) . Images were acquired using an IX71 microscope ( Olympus ) . Cells were counted using Photoshop ( Adobe , San Jose , CA ) . The In Situ Cell Death Detection Kit was purchased from Sigma-Aldrich . Paraffin-embedded tissue sections were deparaffinized and rehydrated . Next , tissue sections were rinsed in distilled water , incubated with proteinase K ( 18 . 5 µg ml–1 in 10 mM Tris·HCl ) at 37°C for 15 min , washed with 1× PBS , and were then incubated with TUNEL reaction mixture in the humidified chamber at 37 °C for 1 hr . Cell nuclei were counterstained with DAPI . Images were acquired with the LSM 780 confocal microscope . TUNEL-positive cells were manually counted using Photoshop . Open field test: Adult male OxtCre , OxtCre;DTA mice , and the OxtCre mice injected with control or hM3Dq AAV were placed in an opaque , square open field ( 40 cm L × 40 cm W × 40 cm H ) , and were then allowed to freely explore for 5 min and monitored with the ImageOF software ( https://cbsn . neuroinf . jp/modules/xoonips/detail . php ? id=ImageOF ) . The open field was divided into a peripheral region and a 13 . 3 cm × 13 . 3 cm central region . Time spent in the central versus peripheral region during the test was presented . Elevated plus maze test: the plus maze had two closed arms ( 35 cm L × 6 cm W × 22 cm H ) and two open arms ( 35 cm L × 6 cm W ) . The maze was elevated 74 cm from the floor . Mice were placed on the center section and allowed to explore the maze freely and monitored with ImageEP software ( Komada et al . , 2008 ) . Time spent in the open versus closed arms during the 5 min period was presented . Light/dark box test: The apparatus was comprised of a dual compartment box ( 20 cm L × 20 cm W × 40 cm H ) with free access between them . The dark box was made of black Plexiglass and the light one was exposed to room light . The exploratory activity was monitored for 5 min using the ImageLD software ( Takao and Miyakawa , 2006 ) . Time spent in the light versus dark box was presented . The detailed procedures have been described previously ( Zhang et al . , 2011 ) . In order to determine the effect of celastrol on Oxt release , PVN tissue slices were dissected from the brain of C57 BL/6 mice and were balanced in normal Locke’s solution supplied with 95% O2 and 5% CO2 at 37 °C . The solution was changed every 5 min for 10 times and the 11th sample was collected to measure the basal Oxt release rate . The slices were then incubated in the same solution containing celastrol ( 5 μM ) for 5 min and this solution was measured to determine the Oxt release rate under celastrol condition . An oxytocin EIA kit ( Enzo Life Sciences , Farmingdale , NY ) was used to determine the Oxt concentration in the solutions . The plasma was collected from mice after the completion of the experiments . Plasma Oxt and corticosterone levels were determined using the Oxt EIA kit and a corticosterone ELISA kit ( Enzo Life Sciences ) , respectively . Plasma ACTH was assessed using an ACTH ELISA kit ( Aviva Systems Biology , San Diego , CA ) . All data are presented as means ± SEM unless otherwise specified . Sample sizes with sufficient power were determined according to our published studies and relevant literature . Animals were assigned to specific experimental groups without bias . Data were analyzed using Prism 8 ( GraphPad Software , San Diego , CA ) or Matlab R2019b . Data distribution was assumed to be normal but this was not formally tested . Two-group comparisons were assessed using two-tailed Student’s t-test . One-way and two-way analysis of variance ( ANOVA ) with Bonferroni’s post hoc test was used for comparisons of more than two groups . Key experiments were repeated at least twice independently . No data were excluded from the analyses . When necessary , experimental performers were blind to group information before data were obtained . A p-value of less than 0 . 05 was considered statistically significant . | Colorectal ( or ‘bowel’ ) cancer killed nearly a million people in 2018 alone: it is , in fact , the second leading cause of cancer death globally . Lifestyle factors and inflammatory bowel conditions such as chronic colitis can heighten the risk of developing the disease . However , research has also linked to the development of colorectal tumours to stress , anxiety and depression . This ‘brain-gut’ connection is particularly less-well understood . One brain region of interest is the hypothalamus , an almond-sized area which helps to regulate mood and bodily processes using chemical messengers that act on various cells in the body . For instance , Oxt neurons in the hypothalamus produce the hormone oxytocin which regulates emotional and social behaviours . These cells play an important role in modulating anxiety , stress and depression . To investigate whether they could also influence the growth of colorectal tumours , Pan et al . used various approaches to manipulate the activity of Oxt neurons in mice with colitis-associated cancer . Disrupting the Oxt neurons in these animals increased anxiety-like behaviours and promoted tumour growth . Stimulating these cells , on the other hand , suppressed cancer progression . Further experiments also showed that treating the mice with celastrol , a plant extract which can act on the hypothalamus , stimulated Oxt neurons and reduced tumour growth . In particular , the compound worked by acting on a nerve structure in the abdomen which relays messages to the gut . These preliminary findings suggest that the hypothalamus and its Oxt-producing neurons may influence the progression of colorectal cancer in mice by regulating the activity of an abdominal ‘hub’ of the nervous system . Modulating the activity of Oxt-producing neurons could therefore be a potential avenue for treatment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience",
"cancer",
"biology"
] | 2021 | Stimulation of hypothalamic oxytocin neurons suppresses colorectal cancer progression in mice |
In the antisaccade task , which is considered a sensitive assay of cognitive function , a salient visual cue appears and the participant must look away from it . This requires sensory , motor-planning , and cognitive neural mechanisms , but what are their unique contributions to performance , and when exactly are they engaged ? Here , by manipulating task urgency , we generate a psychophysical curve that tracks the evolution of the saccadic choice process with millisecond precision , and resolve the distinct contributions of reflexive ( exogenous ) and voluntary ( endogenous ) perceptual mechanisms to antisaccade performance over time . Both progress extremely rapidly , the former driving the eyes toward the cue early on ( ∼100 ms after cue onset ) and the latter directing them away from the cue ∼40 ms later . The behavioral and modeling results provide a detailed , dynamical characterization of attentional and oculomotor capture that is not only qualitatively consistent across participants , but also indicative of their individual perceptual capacities .
Neuroscience aims to explain macroscopic behavior based on the microscopic operation of distinct neural circuits , and this requires carefully designed tasks that expose the relationship between the two . In the case of the antisaccade task ( Coe and Munoz , 2017; Munoz and Everling , 2004 ) , in which participants are instructed to withhold responding to a visual cue in favor of programming a saccade to a diametrically opposed location , performance relies heavily on frontal cortical mechanisms associated with cognitive control ( Guitton et al . , 1985; Everling and Fischer , 1998; Munoz and Everling , 2004; Condy et al . , 2007; Luna et al . , 2008; Hakvoort Schwerdtfeger et al . , 2012 ) , and the paradigm is considered to be a sensitive assay of impulsivity and executive function in general . Indeed , the mean reaction time ( RT ) and overall error rate in the antisaccade task are frequently used as biomarkers for cognitive development ( Klein and Foerster , 2001; Luna et al . , 2008; Coe and Munoz , 2017 ) and , in clinical settings , for mental dysfunction ( Everling and Fischer , 1998; Munoz et al . , 2003; Hutton and Ettinger , 2006; Antoniades et al . , 2015; Wiecki et al . , 2016 ) . The antisaccade task pits against each other two fundamental processes , one involuntary and the other voluntary . On one hand , the sudden appearance of a salient visual stimulus automatically attracts spatial attention ( Theeuwes , 1991; Theeuwes et al . , 1998; Ruz and Lupiáñez , 2002; Busse et al . , 2008; Theeuwes , 2010; Carrasco , 2011; Aagten-Murphy and Bays , 2017 ) to produce either a covert shift ( ‘attentional capture’ ) or an overt saccade ( ‘oculomotor capture’ ) . In either case , the effect is described as bottom-up or exogenous , and is thought to be fast and transient . On the other hand , programming a saccade away from a cue is a top-down or endogenous process that likely summons several mechanisms , such as working memory ( Roberts et al . , 1994; Lavie and De Fockert , 2005 ) and endogenous attention ( Godijn and Theeuwes , 2002; Theeuwes , 2010; Carrasco , 2011 ) , which are thought to be slower and to require a sustained cognitive effort . Thus , the rationale for the task is sound — the timing and intensity of the conflict between bottom-up and top-down mechanisms should correlate with behavior , and with the dynamics of the underlying attentional and oculomotor neural circuits . There is a problem , however: such conflict must unfold very quickly . First , exogenous attention is thought to be mediated by visually-driven responses in oculomotor areas such as the frontal eye field ( FEF ) and superior colliculus ( SC ) , which have latencies of at least 50 ms ( Gottlieb and Goldberg , 1999; Bisley et al . , 2004; Thompson et al . , 2005; Ipata et al . , 2006; Joiner et al . , 2017; White et al . , 2017; Chen et al . , 2018 ) . And second , spatial attention can be endogenously shifted roughly 150 ms after a relevant cue is provided ( Kim and Cave , 1999; Ogawa and Komatsu , 2004; Busse et al . , 2008; Theeuwes , 2010; Markowitz et al . , 2011 ) . This suggests that the competition between exogenous and endogenous responses evolves in less than 100 ms . The usual behavioral metrics of mean RT and overall accuracy are thus unlikely to yield a clean characterization of this competition , because they can be traded against each other and reflect the end results of numerous operations ( perceptual , motor , cognitive ) that contribute to a much longer choice process ( indeed , below we show that such metrics are severely confounded ) . How can this problem be overcome ? The solution is to make the task urgent . The compelled antisaccade task requires subjects , humans in this case , to begin programming a saccade before knowing the direction of the correct response , and to use later arriving information about cue location to appropriately modify the ongoing motor plans . Urgency allows us to generate a special psychometric function , the ‘tachometric curve , ’ which tracks success rate as a function of the perceptually relevant time interval , the raw processing time ( rPT , measured between cue presentation and saccade onset ) . We find that , for the compelled antisaccade task , the tachometric curve takes on a unique shape: within a narrow rPT range , the curve yields a pronounced dip to below-chance performance in which the exogenous capture by the cue is so strong that the success rate approaches 0%; thereafter , however , endogenous control takes over , and the fraction of correct saccades to the ‘anti’ location increases rapidly . The experimental data were comprehensively replicated by a neurophysiologically based model of the saccadic choice process , with the combined results providing a remarkably detailed account of how reflexive and voluntary mechanisms compete over time to determine task performance .
Akin to an athlete anticipating the trajectory of a ball that must be caught or struck , the participant in the compelled antisaccade task must begin programming a movement in advance of the relevant sensory information , and must quickly interpret the later arriving visual cue to modify the developing motor plan ( s ) on the fly . In the sequence of task events ( Figure 1a ) , the key step is the early offset of the fixation spot , which means ‘respond now ! ’ This go signal is given first , before the cue , which is revealed after an unpredictable gap period . The cue appears randomly to the left or right of fixation , and the participant is instructed to make an eye movement away from it , to the diametrically opposite location — but for this response to be correct , the saccade must be initiated within 450 ms of the go signal . Thus , due to urgency , that is , time pressure , the participant must begin planning ( and may even execute ) a motor response before knowing what the correct choice is . This design , in which motor and perceptual processes are meant to run concurrently , stands in contrast to the easy , non-urgent version of the task ( Figure 1b ) , in which delivery of the cue before the go signal allows more time for the perceptual process to be completed before saccade onset . As in other perceptually based urgent tasks ( Becker and Jürgens , 1979; Stanford et al . , 2010; Salinas and Stanford , 2013; Scerra et al . , 2019 ) , the cue viewing time or rPT ( computed as RT - gap , or RT + delay; Figure 1 ) is the crucial variable here , because it specifies how much time is available for detecting and analyzing the cue in each trial . With little or no time to see the cue , the success rate cannot rise above chance , but as the viewing time increases , performance is expected to improve . Using multiple gap values ( 0–350 ms ) ensures full coverage of the relevant rPT range . When the probability of making a correct choice is plotted as a function of rPT — a behavioral metric that we refer to as the tachometric curve — the result is a millisecond-by-millisecond readout of the evolving perceptual decision ( Becker and Jürgens , 1979; Stanford et al . , 2010; Shankar et al . , 2011; Salinas and Stanford , 2013; Seideman et al . , 2018 ) . For the compelled antisaccade task , the tachometric curve exhibits a unique , non-monotonic shape that reflects the interaction between early involuntary and later voluntary processes ( Figure 2 ) . For rPTs shorter than 90 ms , participants perform at chance , as expected . Shortly thereafter , the initial influence of the cue manifests as a pronounced drop in performance , as participants erroneously direct a large proportion of their saccades toward the cue . This dip , which we refer to as the ‘vortex , ’ is short-lived ( visible for rPTs of 100–140 ms approximately; Figure 2 , gray shades ) , but it is so abrupt and occurs so reliably over a consistent range of rPTs , that it reaches nearly 0% correct even in the data pooled from all six participants ( Figure 2 , main panel ) . In trials in which the rPT falls inside this narrow interval , it is almost impossible to avoid looking at the cue . As rPT increases beyond 140 ms , the success rate rises and gradually approaches an asymptote , as participants direct a progressively larger proportion of their saccades to the correct , anti location . This rise in performance is remarkable in that it is extremely fast: for the pooled data ( Figure 2 , main panel ) the tachometric curve goes from 0 . 25 to 0 . 75 in 18 ms , and from 0 . 10 to 0 . 90 in only 37 ms . For some of the participants , the process is faster ( Figure 2 , P1 , P2 , P4 ) . The asymptotic fraction of correct responses is close to 1 ( the lowest across participants was 0 . 978 ) , which indicates that the participants understood the instructions and could perform the task almost perfectly — given enough time . Consistent with this , in easy , non-urgent antisaccade trials ( Figure 1b ) the fraction correct was also close to 1 ( median = 0 . 992 ) . However , because the processing times it generates are so long , the easy version of the task only provides a glimpse of the capture phenomenon , if anything ( Figure 2—figure supplement 1 ) . The characteristic shape of the tachometric curve likely results from the interplay between a reflexive and a voluntary mechanism , both of which depend on the cue . If the vortex indeed reflects the strength of low-level sensory representations that are driven by the cue’s salience , then , consistent with previous demonstrations of attentional/oculomotor capture ( Theeuwes , 1991; Theeuwes , 1992; Theeuwes , 1994 ) , it should become weaker when the salience of the cue is reduced . To investigate this , our participants performed the compelled antisaccade task with cues of three luminance levels , high ( data shown in Figure 2 ) , medium , and low ( Materials and methods ) . The three cues were the same for all participants and were randomly interleaved during the experiment . Because the faintest cue was chosen to be slightly above the detection threshold , we expected it to yield a much shallower vortex . The expectation for the later rise in the tachometric curve was less clear . If the rise is a direct reflexion of the cognitive process that remaps the spatial location of the cue and programs an antisaccade , then its steepness must depend , at least in part , on the speed and the variability of this process . So , if weaker sensory signals are generally processed more slowly or with higher variance , then the tachometric curve should rise more gradually as luminance decreases . The experimental results showed that both the timing and depth of the vortex depend strongly on cue luminance . For the data pooled from all participants ( Figure 3a ) , as luminance decreases from high ( bright green points ) to low ( dark green points ) , the vortex shifts to the right by about 50 ms ( the minimum point shifts from rPT = 111 ± 1 . 3 ms [SE from bootstrap] to rPT = 162 ± 6 . 0 ms ) , suggesting that the time needed to detect the cue increases accordingly . The vortex also becomes much less deep ( the minimum fraction correct goes from 0 . 03 ± 0 . 006 to 0 . 32 ± 0 . 026 ) . These findings are consistent with the expected weakening of ( involuntary ) attentional capture . We also found that , as luminance decreases , the rise of the tachometric curve becomes significantly less steep ( p < 0 . 0001 for all differences in maximum slope between luminance conditions , from bootstrap; see Figure 4—figure supplement 1 ) , consistent with the notion that the voluntary remapping of the cue location proceeds more slowly or less reliably as the cue becomes more dim . Thus , in addition to strongly determining the initial , bottom-up response to the cue , luminance probably also impacts the top-down process at work in the task . Qualitatively similar dependencies on cue luminance were observed in each participant’s data set ( Figure 3b ) , but reliable differences across individuals became evident when the effects were evaluated quantitatively . For any given tachometric curve , quantification was achieved by fitting the empirical data ( Figure 3 , colored data points ) with a continuous analytical function ( Figure 3 , black traces; Equation 2 ) and measuring several features from the fitted curve ( Materials and methods ) . We present results for three such features that were particularly reliable given the size of our samples ( for additional features , see Figure 4—figure supplement 1 ) . The first one is the average value of the tachometric curve for rPTs between 0 and 250 ms , which we refer to as the mean perceptual accuracy ( Figure 4a ) . The second feature is the rPT at which the tachometric curve reaches its minimum , which we designate as the vortex time ( Figure 4b ) . And the third feature is the rPT at which the rising part of the tachometric curve is halfway between its minimum and maximum values , which we call the endogenous response centerpoint , or just the centerpoint of the curve , for brevity ( Figures 2 and 3b , right border of gray shades; Figure 4c ) . These quantities are partially related; the centerpoint , which measures how soon the participant can escape the vortex , is independent of the vortex time ( partial Spearman correlation ρ = 0 . 34 , p = 0 . 2; Materials and methods ) , but is strongly anti-correlated with perceptual accuracy ( ρ = −0 . 85 , p = 10-5 ) . Notably , the separation between the ‘best’ and the ‘worst’ participant within a given luminance condition is statistically large , particularly for the mean perceptual accuracy and the centerpoint of the curve ( Figure 4a , c; note little overlap between 95% confidence intervals for bars of same color ) . The observed effects of cue luminance are highly consistent across participants ( Figure 3b ) , but the quantitative details reveal idiosyncratic variations that distinguish one individual from another ( Figure 4a , c; see below ) . This is significant because cognitive tasks generally produce robust differences either between experimental conditions/treatments or between individuals , but not both ( Borsboom et al . , 2009; Hedge et al . , 2018 ) . Antisaccade performance is often quantified using the mean accuracy . This assumes that the overall success rate in the task directly reflects the degree to which voluntary control can override the involuntary urge to look at a salient stimulus . But is this assumption correct ? Answering this critical question is generally difficult because doing so requires access to an independent assessment of perceptual performance — but that is precisely what the tachometric curve affords . To investigate the relationship between traditional antisaccade performance measures and perceptual capacity , we examined their natural variations accross individual participants . First , we computed the correlation between two variables ( Materials and methods ) , the average perceptual accuracy ( mean value of the tachometric curve ) , and the average observed accuracy ( mean fraction of correct choices ) . We found that , even though both quantites tend to increase with higher luminance , suggesting a positive correlation , they are , in fact , uncorrelated ( Figure 5a ) . The rank of a given participant based on one measure is not predictive of his or her rank based on the other . This may seem surprising . Logic dictates that better perception should translate into better performance — but critically , this is contingent on everything else being equal . The paradox arises because the mean RT also varies across participants , and the two accuracy measures relate to it in opposite ways . The average observed accuracy demonstrates a strong speed-accuracy tradeoff , that is , slower participants are correct more often ( Figure 5b ) . In contrast , the mean perceptual accuracy demonstrates a weaker opposite trend , that is , those participants that exhibit high perceptual ability also tend to respond more quickly ( Figure 5c ) . The results are nearly identical when perceptual performance is quantified with the endogenous response centerpoint ( Figure 5—figure supplement 1 ) . This stark divergence did not arise because our urgent task produced more or different errors , but rather because , unlike the mean accuracy , the tachometric curve is a true metric of perceptual processing . This curve is highly sensitive to the properties of the visual stimuli that must be judged ( such as luminance , in this case ) , and at the same time , when such properties are fixed , it is largely impervious to manipulations that substantially alter the RT ( Stanford et al . , 2010; Shankar et al . , 2011; Salinas et al . , 2014; Scerra et al . , 2019; for evidence that is specific to the compelled antisaccade task , see Figure 5—figure supplement 2 ) . In contrast , the mean observed accuracy depends not only on the shape of the tachometric curve , but also on two factors , unrelated to perception , that determine which parts of the curve are sampled during an experiment , the gap values used and the subject’s urgency . For instance , when only a zero gap is used , most rPTs are beyond the vortex range ( Figure 2—figure supplement 1g , h ) , where differences between participants reflect mainly their asymptotic performance levels , that is , their lapse rates . But even when a wide range of gaps is used ( as in Figure 5 ) , participants that tend to respond quickly ( short RTs ) will generally produce short rPTs and sample more densely the left side of their curves , whereas participants that tend to respond slowly ( long RTs ) will generally produce long rPTs and sample more densely the right side of their curves . This is the source of the speed-accuracy tradeoff found here ( Figure 5b ) . As a result , the mean accuracy provides scarcely any information about the ability of an individual to prevent a captured saccade relative to that of others . We developed a physiologically feasible model ( Materials and methods ) to explore two mechanistic hypotheses about the neural origin of the vortex . This model is a variant of one that replicates both behavioral performance and choice-related neuronal activity ( in the FEF ) in an urgent , two-alternative , color discrimination task ( Stanford et al . , 2010; Shankar et al . , 2011; Costello et al . , 2013; Seideman et al . , 2018 ) . As in that case , the current model considers two variables , rL and rR , that represent oculomotor responses favoring saccades toward left and right locations ( Figure 6 , black and red traces ) . These motor plans compete with each other such that the first one to reach a fixed threshold level ( Figure 6 , dashed lines ) determines the choice: a left saccade if rL reaches threshold first , or a right saccade if rR reaches threshold first . In each trial , after the go signal , rL and rR start increasing with randomly drawn build-up rates . The build-up process is likely to end in a random choice ( i . e . a guess; Figure 6c ) when one of the initial rates is high and/or the gap is long , but otherwise , time permitting , the cue signal modifies the ongoing motor plans ( Figure 6a , b ) . Specifically , once the target has been identified , the plan toward it ( correct ) is accelerated and the other one , toward the opposite , incorrect location , is decelerated ( Figure 6a , note acceleration of black trace and deceleration of red trace after shaded interval ) . This corresponds to the cue content , interpreted according to task rules , informing the correct choice . To adapt this ‘accelerated race-to-threshold’ model to the compelled antisaccade task , we introduced one crucial , task-specific assumption: that the competition is biased in favor of the cue location during a period of time that we refer to as the exogenous response interval , or ERI ( Figure 6 , pink shades ) . During the ERI , the cue has already been detected by the circuit but not yet interpreted as ‘opposite to the target’ ( so it cannot yet drive the endogenous acceleration and deceleration described above ) . We consider two possible mechanisms by which , during the ERI , the detection of the cue may lead to exogenous attentional/oculomotor capture: ( 1 ) it could halt or suppress the ongoing plan toward the anti location ( Figure 6a , b , black traces during pink interval ) or ( 2 ) it could transiently accelerate the ongoing plan toward the cue location ( Figure 6a , b , red traces during pink interval ) . These alternatives are not mutually exclusive . The former is consistent with evidence that salient , abrupt-onset stimuli reflexively interrupt ongoing saccade plans ( Dorris et al . , 2007; Bompas and Sumner , 2011; Hafed and Ignashchenkova , 2013; Buonocore et al . , 2017; Salinas and Stanford , 2018 ) , whereas the latter is consistent with the short-latency , stimulus-driven activation of visually responsive neurons in oculomotor areas ( Gottlieb and Goldberg , 1999; Bisley et al . , 2004; Thompson et al . , 2005; Ipata et al . , 2006; Marino et al . , 2015; Joiner et al . , 2017; White et al . , 2017; Chen et al . , 2018 ) . We found that , to reproduce the psychophysical data accurately , both mechanisms were necessary . To see why , first note that the tachometric curve , which refers to the proportion of correct choices in each rPT bin , can be expressed as a ratio , ( 1 ) C ( rPT ) =fC ( rPT ) fC ( rPT ) +fI ( rPT ) where fC ( rPT ) and fI ( rPT ) describe the frequencies of correct and incorrect choices at each rPT , that is , they are the rPT distributions for correct and incorrect trials ( normalized by the same factor; Materials and methods ) . Each of these distributions demonstrates a distinct feature: fC has a dip ( green shades in Figure 7 , third row ) , whereas fI has a peak ( red shades in Figure 7 , bottom row ) . Both features contribute to the vortex , as dictated by the above expression . The critical mechanistic observation is that acceleration of the motor plan toward the cue during the ERI accounts for the peak in fI ( Figure 7a ) , whereas interruption of the competing motor plan away from the cue produces the dip in fC ( Figure 7b ) . Thus , when the model was implemented with either one of the mechanisms alone , it failed to replicate the experimental feature associated with the other ( Figure 7 , bottom two rows , compare black traces in a vs . b ) . However , with the two mechanisms acting simultaneously , in coordination , the model reproduced the full data set in quantitative detail ( Figures 7c and 8 ) . First , for the data pooled across participants , the model fitted the tachometric curve ( Figure 8a ) and the rPT distributions for correct and incorrect responses ( Figure 8b ) . Second , for individual gap conditions , the model matched the variations in mean success rate and mean RT ( Figure 8—figure supplement 1 ) , but more importantly , it reproduced the shapes of the RT distributions for correct and incorrect choices , which were typically bimodal ( Figure 8c ) . Third , the model accurately captured all the dependencies on luminance ( Figure 8 , compare results across columns ) . Importantly , in doing so , the values of the parameters that correspond to pure motor performance ( the distribution of initial build-up rates for rL and rR , and the distribution of afferent delays associated with the go signal ) were the same across luminance conditions ( Table 1 ) , in correspondence with the fact that all trials proceeded identically up to cue presentation , and that trials with different gap and cue luminance were interleaved during the experiment . And fourth , the model also fitted the ( noisier ) data from individual participants , even though they showed large , idiosyncratic variations in motor performance , as well as in the dips and peaks of their rPT distributions ( Figure 8—figure supplements 2 and 3 ) . For all comparisons across participants , the empirical ( Figures 4 and 5 ) and simulated results ( Figure 4—figure supplement 2; Figure 5—figure supplement 3 ) were nearly indistinguishable . Mechanistically , the best-fitting parameter values of the model ( Table 1 ) provide further insight about the crucial element that gives rise to the vortex — the exogenous bias during the ERI ( Figure 6 , pink shades ) . Consider the following values based on the pooled data . According to the model , the onset of the ERI , which corresponds to the time at which the cue is detected , is highly sensitive to luminance . For the high , medium , and low conditions , the oculomotor circuitry detects the cue 76 ± 5 ms ( mean ± SD for simulated trials ) , 104 ± 13 ms , and 126 ± 19 ms after its presentation . This variation from high to low luminance ( 50 ms ) corresponds closely with the rightward shift of the vortex observed experimentally ( 51 ms; Figure 3a ) and is consistent with the ubiquitous dependence of visual response latency on luminance and contrast ( Purushothaman et al . , 1998; Bisley et al . , 2004; van Rossum et al . , 2008; White et al . , 2008; Oram , 2010; Marino et al . , 2015 ) . Remarkably , the ERI lasts only 24 ms ( on average ) in all three conditions , and the exogenous acceleration of the plan toward the cue occurs only during the last 14 ms ( high luminance ) , or only during the last 10 ms ( medium and low luminance ) ; before that , the plan toward the cue halts just like its counterpart toward the anti location ( Figure 6a , b , note that red trace is initially flat during shaded interval ) . The model suggests that the exogenous acceleration favoring the cue location is very brief but very powerful , which explains why the left edge of the vortex can be so steep . Finally , the parameter values ( Supplementary file 1 ) also point to specific neural mechanisms that likely underlie the individual differences in perceptual capacity . In general , identifying those mechanisms is complicated because their variations across ( random ) participants and across ( controllable ) experimental conditions are not necessarily correlated ( Borsboom et al . , 2009; Hedge et al . , 2018 ) . The cue latency discussed above is a perfect example: it demonstrates ( via parameter μCUEaff ) a strong , consistent dependence on luminance in each participant’s data set , and yet , for a given luminance level , it is not predictive of individual perceptual accuracy ( Figure 5—figure supplement 4a ) . By contrast , we hypothesize that the magnitudes of the exogenous and endogenous acceleration ( via parameters aEX and aEND ) are major sources of individual variation , because although they have weaker dependencies on luminance , they are reliable predictors of perceptual accuracy ( Figure 5—figure supplement 4b–d ) .
By design , the antisaccade task creates a conflict between exogenous and endogenous mechanisms , the former driven by the saliency of the cue and the latter by task instructions followed willfully . Other tasks ( e . g . Kim and Cave , 1999 ) , most notably the singleton-distracter task employed by Theeuwes and colleagues ( Theeuwes , 1991; Theeuwes , 1992; Theeuwes , 1994; Theeuwes et al . , 1998; Theeuwes et al . , 1999; Nissens et al . , 2017 ) , have also revealed such conflict in the form of attentional or oculomotor capture , but its manifestation in those cases consists primarily of small variations ( ∼ tens of ms ) in RT around mean values that are much longer ( >> 250 ms ) than typical intersaccadic intervals , and the results can only provide a crude estimate of the underlying temporal dynamics ( Mulckhuyse et al . , 2008; see also Markowitz et al . , 2011 ) . Those tasks also require more complex visual displays with multiple items , and a secondary discrimination to serve as a probe of the effect . In principle , a minimalistic task typifying such an essential phenomenon would be extremely useful; it could serve to determine the neural correlates of volitional versus reflexive action , or pinpoint the consequences of disease on specific cognitive abilities , for example . Numerous studies based on the traditional antisaccade task have , in fact , reported large differences in overall performance between distinct populations of participants ( Guitton et al . , 1985; Klein and Foerster , 2001; Munoz et al . , 2003; Condy et al . , 2007; Hakvoort Schwerdtfeger et al . , 2012; Antoniades et al . , 2015 ) — but do those differences relate to the conflict at the heart of the task ? This is unclear . While a distinction between fast and slow errors has been drawn based on theoretical considerations ( Lo and Wang , 2016 ) , the tachometric curve reveals three types of error: fast , incorrect guesses ( rPT ≲ 100 ms ) , saccades captured by the cue ( vortex ) , and lapses ( rPT ≳ 200 ms ) , which probably depend on distinct cognitive processes or states that vary over long time scales ( Harris and Thiele , 2011; Lo and Wang , 2016; Nir et al . , 2017 ) . When antisaccade performance is evaluated via the mean accuracy , the most common metric , the three error types are combined in proportions that are unpredictable , because they depend on each participant’s urgency ( how quickly they tend to respond; Figure 5b ) and on the gap values used in the experiment . In particular , when the cue is presented before or simultaneously with the go signal , most rPTs are sampled in the asymptotic performance range , where most errors are lapses ( Figure 2—figure supplement 1 ) . The captured saccades are the essential manifestation of the conflict , but they cannot be reliably quantified unless performance is tracked with high temporal resolution . Insight about the visuo-motor interactions that determine the shape of the tachometric curve can be gleaned by realizing that , for each trial , the rPT conveys information not only about how much time was available for perceptual deliberation , but also about the state of the motor build-up at the time when the cue was detected by the oculomotor circuitry . Consider what must happen for a saccade to be triggered at rPT = 111 ms , when the capture is nearly certain: at the moment that the cue is detected ( left edge of pink interval in Figure 6b ) , the motor activity toward the cue location must be just below threshold , so that the exogenous drive can reliably propel it past threshold . This can happen in many ways , such as when the motor plans build-up slowly and the gap is long , or when the build-up is fast and the gap is short — but whatever the build-up history , an rPT of 111 ms corresponds to the requisite level of subthreshold activity . The same is true for other parts of the tachometric curve . Short rPTs ( guesses ) correspond to insufficient perceptual deliberation and to activity that exceeded threshold before the cue was detected ( Figure 6c ) , whereas long rPTs ( informed choices ) correspond to successful deliberation guiding activity that was still far from threshold at the time of cue detection ( Figure 6a ) . Captured saccades are reliably found within a narrow range of rPTs because their expression requires certain combinations of sensory ( cue exposure ) and motor ( degree of build-up ) conditions to be met . This explanation simply recounts what the model does , so it is worth discussing how our model is different from previous ones , and why we think it is largely credible . Previous models of antisaccade performance ( e . g . Wiecki and Frank , 2013; Lo and Wang , 2016; Aponte et al . , 2017 ) applied to non-urgent conditions , so they provide limited insight about the shape of the tachometric curve . Furthermore , such models were concerned with the neural basis of inhibitory control more generally , so they involve , explicitly or implicitly , multiple brain areas; for example , one for producing motor responses and another for inhibiting the reflexive movement toward the cue . In contrast , our race-to-threshold model is agnostic as to where the relevant perceptual or control signals come from , or how they are computed; it simply deals with their dynamical impact ( i . e . acceleration , deceleration , halting ) on the developing motor activity that must ultimately communicate the urgent choice . As such , the fast variations of the model firing rates are meant to be directly comparable to those of saccade-related oculomotor neurons ( in FEF , and perhaps SC ) . Indeed , previous single-neuron studies in monkeys support key elements of our modeling framework . First , the initial level of motor activity contributes as proposed: during urgent saccadic choices , the build-up of activity in FEF starts after the go signal , regardless of when the cue information arrives ( Stanford et al . , 2010; Costello et al . , 2013 ) , and during antisaccade performance , the ongoing activity in FEF and SC is higher before erroneous saccades toward the cue than before correct antisaccades ( Everling et al . , 1998; Everling and Munoz , 2000 ) . Second , in an urgent color discrimination task , perceptual information does produce acceleration and deceleration of motor activity in an rPT-dependent way ( Stanford et al . , 2010; Costello et al . , 2013 ) . Third , the timing of the vortex and its dependence on luminance parallel those of visual bursts in the oculomotor system ( Gottlieb and Goldberg , 1999; Bisley et al . , 2004; Thompson et al . , 2005; Ipata et al . , 2006; Joiner et al . , 2017; White et al . , 2017; Chen et al . , 2018 ) . In particular , the captured saccades in our experiment resemble so-called ‘express saccades’ in many ways: both result in movements triggered after 100 ms or less of stimulus viewing time; both are more likely with higher luminance; and both are facilitated by the early removal of the fixation requirement , such that a visually evoked response is superimposed on advancing motor activity ( Paré and Munoz , 1996; Dorris et al . , 1997; Marino et al . , 2015 ) . And fourth , the sudden presentation of a salient distracter stimulus has a robust impact on a developing saccade plan , with the effect depending strongly on their spatial congruence: when the saccade target is diametrically opposite to the stimulus , there is ample evidence ( reviewed by Salinas and Stanford , 2018 ) indicating that the developing plan is transiently halted or suppressed , whereas when the saccade target is near the abrupt-onset stimulus , the developing plan is boosted ( Dorris et al . , 2007; Edelman and Xu , 2009; White et al . , 2013; Marino et al . , 2015 ) . These observations are consistent with the exogenous and endogenous mechanisms implemented by the model . Our results are pertinent to a mechanistic question that is central to the ‘premotor theory’ of attention: to what degree is the neural substrate of the deployment of spatial attention the same as that of saccade planning ? There is strong evidence that the rise in oculomotor activity associated with planning a saccade inevitably implies that attentional resources are at least partially allocated to the intended saccade endpoint ( Kowler et al . , 1995; Deubel and Schneider , 1996; Moore and Fallah , 2001; Godijn and Theeuwes , 2003; Moore and Armstrong , 2003; Cavanaugh and Wurtz , 2004; Steinmetz and Moore , 2014; Klapetek et al . , 2016 ) . The converse relationship — that is , whether the covert deployment of spatial attention must be accompanied by saccade planning — has been more contentious ( Juan et al . , 2004; Thompson et al . , 2005 ) . It appears , however , that the hypothesized motor plan associated with attentional allocation is just very difficult to observe when fixation must be actively maintained ( Belopolsky and Theeuwes , 2012 ) . During fixation , such a plan may manifest only as a subtle increase in baseline activity , rather than via the more typical steady rise in firing rate ( Hauser et al . , 2018 ) , but it can be uncovered through experimental manipulations ( Theeuwes et al . , 1998; Theeuwes et al . , 1999; Katnani and Gandhi , 2013; Nissens et al . , 2017 ) , and is evident in microsaccades ( Chen et al . , 2015; Lowet et al . , 2018 ) . Our results are consistent with the idea that attentional and oculomotor capture are different behavioral manifestations of the same underlying neuronal dynamics ( in , say , the FEF or SC ) . When a salient cue is detected , a bias favoring a motor plan toward its location is always generated ( with the bias consisting of acceleration of the plan toward the cue and halting of any plans away from it ) . However , the impact of the exogenous biasing signal depends on the current state of the oculomotor circuitry . When the motor activity congruent with the cue location is far from threshold and is competing with other developing motor plans , the bias corresponds to the covert ( and transient ) deployment of attention to the cue . In contrast , when that activity has already developed to a substantial degree , the bias can quickly propel it past threshold . Then , the exogenous attraction of the cue becomes observable as an overt , captured saccade . The detailed mechanistic framework presented here should provide ample opportunity to test these ideas in future experiments .
Experimental subjects were six healthy human volunteers , two male and four female , ages 21–30 . They were recruited from the Wake Forest School of Medicine and Wake Forest University communities . All had normal or corrected-to-normal vision . All participants provided informed written consent before the experiment . All experimental procedures were conducted with the approval of the Institutional Review Board ( IRB ) of Wake Forest School of Medicine . The experiments took place in a semi-dark room . The participants sat on an adjustable chair , with their chin and forehead supported , facing a VIEWPixx LED monitor ( VPixx Technologies Inc , Saint Bruno , Quebec , Canada; 1920 × 1200 screen resolution , 120 Hz refresh rate , 12 bit color ) at a distance of 57 cm . Viewing was binocular . Eye position was recorded using an EyeLink 1000 infrared camera and tracking system ( SR Research , Ottawa , Canada ) with a sampling rate of 1000 Hz . Stimulus presentation and data collection were controlled using the system’s integrated software package ( Experiment Builder ) . The sequence of events in the antisaccade task is described in Figure 1 . The inter-trial interval was 1 s . The gap values used were −200 , −100 , 0 , 75 , 100 , 125 , 150 , 175 , 200 , 250 , and 350 ms , where negative numbers correspond to delays in the easy antisaccade task ( Figure 1b ) . Thus , compelled and easy , non-urgent trials were interleaved . In each trial , the gap value , cue location ( −10° or 10° ) , and luminance level ( see below ) were randomly sampled . Auditory feedback was provided at the end of each trial: a beep to indicate that the saccadic response was made within the allowed RT window ( 450 ms ) , or no sound if the limit was exceeded . This was independent of the choice . Feedback about the choice itself was unnecessary , as participants easily understood the rules of the task . The task was run in blocks of 150 trials . After 50–150 trials of practice , each participant completed 30 blocks over six experimental sessions ( days ) . Within each session , 2–3 min of rest were allowed between blocks . The cue was a green circle ( 0 . 5° diameter ) appearing on a black background . Each participant performed the task with cues of three luminance levels , high ( 17 . 6 cd m-2 ) , medium ( 0 . 35 cd m-2 ) , and low ( 0 . 22 cd m-2 ) . Luminance was measured with a spectrophotometer ( i1 Pro 2 from X-Rite , Inc , Grand Rapids , MI ) . The cues were generated in Adobe Illustrator using the 8-bit RGB vectors [15 168 40] , [3 28 7] , and [1 12 3] . The lowest luminance was chosen to be close to the detection threshold based on detection curves generated previously for two participants . All data analyses were carried out using customized scripts written in Matlab ( The MathWorks , Natick , MA ) . Except where explicitly noted , results are based on the analysis of urgent trials ( gap ≥ 0 ) only; that is , easy trials ( delay trials with gap < 0 ) were excluded . In each trial , the rPT was computed by subtracting the gap value from the RT value recorded in that trial . We refer to this processing time as ‘raw’ because it includes any afferent or efferent delays in the circuitry ( Stanford et al . , 2010 ) . To compute the tachometric curve and rPT distributions , trials were grouped into rPT bins of 15 ms , with bins shifting every 1 ms . Normalized rPT distributions , fC ( rPT ) and fI ( rPT ) , were obtained by counting the numbers of correct and incorrect trials , respectively , in each rPT bin , and dividing both functions by the same factor . The tachometric curve , which gives the proportion of correct trials in each bin , was then computed using Equation 1 . For display purposes , the normalization factor used was the maximum value of fC or fI , whichever was largest , but the factor has no effect on the tachometric curve . In order to quantify perceptual performance , each tachometric curve was fitted with a continuous analytical function , v ( x ) , which was defined as ( 2 ) v ( x ) =max ( sL ( x ) , sR ( x ) , 0 ) where the maximum function max ( a , b , c ) returns a , b , or c , whichever is largest , and sL and sR are two sigmoidal curves . These are given by ( 3 ) sL ( x ) =B+AL−B1+exp ( x−CLDL ) ( 4 ) sR ( x ) =B+AR−B1+exp ( −x−CRDR ) where sL tracks the left ( decreasing ) side of the tachometric curve and sR tracks the right ( increasing ) side . The asymptotic value on the left side was fixed at AL= 0 . 5 , to enforce the constraint that , for very short processing times , performance must be at chance . For any given empirical tachometric curve , the six remaining parameters defining v , the coefficients B , AR , CL , CR , DL , and DR , were adjusted to minimize the mean absolute error between the experimental and fitting functions . The minimization was done using the Matlab function fminsearch . Once the best-fitting v ( rPT ) function for a given tachometric curve was found , we numerically calculated eight quantities , or features , from it: the asymptotic value ( equal to AR ) , the minimum value ( vortex depth ) , the rPT at which the minimum was found ( vortex time ) , the most negative slope , the most positive slope , the rPT for which v is exactly between 0 . 5 ( chance ) and the minimum ( left edge of the curve ) , the rPT for which v is exactly between the minimum and the asymptote ( the curve’s centerpoint ) , and the average of the curve for rPTs between 0 and 250 ms ( mean perceptual accuracy ) . In Figures 2 and 3 , the gray shades demarcating the vortex correspond to the interval between the left edge and the centerpoint of each tachometric curve . Confidence intervals for all of these quantities were obtained by bootstrapping ( Davison and Hinkley , 2006; Hesterberg , 2014 ) ; that is , by resampling the data with replacement and recalculating all the quantities many times to generate distributions for them . This was done in five steps: ( 1 ) resample the original trials with replacement , keeping the original number of contributing trials , ( 2 ) recompute the empirical tachometric curve from the resampled trials , ( 3 ) fit the new tachometric curve with a continuous v function , ( 4 ) recompute the eight characteristic features from the new v function , and ( 5 ) repeat steps 1–4 10 , 000 times to generate distributions for all the features . Reported confidence intervals correspond to the 2 . 5 and 97 . 5 percentiles obtained from the bootstrapped distributions . To quantify the association between average quantities computed for individual participants , such as the mean RT or mean observed accuracy ( Figure 5 ) , we considered three data points per participant , one for each luminance condition . The strength of association and its significance were calculated with three methods . First we computed the partial Pearson correlation coefficient , which is the standard linear correlation between two variables but controlling for the effect of a third one , luminance in this case . This was implemented via the Matlab function partialcorr . We also computed the partial Spearman correlation coefficient , which involves a similar calculation but based on the ranks of the data points . This was using partialcorr too . Finally , using the Matlab function fitlm , we fitted the data to a linear regression model that also included luminance as a variable . The three methods typically produced similar results . We report those obtained with the partial Spearman correlation , which is denoted as ρ , because they were generally the most conservative . The model for the compelled antisaccade task is a straightforward extension of one developed previously for a two-alternative , urgent , color discrimination task ( Stanford et al . , 2010; Shankar et al . , 2011; Costello et al . , 2013; Seideman et al . , 2018 ) . In that earlier model , both motor plans halt briefly when the relevant cue is detected ( i . e . during the ERI ) ; the dynamics are otherwise identical . As explained in the main text , the idea is that two motor plans ( in the FEF ) , represented by firing rate variables rL and rR , compete with each other to trigger an eye movement with a saccade vector pointing either to the left or to the right . Because the acceleration , deceleration , and halting of these plans depends on the cue location , it is useful to relabel the two variables as rC and rA , where the subscripts now refer to the cue and anti locations ( keeping in mind that the C and A labels are randomly assigned to left and right directions in each trial ) . Note , however , that the following description applies identically if the A and C labels are replaced everywhere by L and R , respectively , and we assume that the cue appears on the left side . Over time , the two motor plans advance toward a fixed threshold ( equal to 1000 arbitrary units , or AU ) . If rC exceeds threshold first , the saccade is incorrect , toward the cue , whereas if rA exceeds threshold first , the saccade is correct , away from the cue . The saccade is considered to be triggered a short efferent delay ( equal to 20 ms ) after threshold crossing . The fixed threshold is a reasonable approximation of the triggering mechanism for saccades ( Hauser et al . , 2018 ) . The two rate variables evolve as follows ( 5 ) rC ( t+Δt ) =rC ( t ) +bCΔtrA ( t+Δt ) =rA ( t ) +bAΔtwhere bC and bA are their respective build-up rates and the time step Δt is equal to 1 ms . When the build-up rates are constant , the firing rates rC and rA increase linearly over time . Periods during which the activity accelerates or decelerates are those during which the build-up rates themselves change steadily , as described below . Any negative rC and rA values are reset to zero . Each simulated trial can be subdivided into three epochs with different model dynamics . Epoch 1: before the ERI . Each trial starts with the two activity variables , rC and rA , equal to zero . The go signal occurs at t=0 , but the two motor plans start building up later , after an afferent delay . This afferent delay is drawn from a Gaussian distribution with mean μGOaff and SD σGOaff , where values below 20 ms are excluded . The initial build-up rates , bC0 and bA0 , are drawn from a two-dimensional Gaussian distribution with mean μb , SD σb , and correlation coefficient ρb . During this epoch , after the initial afferent delay has elapsed , rC and rA evolve according to Equations 5 , with bC=bC0 and bA=bA0 . If during this period one of the motor plans exceeds the threshold , a saccade is produced and the trial ends . Otherwise , the trial continues . Epoch 2: during the ERI . The start of the ERI corresponds to the time point at which the cue is detected by the model circuit ( we stress that this is a local event , and make no claims about the participant’s perceptual experience ) . Cue detection occurs after an afferent delay relative to the time of cue presentation , which is at t= gap . This delay is drawn from a Gaussian distribution with mean μCUEaff and SD σCUEaff , where values below 20 ms are excluded . The duration of the ERI also varies normally across trials . It is drawn from a Gaussian distribution with mean μERI and SD σERI , with negative values reset to zero . The two motor plans behave differently during the ERI . For the plan toward the anti location , rA , the build-up rate is bA=gERIbA0 , where the constant gain factor gERI is either zero ( i . e . the plan halts ) or negative ( i . e . the plan is suppressed ) . This factor was set to zero for the pooled data , but negative values were allowed when fitting the data from individual participants . Whether zero or negative , the build-up rate of the anti plan is the same throughout the whole ERI . In contrast , for the motor plan toward the cue , rC , the build-up rate is bC=gERIbC0 but only during the first ΔERI ms of the ERI; thereafter this build-up rate instantly recovers its initial value ( so bC=bC0 ) and then increases steadily , such that ( 6 ) bC ( t+Δt ) =bC ( t ) +aEXΔtuntil the end of the ERI , where the term aEX is the exogenous acceleration of the cue plan . In this way , the plan toward the cue , rC , first halts for ΔERI ms and then accelerates . If rC exceeds threshold during the ERI , a saccade toward the cue is triggered . Otherwise , the trial continues . Epoch 3: after the ERI . During this last period , the plan toward the anti location first recovers its initial value ( instantly , so bA=bA0 ) and then accelerates , whereas the plan toward the cue decelerates . That is , ( 7 ) bC ( t+Δt ) =bC ( t ) +dENDΔtbA ( t+Δt ) =bA ( t ) +aENDΔtwhere the endogenous deceleration dEND is negative and the endogenous acceleration aEND is positive . The process continues until one of the plans reaches the threshold . Finally , the model also considers lapses , trials in which errors are made for reasons other than insufficient cue viewing time . Lapses occur with a probability λ , and are implemented as trials in which the endogenous acceleration and deceleration are equal to zero . In other words , a lapse corresponds to a trial in which the information about the correct target never reaches the circuit . During lapses , after the ERI ( epoch 3 ) , the motor plan toward the anti location continues building up at its initial rate , bA0 , whereas the plan toward the cue continues advancing at whatever build-up rate it achieved at the end of the ERI . In all , the model has 15 parameters that were adjusted to fit the pooled data set or the data from individual participants . Best-fitting values are listed in Table 1 and Supplementary file 1 . These were obtained by searching over a multidimensional parameter space , gradually reducing its volume , seeking to minimize the mean absolute error between the simulated and the experimental data . For each parameter vector tested , the error consisted of a sum of terms , each representing one target function to be fitted . These functions were the RT distributions for correct choices at individual gaps , the RT distributions for incorrect choices , also at individual gaps , and the tachometric curve . The search/minimization procedure was repeated multiple times with different initial conditions to ensure that solutions were found near the global optimum . | How do you decide what to do next ? Your behavior at any given moment is usually the result of a competition between internal and external factors . Internal factors include your existing plans , goals and knowledge . External factors include events happening in the world around you . When out driving , for example , you check zebra crossings because you know that pedestrians could be present . But you look at stoplights because your eyes are drawn automatically to their changing colors . Scientists can study this competition between internal and external factors using a simple laboratory task . A single spot of light appears in the dark , and your job is to look away from it . The instruction is simple and yet carrying it out requires willful effort . This is because your automatic response is to look at any stimulus that suddenly appears . Overcoming this automatic response requires similar thought processes to those that help someone resist eating that second piece of chocolate . However , the competition between automatic and voluntary visual processes is over in a fraction of a second , which makes it difficult to analyze . Salinas et al . therefore modified the “look-away” task by asking participants to respond under time pressure . This tweak makes it possible to track – with millisecond precision – voluntary and automatic influences on performance . The results revealed that the eyes are automatically drawn to the cue about 100 milliseconds after it appears . The separate voluntary process that directs the eyes away from the cue arises about 40 milliseconds later . Salinas et al . observed these voluntary and involuntary components in every healthy volunteer tested . But there were also differences between individuals in how effectively they could look away from the cue . This is important because the automatic draw of salient stimuli determines what you pay attention to , as well as what you look at . Future studies could use the modified version of the look-away task to examine whether this automatic pull of attention , and the ability to resist it , differs in individuals with disorders like ADHD . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Voluntary and involuntary contributions to perceptually guided saccadic choices resolved with millisecond precision |
Holliday junctions ( HJs ) are key DNA intermediates in homologous recombination . They link homologous DNA strands and have to be faithfully removed for proper DNA segregation and genome integrity . Here , we present the crystal structure of human HJ resolvase GEN1 complexed with DNA at 3 . 0 Å resolution . The GEN1 core is similar to other Rad2/XPG nucleases . However , unlike other members of the superfamily , GEN1 contains a chromodomain as an additional DNA interaction site . Chromodomains are known for their chromatin-targeting function in chromatin remodelers and histone ( de ) acetylases but they have not previously been found in nucleases . The GEN1 chromodomain directly contacts DNA and its truncation severely hampers GEN1’s catalytic activity . Structure-guided mutations in vitro and in vivo in yeast validated our mechanistic findings . Our study provides the missing structure in the Rad2/XPG family and insights how a well-conserved nuclease core acquires versatility in recognizing diverse substrates for DNA repair and maintenance .
Homologous recombination ( HR ) is a fundamental pathway ensuring genome integrity and genetic variability ( Heyer , 2015 ) . In mitotic cells , double-strand breaks ( DSBs ) can be repaired by HR using the sister chromatid as a template to restore the information in the complementary double strand . In meiosis , the repair of programmed DSBs by HR and the formation of crossovers are crucial to provide physical linkages between homologs and to segregate homologous chromosomes . Furthermore , HR during meiosis creates sequence diversity in the offspring through the exchange between homologs ( Petronczki et al . , 2003; Sarbajna and West , 2014 ) . HR proceeds by pathways that may lead to the formation of DNA four-way junctions or Holliday junctions ( HJs ) that physically link two homologous DNA duplexes ( Heyer , 2015; Holliday , 1964; Schwacha and Kleckner , 1995; Szostak et al . , 1983 ) . Faithful removal of HJs is critical to avoid chromosome aberrations ( Wechsler et al . , 2011 ) and cells have evolved sophisticated measures to disentangle joint molecules . One basic mechanism is resolution mediated by HJ resolvases that introduce precise symmetrical nicks into the DNA at the branch point . Nicked DNA strands are then rejoined by endogenous ligases leading to fully restored or recombined DNA strands . This mechanism is well studied for bacterial and bacteriophage resolvases such as Escherichia coli RuvC , T7 endonuclease I , T4 endonuclease VII ( Benson and West , 1994; Lilley and White , 2001 ) . These resolvases operate as dimers and show a large degree of conformational flexibility in substrate recognition and in aligning both active sites for coordinated cleavage . Interestingly , T4 endonuclease VII and RuvC reach into and widen the DNA junction point whereas T7 endonuclease I binds DNA by embracing HJs at the branch point ( Biertümpfel et al . , 2007; Górecka et al . , 2013; Hadden et al . , 2007 ) . In eukaryotes , HR is more complex and tightly regulated . In somatic cells , HJ dissolution by a combined action of a helicase and a topoisomerase ( BLM-TOPIIIα-RMI1-RMI2 complex in humans ) is generally the favored pathway , possibly to restore the original ( non-crossover ) DNA arrangement ( Cejka et al . , 2010 , 2012; Ira et al . , 2003; Putnam et al . , 2009; Wu and Hickson , 2003 ) . In contrast , HJ resolution generates crossover and non-crossover arrangements depending on cleavage direction . Several endonucleases such as GEN1 , MUS81-EME1 , and SLX1-SLX4 have been implicated as HJ resolvases in eukaryotes ( Andersen et al . , 2011; Castor et al . , 2013; Fekairi et al . , 2009; Garner et al . , 2013; Ip et al . , 2008; Muñoz et al . , 2009; Svendsen and Harper , 2010; Svendsen et al . , 2009; Wyatt et al . , 2013 ) . Interestingly , these resolvases are not structurally related and have different domain architectures , giving rise to variable DNA recognition and regulation mechanisms . The interplay between resolution and dissolution mechanisms is not fully understood yet , however , cell cycle regulation of resolvases seems to play an important role ( Blanco et al . , 2014; Chan and West , 2014; Eissler et al . , 2014; Matos et al . , 2011 ) . GEN1 belongs to the Rad2/XPG family of structure-selective nucleases that are conserved from yeast to humans ( Ip et al . , 2008; Lieber , 1997; Yang , 2011 ) . The Rad2/XPG family has four members with different substrate preferences that function in DNA maintenance ( Nishino et al . , 2006; Tsutakawa et al . , 2014 ) . They share a conserved N-terminal domain ( XPG-N ) , an internal domain ( XPG-I ) and a 5’->3’ exonuclease C-terminal domain containing a conserved helix-hairpin-helix motif . C-terminal to the nuclease core is a regulatory region that is diverse in sequence and predicted to be largely unstructured . Although the catalytic cores are well conserved in the superfamily , substrate recognition is highly diverse: XPG/Rad2/ERCC5 recognizes bubble/loop structures during nucleotide-excision repair ( NER ) , FEN1 cleaves flap substrates during Okazaki fragment processing in DNA replication , EXO1 is a 5'->3' exonuclease that is involved in HR and DNA mismatch repair ( MMR ) and GEN1 recognizes Holliday junctions ( Grasby et al . , 2012; Ip et al . , 2008; Nishino et al . , 2006; Tomlinson et al . , 2010; Tsutakawa et al . , 2014 ) . A common feature of the superfamily is their inherent ability to recognize flexible or bendable regions in the normally rather stiff DNA double helix . Interestingly , GEN1 shows versatile substrate recognition accommodating 5’ flaps , gaps , replication fork intermediates and Holliday junctions ( Ip et al . , 2008; Ishikawa et al . , 2004; Kanai et al . , 2007 ) . According to the current model , however , the primary function of GEN1 is HJ resolution ( Garner et al . , 2013; Sarbajna and West , 2014; West et al . , 2015 ) and it is suggested to be a last resort for the removal of joint molecules before cytokinesis ( Matos et al . , 2011 ) . To date , structural information is available for all members of the family but GEN1 ( Miętus et al . , 2014; Orans et al . , 2011; Tsutakawa et al . , 2011 ) . A unified feature of these structures is the presence of two DNA-binding interfaces separated by a hydrophobic wedge . This wedge is composed of two protruding helices that induce a sharp bend into flexible DNA substrates . Rad2/XPG family members also share a helix-two-turn-helix ( H2TH ) motif that binds and stabilizes the uncleaved DNA strand downstream of the catalytic center . However , the comparison of DNA recognition features within the Rad2/XPG family has been hampered because of the lack of structural information on GEN1 . To understand the molecular basis of GEN1's substrate recognition , we determined the crystal structure of human GEN1 in complex with HJ DNA . In combination with mutational and functional analysis using in vitro DNA cleavage assays and in vivo survival assays with mutant yeast strains , we highlight GEN1’s sophisticated DNA recognition mechanism . We found that GEN1 does not only have the classical DNA recognition features of Rad2/XPG nucleases , but also contains an additional DNA interaction site mediated by a chromodomain . In the absence of the chromodomain , GEN1’s catalytic activity was severely impaired . This is the first example showing the direct involvement of a chromodomain in a nuclease . Our structural analysis gives implications for a safety mechanism using an adjustable hatch for substrate discrimination and to ensure coordinated and precise cleavage of Holliday junctions .
In order to structurally characterize human GEN1 , we crystallized the catalytically inactive variant GEN12-505 D30N , denoted GEN1 for simplicity , in complex with an immobile Holliday junction having arm lengths of 10 bp ( Figure 1 ) . The structure was determined experimentally and refined up to 3 . 0 Å resolution with an Rfree of 0 . 25 ( Table 1 ) . The HJ crystallized bridging between two protein monomers in the asymmetric unit ( Figure 1—figure supplement 1 ) . The overall structure of GEN1 resembles the shape of a downwards-pointing right hand with a 'thumb' extending out from the 'palm' and the DNA is packed against the ball of the thumb ( Figure 1 ) . The palm contains the catalytic core , which is formed by intertwined XPG-N and XPG-I domains ( Figure 1A/B , green ) . They consist of a seven-stranded β-sheet in the center surrounded by nine helices harboring the conserved active site ( Figure 1B/D , orange ) . The catalytic residues form a cluster of negatively charged residues ( D30 , E75 , E134 , E136 , D155 , D157 , D208 ) that were originally identified by mutational analysis ( Ip et al . , 2008; Lee et al . , 2002; Wakasugi et al . , 1997 ) and are conserved in other Rad2/XPG family members ( Figure 1B/C and Figure 2 ) . The XPG-I domain is followed by a 5'->3' exonuclease C-terminal domain ( EXO; Figure 1B/D , blue ) . The EXO domain consists of a helix-two-turn-helix ( H2TH ) motif ( helices α10-α11 ) accompanied by several α-hairpins ( α12-α13 and α14-α15 ) . A similar arrangement is also found in other proteins , which use a H2TH motif for non-sequence specific DNA recognition ( Tomlinson et al . , 2010 ) . The EXO domain in GEN1 has a 78 amino acid insertion ( residues 245–322 ) , of which only helix α12b ( residues 308–322 ) is ordered in the structure ( Figure 1A , gray and Figure 2 ) . Helix α12b packs loosely with the H2TH helices ( α10-α11 ) and helix α12 at the 'finger' part of GEN1 . Yeast Rad2 , a homolog of human XPG , also contains helix α12b , and it shows a similar arrangement as in GEN1 ( Figure 1F ) . The EXO domain sandwiches the XPG-N/I domains with a long linker reaching from the bottom 'fingers' ( α10-α13 ) along the backside of GEN1 to the top of the XPG-N/I domains at the 'wrist' ( α14-α15 ) . A structure-based sequence alignment of the nuclease core of human GEN1 , FEN1 , EXO1 and yeast Rad2 proteins with functional annotations relates sequence conservation to features in the Rad2/XPG family ( Figure 2 ) . The comparison with members in the Rad2/XPG identified two DNA binding interfaces and a hydrophobic wedge ( ball of the thumb ) that separates the upstream and the downstream interface ( Figure 1C/D and compare Figure 1F ) . GEN1 has two prominent grooves close to the hydrophobic wedge , which we termed upper and lower gate or gateway for comparison ( Figure 1D , orange and blue ellipses , respectively ) . 10 . 7554/eLife . 12256 . 003Figure 1 . Architecture of human GEN1 . ( A ) Domain architecture of human GEN1 . The structurally unknown regulatory domain ( residues 465–908 ) is shown with dotted lines . ( B ) Overview of the catalytic core of GEN1 in complex with HJ DNA . The protein resembles the shape of a downwards-pointing right hand with helix α6 as the thumb . The protein is depicted in half transparent surface representation with secondary structure elements underneath . The DNA is shown in ladder representation with individual strands in different colors . The coloring of GEN1 follows domain boundaries: intertwining XPG-N and XPG-I in green , 5’->3’ exonuclease C-terminal domain ( EXO ) in blue , chromodomain in pink , unassigned regions in gray . Active site residues ( E134 , E136 , D155 , D157 ) are highlighted in orange . ( C ) Electrostatic surface potential of GEN1 . The coloring follows the potential from -5 ( red ) to +5 kT/e ( blue ) . The DNA-binding interfaces and the position of the hydrophobic wedge are marked in yellow . ( D ) Secondary structure elements of the catalytic core of GEN1 in cartoon representation with the same colors as before . Dotted lines represent parts that are not resolved in the crystal structure . The numbering follows a unified scheme for the Rad2/XPG family ( compare Figure 2 ) for α-helices , β-sheets and 310-helices ( η ) . ( E ) Experimental electron density map ( autoSHARP , solvent flattened , contoured at 1σ ) drawn around the HJ in the GEN1 complex . The DNA model is shown in ball-stick representation with carbon atoms of individual strands in different colors ( yellow , light blue , magenta , green ) and oxygen atoms in red , phosphor atoms in orange , nitrogen atoms in dark blue . ( F ) Structural comparison of Rad2/XPG family nucleases . Proteins are shown in a simplified surface representation with important structural elements in cartoon representation and DNA in ladder representation . The color scheme is the same as in B . Figure 1—figure supplement 1 shows the content of the asymmetric unit . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 00310 . 7554/eLife . 12256 . 004Figure 1—figure supplement 1 . Content of the asymmetric unit of the GEN1-HJ crystal . One protein monomer is shown in surface representation with secondary structure cartoons underneath , the other one only in cartoon representation with α-helices as cylinders and β-strands as arrows . The HJ DNA bridges between two protein monomers in the asymmetric unit . The active sites are labeled with a turquoise ball each . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 00410 . 7554/eLife . 12256 . 005Table 1 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 005Data SetG505-4w006nativeG505-4w006Ta peakG505-4w006SeMet peakDiffraction Data StatisticsSynchrotron BeamlineSLS PXII SLS PXII SLS PXII Wavelength0 . 99995 1 . 25473 0 . 97894 Resolution ( Å ) 75-3 . 0 75 . 4-3 . 8 43 . 6-4 . 4 Space GroupP 32 P 32 P 32 Cell dimensions a ( Å ) 86 . 94 87 . 06 87 . 11 b ( Å ) 86 . 94 87 . 06 87 . 11 c ( Å ) 200 . 72 201 . 30 199 . 69 α ( ° ) 90 90 90 β ( ° ) 90 90 90 γ ( ° ) 120 120 120 I/σI*18 . 4 ( 1 . 9 ) 27 . 49 ( 5 . 83 ) 16 . 58 ( 3 . 82 ) Completeness ( % ) *99 . 8 ( 98 . 8 ) 99 . 6 ( 97 . 3 ) 97 . 3 ( 83 . 3 ) Redundancy*6 . 3 10 . 2 5 . 1 Rsym ( % ) *6 . 2 ( 90 . 7 ) 7 . 7 ( 42 . 2 ) 6 . 9 ( 43 . 4 ) Refinement StatisticsResolution ( Å ) 75-3 . 0 Number of Reflections33933 Rwork/Rfree0 . 199/0 . 241 Number of Atoms Protein6298 DNA1589 Water/Solutes27 B-factors Protein123 . 4 DNA150 . 2 Water/Solutes92 . 6 R . M . S Deviations Bond lengths ( Å ) 0 . 010 Bond Angles ( ° ) 0 . 623 Ramachandran Plot Preferred753 ( 97 . 9 % ) Allowed16 ( 2 . 1% ) *Values for the highest resolution shell are shown in parenthesis10 . 7554/eLife . 12256 . 006Figure 2 . Alignment of the nuclease cores of Rad2/XPG-family proteins . The alignment is based on known crystal structures: human GEN1 ( PDB 5t9j , this study ) , yeast Rad2 ( PDB 4q0w ) , human FEN1 ( PDB 3q8k ) , human EXO1 ( 3qe9 ) . Secondary structure elements are depicted on top of the sequence with dark blue bars for α-helices , light blue bars for 310-helices and green arrows for β-sheets . The numbering follows a unified scheme for the superfamily . Functional elements are labeled and described in the main text . Sequences are colored by similarity ( BLOSUM62 score ) and active site residues are marked in red . Mutations analyzed in this study are marked with an orange triangle and DNA contacts found in the human GEN1–HJ structure have a dark green dot . Disordered or missing parts in the structures are labeled in small letters or with x . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 006 Notably , a small globular domain ( residues 390–464 ) was found extending the GEN1 nuclease core at the wrist ( Figure 1 , pink ) . A DALI search ( Holm and Rosenström , 2010 ) against the Protein Data Bank ( PDB ) identified this domain as a chromodomain ( chromatin organization modifier domain ) . The domain has a chalice-shaped structure with three antiparallel β-strands packed against a C-terminal α-helix and it forms a characteristic aromatic cage . The opening of the chalice abuts helix α15 from the EXO domain . Chromodomains are found in many chromatin-associated proteins that bind modified histone tails for chromatin targeting ( reviewed in Blus et al . , 2011; Eissenberg , 2012; Yap and Zhou , 2011 ) , but it has not previously been associated with nucleases . To understand the significance of the chromodomain for the function of GEN1 , we first examined if the chromodomain is conserved in GEN1 homologs using HMM-HMM ( Hidden Markov Models ) comparisons in HHPRED ( Söding et al . , 2005 ) . We found that the chromodomain in GEN1 is conserved from yeast ( Yen1 ) to humans ( Figure 3A ) . The only exception is Caenorhabditis elegans GEN1 , which has a much smaller protein size of 443 amino acids compared to yeast Yen1 ( 759 aa ) or human GEN1 ( 908 aa ) . 10 . 7554/eLife . 12256 . 007Figure 3 . Chromodomain comparison . ( A ) Sequence alignment of GEN1 chromodomains from different organisms: hsGEN1 ( Homo sapiens ) , clGEN1 ( Canis lupus ) , mmGEN1 ( Mus musculus ) , drGEN1 ( Danio rerio ) , atGEN1/2 ( Arabidopsis thaliana ) , cgGEN1 ( Crassostrea gigas ) , scYEN1 ( Saccharomyces cerevisiae ) . The presence of a chromodomain is conserved from yeast to human with Caenorhabditis elegans as an exception . Secondary structure elements of the GEN1 chromodomain are shown on top . The sequence coloring is based on a similarity matrix ( BLOSUM62 ) . The corresponding positions of the DNA-interaction site in human GEN1 is marked with a red box and residues of the aromatic cage are highlighted with a yellow box . ( B ) GEN1 has a canonical chromodomain fold of three antiparallel beta-sheets packed against an α-helix . ( C ) The arrangement of the aromatic cage in GEN1 is comparable to other chromodomains but less aromatic and slightly larger . ( D ) The superposition of different chromodomains places cognate binding peptides of hsMPP8 and mmCBX7 ( and others ) into the aromatic cage . ( E ) The aromatic cage of GEN1 is closed by helix α15 . Panels B–D show the chromodomains of hsGEN1 ( pink , PDB 5t9j ) , hsCBX3 ( gray , PDB 3kup ) hsSUV39H1 ( green , PDB 3mts ) , hsMPP8 ( yellow , PDB 3lwe ) , dmHP1a ( orange , chromo shadow PDB 3p7j ) , dmRHINO ( cyan , PDB 4quc/3r93 ) , mmCBX7 ( light blue , PDB 4x3s; compare Figure 3—source data 1 ) . ( F ) Phylogenetic tree of all known human chromodomains . GEN1 is distantly related to the CBX chromo-shadow domains and CDY chromodomains . The corresponding alignment for calculating the phylogenetic tree is shown in Figure 3—figure supplement 1 . GEN1 is colored in black , chromobox ( CBX ) proteins are colored in red , interspersed by SUV39H histone acetylases ( orange ) and chromodomain Y-linked ( CDY ) proteins ( yellow ) . Chromo-barrel domain proteins are colored in green and chromodomain-helicase DNA-binding ( CHD ) proteins are in blue . Chromodomains and chromo-shadow domains from the same protein are labeled with 1 and 2 , respectively . Stable branches with boostrap values equal or higher than 0 . 8 are marked with a black dot . The binding of the GEN1 chromodomain to a set of histone peptides was tested but no interaction was detected ( Figure 3—source data 2 and Figure 3—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 00710 . 7554/eLife . 12256 . 008Figure 3—source data 1 . Proteins found in a DALI search . Top hits found in a DALI search for protein structure comparison with the human GEN1 chromodomain ( residues 390–464 ) against the Protein Data Bank . The most similar unique chromodomains are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 00810 . 7554/eLife . 12256 . 009Figure 3—source data 2 . N-terminally fluorescein-labeled peptides used for chromodomain binding assays . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 00910 . 7554/eLife . 12256 . 010Figure 3—figure supplement 1 . Sequence alignment of all known human chromodomains . The alignment was used to calculate the phylogenetic tree in Figure 3F . Colors follow the CLUSTAL X coloring scheme . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01010 . 7554/eLife . 12256 . 011Figure 3—figure supplement 2 . Histone peptide pull-down assay . Nickel resin-immobilized GEN1 chromodomain was incubated with the mixtures of fluorescein-labeled histone peptides , washed , bound peptides eluted and separated by 20% SDS-PAGE . Mix 1 and 2 did not show any binding , and non-specific binding to the resin was found with Mix 3 . The smearing of the bands is due to the small size of the peptides ( ~1 . 5 kDa ) . I , C and E represent input , resin control and elution , respectively . Mix 1: H3K9 , H3K9me1 , H3K9me2 , and H3K9me3 . Mix 2: H3K27 , H3K27me1 , H3K27me2 , and H3K27me3 . Mix 3: H3K36me1 , H3K36me2 , H3K36me3 , and H3K36Ac . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 011 To further compare the structural arrangement of the aromatic cage in human GEN1 with other chromodomains , we analyzed the best matches from the DALI search ( Figure 3B ) . We found many hits for different chromo- and chromo-shadow domains with root mean square deviations between 1 . 9 and 2 . 8 Å ( compare Figure 3—source data 1 ) . A superposition of the aromatic cage of the five structurally most similar proteins with GEN1 ( Figure 3C ) showed that residues W418 , T438 , and E440 are well conserved , whereas two residues at the rim of the canonical binding cleft are changed from phenylalanine/tyrosine to a leucine ( L397 ) in one case and a proline ( P421 ) in another ( Figure 3C ) . Instead , Y424 occupies the space proximal to P421 , which is about 1 . 5 Å outwards of the canonical cage and widens the GEN1 cage slightly . The substitution of phenylalanine/tyrosine to leucine is also found in CBX chromo-shadow domains ( see below ) ; however , the rest of the GEN1 aromatic cage resembles rather chromodomains . Chromodomains often recognize modified lysines through their aromatic cage thus targeting proteins to chromatin ( reviewed in Blus et al . , 2011; Eissenberg , 2012; Yap and Zhou , 2011 ) . Given the conserved aromatic cage in GEN1 , we tested the binding to modified histone tail peptides ( Figure 3C/D ) . However , we did not detect any binding despite extensive efforts using various histone tail peptides in pull-down assays , microscale-thermophoresis ( MST ) or fluorescence anisotropy measurements ( compare Figure 3—source data 2 and Figure 3—figure supplement 2 ) . Our structure shows that the aromatic cage is closed by helix α15 ( Figure 3E blue/pink ) , which has a hydrophobic interface towards the aromatic cage with residues L376 , T380 , and M384 reaching into it ( compare Figure 4F ) . This potentially hampers the binding of the tested peptides in this conformation under physiological conditions . 10 . 7554/eLife . 12256 . 012Figure 4 . DNA interactions in the GEN1-DNA complex . ( A ) Schematic of the GEN1-DNA interactions at the upstream interface . The coloring is the same as in Figure 1 . The nuclease core ( green and blue ) interacts with the uncleaved strand and the chromodomain ( pink ) contacts the complementary strand . Hydrogen bonds are shown with blue dashed lines and van-der-Waals contacts are in red dotted lines . ( B ) Interactions at the hydrophobic wedge . The end of the DNA double helix docks onto the hydrophobic wedge formed by helices α2 and α3 . ( C/D ) Interactions with the uncleaved strand in two views . All key residues form sequence-independent contacts to the DNA backbone . R54 reaches into the minor groove of the DNA . The complementary DNA strand has been removed for clarity ( E/F ) Interactions of the chromodomain with the complementary strand in two views . The backbone of residues 406–410 ( β-hairpin β8-β9 ) abuts the DNA backbone . R406 has a supporting role in the interaction and R408 forms a polar interaction with Q65 , which establishes a connection between the chromodomain and the nuclease core . Helix α15 makes hydrophobic interactions with the aromatic cage and thus blocks it . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 012 To explore the functional role of the GEN1 chromodomain , we evaluated its similarity to other chromodomains by comparing all of the 46 known human chromodomains from 34 different proteins . We made pairwise comparisons with HHPRED , PSIBLAST , combined the alignments and generated a phylogenetic tree ( Figure 3F and Figure 3—figure supplement 1 ) . The analysis showed a tree branching into known subfamilies: chromobox proteins ( CBX , red ) , chromodomain Y-linked proteins ( CDY , yellow ) , chromodomain-helicase DNA-binding proteins ( blue ) and chromo-barrel domain proteins ( green ) . The GEN1 chromodomain was found to be distantly related to the CDY chromodomains and chromobox proteins , particularly to the chromo-shadow domains of CBX1 , CBX3 and CBX5 . This agrees with the result from the DALI search , in which CBX chromo-shadow domains and homologs thereof were among the closest structural matches . Together with the observed differences in residues forming the aromatic cage , it indicates that the GEN1 chromodomain forms a new subgroup with features from chromo- and chromo-shadow domains that emerged from a common ancestor within CBX/CDY proteins . The GEN1-HJ structure revealed that the upstream DNA-binding interface acts as a docking site for double-stranded DNA and that the chromodomain secures its position . The DNA is bound at the upstream interface and the hydrophobic wedge but does not extend into the active site or to the downstream interface ( Figure 1B/C/D ) . Comparison of the structure of GEN1 to related structures of FEN1 , Rad2 and EXO1 ( Miętus et al . , 2014; Orans et al . , 2011; Tsutakawa et al . , 2011 ) suggests that a DNA substrate has to extend to the downstream interface to position a DNA strand for cleavage by the active site of GEN1 ( Figure 1B/C and Figure 1F ) . In the GEN1 structure , the end of the DNA arm attaches to the hydrophobic wedge provided by parts of helices α2-α3 and their connecting loop ( Figure 4A/B ) , forming van-der-Waals contacts with the first base pair , which docks perfectly onto the protruding curb of residues 41–51 ( Figure 4B ) . The uncleaved DNA strand is further stabilized and its geometrical arrangement is fixed by the upstream DNA-binding interface . Particularly , the DNA is contacted by a β-pin ( strands β6-β7; Figure 4A/C ) from one side and by R54 and F58 ( Figure 4A/D ) from helix α3 together with Y370 and K374 ( helix α15 ) from the opposite side ( Figure 4A/C ) . The key residues in the β-pin are T171 that forms a hydrogen bridge to the phosphate of the first base ( Figure 4A , 'G1' ) and M172 that makes a van-der-Waals contact to the DNA backbone at the second base ( Figure 4A , 'A2' ) . R54 reaches into the DNA minor groove and forms a hydrogen bond with the ribose ring oxygen at the third base of the uncleaved strand and F58 packs against the same ribose moiety ( Figure 4C/D ) . Y370 and K374 in α15 form hydrogen bonds to the backbone of the third base of the uncleaved DNA strand ( Figure 4D , 'G3' ) . An additional interaction point is provided by a β-hairpin from the chromodomain ( strands β8-β9 ) , one DNA turn upstream of the hydrophobic wedge ( Figure 4A/E/F ) . This β-hairpin interacts with the complementary DNA strand by matching the protein backbone ( residues 406–411 ) to the contour of the DNA backbone in a sequence unspecific manner ( Figure 4A/E ) . The side chains of K404 and R406 project out , and they are in hydrogen bonding distance to the DNA ( Figure 4E ) . Remarkably , R408 forms a polar interaction with Q65 , which establishes a connection between the DNA contact point at the chromodomain and the nuclease core ( Figure 4E ) . The interactions at the chromodomain extend the upstream DNA-binding interface to cover a full DNA turn , reinforcing the binding . The downstream binding interface can be inferred from other Rad2/XPG structures ( Figure 1C/F ) as the nuclease core is well conserved in GEN1 , FEN1 , Rad2 and EXO1 ( root mean square deviations of 0 . 9–1 . 1 Å for 161 Cα atoms , respectively ) . The residues corresponding to the tip of the thumb ( residues 79–92 ) , which are disordered in the GEN1 structure , likely form helix α4 upon DNA binding to the downstream interface as seen in human FEN1 and EXO1 ( Orans et al . , 2011; Tsutakawa et al . , 2011 ) . The missing residues in GEN1 have 35 . 7% identity and 78 . 6% similarity ( BLOSUM62 matrix ) to the corresponding residues in FEN1 ( 90–103 ) , which form helix α4 in the FEN1-DNA complex ( compare Figure 2 ) . The same region is disordered in FEN1 when no DNA is bound ( Sakurai et al . , 2005 ) . This indicates that also GEN1 undergoes such a disorder-to-order transition to form an arch with helices α4 and α6 upon substrate binding ( Patel et al . , 2012 ) and similar to the arrangement in T5 FEN ( Ceska et al . , 1996 ) . GEN1 has versatile substrate recognition features , ranging from gaps , flaps , replication fork intermediates to HJs ( Ip et al . , 2008; Ishikawa et al . , 2004; Kanai et al . , 2007 ) . To understand the functional relevance of the GEN1 structure for DNA recognition we performed a series of mutagenesis studies with single point mutations and truncated protein variants ( Figure 5 and Figure 5—figure supplement 1/2 ) to investigate the effect on the active site ( D30N ) , upstream DNA binding ( R54E ) , downstream DNA binding ( C36E ) , arch at the downstream interface ( R89E , R93E , H109E , F110E ) , and chromodomain ( Δchromo , K404E , R406E ) . We performed nuclease assays by titrating different amounts of GEN1 to a fixed DNA concentration of 40 nM for 15 min and DNA cleavage products were analyzed by native electrophoresis ( Figure 5A and Figure 5—figure supplement 1/2 ) . We used an immobile HJ and a 5’ flap substrate side-by-side to facilitate the comparison of the effects on separate GEN1 functions . Notably , stoichiometric amounts of GEN1 were required to cleave HJ substrates whereas 5’ flaps were readily processed with catalytic amounts ( Figure 5A ) . 10 . 7554/eLife . 12256 . 013Figure 5 . Functional analysis of GEN1 . ( A ) Nuclease activity of GEN1 with HJ and 5’flap DNA . 40 nM 5’ 6FAM-labeled substrates were mixed with indicated amounts of GEN1 . Reactions were carried out at 37°C for 15 min , products were separated by native PAGE and analyzed with a phosphoimager . Figure 5—source data 1 gives the sequences of DNA oligos used in biochemical assays and Figure 5—source data 3 shows activity measurements . ( B ) Quantification of nuclease assays of wild type GEN1 and variants with mutated residues located at the protein-DNA interfaces . Percentage of cleavage was plotted against the enzyme concentration . Error bars depict the standard deviation calculated from at least three independent experiments . Figure 5—figure supplement 1 shows representative gels from the PAGE analysis . ( C ) Quantification of nuclease assays of wild type GEN1 and variants with mutated residues located at the chromodomain . Error bars depict the standard deviation calculated from at least three independent experiments . Figure 5—figure supplement 2 shows representative gels from the PAGE analysis . ( D ) GEN1 mutations used in this study . Locations of human GEN1 mutations used in biochemical assays and corresponding residues in yeast MMS survival assays are highlighted in red . Active site residues E134 , E136 , D155 , D157 are marked in turquoise . ( E ) Schematic of the cruciform plasmid cleavage assay . A cruciform structure can be formed in plasmid pIRbke8mut , which harbors an inverted-repeat sequence and is stabilized by negative supercoiling . Introducing two cuts across the junction point within the lifetime of the resolvase-junction complex yields linear products whereas sequential cleavage generates nicked products and the relaxed plasmid cannot be a substrate for the next cleavage . ( F ) Cruciform plasmid cleavage assay with different GEN1 variants . Plasmid pIRbke8mut was treated with 256 nM GEN1 each and reactions were carried out at 37°C for 15 min . Supercoiled , linear and nicked plasmids were separated by native agarose gel electrophoresis and visualized with SYBR safe under UV light . ( G ) MMS survival assays with yeast yen1 variants . The survival of yen1 mutants was tested under a yen1Δ mus81Δ background with indicated amounts of MMS . The top part shows mutations at GEN1-DNA interfaces and the bottom part mutations at the chromodomain ( compare Figure 5—figure supplement 3 for all controls and expression tests ) . Figure 5—source data 2 gives a list of all yeast strains . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01310 . 7554/eLife . 12256 . 014Figure 5—source data 1 . Oligonucleotides used in biochemical assays . Four-way junctions were prepared by annealing CB209 , CB210 , CB211 , CB212 . 5’ flaps were prepared by annealing CB209 , CB212 , and CB218 . The annealing protocol is described in Material and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01410 . 7554/eLife . 12256 . 015Figure 5—source data 2 . Yeast strains used for MMS survival assays . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01510 . 7554/eLife . 12256 . 016Figure 5—source data 3 . In vitro activity measurements of different GEN12-505 variants . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01610 . 7554/eLife . 12256 . 017Figure 5—figure supplement 1 . DNA cleavage assays of different GEN1 mutations . All GEN12-505 mutations were generated by site-directed mutagenesis and purified with the same procedure . Experiments were repeated three times and a representative gel picture is shown for each protein variant in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01710 . 7554/eLife . 12256 . 018Figure 5—figure supplement 2 . DNA cleavage assays of different GEN1 fragments . ( A ) 5’ 6FAM labeled four-way junction or 5’flap DNA ( 40 nM ) were mixed with varying concentrations of GEN1 truncations ( 0 . 25 , 0 . 5 , 1 , 2 , 4 , 8 , 16 , 32 , 64 , 128 , 256 nM , respectively ) . ( B ) Quantification of activity assays . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 01810 . 7554/eLife . 12256 . 019Figure 5—figure supplement 3 . MMS survival assays with yeast yen1 mutants . The survival of yen1 mutants was tested in a yen1Δ or yen1Δ mus81Δ background with indicated amounts of MMS ( compare Figure 5 and Figure 5—source data 2 ) . Mus81 overlaps with Yen1 functionally , therefore yen1Δ knock-out strains are fully viable even in the presence of MMS , and hypersensitivity is only seen in the double knock-out . ( A ) Mutations in the chromodomain . ( B ) Mutations at protein-DNA interfaces . ( C ) Yen1 truncations and chromodomain deletion . ( D ) Protein expression test ( Western Blot analysis ) of 3FLAG tagged Yen1 variants . Asterisk denotes a cross-reactive band . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 019 The active site modification D30N showed that the cleavage activity on both HJ and 5’ flap substrates was lost in agreement with previously published data ( Ip et al . , 2008 ) . According to our structure , R54 in helix α3 at the upstream interface fixes the substrate position by reaching into the minor DNA groove and we observed that R54E had a strongly reduced cleavage activity ( ~50%; Figure 5B ) , indicating a key role in substrate positioning . Residue C36 in helix α2 points towards the downstream interface and likely contacts the DNA upon binding ( compare Figure 5D ) . The corresponding FEN1 Y40 , is a key residue stacking with the -1 base of the 5’ flap at the FEN1 active site ( Tsutakawa et al . , 2011 ) . Therefore , we tested the cleavage ability of a GEN1C36E and found that the mutant protein had completely lost its enzymatic activity for both , HJ and 5’ flap cleavage , to the same degree as the active site modification D30N ( Figure 5B ) . This effect is stronger than for FEN1Y40A , which showed only a partial loss in activity ( Tsutakawa et al . , 2011 ) . Our results suggest that C36 provides a polar interface for orienting and guiding the cleaved strand towards the active site and the lower gateway . We further tested a glutamate modification of the superfamily-conserved R89 and R93 located in the disordered part continuing to helix α6 , presumably forming an arch ( see above ) . The arch was shown to facilitate cleavage by clamping flap substrates in FEN1 and the modification R100A showed a strong decrease in the cleavage activity ( Patel et al . , 2012 ) . The GEN1 R89E mutation , corresponding to residue R100 in FEN1 , showed that the activity of GEN1 with a HJ substrate was not altered . In the case of a 5’ flap substrate , cleavage was slightly reduced and it reached to the full level at enzyme concentrations higher than 10 nM . The effect of the R93E modification was even less pronounced compared to R89E . In contrast , the cleavage of both 5’ flap and HJ substrates depended strongly on F110 at helix α6 ( thumb ) , which points towards the active site . An F110E modification showed a reduction in cleavage by 25% for HJ substrates , and the effect was even stronger for 5’ flap substrates , where the activity is reduced by 65% . The equivalent position in FEN1 is V133 showing a critical involvement in stabilizing 5’ flap DNA by orienting the -1 nucleotide for catalysis ( Tsutakawa et al . , 2011 ) . We have also tested the effect of modifying H109 , which neighbors the critical F110 . Even though it points away from the active site , a glutamate at this position reduced 5’ flap cleavage to 83% and HJ cleavage recovered only at high substrate concentrations of 256 nM . Overall , the results suggest that F110 has a key position for DNA recognition and processing . Classical HJ resolvases introduce two symmetrical incisions across the junction point by coordinating the action of two active sites . The first nick is rate-limiting and the second one takes place near-simultaneously and within the lifetime of the resolvase-DNA complex . This mechanism has been well studied for bacterial and bacteriophage HJ resolvases ( Fogg and Lilley , 2000; Giraud-Panis and Lilley , 1997; Pottmeyer and Kemper , 1992; Shah et al . , 1997 ) . Hence , it is thought that also GEN1 dimerizes upon binding to HJ substrates as indicated by coordinated cleavage and by an increase in hydrodynamic radius compared to protein alone ( Chan and West , 2015; Rass et al . , 2010 ) . In order to further examine the effect of GEN1 modifications on HJ cleavage , we used a cruciform plasmid cleavage assay to evaluate GEN1’s nicking function , as illustrated in Figure 5E . Here , the plasmid pIRbke8mut served as a substrate that contains an inverted-repeat sequence extruding a cruciform structure when supercoiled ( Chan and West , 2015; Lilley , 1985; Rass et al . , 2010 ) . Coordinated dual incision of the cruciform ( by a dimer ) leads to linear duplex products with slow migration , whereas uncoordinated cleavage ( by monomeric enzymes ) results in nicked plasmids that migrate even slower ( Figure 5F ) . Cruciform structures are reabsorbed when the superhelical stress is released upon single nicking and the DNA cannot serve as a substrate anymore . We observed that wild type GEN1 resolved cruciform structures into linear products ( Figure 5F ) in agreement with previous reports ( Chan and West , 2015; Rass et al . , 2010 ) . GEN1C36E ( downstream interface ) and GEN1R54E ( upstream interface ) showed only residual activity confirming their importance for HJ cleavage . The cruciform cleavage by F110E ( thumb ) was strongly reduced in line with our nuclease assays using small DNA substrates ( Figure 5B ) . GEN1R89E ( disordered part of the arch ) did not show any appreciable effect , which suggests that this part of the arch is not directly involved in HJ recognition . Taken together , our results suggest that the positioning of HJ junction substrates both at the upper and the lower gateway is critical for productive cleavage . Furthermore , none of the tested modifications at the different DNA interaction interfaces was able to uncouple the coordinated HJ cleavage . Agreeing with the structural significance for DNA binding , the truncation of the chromodomain ( Δchromo , residues 2-389 ) showed a severe reduction ( ~3-fold ) in HJ cleavage activity whereas all longer GEN1 fragments containing the chromodomain ( 2-464 , 2-505 and 2-551 ) showed full activity ( Figure 5—figure supplement 2 ) . Interestingly , the effect of the chromodomain truncation is even more pronounced for 5’ flap DNA cleavage than for HJs , showing a 7-fold reduction compared to wild type ( Figure 5C ) . The activity of GEN1 in the plasmid-based cruciform cleavage assay was also severely hampered in the absence of the chromodomain ( Figure 5F ) showing only a weak band for linear products and no increase for nicked plasmid , emphasizing the importance of the chromodomain for GEN1 activity . Further , to test the influence of the positively charged side chains K404 and R406 on DNA binding , we introduced charge-reversal mutations to glutamates and assessed their nuclease activities . Even though K404 and R406 are within hydrogen-bonding distance to the DNA , K404E , and R406E showed no appreciable influence on GEN1’s nuclease activity . Only a slight reduction in cleavage of 5’ flap substrates was observed for GEN1R406E , whereas the processing of HJ substrates was not altered significantly ( Figure 5C ) . This reinforces the conclusion from our structural observations that the chromodomain and the DNA interact through their backbones via van-der-Waals interactions . PhosphoSitePlus ( Hornbeck et al . , 2014 ) lists two phosphorylation sites at residues T380 and T438 in GEN1 that were found in a T-cell leukemia and a glioblastoma cell line . These residues are located in helix α15 and at the rim of the aromatic cage , respectively . Both phosphorylation sites are positioned to interrupt hydrophobic interactions between helix α15 and the chromodomain ( Figure 5D and Figure 4F ) . Therefore , we tested if the phosphorylation-mimicking modifications T380E and T438E had an effect on GEN1’s activity . At low enzyme concentrations ( <50 nM ) HJ cleavage was similar to that of wild-type protein but at high concentrations the activity declined to less than 80% ( Figure 5C ) . For a 5’ flap substrate , the assay showed consistently lower activity than wild type , recovering to about 80% cleavage at the highest enzyme concentration ( Figure 5C ) . These results suggest that phosphorylation of GEN1 chromodomain residues may regulate DNA recognition and cleavage . To test the physiological relevance of the identified GEN1-DNA interactions , we investigated the survival of Saccharomyces cerevisiae mutant strains expressing variants of Yen1 ( GEN1 homolog ) after treatment with the DNA-damaging agent MMS ( Figure 5G and Figure 5—figure supplement 3/source data 2 ) . All Yen1 variants were expressed to a similar degree as endogenous Yen1 , which was confirmed by Western Blot analysis ( Figure 5—figure supplement 3 ) . Because of the functional overlap of Mus81 and Yen1 in HR ( Blanco et al . , 2010 ) a double knockout ( yen1Δ mus81Δ ) was used and complemented with different variants of Yen1 . The control strain , complemented with wild type Yen1 , survived MMS concentrations of up to 0 . 01% , consistent with the described hypersensitivity of mus81Δ mutants ( Blanco et al . , 2010; Interthal and Heyer , 2000 ) . In stark contrast , cells containing either the active site mutant Yen1-D41N ( corresponding to GEN1D30N ) or the downstream interface mutant Yen1-F47E ( corresponding to GEN1C36E ) did not grow even at an MMS concentration as low as 0 . 0025% ( Figure 5G ) . After the expression of the upstream interface mutant Yen1-I97E ( corresponding to GEN1R54E ) cells showed a slight but significant growth defect at high MMS concentrations ( see panels for 0 . 0075% and 0 . 01% MMS in Figure 5G ) . These results are therefore consistent with the in vitro cleavage results carried out with GEN1 mutants and showing a reduction in activity for R54E and no activity for C36E ( see Figure 5C ) . As a last mutant in the nuclease core , we tested the K298E mutation which is located in helix α10 of the H2TH motif in the downstream DNA-binding interface , and for which we were unable to obtain the corresponding GEN1K219E modification for cleavage assays ( compare Figure 5D ) . This mutant displayed a strong sensitivity towards MMS but lower than the one observed for the catalytic mutant , indicating that the mutant was partially functional in yeast ( Figure 5G ) . We next investigated the effect of mutations in the aromatic cage of Yen1's chromodomain ( compare Figure 3 ) and found that their severity was strongly position dependent . Mutation of R486E and Y487A in Yen1 , both of which are located near the base of the cage , corresponding to the W418 position in GEN1 ( see Figure 3C ) , showed a strong effect on MMS sensitivity ( see Figure 5G ) , similar to the one observed for the catalytic mutant , presumably due to a dysfunctional chromodomain . In contrast , mutations located further outside of the core ( F478A and K484E ) led to a less pronounced MMS sensitivity . The same was true for the K469E variant , which corresponds to position R406 at the chromodomain-DNA interface in GEN1 ( see Figure 3A and 5F ) , and for residues at the rim of the chromodomain ( yen1-N526A , yen1-L528D and yen1-W529A ) , consistent with our in vitro observation for GEN1T438E ( slightly reduced activity , Figure 5C ) . No effect on MMS sensitivity was detected for yen1-L530A , which corresponds to a conserved glutamate in chromodomains ( E440 in GEN1 ) . Lastly , we found that the deletion of the chromodomain ( Yen1-Δ452–560 ) lead to a severe phenotype comparable to the active site mutant Yen1-D41N ( Figure 5G and Figure 5—source data 2 ) . The Yen1 variant lacking the chromodomain was expressed to levels similar to the full-length protein and we therefore conclude that the chromodomain is crucial for the function of Yen1 . Taken together , the functional data of Yen1 mutants in vivo and GEN1 mutants in vitro point towards an essential and evolutionary conserved role of the chromodomain in GEN1/Yen1 proteins .
The structure of the human GEN1 catalytic core provides the missing structural information in the Rad2/XPG family . The GEN1 structure complements recent reports on the structures of Rad2 , EXO1 and FEN1 , ( Miętus et al . , 2014; Orans et al . , 2011; Tsutakawa et al . , 2011 ) . Thereby , it gives insights how relatively conserved nuclease domains recognize diverse substrates in a structure-selective manner and act in different DNA maintenance pathways . In comparison with other Rad2/XPG nucleases , GEN1 shows many modifications on common structural themes that give the ability to recognize a diverse set of substrates including replication fork intermediates and HJs . The upstream DNA interface of GEN1 lacks the 'acid block' found in FEN1 , instead it has a prominent groove at the same position ( compare Figure 1 , 'upper gate' ) with a strategically positioned R54 nearby . Furthermore , the helical arch in GEN1 misses helix α5 , which forms a cap structure in FEN1 and EXO1 that stabilizes 5’ overhangs for cleavage . These features have implications for the recognition and cleavage of HJ substrates ( see below ) . The most striking difference to other Rad2/XPG family members is that the GEN1 nuclease core is extended by a chromodomain , which provides an additional DNA anchoring point for the upstream DNA-binding interface . The evolutionarily conserved chromodomain is important for efficient substrate cleavage as we showed using truncation and mutation analyses . This finding opens new perspectives for the regulation of GEN1 and for its interactions with other proteins . Chromodomains serve as chromatin-targeting modules ( reviewed in Blus et al . , 2011; Eissenberg , 2012; Yap and Zhou , 2011 ) , general protein interaction elements ( Smothers and Henikoff , 2000 ) as well as dimerization sites ( Canzio et al . , 2011; Cowieson et al . , 2000; Li et al . , 2011 ) . These possibilities are particularly interesting , as chromatin targeting of proteins via chromodomains has been implicated in the DNA damage response . The chromatin remodeler CHD4 is recruited in response to DNA damage to decondense chromatin ( reviewed in O’Shaughnessy and Hendrich , 2013; Stanley et al . , 2013 ) . The chromodomains in CHD4 distinguish the histone modifications H3K9me3 and H3K9ac and determine the way how downstream DSB repair takes place ( Ayrapetov et al . , 2014; Price and D’Andrea , 2013 ) . It is plausible that GEN1 uses its chromodomain not only as a structural module to securely bind DNA but also for targeting or regulatory purposes . Even though it was not possible to find any binding partner with a series of tested histone tail peptides , we cannot exclude that the chromodomain is used as an interaction motif or chromatin reader . It will therefore be interesting to extend our interaction analysis to a larger number of peptides and proteins . Interestingly , the modifications GEN1L397E and GEN1Y424A at the rim of the chromodomain did not alter DNA cleavage activity ( Figure 5—figure supplement 1 ) , however , mutations of residues at the rim of Yen1’s chromodomain show a phenotype , suggesting an additional role like binding to an endogenous factor . Another intriguing aspect of the chromodomain is that the conserved T438 at the rim of the aromatic cage and T380 at the closing helix α15 are both part of a casein kinase II consensus sequence for phosphorylation ( Ser/Thr-X-X-Asp/Glu ) . Ayoub et al . , 2008 showed that the analogous threonine in the chromodomain of CBX1 is phosphorylated in response to DNA damage and phosphorylation disrupts the binding to H3K9me . We observed a reduction in DNA cleavage activity for the phosphorylation mimicking mutations T380E and T438E , which may suggest a regulatory role . They might function together and in combination with other modifications to provide a way of functional switching at the chromodomain . Furthermore , Blanco et al . , 2014 and Eissler et al . , 2014 recently identified several CDK phosphorylation sites in an insertion in the Yen1 chromodomain which affects HJ cleavage and together with phosphorylation of a nuclear localization signal ( NLS ) in the regulatory domain restricts Yen1’s activity to anaphase . The insertion is not found in other chromodomains and it is extended in Yen1 compared to GEN1 , which is lacking these phosphorylation sites ( compare Figure 3A/B ) . Notably , the activity of Yen1 is negatively regulated by CDK-dependent phosphorylation ( Blanco et al . , 2014; Chan and West , 2014; Eissler et al . , 2014; Matos et al . , 2011 ) , suggesting that the chromodomain is targeted by cell cycle kinases . It also provides a likely explanation for the different regulatory mechanisms found in GEN1 and Yen1 ( Blanco and Matos , 2015; Chan and West , 2014; Matos and West , 2014 ) . Exploration of the regulatory function of the GEN1 chromodomain will be an important topic to follow up , and this may lead to the understanding of the precise regulation mechanism of GEN1 as well as its substrate recognition under physiological conditions . It is noteworthy that our analysis also revealed that the human transcription modulator AEBP2 , which is associated with the polycomb repression complex 2 ( PRC2 ) , contains a chromo-barrel domain , which , to our knowledge , has not been reported so far . The GEN1-DNA structure showed a considerable similarity to the other members of the Rad2/XPG family , and this facilitated the generation of a combined model to understand substrate recognition of GEN1 ( Figure 6 ) . This was done by superimposing the protein part of the FEN1-DNA complex ( PDB 3q8k ) onto our GEN1 structure and extending the DNA accordingly ( Figure 6A/B ) . Remarkably , the superimposition of the proteins aligns the DNA from the FEN1 structure in the same register as the DNA in the GEN1 complex at the upstream interface ( Figure 6A and 6B insert ) . Furthermore , the free 5’ and 3’ ends of the double flap DNA from the FEN1 structure point towards the lower and the upper gateway in GEN1 , respectively ( Figure 6B ) . We extended the GEN1 structure by homology modeling of the disordered residues 79-92 ( helix α4 ) in GEN1 ( Figure 6B ) . In addition to the similarity of this part to FEN1 , the model readily showed the arrangement forming an arch structure . This would explain why GEN1 recognizes 5’ flap substrates efficiently , analogous to FEN1 , as the arch can clamp a single-stranded DNA overhang for productive cleavage . This also explains why the F110E modification in the arch at helix α6 hampered 5’ flap cleavage severely . The side chain points directly towards the active site and likely disturbs the stabilization of a 5’ overhang for catalysis by charge repulsion . However , there are two features in GEN1 that vary from the arrangement in FEN1 and EXO1 considerably . Helix α6 is longer ( 24 instead of 15 residues ) and helix α5 is missing in GEN1 . As a result the arch points away from the DNA rather than forming a 'cap' structure as it is observed in FEN1 and EXO1 ( Orans et al . , 2011; Tsutakawa et al . , 2011 ) . Furthermore , the modified arch in GEN1 provides an opening , marked as 'lower gate' in Figure 6B . These differences are likely the basis for GEN1’s versatile DNA recognition features . 10 . 7554/eLife . 12256 . 020Figure 6 . Substrate recognition features of GEN1 . ( A ) Superposition of the protein part of the FEN1-DNA complex ( PDB 3q8k , protein in gray , DNA in black ) onto the GEN1-HJ complex ( protein in green and the DNA strands in different colors ) . The FEN1-DNA aligns with the same register as the GEN1-DNA at the upstream interface . ( B ) Model for the recognition of a 5’ flap substrate by GEN1 . The DNA was extended using the superimposition from A . Homology modeling suggests an additional helix α4 ( disordered residues 79–92 ) forming an arch with helix α6 . The protein is shown in a simplified surface representation with the same colors as in Figure 1 and structural elements are highlighted . The insert shows a zoomed in view of the hydrophobic wedge with the modeled FEN1-DNA in gray . ( C ) Model for the dimerization of GEN1 upon binding to a HJ substrate based on the 5’ flap model in B . The monomers interlock via both arches ( α4-α6 ) and the hydrophobic wedges ( α2-α3 ) contact each other . ( D ) Structure of the Thermus thermophilus RuvC-HJ complex ( PDB 4ld0 ) . ( E ) Structure of the T4 endonuclease VII-HJ complex ( PDB 2qnc ) . ( F ) Structure of the T7 endonuclease I-HJ complex ( PDB 2pfj ) . Individual monomers are in surface representation , colored in light blue and beige , respectively . DNA strands are shown as ladders in different colors . DOI: http://dx . doi . org/10 . 7554/eLife . 12256 . 020 The diverging orientation of the arch ( helices α4 and α6 ) in GEN1 compared to the one in FEN1 and EXO1 ( helices α4 , α5 , and α6 ) may have thus significance for the recognition of HJ substrates . By pointing away from the active site the arch provides an opening to accommodate unpaired , single-stranded DNA to pass along the arch at the lower gate ( groove between α2 and α4 ) ( Figure 6B 'lower gate' ) from one GEN1 monomer to the upper gate ( groove between α2-α3 and α14 ) ( Figure 6B 'upper gate' ) of the other within a GEN1 dimer ( Figure 6B/C ) . R54 is perfectly positioned at the minor groove to guide the second cleavage strand to pass through the upper gate ( compare Figure 4 and Figure 6B/C , marked with a asterisk ) . In FEN1 , this position is occupied by the 'acid block' , which stabilizes a single 3’ flap of the unpaired substrate ( Tsutakawa et al . , 2011 ) and it would not accommodate longer 3’ DNA overhangs . In our model , two GEN1 monomers come together crosswise upon HJ binding ( Figure 6C ) . The helical arches of both proteins likely provide additional protein-protein interactions as well as protein-DNA contacts by packing against the backbone of opposite DNA arms ( Figure 6C ) . As a result , the GEN1 dimer orients both active sites symmetrically across the junction point resembling the situation in bacterial RuvC ( Figure 6D; Bennett and West , 1995a; Górecka et al . , 2013 ) . This arrangement would ensure that both incisions are introduced within the lifetime of the GEN1-HJ complex as observed biochemically by us and others ( Chan and West , 2015; Rass et al . , 2010 ) . The mechanism likely works in a coordinated nick-and-counter-nick fashion , as shown for bacterial or bacteriophage HJ resolvases ( Fogg and Lilley , 2000; Giraud-Panis and Lilley , 1997; Pottmeyer and Kemper , 1992; Shah et al . , 1997 ) and recently for GEN1 ( Chan and West , 2015 ) . The distance between both gates is bridged by unpaired bases in our GEN1-HJ model . This view is supported by the observation that FEN1 unpairs two bases near the active site through interactions with the hydrophobic wedge leading to strongly bent DNA arms between the upstream and downstream DNA interfaces . This mechanism seems to be a common feature of Rad2/XPG nucleases ( Finger et al . , 2013; Grasby et al . , 2012; Tsutakawa et al . , 2011 ) . Consistent with this view , the bacterial RuvC resolvase ( Figure 6D ) has also been shown to unfold HJ junctions ( Bennett and West , 1995b; Górecka et al . , 2013 ) . In the case of GEN1 , the critical step would be the assembly of the dimer around the junction point in a highly restraint way and the introduction of the first nick . This releases the tension on the complex like a spring leading to an immediate second cut and subsequent disassembly of the GEN1-HJ complex . Furthermore , a HJ does not provide free DNA ends and adopts a structure that intrinsically restrains its degrees of freedom , thus inhibiting cleavage by a single GEN1 monomer . Altogether we speculate that the arch ( helix α4-α6 ) acts like a lever or hatch switching between flap and HJ recognition modes . When a free 5’ end is available it closes and clamps the flap , thus positions the DNA for cleavage . For the case of a HJ substrate , the arch adopts an open conformation , allowing unpaired , single-stranded DNA to pass , while preventing the correct positioning of the DNA for catalysis at first . HJ cleavage is inhibited until a second GEN1 monomer binds . This mechanism differs from the one used by bacterial or bacteriophage HJ resolvases , which act as obligate dimers binding to DNA substrates in a concerted way ( compare Figure 6D–F ) . Our model for DNA cleavage by GEN1 describes a conformational switch provided by a flexible arch that can discriminate between substrates containing free 5’ ends or those with a restraint structure like HJs . This aspect may explain our observation that GEN1 cleaves 5’ flap DNA catalytically while stoichiometric amounts are required for HJ substrates ( Figure 5A–C ) . Using a switchable hatch in a spring-loaded mechanism would be an efficient way of preventing a single cut at a HJ junction while allowing GEN1 to adapt to recognize various DNA substrates and perform different functional roles . Thus , GEN1 may have an intrinsic safety mechanism that ensures symmetrical dual incision across a branch point . Further studies have to address the exact engagement mechanism . GEN1’s biological role is not fully understood yet . Yeast cells are viable without the GEN1 homolog Yen1 even in the presence of DNA damaging agents as the Mus81-Eme1 complex can complement the defect ( compare Figure 5—figure supplement 3; Blanco et al . , 2010 ) . Consistently , both proteins can cleave 5’ flaps and HJ substrates in vitro . However , GEN1 can cleave intact HJs symmetrically whereas MUS81-EME1 is much more efficient with nicked DNA four-way junctions ( Castor et al . , 2013; Wyatt et al . , 2013 ) . Matos et al . , 2011 suggested that Yen1/GEN1 might serve as a backup enzyme to resolve persistent HJs that have eluded other mechanisms of joint molecule removal before cytokinesis . Our analysis infers that HJ cleavage is slower than 5’ flap cleavage ( Figure 5B/C ) , bringing interesting implications for a safety control of GEN1’s activity . GEN1 may have to assemble in an accurate way before it can cleave a HJ . Likewise , it increases GEN1’s persistence time on HJs and opens a window for branch migration for extending the length of recombined stretches of DNA . Moreover , GEN1 recognizes various DNA substrates , which may point towards a general role in processing substrates in different DNA maintenance pathways . GEN1 has been shown to cleave replication fork intermediates , and it is implicated in the resolution of replication-induced HJs ( Garner et al . , 2013; Sarbajna et al . , 2014 ) . Like MUS81-EME1 , it might also be important for the processing of fragile sites to ensure proper chromosome segregation ( Ying et al . , 2013 ) . These functions have to be tested systematically to understand GEN1’s biological role . In this context , the regulation of GEN1 is an important factor and needs to be explored . Our study identified a chromodomain extending the GEN1 nuclease core that might have a role in regulating the enzyme . An open question is the function and architecture of the remaining 444 amino acids at the C-terminus of GEN1 . They are thought to regulate the nuclease activity and control subcellular localization ( Blanco et al . , 2014; Chan and West , 2014; García-Luis et al . , 2014 ) . It is very likely that new interaction sites and post-translational modifications in this region will be discovered in future . The presented structure together with additional studies will help to unravel these questions and to obtain a comprehensive view of the functions of the Rad2/XPG nucleases . | Factors like ultraviolet radiation and harmful chemicals can damage DNA inside living cells , which can lead to breaks that form across both strands in the DNA double helix . “Homologous recombination” is one of the major mechanisms by which cells repair these double-strand breaks . During this process , the broken DNA interacts with another undamaged copy of the DNA to form a special four-way structure called a “Holliday junction” . The intact DNA strands are then used as templates to repair the broken strands . However , once this has occurred the Holliday junction needs to be ‘resolved’ so that the DNA strands can disentangle . One way in which Holliday junctions are resolved is through the introduction of precise symmetrical cuts in the DNA at the junction by an enzyme that acts like a pair of molecular scissors . Re-joining these cut strands then fully restores the DNA . Enzymes that generate the cuts in DNA are called nucleases , and the nuclease GEN1 is crucial for resolving Holliday junctions in organisms such as fungi , plants and animals . GEN1 belongs to a family of enzymes that act on various types of DNA structures that are formed either during damage repair , DNA duplication or cell division . However , GEN1 is the only enzyme in the family that can also recognize a Holliday junction and it was unclear why this might be . Lee et al . have now used a technique called X-ray crystallography to solve the three-dimensional structure of the human version of GEN1 bound to a Holliday junction . This analysis revealed that many features in GEN1 resemble those found in other members of the same nuclease family . These features include two surfaces of the protein that bind to DNA and are separated by a wedge , which introduces a sharp bend in the DNA . However , Lee et al . also found that GEN1 contains an additional region known as a “chromodomain” that further anchors the enzyme to the DNA . The chromodomain allows GEN1 to correctly position itself against DNA molecules , and without the chromodomain , GEN1’s ability to cut DNA in a test tube was severely impaired . Further experiments showed that the chromodomain was also important for GEN1’s activity in yeast cells growing under stressed conditions . The discovery of a chromodomain in this human nuclease may provide many new insights into how GEN1 is regulated , and further work could investigate if this chromodomain is also involved in binding to other proteins . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Human Holliday junction resolvase GEN1 uses a chromodomain for efficient DNA recognition and cleavage |
The access of Transcription Factors ( TFs ) to their cognate DNA binding motifs requires a precise control over nucleosome positioning . This is especially important following DNA replication and during mitosis , both resulting in profound changes in nucleosome organization over TF binding regions . Using mouse Embryonic Stem ( ES ) cells , we show that the TF CTCF displaces nucleosomes from its binding site and locally organizes large and phased nucleosomal arrays , not only in interphase steady-state but also immediately after replication and during mitosis . Correlative analyses suggest this is associated with fast gene reactivation following replication and mitosis . While regions bound by other TFs ( Oct4/Sox2 ) , display major rearrangement , the post-replication and mitotic nucleosome positioning activity of CTCF is not unique: Esrrb binding regions are also characterized by persistent nucleosome positioning . Therefore , selected TFs such as CTCF and Esrrb act as resilient TFs governing the inheritance of nucleosome positioning at regulatory regions throughout the cell-cycle .
Gene regulatory processes are frequently governed by sequence-specific Transcription Factors ( TFs ) that recognize specific DNA binding motifs enriched at promoters and enhancers ( Spitz and Furlong , 2012 ) . To gain access to DNA , which in eukaryotes is wrapped around a histone core octamer – the nucleosome ( Luger et al . , 1997 ) – TFs employ different strategies ( Voss and Hager , 2014 ) . Whereas nucleosomal DNA is accessible to pioneer TFs ( Cirillo , 1998; Zaret and Carroll , 2011 ) , other TFs require co-factors to evict or displace nucleosomes that occlude their binding sites . These co-factors can be additional TFs that act cooperatively to destabilize nucleosomes ( Mirny , 2010; Miller and Widom , 2003 ) , and/or ATP-dependent activities that slide or evict nucleosomes and thus remodel cis-regulatory elements ( Becker and Workman , 2013; Li et al . , 2014 ) . Typically , TF binding regions are characterized by Nucleosome Depleted Regions ( NDRs ) centered on TF motifs and flanked by Nucleosome Ordered Arrays ( NOAs ) ( He et al . , 2010; Wang et al . , 2012; Valouev et al . , 2011; Teif et al . , 2012 ) . This is particularly well illustrated by the genomic binding sites of the zinc finger CCCTC-binding protein ( CTCF ) ( Fu et al . , 2008; Teif et al . , 2014 ) , a TF involved in chromatin organization and transcriptional control ( Nora et al . , 2017; Merkenschlager and Nora , 2016 ) . However , the passage of the replication fork during DNA replication and the condensation of the chromatin during mitosis are associated with a general loss of TF binding and nucleosome positioning at their target sites ( Ramachandran and Henikoff , 2016; Festuccia et al . , 2019 ) . Whether TFs are readily required to either maintain or rapidly rebuild local nucleosome architectures during or after replication and mitosis remains unclear . DNA replication leads to a period during which TFs and nucleosomes enter into direct competition; in Drosophila S2 cells , the reconstitution of specific NDRs/NOAs over active regulatory elements , particularly at enhancers , takes much longer than previously anticipated ( Ramachandran and Henikoff , 2016 ) . Similarly , in mouse Embryonic Stem ( ES ) cells , chromatin accessibility over TF binding sites is lost during replication and progressively reacquired as nascent chromatin matures ( Stewart-Morgan et al . , 2019 ) . During mitosis , regulatory elements display strongly attenuated nucleosome phasing and , more strikingly , enhancers are invaded by stable nucleosomes , as shown in ES cells ( Festuccia et al . , 2019 ) . Hence , both replication and mitosis can be seen as a tabula rasa of functional interactions between TFs , their cognate motifs and local nucleosomal architectures . Thus , how proliferating cells maintain or restructure nucleosome arrays over regulatory elements as they undergo cycles of replication and mitosis , is largely unknown . This seems particularly important during early development , when TFs not only instruct but also maintain cell identity ( Soufi and Dalton , 2016; Festuccia et al . , 2017a; Festuccia et al . , 2017b; Egli et al . , 2008 ) . For instance , the TF Zelda was shown to be continuously required during early Drosophila development , suggesting that by means of its pioneering activity it is capable of rapidly rebinding its targets after the passage of the replication fork ( McDaniel et al . , 2019 ) . While direct , nucleosome-based evidence is still lacking , it is likely that Zelda ensures the rapid reestablishment of NDRs/NOAs at its binding sites after replication ( McDaniel et al . , 2019 ) . Moreover , recent evidence does not favor a model in which Zelda directly controls its target sites during mitosis ( Dufourt et al . , 2018 ) . In contrast , the TF Esrrb was shown to act as a mitotic bookmarking factor that binds thousands of regulatory elements in mitotic ES cells ( Festuccia et al . , 2016 ) . At these sites , the nucleosomes preserve an interphase-like configuration whereas at regions losing TF binding nucleosomal arrays are largely disorganized ( Festuccia et al . , 2019 ) . Whether Esrrb also maintains nucleosome positioning during replication remains however unknown . The incomplete correlations that are currently available suggest a model in which specific TFs may govern nucleosome positioning during replication and/or mitosis , a mechanism that can potentially complement the inheritance of gene regulatory states by independent epigenetic mechanisms . Here , we focus on CTCF to show that this TF is strictly required to maintain nucleosome positioning in interphase , immediately after replication and during mitosis , in mouse ES cells . While this is also observed at Esrrb binding regions , those bound by other TFs such as Oct4/Sox2 display significant nucleosome rearrangement . Further , we show that genes rapidly reactivated after replication and mitosis are closely associated with CTCF binding . Therefore , certain , but not all TFs , govern nucleosome positioning and confer chromatin resiliency during replication and mitosis to foster the appropriate re-establishment of transcription profiles .
CTCF binding has been previously shown to take place over large and phased nucleosomal arrays displaying a prominent NDR ( Fu et al . , 2008; Teif et al . , 2014; Barisic et al . , 2019; Wiechens et al . , 2016 ) . Therefore , we established the repertoire of CTCF binding regions in ES cells ( cultured in serum and LIF ) by Chromatin Immunoprecipitation followed by sequencing ( ChIP-seq; Figure 1A; top panel; Supplementary file 1 ) . We identified 52 , 129 regions displaying robust peaks that significantly overlap with previous studies ( Nora et al . , 2017; Davis et al . , 2018; Pękowska et al . , 2018 ) ( Figure 1—figure supplement 1A; Supplementary file 2 ) . At these regions , we observed that the aggregate of CTCF motifs directly correlates with CTCF occupancy ( Figure 1—figure supplement 1B ) , indicating that the presence of its cognate binding sequence drives CTCF recruitment . Next , we used Microccocal Nuclease digestion ( MNase-seq ) , histone H3 ChIP-seq following MNase digestion and Assay for Transposase Accessible Chromatin ( ATAC-seq ) to profile nucleosomes genome-wide ( Festuccia et al . , 2019 ) . At individual CTCF binding regions , we observed highly organized nucleosome arrays , particularly over large CTCF peaks ( Figure 1A; bottom panels ) . To obtain a global picture of how nucleosomes are organized around CTCF binding regions , we generated V-plots ( Kent et al . , 2011; Henikoff et al . , 2011 ) centered on the best CTCF motif of each binding region . We observed a very well defined NOA/NDR/NOA structure surrounding CTCF motifs ( Figure 1B and Figure 1—figure supplement 1C ) . CTCF binding footprints , identified as small MNase ( <100 bp ) and ATAC ( <150 bp ) fragments , were also detected within the NDR except upon H3 ChIP ( Figure 1B and Figure 1—figure supplement 1C ) . These observations indicate that CTCF binding at its motif is closely associated with nucleosome positioning . Next , we explored the relationships between the magnitude of CTCF binding , the presence of its cognate motif and nucleosome positioning . After ranking CTCF binding regions by their peak height , we confirmed that the aggregate of CTCF motifs is correlated with CTCF occupancy ( Figure 2A ) . Further , as the motif score and CTCF binding diminishes , the associated MNase- and ATAC-footprints decrease ( Figure 2B ) , providing independent validation of the direct correlation between CTCF binding and the presence of its motif . Notably , the reduction of CTCF binding across the regions correlates with a progressive loss of nucleosome order ( Figure 2C ) , consistent with what has been previously observed ( Vainshtein et al . , 2017 ) . This is particularly well illustrated by the quantitative analysis of the position of the −1 and +1 nucleosomes , which roll inwards , shrinking the NDR and resulting in increased nucleosome occupancy at the CTCF motif ( Figure 2D ) . These observations argue that the interaction of CTCF with its cognate motif acts as a major force driving the establishment of NDRs/NOAs , as previously suggested by siRNA knock-down of CTCF ( Wiechens et al . , 2016 ) . To further establish whether CTCF is readily required to maintain local NDRs/NOAs , we used an Auxin-inducible depletion strategy enabling assessment of the immediate consequences of a loss of CTCF . Hence , we exploited ES cells expressing CTCF fused to an Auxin-Inducible Degron ( Nora et al . , 2017 ) ( AID; thereafter CTCF-aid ) . In accord with the hypomorphic behavior of this line ( Nora et al . , 2017 ) , CTCF-aid displayed reduced expression ( Figure 3—figure supplement 1A ) , bulk chromatin association ( Figure 3—figure supplement 1B ) and binding levels ( Figure 3A ) that correlated with less prominently positioned nucleosomes compared to wild-type cells ( Figure 3B ) . Upon a short ( 2 hr ) treatment with the Auxin analog Indole-3-Acetic Acid ( IAA ) , CTCF-aid expression ( Figure 3—figure supplement 1A ) , chromatin association ( Figure 3—figure supplement 1B ) and binding ( Figure 3A ) were significantly reduced . This led to a dramatic loss of NOAs , major displacements of the +/- 1 nucleosomes , and an invasion of the NDR by nucleosomes ( Figure 3B , C ) . Since this effect was observed across different classes of functional genetic features ( Figure 3D ) , and rapidly follows loss of CTCF , we conclude that CTCF is a major determinant of local nucleosome positioning in steady-state conditions . During replication , TFs are evicted from their targets and the chromatin has to be reconstituted downstream of the replication fork ( Ramachandran and Henikoff , 2016; Groth et al . , 2007; Stewart-Morgan et al . , 2019 ) . Hence , after replication , TFs and nucleosomes have been shown to be in direct competition , a phenomenon that substantially delays the reconstitution of proper NDRs/NOAs over enhancers , as shown in Drosophila S2 cells ( Ramachandran and Henikoff , 2016 ) . Given that proteomic approaches have identified CTCF on newly synthesized chromatin ( Alabert et al . , 2014 ) , we aimed at studying the consequences of replication over CTCF binding sites . To do this , we used Mapping In vivo Nascent Chromatin with EdU ( MINCE-seq ) , a technique that enables the capture of newly synthetized nucleosomal DNA following MNase digestion ( Ramachandran and Henikoff , 2016 ) . We performed two measurements , one immediately after a short pulse of EdU ( 2 . 5 min pulse ) and another 1 hr after a chase period without further EdU incorporation ( Figure 4—figure supplement 1A; Supplementary file 1 ) ; these two time-points reflect the status of newly replicated ( pulse ) and maturing chromatin ( chase ) . In contrast to S2 cells , which even 1 hr post-replication still display altered nucleosomal structures at enhancers ( Ramachandran and Henikoff , 2016 ) , CTCF binding regions display a remarkable nucleosomal resiliency in ES cells . Indeed , only minor changes , if any , were appreciable just after replication ( Figure 4A; center panel ) . During the following hour , CTCF binding regions acquire a nucleosomal structure indistinguishable from the controls ( Figure 4A; right panel ) . Given the direct role of CTCF in local nucleosome positioning ( Figures 2 and 3 ) , it is therefore likely that CTCF is capable of rapidly rebinding its sites post-replication to efficiently re-establish NDRs and NOAs . Prompted by the fast reorganization of CTCF-binding regions in ES cells , we explored the post-replication nucleosome dynamics over ES cell enhancers , centered on p300 summit . We observed an increase in nucleosome density over the NDRs and attenuated NOAs immediately after replication ( Figure 4A; center panel ) , contrasting markedly with CTCF binding regions . One hour after replication , however , we observed that ES cell enhancers were near completely restored ( Figure 4A; right panel ) , indicating that ES cells are more efficient than S2 cells in reconfiguring nucleosome positioning at enhancers . ES cell enhancers are bound by several TFs , often in complex combinations and stoichiometries , as exemplified by regions binding Esrrb , Oct4 and Sox2 ( Heurtier et al . , 2019 ) , three master TFs of pluripotency . At some of these regions nucleosome positioning is organized around the Esrrb motif , and at others , around the Oct4/Sox2 composite motif ( Festuccia et al . , 2019 ) . Thus , we explored whether the dynamics of nucleosome repositioning after replication are distinct over these regions . At regions bound by Oct4/Sox2 , nucleosome positioning was altered upon replication ( Figure 4A ) . This indicates that the pioneer activity of these two factors ( Soufi et al . , 2015 ) , illustrated here by the detection of a nucleosome overlapping their motif ( Figure 4A ) , is not sufficient to rapidly reorganize nucleosomal arrays after replication . Regions organized by Esrrb , however , displayed a prominent NDR centered on the Esrrb motif , two particularly well positioned flanking nucleosomes and a persistent NOA , just after replication ( Figure 4A; center panel ) . Strikingly , the NDR and the positioning of the +/- 1 and surrounding nucleosomes were more prominent immediately after replication than after 1 hr ( Figure 4A ) . This suggests that following replication , Esrrb is rapidly rebound at these sites and imposes strong nucleosome positioning , which is subsequently modified by the binding of additional TFs – a phenomenon that we previously described when we compared Esrrb binding in interphase and mitosis ( Festuccia et al . , 2019 ) . These observations , together with the analysis of p300-centered enhancers and CTCF binding regions , suggest that different sets of regions substantially differ in post-replication nucleosome dynamics . To obtain quantitative comparisons between these regions , we computed two parameters: the R2 coefficient revealing the similarity of the nucleosome profiles after replication with the controls ( Figure 4B; top panel ) and the spectral density assessing nucleosome periodicity ( Figure 4B; bottom panel ) . Both measurements clearly revealed a comparably delayed restoration of nucleosome positioning at p300-enhancers and Oct4/Sox2 regions after replication . In contrast , Esrrb and CTCF , display , in turn , increasing capacity to reinstate NDRs/NOAs minutes after the passage of the replication fork . These differential post-replication nucleosome dynamics among TF binding regions are independently supported by our analysis of repli-ATAC data recently generated in ES cells ( Figure 4—figure supplement 1B ) , a technique where chromatin accessibility is measured as chromatin matures after replication ( Stewart-Morgan et al . , 2019 ) . Given that both Esrrb and CTCF can re-establish nucleosome positioning following replication , we then turned our attention to mitosis , where Esrrb has already been shown to bind at sites preserving NDRs/NOAs ( Festuccia et al . , 2016; Festuccia et al . , 2019 ) . Using MNase-seq data , we observed the typical NDR/NOA structure at CTCF binding regions in mitotic cells ( Figure 5A ) obtained by nocodazole shake-off with >95% purity ( Festuccia et al . , 2019 ) . Therefore , we hypothesized that CTCF acts as a mitotic bookmarking factor ( Shen et al . , 2015; Sekiya et al . , 2017; Zhang , 2019; Burke et al . , 2005 ) . Thus , we performed CTCF ChIP-seq and observed site-specific interactions in mitotic cells associated with ordered nucleosomal arrays ( Figure 5B; Supplementary file 1 ) . Using a generalized linear model , we identified 29 , 539 bookmarked sites ( 56 . 7% ) and categorized them according to the relative sizes of interphase and mitosis peaks ( Figure 5—figure supplement 1A–C; Supplementary file 2 ) . Cohesin , a recurrent partner of CTCF in interphase ( Merkenschlager and Nora , 2016 ) , was found fully evicted from its targets in mitosis ( Warren et al . , 2000; Waizenegger et al . , 2000 ) ( Figure 5B and Figure 5—figure supplement 2A; Supplementary file 2 ) , underscoring the specificity of our observation for CTCF . Notably , we observed that nearly all regions displaying robust enrichment for Cohesin in interphase are bookmarked by CTCF , even though Cohesin accumulation is not a prerequisite for CTCF bookmarking ( Figure 5—figure supplement 2B , C ) . Moreover , the existence of high quality CTCF motifs appeared to be a better indicator of binding in mitosis than in interphase ( Figure 5—figure supplement 3 ) , suggesting that the conditions for CTCF binding are more stringent in mitosis . Our finding of CTCF bookmarking activity in ES cells contradicts recent data in human somatic cells ( Oomen et al . , 2019 ) . However , our analyses in mouse somatic cell lines ( NIH3T3 and C2C12 ) revealed no or reduced evidence of mitotic bookmarking activity by CTCF ( Figure 5C and Figure 5—figure supplement 4; Supplementary file 1 ) . Therefore , even though CTCF can decorate mitotic chromosomes globally , as revealed by microscopy in cell lines and embryos ( Figure 5—figure supplement 5 ) , this is not necessarily translated into mitotic bookmarking capacity in every cell type . We further validated mitotic CTCF bookmarking by analyzing MNase and ATAC footprints after ranking the regions by the magnitude of mitotic binding determined by ChIP-seq ( Figure 6A ) . Examining the association between CTCF binding and nucleosome positioning in mitosis , and in complement to the analysis in interphase ( Figure 2 ) , we found that the progressive reduction of mitotic CTCF binding correlates with a gradual loss of NDRs and NOAs ( Figure 6A ) . Moreover , when CTCF binding regions were split as bookmarked or lost in mitosis , we could confirm that the nucleosomes are less well positioned ( Figure 6B , C and Figure 5—figure supplement 1D ) and exhibit clear displacements toward the motif ( Figure 6C , D ) . Overall , this establishes that CTCF is a mitotic bookmarking TF that preserves nucleosome order during mitosis . Nevertheless , even CTCF bookmarked regions presented a strong displacement inwards of the +1 nucleosome ( 25 bp for +1 versus 3 bp for −1 nucleosome; Figure 6A , D ) and a more moderate shift of all following nucleosomes ( Figure 6D ) . This indicates that the constraints imposed on the nucleosomes by CTCF are slightly different in interphase and in mitosis . Since in interphase CTCF sites that do not bind Cohesin do not show this repositioning of the +1 nucleosome ( Figure 5—figure supplement 2D ) , it cannot be solely explained by the mitotic loss of Cohesin . Additional factors therefore influence nucleosome positioning at CTCF binding regions in either interphase or mitosis . Next , we aimed at exploiting CTCF-aid ES cells to address the impact of losing mitotic CTCF bookmarking directly . We observed that the hypomorphic nature of CTCF-aid binding was amplified in mitosis; CTCF-aid was barely detectable in bulk mitotic chromatin ( Figure 3—figure supplement 1B ) and more specifically at regions binding CTCF in wild-type mitotic cells ( Figure 6—figure supplement 1A ) . This global loss of mitotic bookmarking was associated with the acquisition of nucleosomal properties characteristic of regions losing mitotic CTCF binding in wild-type cells: nucleosome positioning was strongly attenuated ( Figure 6C and Figure 6—figure supplement 1B ) ; the NDRs were partially invaded by nucleosomes ( Figure 6—figure supplement 1C ) ; nucleosomes , especially upstream of the motif , shifted inwards ( ~55 bp upstream versus 21 bp downstream; Figure 6D and Figure 6—figure supplement 1D ) . These changes in nucleosome positioning were only minimally increased upon IAA treatment ( Figure 6—figure supplement 1B–D ) , which leads to further invasion of the NDR by nucleosomes ( Figure 6E ) . We conclude , therefore , that CTCF behaves as a canonical bookmarking factor in ES cells and actively maintains nucleosome order , mirroring our previous data on Esrrb ( Festuccia et al . , 2019 ) . Gene transcription is strongly inhibited during replication and mitosis , and therefore , genes need to be reactivated in an appropriate manner following these processes . We explored two available datasets to study a potential link between CTCF binding and gene reactivation after mitosis ( Teves et al . , 2018 ) and replication ( Stewart-Morgan et al . , 2019 ) . For mitosis , we focused on chromatin associated RNAs , enriched in nascent pre-mRNA , in mitotic ES cells and upon release in interphase measured by RNA-seq ( Teves et al . , 2018 ) . We used k-means clustering to find 5 groups of genes displaying different reactivation dynamics , measured as the rate of recovery to the level of pre-mRNA detected in bulk asynchronous cells ( Figure 7A ) . Next , we computed the Fisher Exact enrichment of bookmarked and lost CTCF peaks at increasing distances from the promoters of genes in the five clusters ( Figure 7B ) . We observed that the two clusters displaying slow reactivation dynamics ( Clusters 1 and 2 ) did not show any enrichment for CTCF binding . Contrastingly , clusters with faster dynamics ( Clusters 3 , 4 and 5 ) showed significant CTCF enrichment . While the enrichment at Cluster three was similarly low and distal for both bookmarked and lost CTCF sites , at Clusters 4 and 5 more proximal ( <100 kb ) CTCF bookmarked sites were specifically and prominently enriched . Cluster 5 , which exhibits the fastest reactivation dynamics , was found to be particularly enriched in CTCF bookmarked sites compared to regions losing CTCF binding in mitosis . We conclude , therefore , that CTCF binding is associated with faster reactivation dynamics after mitosis , with CTCF bookmarking displaying the most robust association with gene reactivation . We next explored repli-ATAC data generated in ES cells ( Stewart-Morgan et al . , 2019 ) , in an identical manner . We focused on accessibility measurements around Transcription Start Sites ( TSSs ) of ES cell expressed genes , which reflect binding of the pre-initiation complex of transcription . We found four clusters displaying different reactivation dynamics ( Clusters 1 to 4; Figure 7C ) and a further cluster showing minor changes in TSS accessibility after replication ( Cluster 5; Figure 7C ) . Notably , only Cluster 4 , which displays rapid reacquisition of accessibility , and Cluster 5 , where TSSs remain accessible immediately after replication , show robust enrichments ( Figure 7D ) . Therefore , both after replication and mitosis , the presence of CTCF binding sites correlates with the speed and efficiency of gene reactivation .
Here , we have studied the functional relationships between TF binding and nucleosome positioning during replication and mitosis , using CTCF as a paradigmatic example . While the role of CTCF in nucleosome positioning has already been suggested using siRNA knock-down ( Wiechens et al . , 2016 ) , our work provides definitive evidence to conclude that CTCF is continuously required to maintain NDRs/NOAs at its binding sites . First , we describe a predictive relationship between the number and quality of CTCF binding motifs , the level of CTCF occupancy , and the degree of nucleosome positioning . Second , we observe a loss of nucleosome order 2 hr only upon CTCF depletion using the auxin degradation system . Given previous results , it is likely that CTCF positions nucleosomes by acting as a barrier to the nucleosome sliding activity of the chromatin remodeler SNF2H ( Barisic et al . , 2019; Wiechens et al . , 2016 ) . Furthermore , we show that the site-specific loss of CTCF binding in mitosis , which occurs at regions with poor binding motifs and reduced/absent Cohesin recruitment in interphase , is associated with a comprehensive loss of nucleosome order . Conversely , at regions preserving CTCF binding in mitosis , proper NDRs/NOAs are maintained; an observation which is underlined by the severely hypomorphic CTCF-aid which both lacks binding of CTCF and nucleosomal order at CTCF sites in mitosis . Hence , our data in interphasic and mitotic ES cells establish that CTCF plays a key role in maintaining local nucleosome order . In addition , our observations following DNA replication further indicate that CTCF not only maintains NDRs/NOAs , but also contributes to their establishment , as suggested by the rapid reinstatement of nucleosome positioning after the passage of the replication fork . This is independently supported by our analyses of post-replication chromatin accessibility ( Stewart-Morgan et al . , 2019 ) in ES cells and , notably , by previous observations linking CTCF binding to reduced DNA polymerase processivity and increased mutation rates at CTCF sites ( Smith and Whitehouse , 2012; Reijns et al . , 2015 ) . We conclude , therefore , that CTCF is a major determinant of local nucleosome order , which orchestrates the organization of NDRs/NOAs throughout the cell-cycle . The direct and near instant dependence of nucleosome positioning on CTCF binding that we have described here , establishes the molecular basis for CTCF to build local chromatin resiliency . Together with previous work ( Festuccia et al . , 2019 ) , it appears that both CTCF and Esrrb , two zinc finger TFs , confer similar properties to their respective binding sites during replication and mitosis . Whilst CTCF comprises 10 C2H2-type and 1 C2HC-type zinc fingers ( Klenova et al . , 1993 ) , Esrrb contains two C4-type zinc fingers – as is frequently observed by similar nuclear receptors ( Gearhart et al . , 2003 ) . Hence , the number and type of zinc fingers do not seem to represent the intrinsic feature underlying the post-replication and mitotic activity of CTCF and Esrrb . It nevertheless remains possible that zinc finger TFs , which rapidly scan DNA for their motifs to engage in stable binding ( Iwahara and Levy , 2013 ) , are particularly suited to maintain nucleosome positioning after replication and during mitosis . Other C2H2 zinc finger TFs have been suggested to organize nucleosomes: in Drosophila , Zelda ( McDaniel et al . , 2019 ) and Phaser ( Baldi et al . , 2018 ) ; in mammals , REST and YY1 ( Valouev et al . , 2011; Wang et al . , 2012 ) . Whether these factors also organize nucleosomes during replication and mitosis requires further study; however , current data suggests that Zelda may act following replication ( McDaniel et al . , 2019 ) but not during mitosis ( Dufourt et al . , 2018 ) . Properties in addition to their intrinsic DNA binding domains may thus be important for TFs to mediate their function during replication and mitosis . Notably , C2H2 zinc-fingers are typically phosphorylated during mitosis , preventing their site-specific interactions ( Dovat , 2002; Rizkallah et al . , 2011 ) . Given our results , it is possible that either CTCF specifically , or C2H2 zinc fingers more generally , are not mitotically inactivated in ES cells . Moreover , not all mitotic bookmarking TFs ( Festuccia et al . , 2017a ) are zinc finger proteins , as illustrated by FoxA1 ( Caravaca et al . , 2013 ) or Tbp ( Teves et al . , 2018 ) . It will therefore be important to comprehensively identify TFs whose genomic targets exhibit nucleosome resiliency after replication and during mitosis . This will not only identify rules to explain the replication and mitotic function of some TFs , but will also address whether every mitotic bookmarking TF positions nucleosomes in mitosis and after replication . Several studies have thus far assessed the mitotic bookmarking capacity of CTCF , with varying results ( Shen et al . , 2015; Sekiya et al . , 2017; Oomen et al . , 2019; Burke et al . , 2005 ) . Here , we show that CTCF binding in mitosis is cell type-specific , possibly reconciling these conflicting observations . Moreover , this indicates that the mitotic bookmarking activity of CTCF is a developmentally regulated phenomenon , which may be mediated by differential mitotic phosphorylation , as discussed above . In support of developmental regulation , recent results have shown CTCF is capable of mitotic bookmarking in an erythroblast cell line ( Zhang , 2019 ) . Confirming to which cell types CTCF bookmarking is specific to , when during development it loses this activity , and how this impacts long-range chromatin interactions in mitosis and early in the following interphase , represent clear lines of investigation for the future . Nevertheless , since Cohesin is lost during prophase ( Nishiyama et al . , 2013 ) , and is required to assemble Topological Associating Domains , CTCF bookmarking may not lead to the mitotic maintenance of these structures , as recently shown ( Zhang , 2019 ) . Similarly , it remains to be addressed whether the post-replication reestablishment of nucleosome order observed at CTCF binding sites is ES cell-specific . Indeed , we show here that nucleosomes are reorganized rapidly post-replication in ES cells at regulatory elements as compared to other cell types ( Ramachandran and Henikoff , 2016 ) . Nevertheless , our analysis indicates that different TFs exhibit drastic differences in their ability to reposition nucleosomes , with CTCF and Esrrb exhibiting a particularly compelling capacity to restructure nucleosomal arrays within minutes of the passage of the replication fork . Since ES cell self-renewal requires constant proliferation , the mechanisms that we have identified may contribute to the continuous maintenance and regulation of gene activity during successive cell divisions . The globally fast reorganization of the nucleosomes after replication , together with the particular capacity of CTCF and Esrrb to govern nucleosome positioning after replication and during mitosis , may indeed facilitate the reassembly of regulatory complexes in nascent chromatin fibers and in daughter ES cells ( Mirny , 2010 ) . While this hypothesis needs to be directly addressed , the correlation between post-replication/mitosis gene reactivation speed and CTCF binding , provides clear support . Therefore , a comprehensive investigation of TFs building local nucleosome resiliency throughout the cell-cycle , in ES cells and more generally during differentiation and development , will identify their role in the preservation of cell identity in proliferative cells . This will ultimately illuminate whether their activity bypasses the requirement for a robust epigenetic memory of active regulatory elements ( Reinberg and Vales , 2018 ) , particularly in cell types with increased plasticity such as ES cells ( Festuccia et al . , 2017b; Festuccia et al . , 2017a ) .
ES cells ( wild-type E14Tg2a and CTCF-aid derivatives ) were cultured on 0 . 1% gelatin ( SIGMA , G1890-100G ) in DMEM+GlutaMax-I ( Gibco , 31966–021 ) , 10% FCS ( Gibco 10270–098 ) , 100 μM β-mercaptoethanol ( Gibco , 31350–010 ) , 1 × MEM non-essential amino acids ( Gibco , 1140–035 ) and 10 ng/ml recombinant LIF ( MILTENYI BIOTEC , 130-099-895 ) . ES cells were passaged 1:10 every 2–3 days . NIH3T3 and C2C12 cells were cultured in DMEM + GlutaMax-I supplemented with 10% FCS . NIH3T3 and C2C12 were passaged 1:15 and 1/20 , respectively , every 3 days . To obtain mitotic ES cells ( >95% purity as assessed by DAPI staining and microscopy ) , we used a nocodazole shake-off approach , as described before ( Festuccia et al . , 2019 ) . To deplete CTCF-aid in interphase , cells were treated with 0 . 5 mM auxin ( IAA Sigma , I5148 ) for 2 hr; to deplete CTCF-aid in mitosis , the cells were first treated with nocodazole for 4 hr and then with nocodazole and IAA for 2 hr , after which they were harvested by shake-off . C2C12 cells were synchronized like ES cells except that slightly longer nocodazole treatment ( 7 hr ) was required to improve the yield . NIH3T3 cells were synchronized with a triple approach . First , 1 . 5 × 106 cells were seeded in 150 mm dishes and grown with 2 mM thymidine for 18 hr to enrich in cells arrested at the G1/S transition . After two washes with PBS , they were released for 8 hr in regular medium before being arrested again for 17 hr with 2 mM Thymidine . After the double thymidine block the cells were washed twice with PBS and released in regular medium for 7 hr . Finally , cells were incubated with nocodazole for 6 hr and mitotic cells isolated by gentle tapping of the dishes against an immobilized surface . The protocol was developed based on three existing protocols ( Ramachandran and Henikoff , 2016; Kliszczak et al . , 2011; Sirbu et al . , 2011 ) ; for the CLICK reaction we used radical-free aerobic conditions , as previously described ( Presolski et al . , 2011 ) . Pulse/Chase with EdU: 15 × 106 live cells were seeded per 150 mm dish 24 hr prior to the beginning of each experiment ( three dishes per condition ) . The cells were then incubated with pre-warmed medium containing 5 μM EdU for 2 . 5 min at 37°C . Next , the cells were either harvested and fixed ( pulse ) or , in parallel , washed with pre-warmed medium and incubated for 1 hr in the presence of 100 μM Thymidine for 1 hr at 37°C ( chase ) . These samples and the respective controls ( Figure 4—figure supplement 1 ) were checked by FACS after fixation ( 4% para-formaldehyde , 15 min at room temperature protected from light ) , permeabilization ( PBS 1% BSA/0 . 1% Saponin for 15 min at room temperature ) , incubation with 100 μl CLICK reaction Mix ( see below ) , and labeling with Streptavidin-Alexa Fluor 488 ( S32354 , Thermo Fisher; diluted 1:1300 in permeabilization solution ) for 1 hr at room temperature in the dark . Microccocal Nuclease digestion ( MNase ) : after trypsinization , 120 × 106 cells per sample were fixed with 1% formaldehyde ( Thermo , 28908; 10 min at room temperature with occasional mixing ) and quenched with 0 . 125M Glycine ( 5 min ) . Fixed cells , after being washed twice with PBS , were lysed for 10 min on a rotating wheel at 4°C in 5 mL of lysis buffer ( 50 mM HEPES pH 7 . 8/150 mM NaCl/0 . 5% v/v IGEPAL/0 . 25% v/v Triton-X/10% v/v Glycerol ) supplemented with 1 × protease inhibitor cocktail ( PIC-Roche , 04 693 116 001 ) . After centrifugation , the cells were resuspended in 4 ml Wash Buffer ( WB: 10 mM Tris-HCl pH 8/200 mM NaCl ) supplemented with 0 . 5 mM DTT and incubated for 10 min on a rotating wheel at 4°C . Nuclei were pelleted for 5 min at 3000 rpm at 4°C and resuspended in 1 . 5 ml RIPA buffer ( 10 mM Tris-HCl pH 8/140 mM NaCl/1% Triton-X/0 . 1% Sodium-Deoxycholate/0 . 1% SDS ) freshly supplemented with 1 × protease inhibitor cocktail and 1 mM CaCl2 . Samples were equally split in 3 ( 1 . 5 ml tubes ) and incubated for 10 min in a water bath at 37°C . MNase ( Thermo Scientific EN0181 ) was diluted at 80 U/μl in RIPA buffer freshly supplemented with 1 × protease inhibitor cocktail and 1 mM CaCl2 . The digestion was carried out by adding MNase at a final concentration of 6U per 106 cells and incubation for 5 min at 37°C with occasional mixing . Digestions were stopped by placing tubes on ice and by adding 2 × STOP Buffer ( 2% Triton/0 . 2% SDS/300 mM NaCl/10 mM EDTA ) . Digested chromatin was recovered by overnight rotation at 4°C followed by centrifugation at 4°C at maximum speed for 10 min . The chromatin was equally distributed in 1 . 5 ml tubes and the crosslinking reversed overnight at 65°C in the presence of 1% SDS and shaking ( 800 rpm ) . The next day the samples were incubated for 2 hr at 56°C with 5 μl Proteinase K ( GEXPRK01B5 Eurobio ) and processed through phenol/chlorophorm extraction and ethanol precipitation . DNA was resuspended in a total volume of 0 . 4 ml UltraPure Nuclease free water . RNase digestion was carried out with 5 μl RNase A ( EN0531 Thermo Scientific ) for 2 hr at 37°C . The DNA was purified and precipitated again and resuspended in 51 μl UltraPure Nuclease free water . 1 μl of DNA was further diluted and the fragment sizes were analyzed with a D1000 HS Screentape ( Agilent 5067–5584 ) on an Agilent Tapestation . CLICK reaction and Streptavidin Pull-Down: each DNA sample ( 50 μl ) was supplemented , in order , with the following buffers: 342 . 5 μl of 100 mM Potassium Phosphate Buffer ( 15433339 Fisher Scientific ) ; 50 μl of 10 mM Biotin-TEG Azide ( 762024 Sigma ) resuspended in DMSO; 7 . 5 μl of pre-mixed Cupric Sulfate Pentahydrate ( 2 . 5 μl at 20 mM; C8027 Sigma ) and THTPA ( 5 μl at 50 mM; 762342 Sigma ) ; 25 μl of 100 mM Aminoguanidine Hydrochloride ( 396494 Sigma ) ; 25 μl of 100 mM Sodium Ascorbate ( A4034 Sigma ) . After briefly vortexing , the reactions were performed for 1 . 5 hr at room temperature protected from light , and DNA was subsequently ethanol-precipitated and resuspended in 400 μl UltraPure water . 20 μl were kept as input and the rest ( 380 μl ) were used for streptavidin pull down . For this , 25 μl M-280 streptavidin magnetic beads ( 10292593 Thermo Scientific ) per reaction were washed three times ( 10 min rotation at room temperature per wash ) with 1 ml 1 × Wash and Binding buffer ( W and B ) ; ( 2 × W and B: 10 mM Tris-HCl pH 7 . 5; 1 mM EDTA; 2M NaCl; 0 . 2% Tween 20 ) and resuspended in 25 μl 1 × W and B buffer . EdU-labeled DNA was then pulled down with 380 μl 2 × W and B buffer and 25 μl washed beads ( 1 hr at room temperature on a rotating wheel ) . Beads were separated by placing the tubes on a magnetic rack and the unbound fraction was stored at −20°C . The beads were washed 3 times with 1 ml W and B buffer ( 10 min at room temperature on a rotating wheel ) . Washed beads were resuspended in 100 μl TE/1% SDS and DNA was eluted with 20 μl Proteinase K ( 1 hr at 65°C with occasional vortexing ) . After adding TE to a final volume of 200 μl , the DNA was phenol/chloroform-extracted , ethanol-precipitated and resuspended in 20 μl UltraPure water . Library preparation and sequencing: we used 10 ng of DNA from the input samples and the whole pulled-down samples to prepare libraries as previously described ( Festuccia et al . , 2019 ) . Sequencing ( PE150 ) was performed by Novogene Co . Ltd . Only samples from experiments with non-detectable levels of DNA in control samples where biotin was replaced by DMSO during the CLICK reaction were used for library preparation . For interphase , 2 . 5 × 106 fixed cells ( 1% formaldehyde for 10 min followed by 5 min with 0 . 125 M Glycine ) were resuspended in 2 ml swelling buffer ( 25 mM Hepes pH 7 . 95 , 10 mM KCl , 10 mM EDTA; freshly supplemented with 1 × protease inhibitor cocktail – 04 693 116 001 PIC-Roche , and 0 . 5% IGEPAL ) , incubated for 30 min on ice , and passed 40 times in a dounce homogenizer to recover nuclei . For mitotic cells , the homogenization step was omitted . Mitotic and interphase cells were then treated in parallel to sonicate the chromatin in 300 μl of TSE150 ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-HCl pH8 , 150 mM NaCl; freshly supplemented with 1 × PIC ) using 1 . 5 ml tubes ( Diagenode ) and a Bioruptor Pico ( Diagenode; seven cycles divided into 30 s ON–30 s OFF sub-cycles at maximum power , in circulating ice-cold water ) . After centrifugation ( 30 min , full speed , 4°C ) , the supernatant was pre-cleared for 2 hr with 50 μl of protein G Sepharose beads ( P3296-5 ML Sigma ) 50% slurry , previously blocked with BSA ( 500 μg/ml; 5931665103 Roche ) and yeast tRNA ( 1 μg/ml; AM7119 Invitrogen ) . 20 μl were set apart ( input ) before over-night immunoprecipitation at 4°C on a rotating wheel with 4 µl of anti-CTCF ( Active Motif 61311 ) or 4 µl of anti-SMC1 ( Bethyl Laboratories A300-055A ) antibodies in 500 μl of TSE150 . Protein G beads ( 25 μL 50% slurry ) were added for 4 hr rotating on-wheel at 4°C . Beads were pelleted and washed for 15 min rotating on-wheel at 4°C with 1 ml of the following buffers: three washes with TSE150 , one wash with TSE500 ( as TSE150 but 500 mM NaCl ) , one wash with Washing buffer ( 10 mM Tris-HCl pH8 , 0 . 25M LiCl , 0 . 5% NP-40 , 0 . 5% Na-deoxycholate , 1 mM EDTA ) , and two washes with TE ( 10 mM Tris-HCl pH8 , 1 mM EDTA ) . Elution was performed in 100 μl of elution buffer ( 1% SDS , 10 mM EDTA , 50 mM Tris-HCl pH 8 ) for 15 min at 65°C after vigorous vortexing . Eluates were collected after centrifugation and beads rinsed in 150 μl of TE-SDS1% . After centrifugation , the supernatant was pooled with the corresponding first eluate . For both immunoprecipitated and input chromatin , the crosslinking was reversed overnight at 65°C , followed by proteinase K treatment , phenol/chloroform extraction and ethanol precipitation . H3 ChIP-seq was performed on MNase-digested chromatin using 2 . 5 × 106 cells fixed with formaldehyde as described above . Fixed cells were resuspended in 500 µl of MNase buffer ( 50 mM Tris-HCl pH8 , 1 mM CaCl2 , 0 . 2% Triton X-100 ) supplemented with protease inhibitor cocktail ( PIC; 04 693 116 001 Roche ) . Cells were pre-incubated for 10 min at 37°C with 16U of MNase ( Expressed in Kunitz , 1 Kunitz is equivalent to 10 gel units , NEB M0247S ) added to the reaction . Cells were incubated for further 10 min at 37°C , inverting the tubes occasionally . The reaction was stopped on ice by adding 500 μl of 2 × STOP buffer ( 2% Triton X-100 , 0 . 2% SDS , 300 mM NaCl , 10 mM EDTA ) . Tubes were left rotating overnight on a wheel to allow diffusion of the digested fragments . The cell suspension was spun down and the supernatant stored at −80°C . Subsequently , 50 µl of chromatin were brought to a final volume of 500 µl with TSE150 ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-HCl pH8 , 150 mM NaCl ) , freshly supplemented with protease inhibitor cocktail ( PIC-Roche , 04 693 116 001 ) , and ChIP was performed as described above using 5 µg of anti-Histone H3 rabbit polyclonal ( Abcam; ab1791 ) . All libraries were generated as previously described ( Festuccia et al . , 2019 ) and sequenced in the BioMics facility of the Institut Pasteur . For Western Blot analysis cell pellets corresponding to 106 cells were resuspended in 100 µl Laemmli Sample Buffer ( 161–0737 BIO-RAD ) containing β-mercaptoethanol . Lysed pellets were boiled for 10 min at 95°C and centrifuged for 10 min at maximum speed at room temperature . Typically , 10 µl per sample were loaded on 4–15% Mini-PROTEAN TGX Stain-Free Gels ( 4568086 BIO-RAD ) and run in 1 × SDS Running Buffer ( 250 mM Tris/1 . 92M Glycine/1% SDS ) at 10-20mA using the Mini-PROTEAN Tetra System ( BIO-RAD ) . Proteins were transferred on nitrocellulose membranes ( Amersham Protran 10600003 ) for 1 hr ( or 3 hr for CTCF ) at 300mA using the wet transfer system ( BIO-RAD ) in 1 × Transfer Buffer ( 10 × 0 . 25M Tris/1 . 92M Glycine ) prepared with a final concentration of 20% Ethanol . Membranes were blocked in PBST ( PBS 0 . 1% Tween-20 ) 5% BSA for 1 hr at room temperature and incubated overnight at 4°C with primary antibodies diluted in PBST 5% BSA ( 1:5000 anti-H3 Abcam ab1791; 1:500 anti-CTCF Active Motif 61311 ) . Excess antibodies were washed off with PBST ( five washes , 5 min each ) and incubated for 1 hr at room temperature in secondary antibodies HRP-conjugated diluted in PBST 5% BSA ( 1:10000 anti-Rabbit IgG-HRP Thermo Fisher RB230254 ) . Membranes were washed five times , 10 min each at room temperature and incubated with PIERCE ECL2 Western Blotting Substrate ( 80196 Thermo Scientific ) 5 min in dark . After excess reagent was removed , proteins were visualized using the BIO-RAD Chemidoc MP Imaging System and processed using the Image Lab Software ( BIO-RAD ) . Ten million asynchronous or mitotic ES cells , prepared as described above , were washed in PBS1X and either fixed with Formaldehyde or directly resuspended in 200 µL of Buffer A ( 10 mM HEPES pH 7 . 9/10 mM KCl/1 . 5 mM MgCl2/0 . 34 M Sucrose/10% Glycerol/1 mM DTT ) , supplemented with complete protease inhibitors and 0 . 1% Triton X-100 . Cells were incubated for 8 min on ice . After recovering the nuclei by centrifugation ( 1300 rpm; 4°C ) , the pellet was washed in Buffer A and then lysed with 100 µL of Buffer B ( 3 mM EDTA/0 . 2 mM EGTA/1 mM DTT ) supplemented with complete protease inhibitors for 30 min on ice . After spinning down , the pellet ( chromatin fraction ) was washed in Buffer B and resuspended in 60 µL of 2x reducing Laemmli buffer supplemented with 5% Bmercaptoethanol ) . The samples were sonicated 10 cycles ( 30seg ON/30 seg off ) in a Bioruptor Pico and boiled for 35 min at 95°C . After 10 min centrifugation at max speed , the samples were processed for CTCF and H3 western-blot as described above . Immunostaining of ES , C2C12 and NIH3T3 cells: cells were plated at a density of 2 × 104 on IBIDI hitreat plates coated overnight with poly-L-ornithine ( 0 . 01%; Sigma , P4957 ) at 4°C , washed and coated 2 hr with laminin ( 10 μg/ml in PBS; Millipore , CC095 ) . Two days post-seeding , cells were washed twice with PBS , fixed for 10 min at room temperature with PBS 4% Formaldehyde ( Thermo , 28908 ) , washed twice in PBS and permeabilized with PBS/0 . 1% v/v Triton X-100 supplemented with 3% of donkey serum ( Sigma , D9663 ) for 15 min at room temperature . Incubation with primary anti-CTCF ( 1:500; Active Motif 61311 ) and anti-SMC1 ( 1:500; Bethyl Laboratories A300-055A ) antibodies were performed in PBS with 3% donkey serum . After three washes in PBS/0 . 1% Triton X-100 , Alexa Fluor 594 AffiniPure Donkey Anti-Rabbit secondary antibodies ( Jackson ImmunoResearch , 711-585-152 ) were applied for 2 hr at room temperature ( 2 μg/ml ) . Cells were washed three times in PBS/0 . 1% v/v Triton X-100 , nuclei counterstained with 4' , 6-diamidino-2- phenylindole ( DAPI; Sigma , D9542 ) , and imaged with a LSM800 Zeiss microscope using a 64 × or 40 × oil immersion objective . Paired end reads were trimmed by aligning read pairs to discover regions of reverse complementarity surrounded by Nextera sequencing adapters for ATAC-seq . Alignment and trimming was performed with the BioSequences package for Julia 0 . 6 ( Bezanson et al . , 2017 ) . Reads were aligned to mm10 genome using Bowtie 2 ( Langmead and Salzberg , 2012 ) with options ‘-k 10 -I 0 -X 1000 --no-discordant --no-mixed’ , and filtered for reads with a single alignment mean edit distance less than four between read pairs . To generate heatmaps , the two end points ( cut sites ) of fragments in the 0–100 bp range , shifted inward by +/- 4 bp as recommended ( Buenrostro et al . , 2013 ) and piled at base pair resolution . Heatmaps are visualized at a resolution of 100 sites per pixel using inferno colormap scaled to 0 . 5 maximum interphase signal . All heatmaps are metaplots are centered on CTCF maximal motifs ( Supplementary file 2 ) ; SMC1 summits ( Supplementary file 2 ) ; Esrrb motifs ( Figure 4A ) for Esrrb bookmarked regions were Esrrb dictates nucleosomal organization as previously defined ( Festuccia et al . , 2019 ) ; Oct4/Sox2 composite motifs ( Figure 4A ) at Oct4 and Sox2 interphase binding regions as defined ( Festuccia et al . , 2019 ) ; P300 binding regions centered of summits derived from the Encode portal ( ENCODE Project Consortium , 2012; Davis et al . , 2018 ) , experiment identifier: ENCSR000CCD , file: ENCFF179FJG . We further restricted P300 binding regions by ChromHMM ES enhancers ( Ernst and Kellis , 2012 ) ( https://github . com/guifengwei/ChromHMM_mESC_mm1 ) ( Pintacuda et al . , 2017 ) . We retain p300 binding regions that intersected as regions annotated as ‘Enhancer’ , ‘Strong Enhancer’ or ‘Weak/Poised Enhancer’ and no other ChromHMM categories . To asses CTCF binding at different chromatin states ( Figure 3D ) we used the same ChromHMM data and selected insulators , the enhancer set described above , active promoters and active gene bodies ( ‘Transcription Elongation’ ) . Chromatin associated RNA-seq reads were taken from GSE109964 ( Teves et al . , 2018 ) and aligned to an index comprising the mm10 genome and the ERCC spikes , using STAR ( Dobin et al . , 2013 ) as part of the RSEM-STAR pipeline ( Li and Dewey , 2011 ) with additional options '--seed 1618 --star-output-genome-bam --calc-pme --calc-ci --estimate-rspd' . To focus on pre-mRNA signal STAR genome bam files were further quantified to count intronic reads . First duplicates were removed using Picard ( https://broadinstitute . github . io/picard/ ) MarkDuplicates with options "TAGGING_POLICY=All ASO=coordinate READ_NAME_REGEX=null" . Reads arising unambiguously from spliced mRNAs or unspliced pre-mRNAs were counted using VERSE ( Zhu , 2016 ) , we generated a custom gtf file with additional with the feature field transcript entries marking the extent of each isoform in the Ensembl ( Zerbino et al . , 2018 ) 93 release with ERCC spikes added . VERSE was run in strand specific mode with options '-s 2 --ignoreDup --singleEnd --multithreadDecompress -T 12 -z 2 -t 'exon;transcript' . To normalize by ERCC spikes we took a strategy that reduces spike-in technical noise ( Owens et al . , 2016 ) , we first normalized for sequencing depth by the total exonic and transcript reads per library to a mean depth of 15 million reads , we then averaged depth normalized RNA spikes between biological replicates and calculated and applied a correction factor per biological condition . If sCj is the total depth normalized exonic counts of all spikes in condition C sample j , we calculate the average spike per condition sC- and σC=∑csc-/nsC- where n is the total number of conditions . We then correct the depth normalised transcript ( intronic ) count by the spikes , for gene i in sample Cj as g^iCj=σCgiCj . Genes were filtered for those in which all replicates in at least two conditions had a spike-corrected depth normalized transcript count of greater than 10 , and whose interphase mean spike corrected RPKM was greater than 2/15 . RPKMs were calculated by the normalizing counts by the total number of non-exonic bases per transcript model . This resulted in 13 , 233 filtered genes . To determine the groups of genes with differing reactivation behavior we used k-means clustering as offered by the Clustering package of Julia ( Bezanson et al . , 2017 ) . We normalized spike corrected mean counts of mitosis , 30 min release and 60 min release to interphase as log2 fold changes . We clustered with 2≤k≤20 , and compared cluster assignments for k to k+1 we found the Rand Index ( Rand , 1971 ) , defined as the proportion of pairs of genes assigned to the equivalent clusters over the total numbers of pairs exceeded 0 . 85 for k≥5 and we selected k=5 . To determine enrichments of CTCF bookmarked and lost peaks in proximity to these clusters , we calculated Fisher Exact test p-values for the genes of a given cluster within xbp of a CTCF peak to a background of all genes clustered within xbp of a CTCF peak , for x in [1 , 1e+6] bp . Post-replication reactivation - repli-ATAC-seq repli-ATAC-seq reads were taken from GSE128643 ( Stewart-Morgan et al . , 2019 ) . Paired end reads were trimmed by aligning read pairs to discover regions of reverse complementarity surrounded by Nextera sequencing adapters for ATAC-seq . Alignment and trimming was performed with the BioSequences package for Julia 0 . 6 ( Bezanson et al . , 2017 ) . Reads were aligned to mm10 genome using Bowtie 2 ( Langmead and Salzberg , 2012 ) with options ‘-k 10 -I 0 -X 1000 --no-discordant --no-mixed’ , and filtered for reads with a single alignment mean edit distance less than four between read pairs . For comparison to MINCE-seq we quantified the total number of cutsites from 0 to 100 bp repli-ATAC-seq fragments normalized to total 0–100 bp fragments per library , falling within +/- 100 bp of CTCF maximal motif in CTCF peaks , Esrrb maximal motif in Esrrb Peaks , Oct4/Sox2 maximal motif in Oct4/Sox2 peaks and the summit of p300 peaks . To determine groups with differing post-replication reactivation dynamics we took total cutsites of 0–100 bp repli-ATAC-seq fragments , normalized to total 0–100 bp fragments , within [−200 , 0] bp of active promoters as determined by chromHMM ES cell data ( Pintacuda et al . , 2017 ) , as used in Figure 3D . We filtered for promoters with normalized total accessibility over the [−200 , 0]bp region of 0 . 15 in all samples and >0 . 5 in steady state . We employed a strategy identical to our analysis of post-mitotic gene reactivation . We normalized , nascent ( Pulse ) , 30 min , 60 min and 120 min chase to steady state as log2 fold changes and clustered with k-means clustering . Similarly , we selected k = 5 as the first k for which the Rand Index exceeded 0 . 85 ( see previous section ) . To determine enrichments of CTCF peaks in proximity to these clusters , we calculated Fisher Exact test p-values for the genes of a given cluster within xbp of a CTCF peak to a background of all genes clustered within xbp of a CTCF peak , for x in [1 , 1e+6] bp . | A single cell contains several meters of DNA which must be tightly packaged to fit inside . Typically , the DNA is wound around proteins , like a thread around many spools , to form more compact structures called nucleosomes . Before a cell divides in two , however , it needs first to access and replicate its DNA so that each new cell can get a copy of the genetic material . The cell then needs to condense the DNA again so that the two copies can be easily separated via a process called mitosis . These two processes – DNA replication and mitosis – entail major rearrangements of the nucleosomes , which then need to be returned to their original positions . Nucleosomes are also repositioned when cells need to access the coded instructions written in genes . Molecules called transcription factors bind to targets within the DNA to make sure genes are active or inactive at the right times of a cell’s life , but many are evicted from the DNA during its replication and during cell division . Most transcription factors also require nucleosomes to be specifically organized to bind to the DNA , and it remains unclear how the factors re-engage with the DNA and how nucleosomes are managed during and after DNA replication and mitosis . Owens , Papadopoulou et al . set out to understand how nucleosomes are organized immediately after DNA is replicated and while cells divide . Experiments with mouse cells grown in the laboratory showed that certain transcription factors can rebind to their targets within minutes of replication finishing , remain bound to the DNA during cell division , and displace nucleosomes from their binding sites . Owens , Papadopoulou et al . refer to these factors as “resilient transcription factors” and identified two examples , named CTCF and Esrrb . Further experiments showed that , by maintaining the structure of nearby nucleosomes while a cell divides , these resilient transcription factors could quickly reactivate genes immediately after DNA replication and mitosis are complete . These findings show that transcription factors play a fundamental role in maintaining gene regulation from one generation of cells to the next . Further studies on this topic may eventually foster progress in research areas where cell division is paramount , such as regenerative medicine and cancer biology . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2019 | CTCF confers local nucleosome resiliency after DNA replication and during mitosis |
Recently , a small-molecule communication mechanism was discovered in a range of Bacillus-infecting bacteriophages , which these temperate phages use to inform their lysis-lysogeny decision . We present a mathematical model of the ecological and evolutionary dynamics of such viral communication and show that a communication strategy in which phages use the lytic cycle early in an outbreak ( when susceptible host cells are abundant ) but switch to the lysogenic cycle later ( when susceptible cells become scarce ) is favoured over a bet-hedging strategy in which cells are lysogenised with constant probability . However , such phage communication can evolve only if phage-bacteria populations are regularly perturbed away from their equilibrium state , so that acute outbreaks of phage infections in pools of susceptible cells continue to occur . Our model then predicts the selection of phages that switch infection strategy when half of the available susceptible cells have been infected .
For several decades now , it has been recognised that communication between individuals is not limited to multicellular organisms , but is also common among microbes . The best-known example of microbial communication is bacterial quorum sensing , a process in which bacteria secrete signalling molecules to infer the local cell density and consequently coordinate the expression of certain genes ( Nealson et al . , 1970; Miller and Bassler , 2001 ) . A wide variety of bacterial behaviours have been found to be under quorum-sensing control ( Miller and Bassler , 2001; Hense and Schuster , 2015 ) , including bioluminescence ( Nealson et al . , 1970 ) , virulence ( Antunes et al . , 2010 ) , cooperative public good production ( Diggle et al . , 2007; Darch et al . , 2012 ) , and antimicrobial toxin production ( Cornforth and Foster , 2013; Kleerebezem and Quadri , 2001 ) . Remarkably , it has recently been discovered that even some bacterial viruses ( bacteriophages or phages for short ) use signalling molecules to communicate ( Erez et al . , 2017 ) . Here , we use a mathematical model to explore the dynamics of this viral small-molecule communication system . We study under what conditions communication between phages evolves and predict which communication strategies are then selected . Bacteriophages of the SPbeta group , a genus in the order of Caudovirales of viruses that infect Bacillus bacteria , encode a small signalling peptide , named ‘arbitrium’ , which is secreted when the phages infect bacteria ( Erez et al . , 2017 ) . These phages are temperate viruses , meaning that each time a phage infects a bacterium , it makes a life-cycle decision: to enter either ( i ) the lytic cycle , inducing an active infection in which tens to thousands of new phage particles are produced and released through host-cell lysis , or ( ii ) the lysogenic cycle , inducing a latent infection in which the phage DNA is integrated in the host cell’s genome ( or episomally maintained ) and the phage remains dormant until it is reactivated . This lysis-lysogeny decision is informed by the arbitrium produced in nearby previous infections: extracellular arbitrium is taken up by cells and inhibits the phage’s lysogeny-inhibition factors , thus increasing the propenstiy towards lysogeny of subsequent infections ( Erez et al . , 2017 ) . Hence , peptide communication is used to promote lysogeny when many infections have occurred . Similar arbitrium-like systems have now been found in a range of different phages ( Stokar-Avihail et al . , 2019 ) . Notably , these phages each use a slightly different signalling peptide and do not seem to respond to the signals of other phages ( Erez et al . , 2017; Stokar-Avihail et al . , 2019 ) . The discovery of phage-encoded signalling peptides raises the question of how this viral communication system evolved . While the arbitrium system has not yet been studied theoretically , previous work has considered the evolution of lysogeny and of other phage-phage interactions . Early modelling work found that lysogeny can evolve as a survival mechanism for phages to overcome periods in which the density of susceptible cells is too low to sustain a lytic infection ( Stewart and Levin , 1984; Maslov and Sneppen , 2015 ) . In line with these model predictions , a combination of modelling and experimental work showed that selection pressures on phage virulence change over the course of an epidemic , favouring a virulent phage strain early on , when the density of susceptible cells is high , but a less virulent ( i . e . lysogenic ) phage strain later in the epidemic , when susceptible cells have become scarce ( Berngruber et al . , 2013; Gandon , 2016 ) . Other modelling work has shown that if phages , lysogenised cells , and susceptible cells coexist for long periods of time , less and less virulent phages are selected ( Mittler , 1996; Wahl et al . , 2018 ) . This happens because phage exploitation leads to a low susceptible cell density , and hence a virulent strategy in which phages rapidly lyse their host cell to release new phage particles that can then infect other cells no longer pays off ( because few cells are available to infect ) . Erez et al . , 2017 propose that the arbitrium system may have evolved to allow phages to cope with the changing environment during an epidemic , allowing the phages to exploit available susceptible bacteria through the lytic cycle when few infections have so far taken place and hence the concentration of arbitrium is low , while entering the lysogenic cycle when many infections have taken place and the arbitrium concentration has hence increased . This explanation resembles results for other forms of phage-phage interaction previously found in Escherichia coli-infecting phages ( Abedon , 2017; Abedon , 2019 ) . In the obligately lytic T-even phages , both the length of the latent period of an infection and the subsequent burst size increase if additional phages adsorb to the cell while it is infected – a process called lysis inhibition ( Hershey , 1946; Doermann , 1948; Abedon , 2019 ) . In the temperate phage λ , the propensity towards lysogeny increases with the number of co-infecting virions , called the multiplicity of infection ( MOI ) ( Kourilsky , 1973 ) . In both cases , modelling work has shown that the effect of the number of phage adsorptions on an infection can be selected as a phage adaptation to host-cell density , as it allows phages to switch from a virulent infection strategy ( i . e . a short latent period or a low lysogeny propensity ) when the phage:host-cell ratio is low to a less virulent strategy ( i . e . a longer latent period or higher lysogeny propensity ) when the phage:host-cell ratio is high ( Abedon , 1989; Abedon , 1990; Sinha et al . , 2017 ) . Here , we present a mathematical model to test if similar arguments can explain the evolution of small-molecule communication between viruses , and to explore the ecological and evolutionary dynamics of temperate phage populations that use such communication systems . We show that arbitrium communication can indeed evolve and that communicating phages consistently outcompete phages with non-communicating bet-hedging strategies . We however find that communication evolves under certain conditions only , namely if the phages regularly cause new outbreaks in substantial pools of susceptible host cells . Moreover , when communication evolves under such conditions , we predict that a communication strategy is selected in which phages use arbitrium to switch from a fully lytic to a fully lysogenic strategy when approximately half of all susceptible cells have been infected . Finally , we investigate how reliable the arbitrium signal needs to be for such communication to evolve , and show that the results are remarkably robust against variation in the density of bacteria .
Following earlier models ( e . g . Stewart and Levin , 1984; Berngruber et al . , 2013; Sinha et al . , 2017; Wahl et al . , 2018 ) , we use ordinary differential equations to describe a well-mixed system consisting of susceptible bacteria , lysogens ( i . e . lysogenically infected bacteria ) , and free phages , but extend this system to include an arbitrium-like signalling peptide ( Figure 1A ) . For simplicity , we consider phages that do not affect the growth of lysogenised host cells; susceptible bacteria and lysogens hence both grow logistically with the same growth rate r and carrying capacity K . Lysogens are spontaneously induced at a low rate α , after which they lyse and release a burst of B free phages per lysing cell . Free phage particles decay at a rate δ and adsorb to bacteria at a rate a . Adsorptions to lysogens result in the decay of the infecting phage , thus describing the well-known effect of superinfection immunity ( Hutchison and Sinsheimer , 1971; Susskind et al . , 1974; McAllister and Barrett , 1977; Kliem and Dreiseikelmann , 1989; Bondy-Denomy et al . , 2016 ) , whereas adsorptions to susceptible bacteria result in infections with success probability b . We consider the lytic cycle to be fast compared to both bacterial growth and the lysogenic cycle ( Stewart and Levin , 1984; Berngruber et al . , 2013; Sinha et al . , 2017; Wahl et al . , 2018 ) , so that a lytic infection can be modelled as immediate lysis releasing a burst of B free phages . Since the genes encoding arbitrium production are among the first genes to be expressed when a phage infects a host cell ( Erez et al . , 2017; Stokar-Avihail et al . , 2019 ) , each infection leads to an immediate increase of the arbitrium concentration A by an increment c . The lysis-lysogeny decision is effected by the current arbitrium concentration: a fraction φ ( A ) of the infections results in the production of a lysogen , while the remaining fraction ( 1-φ ( A ) ) results in a lytic infection . Arbitrium does not decay spontaneously in the model ( since it is a small peptide , spontaneous extracellular degradation is considered to be negligible ) , but it is taken up by bacteria at a rate u ( e . g . through general bacterial peptide importers such as OPP [Erez et al . , 2017] ) , and then degraded intracellularly , thus reducing the arbitrium concentration A . Consider competing phage variants i that differ in their ( arbitrium-dependent ) lysogeny propensity φi ( A ) . The population densities ( cells or phages per volume unit ) of susceptible bacteria S , phage particles Pi and corresponding lysogens Li , and the concentration of arbitrium A can then be described by: ( 1 ) dSdt=rS ( 1−N/K ) ⏟logistic growth−baS∑iPi⏟infection , ( 2 ) dLidt=rLi ( 1−N/K ) ⏟logistic growth+φi ( A ) baSPi⏟lysogenic infection−αLi⏟induction , ( 3 ) dPidt=BαLi⏟burst from induction+B ( 1−φi ( A ) ) baSPi⏟burst from lytic infection−δPi⏟phage decay−aNPi⏟adsorption , ( 4 ) dAdt=cbaS∑iPi⏟ production upon infection−uNA⏟adsorption and degradation , where N=S+∑iLi is the total density of bacteria . We study two scenarios for the lysis-lysogeny decision: ( i ) a baseline scenario in which the arbitrium concentration does not affect the lysis-lysogeny decision; each phage variant has a constant lysogeny propensity ϕi and ( ii ) a full scenario in which the arbitrium concentration does affect the lysis-lysogeny decision; each phage variant causes lytic infection when the arbitrium concentration is low , but switches to some lysogeny propensity ϕmaxi when the arbitrium concentration exceeds the phage’s response threshold θi ( Figure 1B; note that we use ϕmaxi to denote a constant characteristic of the phage and φi to denote the function describing how phage variant i’s lysogeny propensity depends on the arbitrium concentration ) . In this second scenario , phage variants with constant lysogeny propensity are still included: variants with a response threshold θi=0 cause lysogenic infections with lysogeny propensity ϕmaxi independent of the arbitrium concentration . Scenario ( ii ) is hence an extension of scenario ( i ) . On top of the ecological processes described in Equation 1–4 , the model also allows for evolution of the phages due to mutations that change the characteristics ϕ ( max ) and θ of phage variants . In Equation 1–4 terms describing these mutations were omitted for readability; they are described in detail in Appendix A1 . 1 . In short , replication of any phage variant was assumed to produce mutants with slightly different characteristics ( e . g . a slightly higher or lower lysogeny propensity ) with a small probability μ . Under scenario ( ii ) , mutations changing ϕmax and θ are implemented as independent processes . In natural settings as well as in some laboratory experiments , phages regularly cause large outbreaks in pools of susceptible cells that were previously unavailable to the phages ( e . g . when phages are spread to a new area , or when phages are serially passaged in a lab setting ) . Such outbreaks perturb the phage and cell populations away from their equilibrium . To mimic such repeated perturbations , we expose the system of Equation 1–4 to a phage serial-passaging regime ( mimicking the experimental set-up of , for example , Bull et al . , 1993; Bull et al . , 2004; Bollback and Huelsenbeck , 2007; Betts et al . , 2013; Broniewski et al . , 2020 ) . We initialise the model with a susceptible bacterial population at carrying capacity ( S=K cells per mL ) and a small phage population ( ∑iPi=106 phages per mL ) and numerically integrate Equation 1–4 for a time of T hours . Then a fraction of the phage population is taken and transferred to a new population of susceptible bacteria at carrying capacity and Equation 1–4 are again integrated for T hours . This cycle is repeated to bring about a long series of epidemics . Throughout the manuscript , a dilution factor of D=0 . 01 is used ( i . e . the passaged sample is 1% of the phage population ) . Passaging does not alter the relative frequency of the different phage variants , thus ensuring that the phage variants that were highly prevalent in the phage population at the end of an episode remain at a high relative frequency at the start of the new episode . In this set-up , only phages are passaged from one epidemic episode to the next . To assess the robustness of simulation results to changes in this protocol , a second set-up was considered in which a fraction of the full sample ( susceptible cells , lysogens , phages , and arbitrium ) was passaged . We furthermore tested how the results are affected by variation in the bacterial carrying capacity . For this , at the start of each episode a carrying capacity value was sampled from a gamma distribution with mean K . We control the level of noise through the variance of this gamma distribution . In total , the model ( Equation 1–4 ) has nine parameters ( excluding the phage characteristics ϕi , ϕmaxi , and θi , which vary between phage variants present in any given simulation ) . As far as we are aware , none of these have been estimated for phages of the SPBeta group , but many have been measured for other phages , most of which infect E . coli ( Table 1 , estimates taken from Little et al . , 1999; De Paepe and Taddei , 2006; Wang , 2006; Shao and Wang , 2008; Zong et al . , 2010; Berngruber et al . , 2013 ) . To reduce the number of parameters in our analysis , we nondimensionalised the equations to obtain five scaled parameter values ( Appendix A1 . 3 ) and used the literature estimates to derive default values for these scaled parameters ( Table 1 ) . To account for the uncertainty in these estimates , we performed parameter sweeps consisting of 500 simulations with parameter values randomly sampled from broad parameter ranges ( Table 1 ) . To ensure that low values of the parameters were well-represented , parameter values were sampled log-uniformly . Numerical integration was performed in Matlab R2017b , using the default built-in ODE-solver ode45 . Scripts are available from https://github . com/hiljedoekes/PhageCom . Next to numerical integration results , we analytically found expressions for the model equilibria and derived expressions for the evolutionarily stable strategy ( ESS ) under serial passaging in both scenarios ( excluding and including arbitrium communication ) . Detailed derivations are provided in Appendix A3 .
A common approach to analysing ODE-models such as Equation 1–4 is to characterise the model’s equilibrium states ( Stewart and Levin , 1984; Wahl et al . , 2018; Cortes et al . , 2019 ) . Such an analysis is provided in Appendix A2 . However , we will here argue that to understand the evolution of arbitrium communication , and the lysis-lysogeny decision in general , considering the equilibrium states is insufficient . Firstly , the function of the arbitrium system is to allow phages to respond to changes in the density of susceptible cells and phages as reflected in the arbitrium concentration . But when the system approaches an equilibrium state , the densities of susceptible cells and phages become constant , and so does the arbitrium concentration . Equilibrium conditions hence defeat the purpose of small-molecule communication such as the arbitrium system . Evolution of small-molecule communication must be driven by dynamical ecological processes , and hence can only be studied in populations that are regularly perturbed away from their ecological steady state . Secondly , under equilibrium conditions natural selection can act on the lysis-lysogeny decision only if infections still take place , and hence lysis-lysogeny decisions are still taken . We argue that this is unlikely . If the phage population is viable ( i . e . if the parameter values are such that the phages proliferate when introduced into a fully susceptible host population ) , the model converges to one of two qualitatively different equilibria , depending on parameter conditions ( Appendix A2 ) : either ( i ) susceptible host cells , lysogens and free phages all coexist , or ( ii ) all susceptible host cells have been infected so that only lysogens and free phages remain . The evolution of a constant lysogeny propensity in a host-phage population with a stable equilibrium of type ( i ) was recently addressed by Wahl et al . , who show that under these conditions selection always favours phage variants with high lysogeny propensity ( i . e . ϕ=1 ) ( Wahl et al . , 2018 ) . However , only a narrow sliver of parameter conditions permits a stable equilibrium of type ( i ) ( Sinha et al . , 2017; Cortes et al . , 2019 ) , and when we estimated reasonable parameter conditions based on a variety of well-studied phages , we found that they typically lead to a stable equilibrium of type ( ii ) ( Appendix A2 , parameter estimates based on Little et al . , 1999; De Paepe and Taddei , 2006; Wang , 2006; Shao and Wang , 2008; Zong et al . , 2010; Berngruber et al . , 2013 ) . This is because phage infections tend to be highly effective: their large burst size and consequent high infectivity cause temperate phages to completely deplete susceptible host cell populations , replacing them with lysogens that are immune to superinfection and hence have a strong competitive advantage over the susceptible cells ( Bossi et al . , 2003; Gama et al . , 2013 ) . Under common parameter conditions , after a short epidemic the susceptible cells population is hence depleted and no more infections take place , causing the competition between different phage variants to cease ( see Figure 2A for example dynamics ) . Then there is no long-term selection on the lysis-lysogeny decision , and studying its evolution in this state is pointless . We therefore consider a scenario in which the phage and cell populations are regularly perturbed away from equilibrium . To do so , we simulate serial-passaging experiments by periodically transferring a small fraction of the phages to a new population of susceptible host cells at carrying capacity , thus simulating cycles of repeated outbreaks ( see Materials and methods ) . To form a baseline expectation of the evolution of the lysis-lysogeny decision under the serial-passaging regime , we first considered a population of phage variants that do not engage in arbitrium communication , but do differ in their constant lysogeny propensity ϕi . Under typical parameter conditions ( default values in Table 1 ) , each passaging episode starts with an epidemic in which the susceptible cell population is depleted , followed by a period in which the bacterial population is made up of lysogens only ( Figure 2A , dynamics shown for a passaging episode length T=12 h ) . The composition of the phage and lysogen populations initially changes over subsequent passaging episodes ( Figure 2A ) , but eventually an evolutionarily steady state is reached in which one phage variant dominates the phage population ( ϕ=0 . 04; Figure 2B ) , confirming that the lysis-lysogeny decision is indeed under selection . The distribution of phage variants at evolutionarily steady state depends on the time between passages , T ( Figure 2C ) . If this time is short ( T≤5 h ) , the phage variant with ϕ=0 dominates at evolutionarily steady state . This is an intuitive result: under these conditions phages are mostly exposed to environments with a high density of susceptible cells , in which a lytic strategy is favourable . Surprisingly , however , if the time between passages is sufficiently long ( T>5 h ) , the viral population at evolutionarily steady state always centres around the same phage variant , independent of T ( ϕ=0 . 04; Figure 2C ) . This result can be explained by considering the dynamics within a passaging episode ( see Figure 2A ) : Once the susceptible cell population has collapsed , free phages no longer cause new infections and are hence ‘dead ends’ . New phage particles are then formed by reactivation of lysogens only , so that the distribution of variants among the free phages comes to reflect the relative variant frequencies in the lysogen population . Hence , when the time between passages is sufficiently long , the phage type that is most frequent in the sample that is eventually passaged is the one that is most frequent in the population of lysogens ( Figure 2—figure supplement 1 ) . Under default parameter conditions , these are the phages with a low lysogeny propensity of ϕ=0 . 04 . Note that although a single phage variant clearly dominates the population , some diversity is maintained ( Figure 2B–C ) . This is due to a mutation-selection balance: because mutants with slightly different ϕ-values continuously arise from the dominant phage variant and selection against these mutants is weak ( due to their similarity to the dominant phage variant ) , the balance between influx of mutant variants by mutation and their efflux by selection results in the long-term presence of these mutants in the population . Such a population consisting of a dominant variant and its close mutants is called a quasi-species ( Eigen , 1971 ) . Next , we assessed the robustness of the results to changes in the serial passaging protocol . In the standard protocol , only phages are passaged between episodes . If instead the passaged sample consists of the full system ( susceptible cells , lysogens , and phages ) , almost identical results are obtained ( Figure 2—figure supplement 2 ) . This is again explained by realising that , as long as the time between passages is sufficiently long , the distribution of variants in the free phages is equal to the distribution in the lysogens . Since lysogens need to be induced to contribute to a new outbreak and the induction rate α is the same for all phage variants , the contribution of passaged lysogens to the new outbreak does not alter the relative frequency of phage variants . To examine how these results depend on the model parameters , we determined which phage variant was most abundant at evolutionarily steady state for 500 randomly chosen parameter sets ( see Table 1 for parameter ranges ) , always using a long time between passages ( T=24 h ) . The selected ϕ-values for all parameter settings lie between ϕ=0 and ϕ=0 . 12 ( y-axis of Figure 2D ) . We can hence conclude that selection favours phages with low but usually non-zero lysogeny propensities . These phages employ a bet-hedging strategy: throughout the epidemic they ‘invest’ a small part of their infection events in the production of lysogens , such that they are maximally represented in the eventual lysogen population . To better understand how the lysogeny propensity ϕ that is selected depends on parameter values , we derived an analytical approximation for the evolutionarily stable strategy ( ESS ) under the serial-passaging regime if the time between passages is sufficiently long ( Appendix A3 . 1–2 ) . Because the phage dynamics during an epidemic affect the dynamics of the susceptible cells and vice versa , phage fitness is frequency dependent and the ESS is not found by a simple optimisation procedure , but by identifying the particular ϕ-value , denoted ϕ* , that maximises phage fitness given that this strategy ϕ* itself shapes the dynamics of the epidemic ( Box 1 ) . We find that the ESS can be approximated by the surprisingly simple expression ( 5 ) ϕ*=1- ( bB ) -1log ( BKP0 ) , where P0 is the density of phages at the start of a passaging episode . This approximation corresponds well with the results of the parameter sweep ( Figure 2D ) , indicating that it indeed captures the most important factors shaping the evolution of the lysogeny propensity ϕ . Equation 5 shows that the ESS depends on the initial phage density in a passaging episode , P0 , relative to the burst size B and maximal host-cell density K , and the effective burst size bB , which represents the expected number of progeny phages per phage that adsorbs to a susceptible bacterium . The ESS ϕ* decreases with the dilution factor of the phages upon passage ( i . e . with lower P0 ) . On the other hand , ϕ* increases with the effective burst size bB ( note that ( bB ) -1 decreases when ( bB ) increases ) . Both effects can be intuitively understood by considering how these factors affect the duration of the epidemic , TE . If the phage density is low at the start of a passaging episode or if the phages have a small effective burst size , it takes a while before the phage population has grown sufficiently to cause the susceptible population to collapse . Since a lytic strategy is favoured early in the epidemic , when the susceptible cell density is still high , a longer epidemic favours phages with lower values of ϕ ( see the red line in the figure in Box 1 ) . On the other hand , if the initial phage density is high or if the phages have a high effective burst size , the susceptible cell population collapses quickly , phages have a much shorter window of opportunity for lysogen production and hence phages with higher ϕ-values are favoured . Next , we included the possibility of arbitrium communication and let phage variants be characterised by two properties: their arbitrium response threshold , θi , and their lysogeny propensity when the arbitrium concentration exceeds their response threshold , ϕmaxi ( see Figure 1B ) . We then again considered the dynamics of our model under a serial-passaging regime . In Figure 3A , example dynamics are shown for three competing phage variants , all with ϕmax=1 but with different response thresholds θi . The arbitrium concentration increases over the course of the epidemic , and the phage variants switch from lytic infection to lysogen production at different times because of their different response thresholds . Note that the maximum arbitrium concentration obtained during a passaging episode is approximately A=cK ( Figure 3A ) . This is because during the epidemic the dynamics of the susceptible cell density are mostly determined by infection events and not so much by the ( slower ) bacterial growth . Since the arbitrium concentration increases by an increment c every time a susceptible cell is infected , the infection of all initial susceptible cells will result in an arbitrium concentration of A=cK ( assuming that the degradation of arbitrium is also slow and can be ignored during the growth phase of the epidemic ) . The arbitrium concentration during the early epidemic then is a direct reflection of the fraction of susceptible cells that have so far been infected . To study the evolution of arbitrium communication , we again considered the distribution of phage variants at evolutionary steady state for varying values of the time between passages , T . Similar to the results shown in Figure 2C , we find two main regimes ( Figure 3B ) : if the time between passages is short ( T<4 h , illustrated by T=2 h in Figure 3B ) , selection favours phage variants that only cause lytic infections ( ϕmax=0 ) ; if the time between passages is sufficiently long ( T≥5 h , illustrated by T=12 h in Figure 3B ) , the phage population is dominated by variants with ϕmax=1 and θ≈0 . 65cK . For 4≤T<5 h , we see a transition between these two regimes ( Figure 3—figure supplement 1 ) . If the time between passages is sufficiently long ( T>5 h ) , phage variants are hence selected that switch from a completely lytic to a completely lysogenic strategy when the arbitrium concentration exceeds a certain threshold . In the simulations of Figure 3B , phage variants could have emerged that use the bet-hedging strategy found in the absence of communication ( in phage variants with θ=0 , the lysogeny propensity is always ϕmax , independently of the arbitrium concentration ) , but this did not happen . We can hence conclude that any bet-hedging phage variants were outcompeted by variants that do use arbitrium communication . To underscore this conclusion , we simulated a competition experiment between the bet-hedging phage variant that was selected in the absence of communication and the communicating variant selected when arbitrium dynamics were included ( Figure 3C ) . The communicating phage quickly invades on a population of bet-hedging phages and takes over , confirming that communication is indeed favoured over bet-hedging . If a full sample ( susceptible cells , lysogens , phages , and arbitrium ) is passaged instead of phages only , again almost identical results are found ( Figure 3—figure supplement 2 ) . As was the case for the simulations in which arbitrium was absent , passaged lysogens do not alter the distribution of phage variants in the new outbreak . The passaged arbitrium does not significantly affect the outbreak dynamics either , because its concentration after dilution is much lower than the response threshold θ of the phage variants that are selected . To study how the evolution of phage communication depends on phage and bacterial characteristics , 500 simulations were performed with randomly sampled sets of parameter values ( Table 1 ) , using a long time between serial passages ( T=24 h ) . For each simulation , we determined which phage variant was most prevalent at evolutionary steady state . Although we varied the parameter values over several orders of magnitude , the most prevalent phage variant had a lysogeny propensity of ϕmax=1 and a response threshold of θ=0 . 5cK or θ=0 . 6cK in almost all simulations ( Figure 4A ) . Hence , over a broad range of parameter values , phages are selected that use the arbitrium system to switch from a fully lytic to a fully lysogenic strategy ( i . e . ϕmax=1 ) . This suggests that over the course of an epidemic , there is an initial phase during which the lytic strategy is a ‘better’ choice ( i . e . produces the most progeny on the long run ) , while later in the epidemic the production of lysogens is favoured and residual lytic infections that would results from a lysogeny propensity φ<1 are selected against . To better understand the intriguing consistency in θ-values found in the parameter sweep , we used a similar approach as before to analytically derive an approximation for the response threshold θ* of the evolutionarily stable strategy under the condition that the time between passages is long ( Appendix A3 . 3 ) . Again , we find a surprisingly simple expression for the ESS: ( 6 ) θ*=cK2- ( bB ) -1 . Note that the expression in Equation 6 again depends on the effective burst size bB , which is an indicator of the phage’s infectivity . The evolutionarily stable response threshold θ* declines as the effective burst size increases , converging to a value of θ*=12cK for highly infective phages ( Figure 4B , green line ) . The same result was found for simulations of the competition between phage variants with different θ-values under different effective burst sizes ( Figure 4B , blue dots ) . We see that Equation 6 provides a good prediction for the response threshold value that is selected over evolutionary time , especially for phages with high effective burst size ( Figure 4B ) . For phages with a very small effective burst size , the response threshold selected in the simulations tends to be lower than the analytical approximation . This is due to a violation of one of the simplifying assumptions made to arrive at the analytical approximation of Equation 6 , namely that during the active epidemic the dynamics of the arbitrium concentration are dominated by its production through infections and arbitrium uptake and degradation by susceptible cells can be ignored . While this is a reasonable assumption in a fast progressing epidemic , it breaks down if the dynamics of the epidemic are slow , which is exactly the case if the effective burst size bB is small . Under these conditions , the uptake and degradation of arbitrium by susceptible cells cause the arbitrium concentration to be lower than assumed in the analytical derivation . Consequently , the actual selected response thresholds ( which are essentially arbitrium concentration values ) are lower than the analytically predicted values . The result in Equation 6 can be further understood biologically . Remember that the arbitrium concentration during the epidemic varies between A=0 and A=cK , and is a reflection of the fraction of susceptible cells that have so far been infected . It makes sense that the evolutionarily stable response threshold causes phages to switch infection strategy somewhere in the middle of the epidemic: if a phage variant switches to the lysogenic strategy too early , its free phage population does not expand enough to compete with phages that switch later , but if it switches too late , the susceptible-cell density has decreased to such a degree that the phage has missed the window of opportunity for lysogen production . The ESS results from a balance between the fast production of phage progeny during the initial lytic cycles and the eventual production of sufficient lysogens . For phages with a high effective burst size , this balance occurs around the time that half of the available susceptible cells have been infected . Phages with lower effective burst size are , however , predicted to switch later , because these phages need to invest a larger portion of the available susceptible cells in the production of free phages to produce a sufficient pool of phages that can later form lysogens . Note , however , that the range of biologically reasonable effective burst sizes includes many high values ( range of x-axis in Figure 4B , Table 1 ) , that is , many real-life phages have high infectivity . Hence , for natural phages in general , we predict that if they evolve an arbitrium-like communication system , communication will be used to switch from causing mostly lytic to mostly lysogenic infections when in an outbreak approximately half of the pool of susceptible bacteria has been infected . So far , we have considered the evolution of arbitrium communication under highly predictable settings , with each outbreak taking place in a population of bacteria with the same initial density ( i . e . the bacterial carrying capacity was constant ) . As argued above , in such a set-up the arbitrium concentration provides information on the density of susceptible cells still available for infection , and the phages use this to inform their lysis-lysogeny decision . While the bacterial carrying capacity can be kept constant in lab experiments , it is far from obvious that this would be the case in natural environments . This warrants the question of how robust the results are to variation in the bacterial carrying capacity . We therefore performed simulations in which the carrying capacity varies from outbreak to outbreak . For a long time between passages ( T = 24 h ) , at the start of each passaging episode a random carrying capacity was drawn from a gamma distribution with mean K and a pre-set variance that differs from simulation to simulation . We use the coefficient of variation ( CV ) , which is defined as the standard deviation relative to the mean , to describe the level of noise . Figure 5 summarises the results of these additional simulations . Surprisingly , a communication strategy with ϕmax=1 and θ≈0 . 5cK is selected for a large range of carrying capacity noise up to CV ≤0 . 35 ( illustrated by CV = 0 . 22 in Figure 5A; see Figure 5—figure supplement 1 for full data ) . In other words , even if the carrying capacity varies with a standard deviation up to one third of its mean value , the communication strategy described in the previous section is still selected . As the coefficient of variation increases even further , the arbitrium response threshold value θ of the selected phages decreases , and so does the lysis-lysogeny propensity that is used at high arbitrium concentration ϕmax ( Figure 5B and C ) . These results make sense: if the carrying capacity strongly varies between passaging episodes , the phages regularly cause outbreaks in bacterial populations with low density . Phages with a response threshold value larger than the bacterial carrying capacity do not produce any lysogens during such an outbreak , which is disastrous for their long-term fitness . Hence , lower response thresholds are selected . The corresponding lower ϕmax values likely evolve to compensate for the earlier switch to lysogen production caused by the lower θ-values . In highly variable conditions , phages are hence selected to switch from a lytic strategy very early in the epidemic to a bet-hedging strategy later . While we find much lower response threshold values when the variation in bacterial carrying capacity is high , these threshold values do remain clearly larger than zero ( Figure 5C ) . This is true even if the carrying capacity is exponentially distributed ( CV = 1; see Figure 5—figure supplement 1 ) . Hence , even under very high variation of bacterial density a form of arbitrium communication ( in which phages use the arbitrium signal to switch from a lytic to a bet-hedging strategy ) is still favoured over completely bet-hedging strategies .
We have presented a mathematical model of a population of phages that use an arbitrium-like communication system , and used this model to explore the evolution of the lysis-lysogeny decision and arbitrium communication under a serial-passaging regime . When arbitrium communication was excluded from the model , we found that bet-hedging phages with relatively low lysogeny propensity were selected . But when arbitrium communication was allowed to evolve these bet-hedging phages were outcompeted by communicating phages . These communicating phages switch from a lytic strategy early in the epidemic to a fully lysogenic strategy when approximately half of the available susceptible cells have been infected . The serial-passaging set-up of the model is crucial for the evolution of the lysis-lysogeny decision and arbitrium communication . This has two main reasons . Firstly , it ensures that the phages are regularly exposed to susceptible cells , thus maintaining selection pressure on the lysis-lysogeny decision . Because of their high infectivity ( see Materials and methods section and De Paepe and Taddei , 2006; Wang , 2006 ) , most temperate phage outbreaks will completely deplete pools of susceptible bacteria , resulting in a bacterial population consisting of lysogens only in which the phage no longer replicates through infection ( Bossi et al . , 2003; Gama et al . , 2013 ) . The bet-hedging strategy we found in the absence of phage communication is a mechanism to deal with these ( self-inflicted ) periods of low susceptible cell availability , consistent with earlier studies ( Maslov and Sneppen , 2015; Sinha et al . , 2017 ) . Secondly , the serial-passaging set-up imposes a dynamic of repeated epidemics in which a small number of phages is introduced into a relatively large pool of susceptible cells . Such dynamics are necessary for the arbitrium system to function: the arbitrium concentration provides a reliable cue for a phage’s lysis-lysogeny decision only if it is low at the beginning of an epidemic and subsequently builds up to reflect the fraction of cells that have so far been infected . Based on these considerations , we can stipulate which environments promote the evolution of small-molecule communication such as the arbitrium system . One major factor that can ensure a regular exposure to susceptible cells ( the first requirement ) is spatial structure . If phages mostly infect bacteria that are physically close to them , a global susceptible population can be maintained even though susceptible bacteria may be depleted in local environments ( Kerr et al . , 2006 ) . Indeed , spatial structure has been shown to greatly influence phage evolution , for instance by promoting the selection of less virulent strains that deplete their local host populations more slowly ( Kerr et al . , 2006; Heilmann et al . , 2010; Berngruber et al . , 2015 ) . For small-molecule communication to evolve , however , the phages would also have to undergo repeated , possibly localised , outbreak dynamics ( the second requirement ) . Such dynamics could occur in structured meta-populations of isolated bacterial populations , between which the phages spread at a limited rate . Alternatively , phages might encounter large pools of newly susceptible bacteria if they escape superinfection immunity through mutation ( Zinder , 1958; Bailone and Devoret , 1978; Scott et al . , 1978 ) . Under this scenario , however , any remaining arbitrium signal from previous infection events no longer provides accurate information about the number of susceptible cells available , since cells that were lysogenically infected are once again susceptible to infection with the new phage variant . If the escape mutation occurs after the arbitrium produced during previous epidemics has been degraded , this problem does not occur and the newly produced arbitrium does function as a reliable signal of susceptible cell density for the new phage variant . If , however , the escape mutation occurs while the arbitrium concentration is still high from previous outbreaks , the new phage variant will cause lysogenic infections while in fact the lytic cycle should be favoured . There will then be selection pressure on the new phage variant to acquire additional mutations that change its signal specificity . This might in part explain the large diversity of phage signalling peptides observed ( Erez et al . , 2017; Stokar-Avihail et al . , 2019 ) . The model presented in this paper allows us to put hypotheses about the arbitrium system to the test . For instance , it has been suggested that the arbitrium system would benefit from the production of arbitrium by lysogens , because phages thereby would be warned about the presence of neighbouring lysogens ( which are immune to superinfection; Hynes and Moineau , 2017 ) . Above we have argued , however , that under repeated epidemics , such as caused by serial passaging , selection on the lysis-lysogeny decision and arbitrium signalling is limited to the relatively short window of time in which all ( locally ) present susceptible cells become infected: afterwards no new infections occur and arbitrium therefore has no effect . During this short time window , the density of lysogens is still low , and any arbitrium produced by lysogens contributes little to the information already conveyed by arbitrium produced during infection events . Hence , our model predicts that , under repeated epidemics that completely deplete ( local ) pools of susceptible cells , the effects of arbitrium production by lysogens are likely very minimal . Arbitrium production by lysogens can be effective only if lysogens and susceptible cells coexist over sufficiently long periods of time , such that infection events occur in the presence of lysogens . In the model , we found that such coexistence is highly unlikely . Coexistence between lysogens and susceptible cells might , however , happen under circumstances that were not included in our model , for instance through a constant inflow of susceptible cells because of cell migration , or through the loss of superinfection immunity by lysogens . Intriguingly , our model predicts that phages using small-molecule communication to coordinate their lysis-lysogeny decision would be selected to switch from a lytic to a lysogenic strategy once approximately half of the available susceptible bacteria have been lytically infected . This prediction warrants experimental testing . However , it also raises the question of how the phages would ‘know’ at what bacterial density the susceptible population has been halved . For the arbitrium signal to carry reliable information about the density of remaining susceptible cells , the initial concentration of susceptible bacteria has to be similar from outbreak to outbreak . Hence , one might expect the communication strategy to break down if the density of susceptible bacteria is variable . Surprisingly , this turned out not to be the case . We found that arbitrium-like communication could evolve even if the bacterial carrying capacity was highly variable . The characteristics of the communication system then depend on the level of noise . In highly variable environments , we predict the selection of phages that start their lysogen production earlier in an outbreak ( i . e . phages that have a low response threshold ) , and then do so in a bet-hedging way ( i . e . with a lysogeny propensity much smaller than 1 ) . In fact , few details are known so far about the response curve of phages’ lysogeny propensity to the arbitrium concentration . In the model , we chose to implement the response to arbitrium as a stepwise function . This allowed us to clearly distinguish between strategies that are favoured at low arbitrium concentration ( the lytic cycle ) and at high arbitrium concentration ( the lysogenic cycle ) . In reality , phages might respond more gradually to the arbitrium concentration . While this would alter some of our results ( e . g . pinpointing an arbitrium concentration at which the phages switch infection strategy becomes harder , if not impossible ) , we do not expect the results in general to depend on the precise shape of the response curve: phages will still use the arbitrium signal to adjust their infection strategy to whichever strategy currently yields most progeny phage on the long run . Once more data become available on the actual shape of the response curve , these can be incorporated in the model by adjusting the arbitrium response function φ ( A ) , thus producing a more specific model of the arbitrium system . Next to the arbitrium system , several other examples of temperate phages affected by small signalling molecules have recently been described . For instance , the Vibrio cholerae-infecting phage VP882 ‘eavesdrops’ on a quorum-sensing signal produced by its host bacteria , favouring lytic over lysogenic infections when the host density is high ( Silpe and Bassler , 2019 ) , while in coliphages λ and T4 and several phages infecting Enterococcus faecalis , the induction of prophages , that is , the lysogeny-lysis decision , is affected by bacterial quorum sensing signals ( Ghosh et al . , 2009; Rossmann et al . , 2015; Laganenka et al . , 2019 ) . The model could be adapted to capture these other regulation mechanisms by changing the arbitrium equation to an equation describing the production and degradation of the bacterial quorum sensing signal , and – for the second mechanism – letting the prophage reactivation rate α , rather than the lysogeny propensity ϕ , depend on the signal concentration . Similar analyses to the ones in this paper would then allow us to study under what conditions phage eavesdropping on bacterial quorum sensing can and cannot evolve . Mathematical and computational modelling can thus help to better understand the ecology and evolution of these fascinating regulation mechanisms as well . | Bacteriophages , or phages for short , are viruses that need to infect bacteria to multiply . Once inside a cell , phages follow one of two strategies . They either start to replicate quickly , killing the host in the process; or they lay dormant , their genetic material slowly duplicating as the bacterium divides . These two strategies are respectively known as a ‘lytic’ or a ‘lysogenic’ infection . In 2017 , scientists discovered that , during infection , some phages produce a signalling molecule that influences the strategy other phages will use . Generally , a high concentration of the signal triggers lysogenic infection , while a low level prompts the lytic type . However , it is still unclear what advantages this communication system brings to the viruses , and how it has evolved . Here , Doekes et al . used a mathematical model to explore how communication changes as phages infect a population of bacteria , rigorously testing earlier theories . The simulations showed that early in an outbreak , when only a few cells have yet been infected , the signalling molecule levels are low: lytic infections are therefore triggered and the phages quickly multiply , killing their hosts in the process . This is an advantageous strategy since many bacteria are available for the viruses to prey on . Later on , as more phages are being produced and available bacteria become few and far between , the levels of the signalling molecule increase . The viruses then switch to lysogenic infections , which allows them to survive dormant , inside their host . Doekes et al . also discovered that this communication system only evolves if phages regularly cause large outbreaks in new , uninfected bacterial populations . From there , the model was able to predict that phages switch from lytic to lysogenic infections when about half the available bacteria have been infected . As antibiotic resistance rises around the globe , phages are increasingly considered as a new way to fight off harmful bacteria . Deciphering the way these viruses communicate could help to understand how they could be harnessed to control the spread of bacteria . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"evolutionary",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2021 | Repeated outbreaks drive the evolution of bacteriophage communication |
Actin filaments and microtubules create diverse cellular protrusions , but intermediate filaments , the strongest and most stable cytoskeletal elements , are not known to directly participate in the formation of protrusions . Here we show that keratin intermediate filaments directly regulate the morphogenesis of microridges , elongated protrusions arranged in elaborate maze-like patterns on the surface of mucosal epithelial cells . We found that microridges on zebrafish skin cells contained both actin and keratin filaments . Keratin filaments stabilized microridges , and overexpressing keratins lengthened them . Envoplakin and periplakin , plakin family cytolinkers that bind F-actin and keratins , localized to microridges , and were required for their morphogenesis . Strikingly , plakin protein levels directly dictate microridge length . An actin-binding domain of periplakin was required to initiate microridge morphogenesis , whereas periplakin-keratin binding was required to elongate microridges . These findings separate microridge morphogenesis into distinct steps , expand our understanding of intermediate filament functions , and identify microridges as protrusions that integrate actin and intermediate filaments .
Cytoskeletal filaments are scaffolds for membrane protrusions that create a vast diversity of cell shapes . The three major classes of cytoskeletal elements—microtubules , actin filaments , and intermediate filaments ( IFs ) —each have distinct mechanical and biochemical properties and associate with different regulatory proteins , suiting them to different functions . Actin filaments create a wide variety of well-studied protrusions , including filopodia , lamellipodia , invadopodia , and microvilli ( Blanchoin et al . , 2014; Pollard and Cooper , 2009 ) . Similarly , microtubules form cilia and flagella ( Mirvis et al . , 2018 ) . By contrast , IFs are not commonly believed to directly participate in the formation of cellular protrusions . IFs are diverse , including nuclear lamins , neurofilaments , glial fibrillary acidic proteins , vimentins , and keratins ( Etienne-Manneville , 2018; Herrmann and Aebi , 2016 ) . Although each IF type is biochemically distinct , they all share structural properties . Whereas actin filaments and microtubules lengthen by preferentially adding filaments to one end , IFs are unpolarized . Tetrameric IF subunits incorporate not only at filament ends , but also within filaments , a process called ‘intercalary exchange’ that allows IFs to replace subunits without altering filament structure ( Colakoğlu and Brown , 2009 ) . The viscoelastic properties of IFs make them the strongest of the cytoskeletal elements . IFs deform at low strains , but , whereas actin filaments and microtubules break at high strains , IFs rigidify and resist breakage ( Janmey et al . , 1991 ) . Their stability and strength together make IFs ideal for maintaining cellular integrity . Keratins are the most abundant IFs in epithelial cells . They organize and reinforce epithelial tissues by anchoring cells to one another and to the extracellular matrix at desmosomes and hemi-desmosomes ( Osmani and Labouesse , 2015 ) , and they are bundled and cross-linked during the process of keratinization to create the cornified outer layers of mammalian skin ( Eckhart et al . , 2013; Smack et al . , 1994 ) . Although keratins are not known to be directly involved in the morphogenesis of protrusions , they support microvilli as part of the terminal web at their base , where they interact with F-actin rootlets extending from microvilli , as well as myosin and other actin-binding proteins ( Hirokawa et al . , 1982 ) . Keratins , along with the IF vimentin , have also been detected in long invadopodia ( Schoumacher et al . , 2010 ) . Although keratins themselves are not required for invadopodia morphogenesis , disrupting vimentin prevents their full lengthening , suggesting that IFs may play a role in the late stages of invadopodia extension ( Schoumacher et al . , 2010 ) . Cytolinker proteins bind to multiple cytoskeletal elements to integrate them into cellular structures . For example , keratins are connected to F-actin in the terminal web of intestinal microvilli , potentially by the cytolinker plastin 1 ( Grimm-Günter et al . , 2009 ) . Another family of cytolinkers , the plakins , consist of several large , multi-domain proteins that link cytoskeletal elements to cell junctions or to one another ( Jefferson et al . , 2004; Sonnenberg and Liem , 2007 ) . The plakin family members periplakin ( Ppl ) and envoplakin ( Evpl ) , which dimerize with each other and form hetero-oligomeric complexes ( Kalinin et al . , 2004 ) , localize to desmosomes ( DiColandrea et al . , 2000 ) and are components of the cornified envelope in mammalian skin ( Ruhrberg et al . , 1997; Ruhrberg et al . , 1996 ) . ppl and evpl knockout mice are viable ( Aho et al . , 2004; Määttä et al . , 2001 ) , though skin barrier formation is delayed in evpl mutants ( Määttä et al . , 2001 ) . Evpl and Ppl have large N-terminal regions with direct actin-binding activity ( Kalinin et al . , 2005 ) , as well as domains that associate with actin-binding proteins in other plakin family members ( Jefferson et al . , 2004; Sonnenberg and Liem , 2007 ) . Plakins also have rod domains that form coiled-coils mediating dimerization ( Kalinin et al . , 2004 ) , and C-terminal domains that bind to IFs ( Karashima and Watt , 2002; Kazerounian et al . , 2002 ) . Thus , Evpl and Ppl have the potential to link F-actin with keratin filaments . In this study , we investigated the relationship between keratins , F-actin , and plakins in the morphogenesis of microridges , which are laterally elongated protrusions arranged in maze-like patterns on the apical surface of epithelial cells ( Depasquale , 2018 ) . Microridges are formed on a variety of mucosal epithelial cells , including cells that make up the outer layer of the zebrafish epidermis , called the periderm , where they are required to maintain glycans on the skin surface ( Pinto et al . , 2019 ) . Microridge protrusions are filled with F-actin but are more persistent than several better studied actin-based structures , such as lamellipodia and filopodia . Microridges are formed from the coalescence of finger-like , actin-based precursor protrusions called pegs ( van Loon et al . , 2020 ) , a process dependent on the F-actin nucleator Arp2/3 ( Lam et al . , 2015; Pinto et al . , 2019; van Loon et al . , 2020 ) and the relaxation of surface tension by cortical myosin-based contraction ( van Loon et al . , 2020 ) . Although studies of microridge morphogenesis have exclusively focused on F-actin regulation , like microvilli , microridges have keratins at their base , and ultrastructural studies have reported the occasional presence of IFs within microridges ( Pinto et al . , 2019; Schliwa , 1975; Uehara et al . , 1991 ) . Using a combination of live imaging , mutant analysis , and structure-function studies , we found that keratins are integral components of microridges , and that Evpl and Ppl control microridge stability and length by recruiting keratin cytoskeletal filaments . Thus , F-actin-keratin cytolinkers create a hybrid cytoskeletal scaffold that enables the morphogenesis of microridge protrusions .
To investigate keratin localization in the zebrafish periderm , we tagged six type I keratin proteins expressed in periderm cells ( Cokus et al . , 2019 ) with GFP or mRuby at their C-termini , using bacterial artificial chromosomes ( BACs ) . Imaging periderm cells in live zebrafish expressing these reporters revealed that all keratins localized in two distinct patterns within a cell: As expected , they formed a filamentous network filling cells; remarkably , they also formed what appeared to be thick bundles in the pattern of microridges at the apical surface ( Figure 1A , Figure 1—figure supplement 1 , Video 1 ) . Tagging an allele of one of these keratins , keratin 17 ( Krt17 ) , with CRISPR-facilitated homologous recombination , confirmed that the endogenously expressed protein localized in these two patterns ( Figure 1B ) . To observe keratin localization at higher resolution , we used super-resolution structured illumination microscopy ( SIM ) to image cells expressing the Krt17-GFP BAC reporter . At an early stage of microridge morphogenesis ( 19 hr post-fertilization , hpf ) , when periderm surfaces are dominated by pegs ( van Loon et al . , 2020 ) , Krt17 formed the filamentous pattern in cells but did not localize within pegs . At a later stage ( 48hpf ) , when mature microridges have formed , Krt17 invaded microridges , where it appeared to form filaments alongside F-actin ( Figure 1C ) . Scoring the presence of keratin in protrusions of different lengths confirmed that the smallest protrusions , pegs , are largely devoid of keratin , and that keratin localization to microridges increased as they lengthened ( Figure 1D–E ) . Intriguingly , the average microridge length per cell , as measured by imaging the actin reporter Lifeact-mRuby , was slightly longer in Krt17-GFP over-expressing cells than in wild-type ( WT ) cells ( Figure 1F ) . These observations confirm that keratins are components of mature microridges and suggest that they may play a role at later stages of microridge morphogenesis . Since IFs are the strongest and most stable of the cytoskeletal filaments , we wondered if they might contribute to microridge stability . To destabilize F-actin in microridges , we treated animals for 30 min with the Arp2/3 inhibitor CK666 , which causes the F-actin in microridges to disassemble and redistribute back into peg-like structures ( Figure 2A and D; Lam et al . , 2015; Pinto et al . , 2019; van Loon et al . , 2020 ) . Strikingly , despite the loss of the F-actin microridge pattern , the apical Krt17-GFP microridge pattern was retained ( Figure 2A; Figure 2E ) . Labeling cells with a membrane reporter ( GFP-PH-PLC ) revealed that overexpressing the Krt17-GFP BAC reporter preserved the protrusive membrane topology upon F-actin disruption ( Figure 2B; Figure 2C ) . These results suggest that keratins may play roles in stabilizing and/or elongating microridges at later stages of morphogenesis . If F-actin and keratins both contribute to microridge morphogenesis , we speculated that they may interact via linker proteins . By examining larval periderm cell transcriptomes ( Cokus et al . , 2019 ) , we identified two potential cytolinker proteins , Evpl and Ppl , which are highly expressed and enriched in periderm cells , relative to other epithelial cells . Evpl and Ppl , members of the plakin protein family ( Jefferson et al . , 2004 ) , contribute to keratinization of the mammalian skin ( Ruhrberg et al . , 1997; Ruhrberg et al . , 1996 ) , heterodimerize with each other ( Kalinin et al . , 2004 ) , and can bind both F-actin ( Kalinin et al . , 2005 ) and keratins ( Karashima and Watt , 2002; Kazerounian et al . , 2002 ) , thus potentially linking the two types of cytoskeletal filaments in microridges . To determine the localization of Evpl and Ppl in zebrafish periderm cells , we made Ppl-GFP , Evpl-GFP , and Evpl-mRuby BAC reporter fusions and imaged them in transient transgenic animals . These reporters were expressed in periderm cells throughout the animal and , as expected from genomic analyses ( Cokus et al . , 2019; Liu et al . , 2020 ) , were apparently exclusive to periderm cells . Evpl and Ppl both localized to microridges ( Figure 3A–E ) . Similar to its behavior in mammalian cell culture ( DiColandrea et al . , 2000 ) , when expressed on its own , Evpl formed prominent aggregates , which were reduced when Ppl was co-expressed ( Figure 3A–B , Figure 3—figure supplement 1 ) , consistent with the possibility that the two plakin proteins dimerize or oligomerize . GFP-tagging the endogenous gene with CRISPR-facilitated homologous recombination verified the Ppl localization pattern ( Figure 3C ) . SIM microscopy of the co-expressed BAC reporters revealed that Evpl and Ppl were localized within microridges . In optical sections along the z-axis , Ppl and Evpl appeared to be adjacent to F-actin and keratin filaments but were almost completely overlapping with each other ( Figure 3D ) . In an x-y section , however , Evpl and Ppl formed an apparently alternating pattern , consistent with their ability to assemble into a higher order oligomeric arrangement ( Kalinin et al . , 2004; Figure 3E ) . To determine when plakins first localize to microridges , we observed Ppl localization at an earlier stage ( 19hpf ) , when periderm cell surfaces are dominated by pegs . Unlike Krt17 , at this stage , Ppl associated with pegs and , surprisingly , often appeared to form longer and more continuous structures than F-actin itself ( Figure 3F ) . To observe their localization with greater temporal precision , we imaged Ppl and F-actin reporters in time-lapse movies of cells undergoing cytokinesis . Just before periderm cell division , microridges dissolve , but then re-assemble at the end of cytokinesis ( Lam et al . , 2015 ) , thus allowing us to image the entire process with rapid , predictable kinetics . These movies revealed that Ppl formed elongated , continuous structures immediately before the coalescence of F-actin pegs into microridges , potentially as part of a template for microridge assembly ( Figure 3G , Video 2 ) . Since keratins do not localize to microridges until later in development , these observations suggest that Ppl likely plays a keratin-independent role in the initiation of microridge morphogenesis . To determine the function of Evpl and Ppl in microridge morphogenesis , we created stable zebrafish mutant lines by deleting several exons of each gene with the CRISPR/Cas9 system ( Figure 4 , Figure 4—figure supplement 1 ) . At larval stages , ppl mutants lacked microridges , forming only pegs , whereas evpl mutants formed pegs and short microridges ( Figure 4A–B , Figure 4—figure supplement 2A–B , Supplementary file 1 ) . Periderm cells in double mutants were indistinguishable from those in ppl mutants , projecting only pegs . Evpl and Ppl BAC reporters rescued microridge formation in each mutant ( Figure 4—figure supplement 3 ) , demonstrating that the BAC fusions were functional and verifying that the mutations reduced gene function . This conclusion was further supported by the observation that morpholino antisense oligonucleotides targeting evpl and ppl caused similar microridge defects , which were rescued by expressing cDNAs of each gene ( Figure 4—figure supplement 4 ) . Microridge defects persisted through adulthood in both genetic mutants ( Figure 4—figure supplement 5 ) but all single and double mutant animals were homozygous viable , fertile , and appeared morphologically normal at all stages ( Figure 4—figure supplement 6 ) . Apical cell areas were comparable between the wildtype and mutant larvae , indicating that these mutations do not cause cells to become dysmorphic or compromise apical constriction ( Figure 4—figure supplement 2C , Supplementary file 1 ) . However , apical cell areas were , on average , significantly smaller in double mutants , suggesting that Evl and Ppl may play an unrelated , redundant role in restraining apical constriction . Unexpectedly , in both evpl and ppl heterozygotes , average microridge length per cell was shorter than in WT cells but longer than in homozygous mutants ( Figure 4B , Supplementary file 1 ) . Average microridge length was shorter in trans-heterozygous mutants ( evpl+/−; ppl+/− ) than in either heterozygous mutant alone but longer than in homozygous mutants . These observations reveal that evpl and ppl mutants are semi-dominant and that the dose of plakin proteins dictates microridge length . To determine if Evpl and Ppl are not only required , but also sufficient for lengthening microridges , we co-overexpressed the Evpl-mRuby and Ppl-GFP BAC reporters in WT animals ( Figure 4C ) . Since the reporters were fluorescently tagged , we estimated the relative concentration of overexpressed plakins in each cell from its fluorescence intensity . Plotting fluorescence intensity versus average microridge length per cell revealed that cells expressing higher levels of plakins tended to have longer microridges ( Figure 4—figure supplement 7 ) . Indeed , grouping cells into high- and low-expressing categories demonstrated that microridges in cells with high plakin levels were substantially longer than microridges in WT cells ( Figure 4D , Figure 4—figure supplement 7B ) . Together , these results indicate that Ppl and Evpl function like a molecular rheostat for microridge length: lowering plakin expression shortens microridges whereas increasing plakin expression lengthens microridges . Given their ability to bind F-actin and keratins ( Kalinin et al . , 2005; Karashima and Watt , 2002; Kazerounian et al . , 2002 ) , we hypothesized that Evpl and Ppl serve as linkers between keratins and F-actin in microridges . Previous biochemical studies showed that their N-terminal head domains can bind to actin ( Kalinin et al . , 2005 ) , their rod domains promote dimerization ( DiColandrea et al . , 2000; Kalinin et al . , 2004 ) , and their C-terminal tail domains bind to keratins ( Karashima and Watt , 2002; Kazerounian et al . , 2002 ) . To determine the localization of these domains in zebrafish periderm cells , we GFP- or tdTomato-tagged each domain of each plakin protein and expressed them in periderm cells of WT animals . As expected , full-length plakins localized to microridges , but the isolated head and tail domains of each plakin localized throughout the cytoplasm and nucleus , suggesting that dimerization via the rod domain is required for their localization to microridges ( Figure 5A ) . The Ppl rod domain , which is required for dimerization in vitro ( Kalinin et al . , 2004 ) , localized variably to microridges . However , the Evpl rod domain did not localize to microridges ( Figure 5A ) , suggesting that it is not sufficient for dimerization , and implying that the Ppl rod domain may weakly homodimerize , as previously suggested by biochemical studies ( Kalinin et al . , 2004 ) . Indeed , whereas full-length Ppl localized to the short microridges in evpl mutants ( Figure 5—figure supplement 1A ) , full-length Evpl was cytoplasmic in ppl mutants ( Figure 5—figure supplement 1B ) , consistent with the hypothesis that Ppl , but not Evpl , can homodimerize ( or homo-oligomerize ) , and that dimerization is required for microridge localization . This observation may also explain why ppl mutants have a more severe microridge phenotype than evpl mutants ( Figure 4A–B ) . To determine where dimerized head and tail domains localize , we expressed GFP-tagged head-rod ( i . e . Δtail ) or rod-tail ( i . e . Δhead ) domain fusions . Head-rod fusions of both plakins localized to microridges more robustly than rod domains alone ( Figure 5A ) , suggesting that the dimerized head domain enhances localization to microridges , potentially through F-actin binding . Ppl , but not Evpl , rod-tail domain fusions variably localized in keratin filament-like patterns ( Figure 5A ) , suggesting that the dimerized Ppl tail domain associates with keratins . Since parts of the rod domain can contribute to F-actin binding ( Kalinin et al . , 2005 ) , we made two shorter fusions containing only part of the Ppl rod domain and the entire Ppl tail domain ( sRT1-GFP and sRT2-GFP ) . These shorter fusions were strongly localized in keratin-like patterns that included the thick microridge-associated bundles and the filamentous network throughout the cell ( Figure 5B–C , Figure 5—video 1 ) . Together these findings demonstrate that dimerized plakins have the potential to link F-actin with keratins in microridges . To determine how each plakin domain contributes to microridge morphogenesis , we attempted to rescue plakin mutants with full-length and truncated tagged versions of the plakins . Injecting genes encoding fluorescently tagged full-length Evpl-tdTomato and Ppl-GFP into evpl−/−;ppl−/− double mutant fish rescued microridge development ( Figure 6A ) . Strikingly , the average microridge length per cell correlated with fluorescence intensity ( Figure 6B ) , further illustrating that plakin expression levels determine microridge length . By contrast , truncated fusion proteins lacking the plakin head domains did not rescue microridge length in double mutants ( Figure 6A–B ) . Similarly , Ppl rod-tail fusions did not rescue Ppl single mutants ( Figure 6C–D ) , suggesting that heterodimers containing only the Evpl head domain cannot support microridge morphogenesis . To ask if plakin-keratin interactions play a role in microridge morphogenesis , we expressed tagged Evpl and Ppl head-rod domain fusions , which lack the keratin-binding tail domain , in double mutant embryos . These fusions rescued the initiation of microridge formation but did not rescue microridge length as well as full-length plakins ( Figure 6A–B ) . These results indicate that the Evpl and Ppl dimerized head domains are sufficient for the initiation of microridge morphogenesis but not for their full elongation . Similarly , expressing the Ppl head-rod fusion in ppl single mutants allowed the formation of only short , but not fully elongated , microridges ( Figure 6C–D ) . Together these experiments suggest that plakin-keratin association facilitates microridge elongation . To further test if plakin-keratin interactions promote microridge elongation , we deleted five amino acids in the Ppl tail domain required for keratin binding ( ∆DWEEI ) ( Figure 7A; Karashima and Watt , 2002 ) . Localization of Ppl ( ∆DWEEI ) rod-tail fusions ( sRT2[∆DWEEI]-GFP ) to the keratin network was severely reduced ( Figure 7B ) , confirming in vivo that this site is required for optimal Ppl-keratin binding . Similar to head-rod fusions , Ppl ( ∆DWEEI ) could not rescue microridge length in ppl mutants as well as WT Ppl . Deleting another small domain required for keratin binding ( ∆1/2Box2 ) ( Karashima and Watt , 2002 ) in Ppl yielded similar results ( Figure 7C–D ) . These findings suggest that Ppl-keratin interactions are required for full microridge elongation . IFs are the strongest cytoskeletal elements ( Janmey et al . , 1991 ) , maintain their structure even as they replace subunits ( Colakoğlu and Brown , 2009 ) , and protect cells from mechanical stress ( Leube et al . , 2017 ) . We , therefore , hypothesized that plakin-mediated recruitment of keratins into microridges stabilizes them , and that stabilization permits their elongation . To test our hypothesis , we first compared the stability of short , nascent microridges at early development ( 24hpf ) , when keratins are not found abundantly in microridges , to their stability at a later stage ( 48hpf ) , when microridges are longer and contain keratins . Over the course of 5 minute time-lapse movies , the microridge pattern changed more rapidly at 24hpf than at 48hpf , indicating that shorter microridges likely lacking keratins are less stable than longer keratin-containing microridges in older animals ( Figure 8A–B , Video 3 ) . To test if plakin-keratin binding plays a role in stabilization , we compared cells in double mutant animals expressing full-length Evpl and Ppl to those expressing the Evpl and Ppl head-rod domains ( i . e . ∆tail domain ) at 48hpf . Microridges were more dynamic in cells expressing Evpl and Ppl head-rod fusions than those expressing full-length plakins ( Figure 8C–D , Video 4 ) , similar to microridges in WT cells of younger animals . These results suggest that plakin-keratin interactions stabilize microridges to allow them to elongate .
This study has uncovered new roles for cytoskeletal filaments and cytolinkers in cellular morphogenesis , dissecting a complex morphogenetic process into discrete , mechanistically distinct steps . Of the three classes of cytoskeletal elements , only IFs were not previously thought to directly contribute to the morphogenesis of cellular protrusions . However , our findings suggest that keratin IFs play a key role in stabilizing and elongating microridge epithelial protrusions . Plakin proteins , which can bind to both actin and keratin filaments , are required and sufficient for microridge elongation , implying that this process involves the integration of two distinct cytoskeletal components . In contrast to other membrane protrusions , which grow and shrink as a single unit , microridges form from the coalescence of peg precursors ( Lam et al . , 2015; van Loon et al . , 2020 ) . Dissecting plakin functions in microridge development uncovered an additional , cryptic step of morphogenesis—the elongation of short , dynamic microridges into long , stable microridges . Thus , microridge morphogenesis proceeds through at least three molecularly separable steps: ( 1 ) peg formation , ( 2 ) peg coalescence to form nascent microridges , and ( 3 ) microridge elongation ( Figure 8E ) . Evpl and Ppl are first required for the coalescence of pegs into microridges , since periderm cells in evpl and ppl mutants retain pegs , but microridges are reduced or absent in these mutants ( Figure 4A–B ) . A role for the plakins at this step of morphogenesis was also supported by the observation that Ppl formed longer and more continuous structures than F-actin during the process of peg coalescence ( Figure 3F–G ) . This observation suggests the intriguing possibility that the plakins , which can form oligomers on their own in vitro ( Kalinin et al . , 2004 ) , guide peg coalescence . Plakins lacking their N-terminal head domains failed to rescue peg coalescence in plakin mutants ( Figure 6A–B ) , indicating that this protein region is required at this early morphogenetic step . The plakin head domains have direct actin-binding activity ( Jefferson et al . , 2004; Kalinin et al . , 2005; Sonnenberg and Liem , 2007 ) and also contain spectrin repeats and a SRC homology domain 3 ( SH3 ) domain ( Jefferson et al . , 2004; Sonnenberg and Liem , 2007 ) , which in other proteins can bind to actin interacting proteins , including the Arp2/3 activator WAVE ( Cestra et al . , 2005 ) . Peg coalescence requires not only the plakin proteins , but also the Arp2/3 actin branch nucleating complex ( Lam et al . , 2015; Pinto et al . , 2019; van Loon et al . , 2020 ) and actomyosin contraction ( van Loon et al . , 2020 ) . Thus , it is likely that Evpl and Ppl contribute to peg coalescence by interacting directly with F-actin or actin regulators , helping to incorporate the actin bundles within pegs into the larger actin network of microridges . The next step of microridge morphogenesis , the elongation of dynamic nascent microridges into mature microridges , was revealed by removing the C-terminal domains of Evpl and Ppl ( Figure 6A–B ) , and by mutating a few amino acids in Ppl required for IF binding ( Figure 7 ) . Without this keratin-binding domain , plakins could promote peg coalescence , but not full microridge elongation , indicating that Evpl and Ppl’s role in the earlier step is independent of keratin-binding . Consistent with this notion , we detected keratin in microridges only at later stages , when microridges were longer ( Figure 1C–E ) . We thus propose that Evpl and Ppl recruit keratins into microridges to instigate the subsequent elongation step of morphogenesis ( Figure 8E ) . Evpl and Ppl were both necessary and sufficient for lengthening microridges ( Figure 4 ) , and , remarkably , microridge length correlated tightly with plakin expression levels ( Figure 6 ) . The fact that overexpressing Krt17 also lengthened microridges ( Figure 1F ) lends support to the notion that plakins determine microridge length by recruiting keratins . How do keratins promote microridge elongation ? Our observation that short protrusions , which tend to lack keratin , are less stable than long microridges , which contain keratin ( Figure 8A–B ) , suggests that keratin stabilization of microridges might permit their elongation . Indeed , although plakin proteins lacking IF-binding domains enabled the formation of short , relatively unstable microridges , they never matured into long , stable structures ( Figure 8C–D ) . Perhaps most compellingly , overexpressing Krt17 preserved the three-dimensional structure of microridges upon F-actin disruption ( Figure 2A–C ) . Together these observations indicate that keratins provide microridges with stability . Notably , the IF vimentin has also been hypothesized to play a role in the growth of invadopodia at a late morphogenetic step ( Schoumacher et al . , 2010 ) . Perhaps vimentin , similar to keratin in microridges , stabilizes invadopodia to allow them to elongate . A role for IFs as microridge stabilizers is consistent with the fact that IFs are intrinsically stronger than the other cytoskeletal elements , and can preserve their form even as subunits are replaced ( Colakoğlu and Brown , 2009; Janmey et al . , 1991 ) . By contrast , actin filaments are more dynamic , consistent with their role in forming transient protrusions , like lamellipodia , filopodia , and dorsal ruffles . Microridges remodel but do not form and disassemble as rapidly as the aforementioned protrusions . In terms of stability , microridges may be more similar to stereocilia or microvilli , although each of these structures likely use different strategies to maintain their forms . The extreme stability of stereocilia reflects the fact that there is little F-actin turnover within them ( Narayanan et al . , 2015; Zhang et al . , 2012 ) , whereas microvilli maintain their morphologies despite constant actin turnover ( Loomis et al . , 2003; Meenderink et al . , 2019; Tyska and Mooseker , 2002 ) . Like microvilli , microridges constantly replace their F-actin scaffolds , since inhibiting Arp2/3 leads to disassembly of the microridge F-actin network within thirty minutes ( Lam et al . , 2015; van Loon et al . , 2020 ) . However , in contrast to microvilli , microridges require keratins to preserve their forms in the face of F-actin turnover , perhaps because their more extended morphologies and less organized F-actin networks ( Pinto et al . , 2019 ) require additional stabilization . A landmark study comparing the three classes of cytoskeletal elements speculated that the mechanical ‘differences between F-actin and vimentin are optimal for the formation of a composite material with a range of properties that cannot be achieved by a single polymer network’ ( Janmey et al . , 1991 ) . By combining F-actin with IFs , microridges may achieve an optimal balance between plasticity and stability .
Zebrafish ( Danio rerio ) were grown at 28 . 5°C on a 14 hr/10 hr light/dark cycle . Embryos were raised at 28 . 5°C in embryo water ( 0 . 3 g/L lnstant Ocean salt , 0 . 1% methylene blue ) . For live confocal imaging , pigmentation was blocked by treating embryos with phenylthiourea ( PTU ) at 24hpf . All experimental procedures were approved by the Chancellor’s Animal Research Care Committee at UCLA . To generate guide RNAs ( gRNAs ) we used the ‘short oligo method to generate gRNA’ , as previously described ( Talbot and Amacher , 2014 ) . Two Cas9 binding sites were selected for each gene . Evpl targeting sequences were located in exons 3 and 7 , ppl targeting sequences were in exons 2 and 5 . The DNA template was PCR amplified to make a product containing a T7 RNA polymerase promoter , the gene targeting sequence , and a gRNA scaffold sequence . PCR products were used as a template for RNA synthesis with T7 RNA polymerase ( New England Biolabs ) and purified ( QIAGEN RNA purification kit ) to generate gRNAs . Injection mixes contained Cas9 protein ( 1 mg/mL; IDT ) , gRNAs ( 0 . 5–1 ng/µL ) , and 300 mM KCI . Injection mixes were incubated on ice 15 min before injection . Embryos were injected at the 1-cell stage with 2–5 nL of injection mix . To identify germline founders , F0 fish were crossed to wild-type fish and 48hpf embryos were collected for PCR genotyping . Founder progeny were raised to adulthood to establish stable mutant lines . RNA was extracted with TRIzol ( Thermo FIsher Scientific ) from five scales per adult for each genotype . RNA was purified ( QIAGEN RNA purification kit ) , reverse transcribed with Superscript three using an Oligo ( dT ) 20 primer to make cDNA ( Invitrogen ) , and PCR amplified . Primers for RT-PCR are listed in the Key Resources Table . About 4 ng of morpholino antisense oligonucleotides ( Gene Tools ) , targeting splice sites in evpl or ppl , were injected into 1-cell stage embryos . A GFP-tagged cDNA for evpl and a tdTomato-tagged cDNA for ppl were injected at the 1-cell stage for rescue experiments . evpl MO sequence: 5’-GTGTCTTTAGTGTCACTCACTCATT-3’; ppl MO sequence: 5’-TCTGAGGTGAAACAACAGCGAGTTT-3’ ( also listed in the Key Resources Table ) . Tg ( Krt5:Lifeact-GFP ) LA226 and Tg ( Krt5:Lifeact-mRuby ) LA227 lines were previously described ( van Loon et al . , 2020 ) . To create translational fusion transgenes , GFP or mRuby reporter gene cassettes were recombined into a site directly preceding the stop codon of target genes in bacterial artificial chromosomes ( BACs ) , as previously described ( Cokus et al . , 2019; Suster et al . , 2011 ) . BAC identifiers are listed in the Key Resources Table . Plasmids were constructed using the Gateway-based Tol2kit ( Kwan et al . , 2007 ) . Primer sequences are listed in the Key Resources Table . The following plasmids were previously described: p5E-Krt5 ( Rasmussen et al . , 2015 ) , pME-EGFPpA , p3E-EGFPpA , p3E-tdTomato , and pDestTol2pA2 ( Kwan et al . , 2007 ) , Krt5-Lifeact-GFP and Krt5-Lifeact-mRuby ( van Loon et al . , 2020 ) . Krt5:Ppl ( ∆DWEEI ) and Krt5:Ppl ( ∆1/2Box2 ) transgenes were created with PCR from Krt5-Ppl-GFP or Krt5-Ppl ( sRT2 ) -GFP plasmids with SuperFi DNA Polymerase ( Invitrogen ) . PCR products were gel extracted and transformed , and selected colonies were sequenced . To create endogenously tagged Ppl and Krt17 alleles in transient transgenics , CRISPR gRNA target sites were selected 676 bp ( Ppl ) and 895 bp ( Krt17 ) downstream of the stop codon . Donor plasmids for recombination were generated using the Gateway-based Tol2kit . These plasmids contained a 5’ homology arm consisting of an ~1 Kb sequence upstream of the stop codon , GFP with a polyA termination sequence from EGFP-SV40 , and a 3’ homology arm consisting of an 800 bp ( Krt17 ) or 1 Kb ( Ppl ) sequence downstream of the gRNA target site . Primer sequences used to amplify homology arms are listed in the Key Resources Table . To linearize the donor plasmid , a gRNA with a target site on the plasmid was created ( see primers in the Key Resources Table ) ; 2–5 nL of this mix were injected into 1-cell stage embryos and fluorescence was observed with confocal microscopy at 24hpf . Injection mixes contained Cas9 mRNA ( 250 ng/µL ) , gRNAs for each gene ( 25 ng/µL ) , a gRNA for the donor plasmid ( 25 ng/µL ) , and the donor plasmid ( 25 ng/µL ) . Cas9 mRNA was synthesized as previously described ( Julien et al . , 2018 ) . Live zebrafish embryos were anaesthetized with ~0 . 2 mg/mL MS-222 ( tricaine ) in system water prior to mounting . Embryos were embedded in 1 . 2% agarose on a cover slip ( Fisher Scientific ) and a plastic ring was sealed with vacuum grease onto the coverslip to create a chamber that was filled with 0 . 2 mg/mL MS-222 solution , as previously described ( O'Brien et al . , 2009 ) . High precision cover glasses ( Marienfeld ) were used for Airyscan and Elyra microscopy . Confocal imaging was performed on an LSM 800 or LSM 880 microscope with Airyscan ( Carl Zeiss ) using a 40× oil objective ( NA = 1 . 3 ) or 60× oil objective ( NA = 1 . 4 ) . SIM imaging was performed on an Elyra microscope ( Carl Zeiss ) using a 60× oil objective ( NA = 1 . 4 ) . CK666 ( Fisher Scientific ) was dissolved in DMSO . Treatment solutions were created by adding CK666 or an equivalent volume of DMSO ( ≤2% ) to Ringer’s Solution with 0 . 2 mg/mL MS-222 . Zebrafish larvae were treated with 200 µM CK666 or DMSO just before imaging . During imaging , larvae were mounted in agarose in sealed chambers , as described above , and chambers were filled with treatment solutions . Fish were anesthetized in 0 . 016% MS-222 ( wt/vol ) dissolved in system water to remove scales . Scales were removed from the lateral trunk region of 3 month old fish with forceps . Isolated scales were fixed in 4% PFA for 30 min at room temperature on a shaker . Scales were washed twice in 0 . 01% Tween in PBS ( PBST ) , then permeabilized for 10 min at room temperature with 0 . 1% TritonX-100 in PBS . Scales were incubated for 2 hr at room temperature with AlexaFluor 488 phalloidin ( Thermo Fisher Scientific ) diluted 1:250 in PBST . Scales were washed 2 × 10 min with PBST on a shaker , mounted inside reinforcement labels ( Avery 5722 ) on a slide , and filled with PBST . The coverslips ( Fisher Scientific ) were sealed with nail polish over the reinforcement labels . Image analysis was performed with FIJI ( Schindelin et al . , 2012 ) . For display purposes , confocal z-stack images were projected ( maximum intensity projection ) and brightness and contrast were optimized . The Image Stabilizer plugin was used to adjust for cell drift . An automated pipeline implemented in FIJI was used to analyze average microridge length per cell , as previously described ( van Loon et al . , 2020 ) . To analyze fluorescence intensity , all images were acquired with identical imaging parameters . Cells were outlined by hand , and the background was subtracted using the ‘rolling ball’ radius 50 . 0 pixels method . The area outside cells was cleared before the mean fluorescence intensity was measured . To analyze overlap coefficients , cells were outlined by hand , brightness and contrast were automatically enhanced , and the area around cells was cleared . Lifeact-GFP images were blurred using the Smoothen function three times , and passed through a Laplacian morphological filter from the MorphoLibJ plugin , using the square element and a radius of 1 , as previously described ( van Loon et al . , 2020 ) . Images were thresholded using the Triangle method for Lifeact-GFP images and the Percentile method for Krt17-GFP images . Thresholded images were analyzed to obtain overlap coefficients using the JACoP FIJI plugin . Statistical analyses and graphs were generated with RStudio . Details of statistics for each experiment are listed in Figure Legends . | Cells adopt a wide array of irregular and bumpy shapes , which are scaffolded by an internal structure called the cytoskeleton . This network of filaments can deform the cell membrane the way tent poles frame a canvas . Cells contain three types of cytoskeleton elements ( actin filaments , intermediate filaments , and microtubules ) , each with unique chemical and mechanical properties . One of the main roles of the cytoskeleton is to create protrusions , a range of structures that ‘stick out’ of a cell to allow movement and interactions with the environment . Both actin filaments and microtubules help form protrusions , but the role of intermediate filaments remains unclear . Microridges are a type of protrusion found on cells covered by mucus , for instance on the surface of the eye , inside the mouth , or on fish skin . These small bumps are organised on the membrane of a cell in fingerprint-like arrangements . Scientists know that actin networks are necessary for microridges to form; yet , many structures supported by actin filaments are not stable over time , suggesting that another component of the cytoskeleton might be lending support . Intermediate filaments are the strongest , most stable type of cytoskeleton element , and they can connect to actin filaments via linker proteins . However , research has yet to show that this kind of cooperation happens in any membrane protrusion . Here , Inaba et al . used high-resolution microscopy to monitor microridge development in the skin of live fish . In particular , they focused on a type of intermediate filaments known as keratin filaments . This revealed that , inside microridges , the keratin and actin networks form alongside each other , with linker proteins called Envoplakin and Periplakin connecting the two structures together . Genetic experiments revealed that Envoplakin and Periplakin must attach to actin for microridges to start forming . However , the two proteins bind to keratin for protrusions to grow . This work therefore highlights how intermediate filaments and linker proteins contribute to the formation of these structures . Many tissues must be covered in mucus to remain moist and healthy . As microridges likely contribute to mucus retention , the findings by Inaba et al . may help to better understand how disorders linked to problems in mucus emerge . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2020 | Keratins and plakin family cytolinker proteins control the length of epithelial microridge protrusions |
Translation is a core cellular process carried out by a highly conserved macromolecular machine , the ribosome . There has been remarkable evolutionary adaptation of this machine through the addition of eukaryote-specific ribosomal proteins whose individual effects on ribosome function are largely unknown . Here we show that eukaryote-specific Asc1/RACK1 is required for efficient translation of mRNAs with short open reading frames that show greater than average translational efficiency in diverse eukaryotes . ASC1 mutants in S . cerevisiae display compromised translation of specific functional groups , including cytoplasmic and mitochondrial ribosomal proteins , and display cellular phenotypes consistent with their gene-specific translation defects . Asc1-sensitive mRNAs are preferentially associated with the translational ‘closed loop’ complex comprised of eIF4E , eIF4G , and Pab1 , and depletion of eIF4G mimics the translational defects of ASC1 mutants . Together our results reveal a role for Asc1/RACK1 in a length-dependent initiation mechanism optimized for efficient translation of genes with important housekeeping functions .
Ribosomes are universal protein-synthesizing machines that are highly conserved in their structure and function throughout all kingdoms of life . However , each domain of life has evolved unique ribosomal proteins that are added to the conserved core . The fundamental tasks of ribosomes — deciphering the genetic code and synthesizing peptide bonds — are the same in all organisms , so the functions of these ‘extra’ ribosomal proteins are intriguing , yet almost entirely unknown . Eukaryotic ribosomes contain 13 domain-specific proteins that may play roles in translation initiation , which is both more complicated and more highly regulated in eukaryotes than in prokaryotes ( Ban et al . , 2014; Sonenberg and Hinnebusch , 2009 ) . Recruitment of prokaryotic ribosomes to mRNAs requires only three initiation factors , IF1 , 2 , and 3 , and relies on base-pairing between the RNA of the small ribosomal subunit and the anti-Shine-Delgarno sequence of the mRNA ( Boelens and Gualerzi , 2002 ) . In contrast , translation initiation in eukaryotes requires at least 12 initiation factors and proceeds by a complex series of steps beginning with recognition of the mRNA 5′ cap structure , followed by unwinding of mRNA secondary structure , recruitment of the small ( 40S ) ribosomal subunit , scanning , recognition of the initiation codon , and finally , joining of the large ( 60S ) ribosomal subunit to form a functional ribosome ( Aitken and Lorsch , 2012 ) . Although the eukaryotic ribosome is generally considered to be a passive player during canonical initiation , several of its proteins have been implicated in mRNA recruitment . For example , RPL38 is required for the translation of the Hox body-patterning genes during embryonic development , allowing spatiotemporal regulation of gene expression through translational control ( Kondrashov et al . , 2011 ) . Other proteins including RPS25 , RPL40 , and RACK1 are essential for the translation of viral mRNAs that are recruited to the ribosome via alternative initiation pathways ( Cherry et al . , 2005; Landry et al . , 2009; Lee et al . , 2013; Majzoub et al . , 2014 ) . The eukaryote-specific ribosomal protein RACK1 is a WD40-repeat β-propeller protein that binds the solvent-exposed face of the 40S subunit near the mRNA exit channel , in close proximity to proteins that contact the mRNA during translation initiation ( Pisarev et al . , 2008; Sengupta et al . , 2004 ) . In addition to its function as a core ribosomal protein , in mammalian cells , RACK1 has been found in complex with several proteins involved in signal transduction including protein kinase C , Src kinase , and cAMP phosphodiesterase , among many others ( Adams et al . , 2011 ) . The location of RACK1 on the ribosome together with its interactions with signaling proteins suggests a possible role in conveying stimulus-dependent information to the translation machinery ( Nilsson et al . , 2004 ) . However , signaling pathways in yeast and human have diverged significantly compared to genes required for ribosomal function ( Stuart et al . , 2003 ) , suggesting that RACK1 might have another , more conserved function during translation . Loss of RACK1 causes widespread and pleiotropic defects in many organisms . Deletion of the RACK1 homolog in budding yeast , ASC1 , leads to slow growth , loss of invasive growth , loss of cell wall integrity , and decreased 60S subunit levels , among many described effects ( Li et al . , 2009; Melamed et al . , 2010; Valerius et al . , 2007; Yoshikawa et al . , 2011 ) . In metazoans , RACK1 is required for cell migration , neural tube closure , and control of post-synaptic excitation in the brain ( Kiely et al . , 2009; Ron et al . , 1999; Wehner et al . , 2011; Yaka et al . , 2002 ) . These cellular functions may explain why homozygous RACK1 loss-of-function mutations cause early developmental lethality in mouse and flies ( Kadrmas et al . , 2007; Volta et al . , 2013 ) . However , it is not known whether and how the effects of RACK1 on ribosome function contribute to the myriad cellular and organismal phenotypes observed in RACK1/ASC1 mutants ( Gibson , 2012 ) . Here we have examined the translational functions of Asc1/RACK1 genome-wide by ribosome footprint profiling in yeast ASC1 mutants . We show that Asc1 is required for the efficient translation of short mRNAs , including those encoding cytoplasmic and mitochondrial ribosomal proteins . This requirement is specific as deletion of other ribosomal proteins does not cause similar translation defects . Using translational reporters we demonstrate that length per se determines translational sensitivity to Asc1 , thus confirming a role for Asc1 in the translational privileging of short mRNAs , which is a dominant trend in genome-wide translational efficiency data from diverse eukaryotes . Remarkably , mRNA enrichment with proteins that mediate the formation of a ‘closed loop’ during translation — eIF4E , eIF4G , and Pab1 — is strongly biased towards short mRNAs and predicts Asc1-sensitivity , suggesting a role for Asc1 in closed-loop-dependent ribosome recruitment . Consistent with this prediction , we find that depletion of the central closed loop factor eIF4G mimics the translational effects of mutating ASC1 . Finally , we show that loss of ASC1 reduces mitochondrial translation and renders cells unable to use alternative carbon sources that require enhanced mitochondrial function , demonstrating the functional significance of translational perturbation in ASC1 mutants . Together , our results reveal a role for Asc1 in the enhanced translation of short mRNAs and establish a direct connection between gene-specific effects of Asc1 on translation and defects in cellular physiology . Furthermore , because mitochondria are essential for energy generation and regulation of many cellular networks , our results suggest that the pleiotropic phenotypes associated with the Asc1/RACK1 protein should be re-examined in the context of mitochondrial health .
The ASC1 locus encodes two distinct gene products — the Asc1 protein and an intronic small nucleolar RNA , snR24 . Because snR24 directs 2′-O-methylation of 25S rRNA at positions C1437 , C1449 , and C1450 , some of the reported phenotypes of ASC1 null mutants ( asc1Δ ) could be due to effects of deleting SNR24 on ribosome biogenesis or function . In addition , Asc1/RACK1 may have functions off the ribosome ( Baum et al . , 2004; Coyle et al . , 2009; Warner and McIntosh , 2009 ) . We therefore created an allelic series of yeast mutants with altered Asc1 function to enable direct comparison of cellular and translational effects of Asc1/RACK1 ( Figure 1A ) . We created protein null alleles by mutating a codon early in the ASC1 ORF to a stop codon ( asc1-M1X and asc1-E5X , where X denotes a stop codon ) , which abolished Asc1 protein expression but maintained wild type levels of SNR24 ( Figure 1B , C ) . Although bulk polysomes appeared normal in these strains , both asc1∆ and asc1-M1X showed reduced levels of free 60S subunits ( Figure 1D ) . This slight discrepancy between our results and the literature ( Li et al . , 2009 ) may stem from differences in strain backgrounds because the Sigma1278b strain used here has higher free 60S subunit levels than S288C . Restoring SNR24 expression rescued the temperature-sensitive polysome defect of the asc1∆ mutant in agreement with previous observations ( Figure 1—figure supplement 1A–D ) ( Li et al . , 2009 ) . Both asc1-M1X and asc1∆ grow slowly under standard laboratory conditions , whereas a mutant lacking only snR24 grows as well as wild type , further demonstrating the importance of the Asc1 protein ( Figure 1—figure supplement 1E ) . 10 . 7554/eLife . 11154 . 003Figure 1 . Loss of the Asc1 protein causes widespread changes in translation efficiency . ( A ) Gene model of ASC1 , showing the SNR24 snoRNA and location of protein null ( M1X and E5X ) and ribosome binding ( DE and D109Y ) mutations . ( B ) Asc1 protein levels quantified by Western blot . Pgk1 blot on the same membrane is shown as a loading control . Dilutions of the WT sample are shown on the left . Data is representative of three biological replicates . ( C ) ASC1 mRNA and SNR24 snoRNA levels quantified by qRT-PCR . Levels were normalized to ACT1 mRNA levels . Error bars represent s . d . from three technical replicates . Data is representative of three biological replicates . ( D ) Polysome profiles of the ASC1 mutants at 30˚C . The polysome/monosome ( P/M ) ratio and 60S/40S ( 60/40 ) ratio are shown with s . d . from two biological replicates . ( E ) Calculation of translation efficiency as the ratio of ribosome-protected mRNA fragments to total mRNA abundance . ( F ) Distribution of changes in TE comparing two biological replicates from WT cells ( i . e . replicate error ) or asc1-M1X or asc1∆ to its corresponding WT comparison . #1 and #2 denote biological replicate experiments . ( G ) Scatterplot of TE changes between the two ASC1 null mutants . The Pearson correlation coefficient is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 00310 . 7554/eLife . 11154 . 004Figure 1—figure supplement 1 . The M1X mutation rescues the temperature-sensitive ribosome biogenesis defect of asc1∆ . ( A–D ) Polysomes of ASC1 mutants grown at 37˚C , showing halfmer formation in mutants lacking SNR24 ( inset ) , but rescued in asc1-M1X . Halfmers result from loading of a 40S ribosomal subunit onto an mRNA without subsequent 60S joining . Because the mRNA is loaded with other ribosomes , it migrates near mRNAs loaded with a whole number of ribosomes , but shifted slightly deeper into the gradient due to the extra 40S subunit . ( E ) Growth of the ASC1 mutants on YPAD plates at 30˚C . Growth of five-fold dilutions of the culture is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 004 To define the translational function of Asc1 , we subjected the ASC1 mutants to ribosome footprint profiling and RNA-seq . Together , these techniques allow quantification of ribosome densities transcriptome-wide and can be used to infer changes in gene-specific translation activity ( Ingolia et al . , 2009 ) . Loss of the Asc1 protein caused changes in translation activity for many mRNAs as measured by translational efficiency ( TE ) — the number of ribosomal footprints normalized by the number of total RNA fragments for each mRNA ( Figure 1E–G ) . The magnitude and pervasiveness of translation changes in asc1-M1X and asc1Δ are notable given the normal appearance of bulk polysomes in ASC1 mutants ( Figure 1D ) . Thus superficially normal polysomes can conceal significant perturbations of cellular translation . Together , these results demonstrate that the lack of Asc1 substantially alters the translational landscape of yeast cells . Next we examined isogenic yeast strains that express normal levels of Asc1 protein with perturbed association to the ribosome . Asc1 is a WD-repeat protein that interacts with helices 39 and 40 of the 18S rRNA primarily through its N-terminal blade ( Sengupta et al . , 2004 ) . Directed mutation of several basic residues in this region interferes with the ribosome-binding capacity of the protein , with the strongest defect observed in the R38D K40E ( DE ) mutant ( Coyle et al . , 2009; Sengupta et al . , 2004 ) . Another Asc1 ribosome-binding mutant , D109Y , was discovered serendipitously in a forward genetic screen for mutants with defects in no-go decay , a ribosome-associated RNA quality control mechanism ( Kuroha et al . , 2010 ) . These mutant proteins were expressed at near wild type levels ( Figure 1B ) , and both mutations substantially decreased co-sedimentation of Asc1 with ribosomes in sucrose gradients , with D109Y having a markedly stronger effect ( Figure 2A ) that is consistent with previous reports ( Kuroha et al . , 2010 ) . 10 . 7554/eLife . 11154 . 005Figure 2 . Asc1 ‘ribosome-binding’ mutants retain ribosomal association in vivo . ( A ) Association of Asc1 mutant proteins with the ribosome assayed by Western blot of fractions isolated after velocity gradient sedimentation . ( B , C ) Scatterplot of TE changes between the two ASC1 null mutants and the asc1-D109Y and asc1-DE ribosome-binding mutants . The Pearson correlation coefficients are shown . ( D ) The same as ( A ) but proteins were crosslinked with formaldehyde in vivo before sample processing . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 005 The D109Y strong ribosome-binding mutant showed translational defects that , although correlated with those observed in the ASC1 null mutants ( r=0 . 43 , p=10–221 for asc1∆; r=0 . 42 , p=10–204 for asc1-M1X ) , were much smaller in magnitude ( Figure 2B ) , while the DE mutant showed almost negligible effects on translation ( Figure 2C ) . These findings suggest that either Asc1 primarily affects translation from a location off the ribosome , or that the ribosome-binding assay underestimates the extent of in vivo association of the D109Y and DE mutant proteins because ribosome-bound factors can dissociate during ultracentrifugation ( Valásek et al . , 2007 ) . To test this second possibility , we performed formaldehyde crosslinking before ultracentrifugation . In the presence of crosslinking , we observed significant co-sedimentation of the DE and D109Y proteins with polysomes ( Figure 2D ) . Crosslinking is not quantitative ( Orlando , 2000 ) ; thus this assay underestimates the extent of ribosome binding by the mutant Asc1 proteins in vivo , which are likely much closer to wild type than previously appreciated . An important implication of these findings is that phenotypic differences between ‘ribosome-binding’ alleles and ASC1 null mutants likely reflect different degrees of perturbing ribosome function and do not constitute strong evidence for ‘extra-ribosomal’ activity by Asc1/RACK1 . We attempted to generate stronger ribosome-binding-defective alleles by combining multiple mutations , but these proteins were expressed at very low levels potentially due to misfolding ( data not shown ) . Given the overall correlation between asc1-D109Y and ASC1 null alleles for translation changes transcriptome-wide , we infer that many of the translational changes in asc1-M1X and asc1Δ are likely to be caused by direct effects of Asc1 on ribosome function . To probe the mechanism by which Asc1 promotes translation of specific mRNAs , we searched for shared attributes among mRNAs with decreased TE in the asc1-M1X mutant . Motif analysis of 5′ UTRs revealed the presence of a U-rich sequence in mRNAs sensitive to loss of Asc1 ( Figure 3—figure supplement 1 ) , but not found in mRNAs resistant to loss of Asc1 . However , this motif was present in only 11% of Asc1-sensitive mRNAs and so cannot be generally required for translational enhancement by Asc1 . We next examined various physical properties of Asc1-sensitive mRNAs ( Table 1 ) . Among the tested attributes , ORF length was notably well-correlated with ∆TE in asc1-M1X ( r=0 . 27 , p=10–84 , Table 1 ) and ORFs <500 nts were the most strongly affected ( Figure 3A ) . Short ORFs are highly translated in wild type cells ( Figure 3B and [Arava et al . , 2003] ) , an effect that has been hypothesized to reflect a higher rate of translation initiation on short mRNAs for reasons that are mechanistically mysterious ( Figure 3C and Arava et al . , 2005; Shah et al . , 2013 ) . Because short ORFs are among the most highly expressed , the loss of Asc1/RACK1 significantly alters the gene expression landscape of the cell . 10 . 7554/eLife . 11154 . 006Table 1 . Properties of Asc1-sensitive mRNAs . Gene or mRNA attributes were correlated with ∆TE in the asc1-M1X mutant . The spearman correlation coefficients and p-values are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 006attributeSpearman r ( ∆TE asc1-M1X vs . attribute ) p-valuewild type protein level0 . 1031 . 49e-2wild type translation efficiency-0 . 0911 . 55e-10tRNA adaptation index ( tAI ) 0 . 0231 . 17e-15′ UTR length-0 . 0047 . 73e-13′ UTR length0 . 0799 . 62e-8ORF length0 . 2723 . 05e-845′ folding energy ( MFE ) 0 . 0303 . 97e-23′ folding energy ( MFE ) -0 . 0771 . 83e-7poly ( A ) tail length0 . 0297 . 38e-210 . 7554/eLife . 11154 . 007Figure 3 . Asc1 is required for efficient translation of short ORFs that form closed loop complexes . ( A ) Relationship between ORF length and TE changes in asc1-M1X . The values shown represent the average percent change in TE for bins of 100 genes arranged by length . The ORF lengths shown correspond to the point at which the average ORF length of the bin exceeds the indicated value . Shaded areas represent +/- 1 s . d . from the average change . The ASC1 gene is excluded from the plot . ( B ) Relationship between ORF length and translational efficiency in WT yeast cells ( data from this study ) . The Spearman correlation coefficient is shown . ( C ) Model showing the expected effect of a higher initiation rate on short mRNAs compared to long mRNAs on translation efficiency measurements . ( D ) Diagram of ORF length reporter constructs . The I27 monomer was repeated to make the octamer and each ORF was fused to a C-terminal V5 epitope tag . ( E ) Result of ORF length reporter experiment . TE is calculated as the normalized protein ( V5 tag/Pgk1 ) to mRNA ratio ( V5 mRNA/18S ) and the ∆TE ( ratio between mutant and WT ) is shown . Relative protein concentration was obtained from quantitative Western blotting and mRNA concentration from qRT-PCR . *p=0 . 002 , two-tailed Student’s t-test ( monomer vs . octamer ) . Error bars are SEM from 3 biological replicates derived from independent genetic isolates of asc1-M1X . ( F ) The structure of the mammalian 48S pre-initiation complex is shown ( Lomakin and Steitz , 2013 ) with the mRNA , RACK1 , and Rps28 , which crosslinks to the -7 and -10 positions of the mRNA relative to the AUG ( Pisarev et al . , 2008 ) , indicated . The outline of eIF3 from Hashem et al . ( 2013 ) is shown . eIF4G is placed on the left arm of eIF3 based on electron microscopy data from Siridechadilok et al . ( 2005 ) . ( G , H , I ) The relationship between closed loop complex association and ORF length ( p=10–172and 10–135 for strong closed loop and closed loop groups vs . other mRNAs , respectively ) ( G ) , ∆TE in asc1-M1X ( p=10–71and 10–42 for strong closed loop and closed loop vs . other mRNAs , respectively ) ( H ) , and ∆TE after eIF4G depletion ( p=10–70and 10–73 for strong closed loop and closed loop vs . other mRNAs , respectively . Data from Park et al . , 2011 ) ( I ) . In ( H ) and ( I ) , the dotted lines show the results after accounting for the relationship between ORF length and ∆TE using linear regression . For asc1-M1X , ORF length corrected p-values are 10–30 and 10–14 for strong closed loop and closed loop groups , respectively . For eIF4G depletion , ORF length corrected p-values are 10–17 and 10–28 for strong closed loop and closed loop groups , respectively . p-values are from the one-sided Mann-Whitney U test . Closed loop association groups are from Costello et al . ( 2015 ) . For G-I , ***p<10–18 , **p<10–9 , *p<10–3DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 00710 . 7554/eLife . 11154 . 008Figure 3—figure supplement 1 . Identification of properties of Asc1-sensitive mRNAs . Motif analysis of 5′ UTRs of mRNAs with decreased TE in asc1-M1X , defined as having a z-score ≤ -1 . ( motif present in 37/325 genes , E-value= 8 . 3e-11 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 00810 . 7554/eLife . 11154 . 009Figure 3—figure supplement 2 . Partial correlation analysis showing the relationship between wild type ORF length , transcript length , and TE . ( A ) TE vs . ORF length ( B ) TE vs . transcript length ( C ) transcript length vs . ORF length ( D ) TE vs . ORF length , partial correlation controlling for transcript length ( E ) TE vs . transcript length , partial correlation controlling for ORF length . Spearman correlation coefficients are shown between the indicated values or the residuals after linear regression . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 00910 . 7554/eLife . 11154 . 010Figure 3—figure supplement 3 . Evidence for ORF-length-dependent translational regulation . ( A ) Representative Western blots showing V5-tagged I27 monomer ( top ) or octamer ( bottom ) in WT and asc1-M1X cells . The standard curve is a two-fold dilution series of WT extract . Protein bands display variable brightness due to transfer efficiency and membrane binding differences between proteins of different molecular weights ( Bolt and Mahoney , 1997 ) . Therefore , we quantify the relative difference between mutant and WT at each protein size and cannot draw conclusions about absolute protein concentrations across the molecular weight range from Western blotting analysis . ( B–D ) Scatterplots showing the relationships between TE and ORF length in diverse eukaryotes: C . elegans , dauer stage ( Stadler and Fire , 2013 ) ( B ) , M . musculus , neutrophils ( Guo et al . , 2010 ) ( C ) , and H . sapiens , HeLa cells ( Guo et al . , 2010 ) ( D ) . ( E–G ) The effect of closed loop complex association on ORF length as in Figure 3 ( G–I ) but with all groups from Costello et al . ( 2015 ) , in which mRNAs were subdivided by hierarchical clustering into groups with similar translation factor enrichment profiles . For Figure 3 ( G–I ) , Group 3A and 3B were combined and labeled ‘strong closed loop’ . Group 4A was labeled ‘closed loop’ and all other groups were combined and labeled ‘other’ based on their association with closed-loop factors eIF4E , eIF4G , and Pab1 , and de-enrichment with 4E-binding protein ( 4E-BP ) repressors whose association with an mRNA should be mutually exclusive with the closed loop complex . Group 3A and 3B consist of mRNAs enriched for the closed loop factors and de-enriched for the 4E-BPs . Group 4A is similarly enriched for the closed loop factors but not de-enriched for the 4E-BPs . Groups in the ‘other’ category either show enrichment for the 4E-BPs or de-enrichment for the closed loop factors . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01010 . 7554/eLife . 11154 . 011Figure 3—figure supplement 4 . Relationship between ORF length and changes in mRNA polysome association after eIF4G depletion ( data from Park et al . , 2011 ) . Plot parameters are as described for Figure 3A . The genes encoding the two eIF4G isoforms are excluded from the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 011 We then sought to determine whether ORF length or transcript length is more predictive of translational efficiency . ORF length was slightly more predictive of wild type translation efficiency than transcript length , ( Figure 3—figure supplement 2A–E , partial correlation r=-0 . 09 ( p=10–8 ) vs . r=0 . 03 ( p=10–1 ) ) . For simplicity , and because transcript boundary annotations are not available for all yeast genes , we have used the ORF length metric in subsequent analyses . To test whether length per se , and not some other feature common to short mRNAs , is responsible for Asc1-sensitive translation , we generated two constructs with identical regulatory regions ( promoter , 5′ UTR , 3′ UTR ) that differed only in the length of the ORF ( Figure 3D ) . These ORF length reporters contain either one or eight repeats of the I27 domain from the human cardiac protein titin , which has been used extensively in studies of protein folding because the small globular domains fold and unfold independently of each other ( Hoffmann and Dougan , 2012 ) . This modular architecture allows the construction of proteins of different lengths that resemble linear chains and minimizes the potential for differential protein folding or stability to impact the abundance of the reporter proteins . We performed quantitative Western blotting by fluorescent detection of a common C-terminal epitope tag in combination with qRT-PCR measurements of mRNA levels to determine the translational efficiency ( protein/mRNA ) of each construct ( Figure 3—figure supplement 3A ) . Remarkably , the translational efficiency of the short ORF ( ~300 nt ) was two-fold lower in the asc1-M1X mutant compared to the long ORF ( ~2400 nt ) ( p=0 . 002 , Figure 3E ) . Together with the genome-wide trend , these reporter results demonstrate a role for Asc1 in the translational advantage of short mRNAs . Given that ORF length is strongly anti-correlated with translational efficiency in diverse eukaryotes ( Figure 3—figure supplement 3B–D , data from Guo et al . , 2010; Stadler and Fire , 2011 ) , this function of Asc1/RACK1 may be conserved . How might short ORFs be translationally privileged and sensitive to loss of Asc1 ? Asc1’s position near the mRNA exit channel places it in close proximity to translation initiation factors that interact with the 5′ end of the mRNA during initiation , including eIF3 and eIF4G ( Kouba et al . , 2012 ) ( Figure 3F , note that the structure shown is the mammalian ribosome , for which structural information regarding the orientation of eIF3 and eIF4G has been reported [Hashem et al . , 2013; Lomakin and Steitz , 2013; Siridechadilok et al . , 2005] ) . eIF4G has a well-characterized role in promoting a circularized form of the mRNA in which the 5′ and 3′ regions of the mRNA are bundled together via the interaction between the eIF4G protein , associated with the mRNA 5’ cap through the eIF4F complex , and Pab1 , an RNA-binding protein that binds the poly ( A ) tail . The mRNA in this conformation is known as the closed loop , and closed loop formation is thought to enhance translation ( Kahvejian et al . , 2001 ) . We hypothesized that mRNAs with short ORFs might form closed loop structures more efficiently than mRNAs with longer ORFs , and that Asc1 could promote the function of the closed loop in translation . According to this model , mRNAs with short ORFs should be more highly associated with the closed loop factors — eIF4E , eIF4G , and Pab1 — than other mRNAs . To test this prediction , we analyzed data quantifying the association of specific mRNAs with the closed loop factors and the eIF4E-binding proteins ( 4E-BPs ) by RNA immunoprecipitation and sequencing ( Costello et al . , 2015 ) . We grouped mRNAs into ‘closed loop’ , ‘strong closed loop’ , and ‘other’ categories based on the following enrichment profiles: ‘Strong closed loop’ mRNAs are enriched in immunoprecipitations of eIF4E , eIF4G , and Pab1 , and de-enriched in immunoprecipitations of the 4E-BPs , which should not be associated with mRNAs in closed loops because 4E-BPs and eIF4G compete for binding to eIF4E ( Haghighat et al . , 1995 ) . ‘Closed loop’ mRNAs have similar enrichment profiles to ‘strong closed loop’ mRNAs , but are not de-enriched for association with the 4E-BPs . Remarkably , we found that both ‘closed loop’ and ‘strong closed loop’ mRNAs were dramatically shorter than other mRNAs ( median ORF lengths= 489 , 774 , and 1694 nt for ‘strong closed loop’ , ‘closed loop’ , and ‘other’ mRNAs , respectively ) . This association between ORF length and closed loop association was observed regardless of whether mRNAs encoding ribosomal proteins were included in the analysis ( Figure 3G and Figure 3—figure supplement 3E ) . Thus , although ~30% of the ‘strong closed loop’ mRNAs encode ribosomal proteins , a specialized mechanism for enhancing the translation of ribosomal protein mRNAs cannot explain the ORF length bias of the ‘strong closed loop’ group . This discovery — that closed-loop-associated mRNAs are much shorter than other mRNAs — provides a plausible biochemical explanation for the preference for higher translation efficiency of mRNAs with short ORFs observed here and previously ( Arava et al . , 2003 ) . Remarkably , loss of Asc1 or eIF4G depletion ( data from Park et al . , 2011 ) similarly decreased the translation of closed-loop-associated mRNAs ( Figure 3H and I ) . Although ORF length , closed loop enrichment , and ∆TE in ASC1 and eIF4G mutants are correlated , some longer RNAs are strongly associated with the closed loop and require Asc1 for efficient translation while some short mRNAs are neither enriched with closed loop factors nor particularly dependent on Asc1 for their translation . Accounting for the global relationship between ORF length and ∆TE by linear regression showed that closed loop association has additional explanatory power for translational sensitivity to Asc1 and eIF4G: the observed reductions in translation efficiency for closed-loop-enriched mRNAs were significantly more than would be expected if ORF length alone determined their translation efficiencies ( p=10–14 and 10–30 for ‘closed loop’ and ‘strong closed loop’ groups , respectively , Figure 3H and I ) . These results suggest that Asc1 is important for closed loop formation and/or stability or for closed-loop-dependent ribosome recruitment , a process that is apparently biased towards short ORFs . What are the potential consequences of impairing translation of short mRNAs ? Using gene ontology analysis , we found that transcripts annotated to the category ‘ribosomal subunit’ had significantly decreased TE in the ASC1 null mutants ( asc1-M1X , p=10–35 , Figure 4A and Figure 4—source data 1 ) . This category is composed of short mRNAs encoding both cytoplasmic and mitochondrial ribosomal proteins ( RPs , MRPs ) , which both displayed ~20% decreased TE in ASC1 null mutants ( asc1-M1X , p=10–37 and p=10–10; asc1∆ , p=10–35 and p=10–10 , respectively , Figure 4B ) . As the median RP and MRP ORF lengths are 434 and 716 nt , respectively , their translational defects are within the range predicted by their length . Indeed , removing RP and MRP genes does not significantly alter the global relationship between ∆TE and ORF length in asc1-M1X ( r=0 . 23 , p=10–58 , Figure 4—figure supplement 1A ) indicating that all classes of genes with short ORFs have decreased TE in asc1-M1X . Although these GO categories were the clear outliers , most GO categories with short median ORF lengths also displayed decreased TE in the ASC1 null mutants , including several additional groups of genes whose protein products function in mitochondria ( Figure 4—figure supplement 1B ) . Because short ORF length is associated with specific functional categories , loss of Asc1 — and , potentially , modulation of its activity — leads to coherent changes in gene expression . 10 . 7554/eLife . 11154 . 012Figure 4 . Loss of Asc1 causes decreased translational efficiency of cytoplasmic and mitochondrial ribosomal protein mRNAs . ( A ) GO Component category enrichments for mRNAs with decreased TE in the ASC1 mutants . GO categories related to the top category ‘ribosomal subunit’ for the asc1-M1X mutant are displayed . ( B ) Violin plot showing the decreases in TE for the cytosolic ribosomal protein ( RP ) and mitochondrial ribosomal protein ( MRP ) gene sets in the ASC1 mutants . The violin shape represents a kernel density estimation and the top and bottom of the plot extend to the most extreme data point within 1 . 5x of the inner quartile range . Midlines represent the medians . ***p<10–18 , **p<10–9 , *p<10–3 . ( C ) Scatterplot showing the decrease in both the footprint ( FP ) and total RNA pool for RP and MRP mRNAs . The Pearson correlation coefficient in shown . ( D ) As in ( B ) , but with the change in ribosome association ( FP ) shown . ( E , F ) Polysome qRT-PCR showing decreased association of MRP genes with heavy polysomes . Values are normalized to an RNA spike-in control in each fraction and then set so that the sum of all fractions=1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01210 . 7554/eLife . 11154 . 013Figure 4—source data 1 . GO category enrichments for mRNAs with changes in FP , total , or TE in ASC1 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01310 . 7554/eLife . 11154 . 014Figure 4—figure supplement 1 . Exploring potential effects of mRNA functional categories , decay rates , and poly ( A ) tail length on translation efficiency measurements . ( A ) Relationship between ORF length and TE change in asc1-M1X showing that excluding mRNAs encoding RPs and MRPs from the analysis ( no RPs ) does not change the results . Plot made as in Figure 3A . ( B ) The percentage change in TE for all GO component categories containing >20 genes and arranged with equal spacing by median ORF length . The x-axis denotes the point at which the median ORF length of the group exceeds the indicated value . ( C , D ) TE values from our data correlated with steady-state mRNA half-life measurements obtained using 4-thiouracil labeling from ( Miller et al . ( 2011 ) ( C ) or ( Neymotin et al . , 2014 ) ( D ) . The Spearman correlation coefficients are shown . ( E ) Violin plot showing TE changes of the RP and MRP groups in asc1-M1X using either poly ( A ) selection or rRNA subtraction ( Ribo-Zero ) during preparation of the total RNA libraries . ***p<10–18 , **p<10–9 , *p<10–2 . ( F ) Scatterplot showing well-correlated global changes in ∆TE whether using rRNA subtraction or poly ( A ) selection . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01410 . 7554/eLife . 11154 . 015Figure 4—figure supplement 2 . Changes in mRNA levels are unlikely to explain observed translation efficiency effects in the asc1-M1X mutant . Changes in ribosome footprint levels ( ∆FP ) and total mRNA levels ( ∆total mRNA ) for mRNAs encoding cytosolic ribosomal proteins ( RPs ) and mitochondrial ribosomal proteins ( MRPs ) ( A and B ) or the ‘strong closed loop’ and ‘closed loop’ mRNAs ( C and D ) in the asc1-M1X mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 015 We noted that RP and MRP mRNAs decreased in both the total RNA pool and the ribosome-protected footprint ( FP ) pool ( Figure 4C , D ) . The additional decrease in the FP pool shows that these mRNA substrates are translationally disadvantaged in the ASC1 mutants . In support of this interpretation , qRT-PCR analysis of polysome gradient fractions demonstrated that representative MRP mRNAs associated with fewer ribosomes in asc1-M1X ( Figure 4E , F ) , which specifically indicates a defect in translation initiation . Because inhibiting translation initiation can induce mRNA degradation ( Coller and Parker , 2005; LaGrandeur and Parker , 1999; Schwartz and Parker , 1999 ) , decreased translation may account for the reduction in total mRNA levels although we cannot exclude the possibility of transcriptional effects or translation-independent effects of Asc1 on mRNA stability . We note that our translation efficiency measurements are correlated with steady-state mRNA half-life estimates using non-invasive metabolic labeling approaches ( r=0 . 43 , p=10–194 , Figure 4—figure supplement 1C , data from Miller et al . , 2011 and r=0 . 39 , p=10–168 , Figure 4—figure supplement 1D , data from Neymotin et al . , 2014 ) , consistent with the hypothesis that the decay rates of mRNAs are coupled to their translational status . The same trends of decreased TE for the RP and MRP genes were observed using an rRNA-depletion strategy instead of poly ( A ) selection ( Figure 4—figure supplement 1E , F ) , ruling out a significant effect of poly ( A ) tail length on our ∆TE calculations ( Subtelny et al . , 2014 ) . Thus , Asc1 is required for efficient translation of short ORFs , which includes most ORFs encoding cytosolic and mitochondrial ribosomal proteins . Although Asc1 has been implicated in the ribosome-dependent no-go decay pathway ( Kuroha et al . , 2010 ) , the observed co-directional changes in mRNA abundance and translational efficiency are not consistent with widespread defects in no-go decay as a driver of changes in translation efficiency . If decreases in translation efficiency were caused by defects in no-go decay stabilizing mRNAs , thus inflating the denominator in the footprint RNA/total RNA calculation , then the levels of affected mRNAs should increase in the total RNA pool . However , the overall trend was for the levels of total mRNA for genes with decreased TE in the asc1-M1X mutant to go down or remain constant rather than increase ( Figure 4C , Figure 4—figure supplement 2A–D ) . The mRNAs that are sensitive to the loss of Asc1 are among the most efficiently translated in a cell . We therefore considered the possibility that reduced translation of these mRNAs might be a general consequence of perturbing the ribosome . To assess the specificity of the translational phenotypes of ASC1 null mutants , we tested four additional ribosomal protein mutants , rpl23b∆ rpp1a∆ , rps0b∆ , and rps16b∆ , each of which deletes one paralog encoding a core ribosomal protein . Like asc1-M1X , rpl23b∆ and rpp1a∆ show reduced growth on glucose and decreased 60S subunit levels ( Figure 5—figure supplement 1A , B ) . RPS0B and RPS16B encode small ribosomal subunit proteins that bind the ribosome near the mRNA exit channel in the vicinity of Asc1/RACK1 and deletion of these loci results in increased 60S subunit levels relative to 40S levels ( Figure 5—figure supplement 2A and B ) . However , none of the tested ribosomal protein mutants showed notable similarity to asc1-M1X in their translational dysregulation genome-wide ( r= -0 . 06 to 0 . 18 , Figure 5A and B ) , and they did not display decreased translation efficieny of ‘closed loop’ mRNAs ( Figure 5C ) . Because the growth and bulk translation phenotypes of these other ribosomal protein mutants are more severe than asc1-M1X , any shared defects in gene-specific translation should have been readily detected . Thus , decreased translation of RP genes is not a general feature of slow-growing mutants , ribosomal subunit imbalance , or perturbations in the vicinity of the mRNA exit channel near RACK1 . 10 . 7554/eLife . 11154 . 016Figure 5 . Other ribosomal protein mutants do not share translational phenotypes with the ASC1 mutants . ( A ) Correlations between ∆TE among asc1-M1X and mutants with reduced expression of large ribosomal subunit proteins , rpl23b∆ and rpp1a∆ . The Pearson correlation coefficient is shown . ( B ) Correlations between ∆TE among asc1-M1X and mutants with reduced expression of small ribosomal proteins in the vicinity of Asc1 , rps0b∆ and rps16b∆ . The Pearson correlation coefficient is shown . ( C ) Violin plots showing the change in TE for the ‘strong closed loop’ and ‘closed loop’ mRNAs in asc1-M1X and the other ribosomal protein mutants . Violin plot parameters are described in Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01610 . 7554/eLife . 11154 . 017Figure 5—figure supplement 1 . Phenotypes of selected large ribosomal protein mutants . ( A ) Growth curve comparing growth of asc1-M1X with other RP mutants rpl23b∆ and rpp1a∆ at 30˚C in glucose . ( B ) Polysome profiles of the rpl23a∆ and rpp1a∆ mutants at 30˚C . The polysome/monosome ( P/M ) and 60S/40S ( 60/40 ) ratios are shown with s . d . from two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01710 . 7554/eLife . 11154 . 018Figure 5—figure supplement 2 . Ribosomal location and phenotypes of selected small ribosomal protein mutants . ( A ) Structure of the yeast 40S ribosome with the positions of Rps0 and Rps16 shown relative to Asc1 . Structure taken from Ben-Shem et al . ( 2011 ) . ( B ) Polysome profiles from WT , rps0b∆ , and rps16b∆ . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 018 To assess the physiological significance of gene-specific translation defects in ASC1 mutants , we looked for phenotypes related to gene categories with significantly impaired translation . In particular , the requirement of Asc1 for efficient MRP translation suggested the possibility of impaired mitochondrial function in ASC1 mutants . To assess mitochondrial health , we measured growth on the non-fermentable carbon source glycerol , which requires the activity of the mitochondrial respiratory chain to generate energy ( Dimmer et al . , 2002 ) . When shifted to glycerol-containing media , wild type yeast resumed rapid growth after an initial adaptation phase , but the asc1-M1X mutant completed only ~3 doublings before ceasing growth ( Figure 6A ) . In contrast , the rpl23b∆ and rpp1a∆ mutants grew better in glycerol than asc1-M1X , demonstrating the specificity of this phenotype ( Figure 6—figure supplement 1A ) . Consistent with our results , a proteomic survey of asc1∆ cells showed a shift away from respiration and towards fermentative metabolism ( Rachfall et al . , 2012 ) . Because mitochondrial ribosomes are required for mitochondrial biogenesis and function , it is plausible that the growth and metabolic defects of ASC1 mutants are consequences of the translation defects observed for MRP genes . 10 . 7554/eLife . 11154 . 019Figure 6 . Asc1 is required for adaptation to a non-fermentable carbon source . ( A ) Growth curves of WT and asc1-M1X cells after a shift from YPAD to fresh media containing either glucose ( left ) or glycerol ( right ) . Curves are averages of two biological replicates , error bars are s . d . ( B , C ) Polysome profiles of WT ( B ) and asc1-M1X ( C ) yeast after a shift from glucose- to glycerol-containing media . Yeast were shifted at OD600=0 . 5 . The polysome/monosome ( P/M ) and 60S/40S ( 60/40 ) ratios are shown with s . d . from two biological replicates . ( D ) Violin plot of the FP , total mRNA , and TE changes in asc1-M1X for MRP transcripts during growth in glucose ( left , FP and TE data is also represented in Figure 4B , D ) or after 6 hr of growth in glycerol ( right ) . Violin plot parameters are as described for Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 01910 . 7554/eLife . 11154 . 020Figure 6—figure supplement 1 . Loss of Asc1 compromises mitochondrial function . ( A ) Growth curves of rpl23b∆ and rpp1a∆ after a shift from glucose- to glycerol-containing media , as described in Figure 6A . Data shown are averages and s . d . from two biological replicates . WT and asc1-M1X curves ( also shown in Figure 6 ) are shown for comparison . ( B ) Measurement of mitochondrial translation in WT and asc1-M1X . Left: Coomassie stain , used for total protein quantification; Right: 35S-labeled mitochondrial proteins . A two-fold dilution series of each sample was loaded to ensure accurate quantification . The six mitochondrial proteins used for quantification are indicated . ( C ) Quantification of mitochondrial translation in WT and asc1-M1X . The average change for each of six bands was averaged . Error bars are s . d . from two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 11154 . 020 To directly determine the impact of the MRP translation defects on mitochondrial translation activity , we performed 35S metabolic labeling assays in the asc1-M1X mutant in the presence of cycloheximide , which inhibits cytosolic but not mitochondrial ribosomes . Synthesis of all mitochondrially-translated proteins was reduced >two-fold in asc1-M1X compared to wild type ( Figure 6—figure supplement 1B and C ) . Thus pervasive , moderate impairment of MRP translation is associated with substantial defects in mitochondrial protein synthesis . Given the severe growth defects of the asc1-M1X mutant in glycerol , we wondered whether the moderate impairment of MRP translation observed in glucose would worsen under conditions of increased MRP expression . Adaptation to growth in non-fermentable carbon sources or low glucose is accompanied by a rewiring of the transcriptional network in yeast ( Galdieri et al . , 2010 ) and widespread reprogramming of translation ( Vaidyanathan et al . , 2014 ) . To investigate whether the glycerol growth defect of asc1-M1X is linked to inadequate translational adaptation , we examined translation genome-wide 6 hr after transfer from glucose to glycerol , a point just before the resumption of rapid growth in wild type cells ( Figure 6A ) and coincident with the recovery of polysomes after the initial collapse upon glucose withdrawal ( Figure 6B ) . In the asc1-M1X mutant , polysomes recovered only partially ( Figure 6C ) . Moreover , the ribosome-associated pool was strongly depleted of MRP mRNAs compared to wild type ( Figure 6D ) . However , the magnitude of translational defect ( ΔTE ) for this class of mRNAs was similar in both glucose and glycerol , supporting a constitutive rather than regulatory role for Asc1 in translation of MRP mRNAs ( Figure 6D ) . The fact that asc1-M1X shows a bulk translation defect in glycerol but not in glucose may reflect the fact that MRP mRNAs make up a larger portion of the translatome in glycerol . Thus , the cellular context is an important factor in determining the phenotypic consequences of translational perturbations in asc1-M1X and likely in other ribosomal protein mutants as well . Taken together , our results suggest an important role for Asc1 in supporting cellular respiration by promoting synthesis of mitochondrial ribosomal proteins .
Here we have demonstrated that the eukaryote-specific ribosomal protein Asc1/RACK1 is required for efficient translation of short mRNAs , a category that includes functionally related groups of genes required for vital cellular processes ( e . g . cytoplasmic and mitochondrial ribosomal proteins ) . A correlation between ORF length and translation efficiency or ribosome density has been observed since the advent of genome-wide translation profiling ( Arava et al . , 2003; Ingolia et al . , 2009 ) and we observed this relationship in data collected from diverse eukaryotes including yeast , nematodes , mice , and humans ( Guo et al . , 2010; Stadler and Fire , 2011 ) . To account for this trend , it was proposed that the rate of translation initiation is higher for short ORFs ( Arava et al . , 2005; Shah et al . , 2013 ) , but the mechanism ( s ) underlying length-dependent initiation rate differences were unknown . It has been suggested that the increased probability of mRNA circularization by diffusion could make initiation more efficient on short mRNAs ( Chou , 2003; Guo et al . , 2015 ) . Our results add an additional nuance to these physical models — the presence of a ribosome-dependent regulatory mechanism that specifically enhances the translation of short mRNAs by promoting the formation and/or function of the closed loop . Our analyses reveal a clear trend that short mRNAs preferentially associate with closed loop factors in vivo , and consistent with these observations , short mRNAs form more stable closed loop complexes than longer mRNAs in vitro ( Amrani et al . , 2008 ) . A challenge for the future will be to determine how the mRNA , the closed loop factors , and the ribosome cooperate to privilege the translation of short mRNAs . How might Asc1/RACK1 promote closed loop formation ? RACK1’s placement on the solvent exposed side of the head of the small subunit puts it in close proximity to the mRNA exit channel , in a position with the potential to interact with the mRNA-bound closed loop factors during initiation . Intriguingly , eIF4G co-purifies with Asc1 from yeast lysates under stringent conditions in which most other initiation factors do not ( Gavin et al . , 2002; 2006 ) , suggesting that Asc1 may interact directly with the closed loop via eIF4G . Our results also raise the possibility that translation of many mRNAs could be co-regulated by mechanism ( s ) that target Asc1/RACK1’s function in closed-loop-dependent initiation . Moving forward , it will be important to determine how the many signaling pathways that have been linked to Asc1/RACK1 impact the translation of closed-loop-dependent mRNAs . More generally , our study highlights the fact that individual ribosomal proteins can contribute to efficient translation of subsets of mRNAs with important consequences for cellular physiology . In particular we show that loss of the non-essential ribosomal protein Asc1/RACK1 causes a concerted decrease in MRP expression that leads to mitochondrial insufficiency . Given the central role of mitochondria in energy and metabolite production in eukaryotic cells , it is not surprising that mitochondrial defects elicit pleiotropic consequences ( Calvo and Mootha , 2010; Fleming et al . , 2001; Kushnir et al . , 2001; Shoffner et al . , 1990; Vafai and Mootha , 2012 ) . In light of our findings , many of Asc1/RACK1’s ascribed cellular functions should be re-evaluated for potential connections to mitochondrial dysfunction . Finally , it is intriguing that several distinct mutations in human ribosomal proteins and ribosome biogenesis factors result in anemia , the cause of which is currently the source of much debate ( Freed et al . , 2010; Narla et al . , 2011 ) . Given that many forms of heritable anemia have been traced to defects in mitochondrial iron metabolism ( Dailey and Meissner , 2013; Huang et al . , 2011 ) , it will be interesting to see whether translation of nuclear-encoded mitochondrial proteins is affected in these diseases and whether these defects contribute to pathogenesis .
The cDNA encoding the I27 domain monomer from human cardiac titin was a generous gift from Julio Fernandez . The I27 monomer was fused to a serine-glycine linker ( SGGGGG ) followed by the V5 epitope tag . The I27 octamer was made using the iterative subcloning method that relies upon the compatible cohesive ends of BamHI and BglII and results in an arginine-serine linker added between individual domains ( Hoffmann and Dougan , 2012 ) . I27 proteins were expressed under the GAL1 promoter and followed by the CYC1 terminator in the pRS415 low-copy yeast vector . Deletion strains of the ASC1 , RPL23B , and RPP1A loci were obtained by homologous recombination using the pFA6a-kanMX6 plasmid as a template and PCR product adding 40 nt of homology to each side of the kanMX6 cassette ( Longtine et al . , 1998 ) . Isolates were confirmed by PCR . Deletion strains of RPS0B , RPS16B , and their isogenic wild type were obtained from the Sigma1278b deletion collection ( Dowell et al . , 2010 ) . To make the ASC1 protein null alleles , a codon early in the ASC1 open reading frame was mutated to a stop codon , denoted as X ( i . e . M1X , E5X ) . Integration of mutant ASC1 alleles was performed using the two-step gene replacement strategy . First , the URA3 marker was integrated at the ASC1 locus . Subsequently , ASC1 mutant alleles were amplified by PCR from plasmid templates and integrated into the asc1::URA3 strain at the ASC1 locus . Isolates were identified by 5-FOA resistance and correct integration was confirmed by sequencing . All strains were constructed in the Sigma1278b strain background . Yeast were cultivated in liquid or on solid ( 2% agar ) YPAD media ( yeast extract , peptone , dextrose ( 2% w/v ) supplemented with adenine hemisulfate ) . Liquid cultures were grown with rapid agitation at 30˚C , unless otherwise noted , and harvested at OD 0 . 6–0 . 9 ( 0 . 6-0 . 7 for ribosome footprint profiling experiments in YPAD ) . For glycerol shift polysome experiments , yeast were grown to mid log phase ( OD 0 . 5–0 . 6 ) in YPAD and then media was removed by brief centrifugation and replaced with YPAG ( YPA + 3% ( w/v ) glycerol ) . For the yeast growth curves , yeast were diluted from saturated cultures into fresh media and allowed to double 1–2 times before rediluting to an OD of 0 . 1 in glucose- or glycerol-containing media . Cycloheximide ( CHX , Sigma-Aldrich , St . Louis , Missouri ) was added to a final concentration of 0 . 1 mg/ml to cells and incubated an additional 2 min at the growth temperature with shaking . Cells were rapidly cooled and washed with polysome lysis buffer ( PLB: 20 mM Hepes-KOH , pH 7 . 4 , 2 mM Mg acetate , 100 mM K acetate , 3 mM DTT , 0 . 1 mg/ml CHX + 1% Triton X-100 ) . Formaldehyde crosslinking experiments were performed as described ( Valásek et al . , 2007 ) . 10–15 OD260 units were loaded on 10–50% sucrose gradients in polysome gradient buffer ( PGB: PLB –Triton ) and centrifuged in an SW 41 rotor ( Beckman Coulter , Brea , California ) at 35 , 000 rpm for 3 hr . Fractions were collected from the top using a BioComp Gradient Station ( Biocomp Instruments , Canada ) . To calculate the ratio of free 60S/40S subunits , A254 traces of the native polysome profiles ( without dissociation into free subunits ) were quantified with a custom script , available on github: https://github . com/marykthompson/Thompson_eLife_2016/ . Minima were identified and used as boundaries for each peak . Values are the integral under the curve to the baseline , which was set as a line connecting the lowest minimum in the first half of the trace with the lowest minimum in the second half of the trace . Ribosome footprint profiling was performed essentially as described ( Ingolia et al . , 2009 ) with the following modifications: monosomes were isolated manually from 10–50% sucrose gradients . 50 A260 units were digested with 750 U of RNAse I ( Ambion , Waltham , Massachusetts ) . Selective precipitation was used to enrich for small RNA fragments prior to size selection of 28mers on denaturing gels . In brief , RNA samples were resuspended in GuHCl buffer ( 8 M guanidine HCl , 20 mM MES hydrate , 20 mM EDTA ) and brought to 33% ethanol before binding to a silica-based column ( Zymoprep-V , Zymoresearch , Irvine , California ) to precipitate and remove large RNAs . The eluate was brought to 70% ethanol to precipitate small RNAs . Total RNA for accompanying RNA-seq samples was isolated from the same cell extracts used for footprint library generation using the hot acid phenol method . Poly ( A ) selection was performed using oligo-dT cellulose ( Sigma-Aldrich or NEB , Ipswich , Massachusetts ) as previously described ( Sambrook et al . , 2001 ) . For experiments using rRNA-depletion to enrich for coding transcripts , the Ribo-Zero kit ( Epicentre , Madison , Wisconsin ) was used . The asc1-DE and matched WT libraries were constructed using an earlier version of the protocol that used Microcon YM-100 ( EMD Millipore , Billerica , Massachusetts ) filters to enrich for small RNA fragments , poly ( A ) tailing to capture the small RNA fragments , and downstream library construction steps as previously described ( Ingolia et al . , 2009 ) . For all other libraries , we ligated a pre-adenylated 3’ adaptor ( 5Phos/TGGAATTCTCGGGTGCCAAGG/3ddC/ ) to the fragments using T4 RNA Ligase 1 ( NEB ) . First strand synthesis was performed with Superscript III ( Life Technologies , Carlsbad , California ) or AMV ( Promega , Madison , Wisconsin ) using primer OJA225 ( /5Phos/GATCGTCGGACTGTAGAACTCTGAACCTGTCGGTGGTCGCCGTATCATT/iSp18/CACTCA/iSp18/GCCTTGGCACCCGAGAATTCCA ) . cDNA was amplified using primer oNTI230 ( AATGATACGGCGACCACCGA ) and ( CAAGCAGAAGACGGCATACGAGATXXXXXXGTGACTGGAGTTCCTTGGCACCCGAGAATTCCA ) , where XXXXXX denotes a six nucleotide barcode used to distinguish samples run in the same lane . Samples were run on an Illumina HiSeq 2000 instrument or an Illumnina GAIIx . RNA was extracted using the hot acid phenol method . RNA was treated with TURBO DNase ( Life Technologies ) . First strand synthesis was performed with AMV Reverse Transcriptase ( Promega ) using an anchored oligo-dT primer ( for coding transcripts ) or a random hexamer primer ( for SNR24 ) . Quantitative PCR was performed with SYBR Fast reagents ( Kapa Biosystems , Wilmington , Massachusetts ) using a Lightcycler 480 ( Roche , Switzerland ) . Gene-specific primer sequences are: ACT1: ( TTCTGAGGTTGCTGCTTTGG , CTTGGTGTCTTGGTCTACCG ) , ASC1: ( ATGTTTGGCCACTTTGTTGG , GTTACCGGCAGAAATGATGG ) , MRP2: ( AATAGGTGCGTGGACTCTGG , CTGGCAAATTACCCTTCAGAGC ) , SNR24: ( TTGCTACTTCAGATGGAACTTTG , TCAGAGATCTTGGTGATAATTGG ) , V5: ( AGATCTTCCGGAGGCGGG , GGATCTATTACGTAGAATCGAGACC ) , YML6: ( AGAGTAGGCGCCTCAAATCC , TTGGAGAGTTAGCATCCCCG ) , 18S: ( TGGCGAACCAGGACTTTTAC , CCGACCGTCCCTATTAATCAT ) , FLUC: ( GTACCAGAGTCCTTTGATCGTGA , ACCCAGTAGATCCAGAGGAATTC ) . Total protein levels were determined using the BCA assay ( Thermo Scientific , Waltham , Massachusetts ) . For total Asc1 level quantification , 1 μg of total protein obtained by TCA precipitation followed by cell lysis was loaded onto 12% SDS-PAGE gels . For polysome Westerns , the same volume of each fraction was loaded per well . Blots were overexposed to show the remaining ribosome-associated protein for the ribosome-binding mutants . Membranes were blotted with α-Asc1 ( Coyle et al . , 2009 ) and α-Pgk1 ( Life Technologies 22C5D8 ) . After secondary antibody incubation , blots were incubated with ECL ( GE Healthcare Life Sciences , United Kingdom ) and exposed to X-ray film . Yeast grown overnight in SC-Leu ( synthetic complete media lacking leucine ) were diluted to OD 0 . 2 in YPA + 2% galactose and grown for 8 hr before harvest . Cells were lysed in PBS pH 7 . 4 supplemented with protease inhibitors ( 1X Roche complete mini EDTA-free , 1 mM PMSF ) with glass beads . Total protein was quantified by the BCA assay ( Thermo Scientific ) and 1 ug ( octamer ) or 2 ug ( monomer ) total protein was loaded per lane with each sample loaded in 4 different lanes as technical replicates for each of three biological replicates . A standard curve encompassing 2X , 1X , 0 . 5X and 0 . 25X of the WT extract concentration was loaded on each gel . Western blotting was performed using the ECL Plex kit ( GE ) according to the manufacturer’s instructions and blots were scanned with a Typhoon imager ( FLA 9500 , GE ) . Primary antibodies were α-Pgk1 ( Life Technologies 22C5D8 ) and α-V5 ( Sigma-Aldrich V8137 ) . Images were quantified with ImageStudio ( LI-COR Biosciences , Lincoln , Nebraska ) and the values of each sample were calculated relative to the standard curve . Although all standards were in linear range ( linear fit of signal vs . concentration r2 ≥ 0 . 95 for all blots ) , we used a quadratic fit as it fit the standards slightly better . RNA was extracted from the extracts in parallel , and the mRNA levels of each reporter were quantified by qRT-PCR using primers recognizing the region encoding the V5 tag and normalized to 18S levels also determined by qRT-PCR . For each sample , a translation efficiency was calculated from the ratio of the normalized protein levels of the reporter ( V5 protein/Pgk1 protein ) to the normalized mRNA levels of the reporter ( V5 mRNA/18S rRNA ) . Mitochondrial translation products were labeled with 35S-methionine as previously described ( Funes and Herrmann , 2007 ) . In brief , cells were grown overnight in SC-Met ( with glucose ) to OD 0 . 4 then transferred to SC-Met with glycerol for 3 hr . Equal OD units of cells were then incubated with 35S-methionine ( EasyTag L-[35S]-Methionine , PerkinElmer , Waltham , Massachusetts ) and cycloheximide to inhibit cytoplasmic protein synthesis . After 30 min , total protein synthesis was halted by the addition of puromycin . TCA-precipitated protein was visualized by Coomassie staining ( total protein normalization ) and autoradiography on a Typhoon imager ( mitochondrial proteins ) . Total protein in each sample was quantified with ImageJ using Coomassie signal across the whole lane . Six bands corresponding to mitochondrial translation products were quantified with ImageQuant ( GE ) . Yeast reads were aligned to the Sigma1278b ( Dowell et al . , 2010 ) genome downloaded from the Saccharomyces Genome Database on June 29 , 2014 . We used Tophat to map first to annotated splice junctions and then to the genome . We used only uniquely-mapping reads for all downstream analyses . Because ribosome footprint reads generally start 12 nt upstream of start codons and end 18 nt upstream of stop codons ( Ingolia et al . , 2009 ) , we used only reads mapping within these boundaries . Additionally , to avoid potential variability that can be present at the 5’ end of mRNAs , we excluded the first 30 codons from counting for quantification of gene expression . For comparisons between libraries , we used normalized values obtained from running count data through the DE-Seq package ( Anders and Huber , 2010 ) because RPKM values are strongly biased by the transcript lengths of the RNA pool ( Wagner et al . , 2012 ) . For gene expression measurements , we only included genes for which there were at least 128 mapping reads total among the libraries used for the analysis ( Ingolia et al . , 2009 ) . All analyses were performed with custom Bash and Python scripts written in-house , available on github: https://github . com/marykthompson/Thompson_eLife_2016/ . Data in figures represent the average of two biological replicates . Figures were constructed using Matplotlib ( Hunter , 2007 ) . To determine whether the decrease in translation efficiency of the ‘closed loop’ mRNAs in asc1-M1X could be accounted for completely by the relationship between ∆TE and ORF length , we first regressed ∆TE against ORF length . We then took the residuals from this correlation ( i . e . the part of ∆TE that cannot be accounted for by the global correlation between ∆TE and ORF length ) and plotted these values among the ‘strong closed loop’ , ‘closed loop’ and ‘other’ mRNAs , as shown by the dashed lines in Figure 3H and I . Note that this analysis assumes linear relationships between ∆TE and ORF length . These results demonstrate that the decrease in translation efficiency of the ‘closed loop’ mRNAs in asc1-M1X is more than would be expected if ∆TE was determined entirely by ORF length , thus suggesting that ‘closed loop’ enrichment may be more important . However , as with all correlative analyses , the results cannot assign causality . For correlations of TE with ORF length shown in Figure 3—figure supplement 3 , processed data files were downloaded from NCBI GEO and used to calculate TE . Only genes in which the pooled the reads or scaled reads ( for Stadler and Fire , 2013 ) from footprint and total RNA libraries reached 128 reads were included . To identify groups of genes with significantly altered TE in yeast mutants , we used the Mann Whitney U test and report one-sided p-values for groups of genes with significantly altered TE each condition . For this analysis , we included all genes without filtering for read cutoff and added a pseudocount of one read in cases where >1 read was detected in some but not all libraries . We used MEME ( Bailey and Elkan , 1994 ) to identify motifs present in 5′ UTRs of the selected groups of mRNAs . 5′ UTR boundaries were taken from the median UTR length reported in Pelechano et al . ( 2014 ) . UTRs <8 nt were excluded from the motif analysis . WebLogo ( Crooks et al . , 2004 ) was used to generate sequence logos . 5′ and 3′ UTR lengths were taken as the median length from Pelechano et al . ( 2014 ) . MFEs were calculated by running these sequences through RNAfold ( Gruber et al . , 2008 ) with temperature set to 30˚C and otherwise default parameters . Translation adaptation index values per gene were calculated by Eckhard Jankowsky and colleagues using values from Tuller et al . ( 2010 ) . Poly ( A ) tail length was taken from Subtelny et al . ( 2014 ) . Wild type protein levels were taken from de Godoy et al . , 2008 . | Ribosomes are structures within cells that are responsible for making proteins . Molecules called messenger RNAs ( or mRNAs ) , which contain genetic information derived from the DNA of a gene , pass through ribosomes that then “translate” that information to build proteins . Although all living cells contain ribosomes , the protein building blocks that make up the structure of the ribosome are not the same in all species . Furthermore , the exact roles that each building block plays during translation are not known . The ribosomes of plants , animals , and budding yeast contain the same protein , known as Asc1 in budding yeast and RACK1 in plants and animals . Thompson et al . have now explored the role of Asc1 in yeast cells by measuring translation in the absence of Asc1 using a technique called ribosome footprint profiling . This analysis revealed that cells lacking Asc1 translate fewer short mRNA molecules than normal cells . Short mRNAs encode small proteins that tend to play important ‘housekeeping’ roles in the cell — by forming the structural building blocks of ribosomes , for example . It has been observed previously that short mRNAs are translated at a higher rate than longer mRNAs on average , although the reasons behind this bias are still mysterious . The findings of Thompson et al . suggest that the ribosome itself may discriminate between short and long mRNAs and that the Asc1 protein is involved in calibrating the ribosome’s preference for short mRNAs . Cells need differing amounts of small proteins in different growth conditions . It will therefore be interesting to investigate whether mRNA length discrimination can be regulated by Asc1 and/or other components of the ribosome to tune gene expression to the environment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2016 | The ribosomal protein Asc1/RACK1 is required for efficient translation of short mRNAs |
Autism Spectrum Disorder ( ASD ) is the most prevalent neurodevelopmental disorder in the United States and often co-presents with sleep problems . Sleep problems in ASD predict the severity of ASD core diagnostic symptoms and have a considerable impact on the quality of life of caregivers . Little is known , however , about the underlying molecular mechanisms of sleep problems in ASD . We investigated the role of Shank3 , a high confidence ASD gene candidate , in sleep architecture and regulation . We show that mice lacking exon 21 of Shank3 have problems falling asleep even when sleepy . Using RNA-seq we show that sleep deprivation increases the differences in prefrontal cortex gene expression between mutants and wild types , downregulating circadian transcription factors Per3 , Bhlhe41 , Hlf , Tef , and Nr1d1 . Shank3 mutants also have trouble regulating wheel-running activity in constant darkness . Overall , our study shows that Shank3 is an important modulator of sleep and clock gene expression .
Autism Spectrum Disorder ( ASD ) is the most prevalent neurodevelopmental disorder in the United States ( diagnosed in 1 in 59 children [Baio et al . , 2018] ) . The core symptoms of ASD include social and communication deficits , restricted interests , and repetitive behaviors ( American Psychiatric Association , 2013 ) . In addition , several studies show that individuals with ASD report a variety of co-morbid conditions including sleep problems and altered circadian rhythms ( Glickman , 2010 ) . It is estimated that 40–80% of the ASD population experience sleep disorders that do not improve with age ( Johnson et al . , 2009 ) . More specifically , people with ASD have problems falling and staying asleep ( Hodge et al . , 2014 ) . A recent study showed that sleep problems co-occur with autistic traits in early childhood and increase over time , suggesting that sleep problems are an essential part of ASD ( Verhoeff et al . , 2018 ) . Indeed , sleep impairments are a strong predictor of the severity of ASD core symptoms as well as aggression and behavioral issues ( Cohen et al . , 2014; Tudor et al . , 2012 ) . Although a great number of studies documented sleep problems in ASD , little is known about the underlying molecular mechanisms . To better understand the mechanisms underlying sleep problems in ASD , we need animal models that closely recapitulate sleep phenotypes observed in clinical populations . The study of genetic animal models of ASD , in which a genetic abnormality that is known to be associated with ASD is introduced , has provided valuable insight into the molecular mechanisms underlying ASD ( de la Torre-Ubieta et al . , 2016 ) . These models include Fragile X syndrome , 16p11 . 2 deletion syndrome , cortical dysplasia-focal epilepsy ( CDFE ) syndrome , and mutations in neuroligins , neurexins , and shank genes among others . However , sleep research in animal models of ASD is limited and has not yet revealed the underlying mechanisms of sleep issues associated with ASD . Studies using a fly model of Fragile X syndrome reported an increase in sleep which is in contrast to what is observed in the clinical population ( Bushey et al . , 2009 ) . The opposite phenotype was reported in a Fragile X mouse model , displaying instead an age-dependent reduction in activity during the light phase ( i . e the mouse subjective night ) ( Boone et al . , 2018 ) . Neuroligin 1 knockout mice spend more time asleep ( El Helou et al . , 2013 ) , but mice with mutations in Neuroligin 2 spend less time asleep and more time awake ( Seok et al . , 2018 ) . Mice with a missense mutation in Neuroligin 3 show normal sleep behavior ( Liu et al . , 2017 ) , but rats with a deletion mutation in Neuroligin 3 spend less time in non-rapid eye movement ( NREM ) sleep than wild type rats ( Thomas et al . , 2017 ) . Mutant rat models of CDFE syndrome show longer waking periods while the mutant mice show fragmented wakefulness ( Thomas et al . , 2017 ) . Mice carrying a 16p11 . 2 deletion syndrome sleep less than wild type mice , but only males are affected ( Angelakos et al . , 2017 ) . More importantly , issues with sleep onset , the most prominent feature of sleep problems in ASD patients , have not been evaluated in animal models of ASD . In this study , we examined sleep in Phelan-McDermid syndrome ( PMS ) patients with SHANK3 mutations and in a mutant mouse with a deletion in Shank3 exon 21 ( Shank3ΔC ) . PMS is a syndromic form of ASD characterized by gene deletions affecting the human chromosomal region 22q13 . 3 ( Phelan and McDermid , 2012 ) , particularly the neuronal structural gene SHANK3 . Individuals with PMS have high rates of intellectual disability , difficulties in communication and motor function , and approximately 84% fit the core diagnostic criteria for ASD ( Soorya et al . , 2013 ) . There is also a high rate of sleep problems in PMS ( Bro et al . , 2017 ) . Mice with mutations in Shank3 recapitulate multiple features of both ASD and PMS ( Bozdagi et al . , 2010; Dhamne et al . , 2017; Jaramillo et al . , 2017; Jaramillo et al . , 2016; Kouser et al . , 2013; Peça et al . , 2011; Speed et al . , 2015 ) , including cognitive impairment , deficits in social behavior , and impaired motor coordination . We show that PMS patients have trouble falling and staying asleep similar to what is observed in the general ASD population . We also show that Shank3ΔC mice sleep less than wild type mice when sleep pressure is high , have reduced sleep intensity ( using an accepted electroencephalographic ( EEG ) metric ) , and have delayed sleep onset following sleep deprivation . To identify molecular mechanisms underlying sleep changes in Shank3ΔC mice , we carried out genome-wide gene expression studies . We previously showed that genome-wide gene expression analysis is a valuable approach to understand the molecular mechanisms underlying the detrimental effects of sleep deprivation ( Gerstner et al . , 2016; Vecsey et al . , 2012 ) . In this study , we found that sleep deprivation sharply increases the differences in gene expression between Shank3ΔC mutants and wild type mice , downregulating circadian transcription factors Per3 , Bhlhe41 ( Dec2 ) , Hef , Tlf , and Nr1d1 ( Rev-erbα ) . We also show that Shank3ΔC mice are unable to sustain wheel-running activity in constant darkness . Overall , these studies demonstrate that Shank3 is an important modulator of sleep that may exert its effect through the regulation of circadian transcription factors . Our findings may lead to a deeper understanding of the molecular mechanisms underlying sleep problems in ASD . This may one day lead to the development of successful treatments or interventions for this debilitating comorbidity .
Recent studies suggest that sleep problems may be present in a substantial number of PMS patients and may be an important factor for caregivers' well-being ( Bro et al . , 2017 ) . We obtained genetic and sleep questionnaire data from the Phelan-McDermid Syndrome International Registry ( PMSIR ) to estimate the frequency and age of onset of sleep problems in PMS individuals carrying a SHANK3 deletion . In parallel , we surveyed the clinical literature to estimate the prevalence of sleep problems in ASD ( Andersen et al . , 2008; Cotton and Richdale , 2006; Gail Williams et al . , 2004; Giannotti et al . , 2008; Giannotti et al . , 2006; Krakowiak et al . , 2008; Liu et al . , 2006; Miano et al . , 2007; Paavonen et al . , 2008; Polimeni et al . , 2005; Richdale and Prior , 1995; Tani et al . , 2003; Thirumalai et al . , 2002; Wiggs and Stores , 2004 ) and typically developing populations ( Anders and Eiben , 1997; Baweja et al . , 2013; Bixler et al . , 2009; Hysing et al . , 2013; Leger et al . , 2012; Loessl et al . , 2008; Lozoff et al . , 1985; Lumeng and Chervin , 2008; Ohayon et al . , 2000; Pallesen et al . , 2008; Patzold et al . , 1998; Sadeh et al . , 2000 ) . Figure 1 shows that PMS patients have trouble falling asleep and experience multiple night awakenings starting at about 5 years of age . Those difficulties translate to reduced time asleep particularly during adolescence . Although total sleep time seems to improve in adulthood for PMS patients , that improvement is accompanied by an increase in parasomnias . Problems falling and staying asleep persist regardless of age . The frequency of problems falling and staying asleep in PMS patients is similar to what is observed in the general ASD population and much higher than in typically developing individuals ( Figure 1—source data 1 ) . To determine if Shank3ΔC mice have deficits in spontaneous sleep , undisturbed baseline EEG and electromyographic ( EMG ) recordings were obtained from wild type ( WT ) and Shank3ΔC mice . There was a significant period x genotype interaction for total sleep time ( i . e . total time spent in NREM and rapid eye movement ( REM ) sleep ) during the light ( hours 1–12 ) and dark ( hours 13–24 ) periods ( F ( 1 , 28 ) = 5 . 198 , p=0 . 036 ) . Posthoc pairwise comparisons using Sidak correction showed that Shank3ΔC mice slept less than WT mice during the dark period ( p=0 . 045; Table 1 ) . To determine in which arousal states and hours this effect was most pronounced , we examined hourly time in state data for wakefulness , NREM sleep , and REM sleep . Repeated measures ANOVA over the full 24 hr recording period found significant time x genotype interactions for wakefulness ( F ( 23 , 345 ) = 2 . 419 , p<0 . 0001 ) , NREM sleep ( F ( 23 , 345 ) = 2 . 357 , p=0 . 001 ) , and REM sleep ( F ( 23 , 345 ) = 2 . 175 , p=0 . 002 ) ( Figure 2A ) . Posthoc comparisons found the most pronounced differences at hour 19 when Shank3ΔC mice spent more time in wakefulness ( p<0 . 0001 ) and less time NREM ( p<0 . 0001 ) and REM ( p=0 . 040 ) sleep ( Figure 2A ) marking the beginning of an overall trend for increased wakefulness during the last half of the dark period . Analysis of the sleep architecture found time x genotype interactions for bout number during REM sleep ( F ( 3 , 51 ) = 2 . 832 , p=0 . 047 ) and bout duration during NREM ( F ( 1 . 749 , 29 . 734 ) = 4 . 077 , p=0 . 032 ) and REM ( F ( 1 . 897 , 32 . 243 ) = 6 . 536 , p=0 . 005 ) sleep . Posthoc comparisons showed these effects were driven by differences in the last half of the dark period ( hours 19–24 ) when Shank3ΔC mice had fewer REM bouts ( p=0 . 003; Figure 2—figure supplement 1A ) and shorter NREM ( p=0 . 005 ) and REM ( p=0 . 013 ) bouts ( Figure 2—figure supplement 1B ) than WT mice . Overall , these data show that Shank3ΔC mice spend more time awake at the end of the dark phase compared to WT mice under baseline conditions . Fourier analysis of the EEG indicates that Shank3ΔC mice also sleep differently than WT mice ( Figure 2B ) . Significant genotypic differences in spectral frequencies occur at points of non-overlaping 95% confidence intervals . During NREM sleep , power in the delta frequencies ( 0 . 5–4 Hz ) was blunted and alpha frequencies ( 10–15 Hz ) were enhanced for Shank3ΔC mice relative to WT during the light and dark periods ( Figure 2B ) . NREM delta power is a measure of synchrony of the neural network and can be a measure of sleep intensity or depth ( Achermann and Borbely , 2017 ) . These data indicate that Shank3ΔC mice exhibit disrupted neural connectivity during NREM sleep which may suggest that Shank3ΔC mice sleep less deeply under baseline conditions . To investigate sleep homeostasis in Shank3ΔC mice we sleep deprived mutant and WT mice for 5 hr via gentle handling starting at light onset ( hour 1 ) . Sleep deprivation ( SD ) was effective resulting in WT and Shank3ΔC mice spending 97 . 37 ± 0 . 76% and 95 . 08 ± 1 . 08% time in wakefulness , respectively , during the 5 hr SD period ( t ( 18 ) = 1 . 732 , p=0 . 100 ) . Both WT and Shank3ΔC mice showed homeostatic responses to SD ( Figure 3—figure supplement 1 ) with reduced wakefulness and increased time spent in NREM and REM sleep during the recovery phase ( Figure 3—source data 2 ) . Sleep was also more consolidated after SD for both WT and Shank3ΔC mice during the light period , and changes in sleep architecture did not statistically differ between WT and Shank3ΔC mice ( Figure 3—figure supplement 2 ) . However , there were time x genotype interactions for wakefulness ( F ( 18 , 288 ) = 2 . 025 , p=0 . 009 ) , NREM sleep ( F ( 18 , 288 ) = 1 . 928 , p=0 . 014 ) , and REM sleep ( F ( 18 , 288 ) = 1 . 716 , p=0 . 036 ) over the SD recovery phase ( hours 6–24 ) . This effect was most pronounced near the end of the dark period where Shank3ΔC mice spent more time in wakefulness and less time in NREM and REM sleep compared to WT ( Figure 3—figure supplement 1C ) which is consistent with the trend seen in the baseline data ( Figure 2A ) . Changes in NREM EEG delta power were similar between WT and Shank3ΔC mice , indicating that mutant mice accumulate sleep pressure similarly to WT mice ( Figure 3A ) . However , Shank3ΔC mice showed a transient enhancement of NREM EEG spectral power ( F ( 1 , 18 ) = 12 . 07 , p=0 . 003 ) compared to WT mice in the first 2 hr post-SD ( hours 6–7 ) . This enhancement was found in the higher frequencies ( 3 . 9–19 . 5 Hz; Figure 3B ) and resolved to baseline and WT values during hours 11–12 ( F ( 1 , 18 ) = 0 . 59 , p=0 . 454; Figure 3C ) . These data suggest that some aspects of the Shank3ΔC homeostatic response to sleep deprivation are abnormal . Remarkably , Shank3ΔC mice took longer to enter NREM sleep post-SD compared to WT ( t ( 10 . 44 ) = −2 . 31 , p=0 . 043; Figure 3D ) . As a consequence , Shank3ΔC mice spent less time in NREM sleep compared to WT during the first 2 hr of the recovery phase ( t ( 13 . 26 ) = 2 . 85 , p=0 . 014; Figure 3E ) . During the subsequent 5 hr of the light period ( hours 8–12 ) , Shank3ΔC mice spent more time in NREM sleep compared to WT ( t ( 18 ) = −2 . 46; p=0 . 024; Figure 3F ) . Overall , our data show that Shank3ΔC mice have difficulties falling asleep despite heightened sleep pressure . We conducted a genome-wide gene expression study to investigate the molecular basis for the Shank3ΔC mouse sleep phenotype . Shank3ΔC and WT adult male mice were subjected to 5 hr of SD starting at light onset and sacrificed immediately post-SD . Additional mice from both genotypes were sacrificed at the same time of day to determine differences in gene expression under homecage conditions ( HC ) . Prefrontal cortex was collected for all animals and subjected to RNA sequencing ( RNA-seq , n = 5 per group ) . As expected , sleep deprivation is the greatest source of variation in the data ( principal component 1 ) , followed by the genotype effect ( principal component 2; Figure 4A ) . Furthermore , the difference between genotypes is enhanced after sleep deprivation ( Figure 4A ) greatly increasing the number of differentially expressed genes between Shank3ΔC and WT mice – starting at 69 genes ( HC ) and doubling to 134 genes ( SD ) ( Figure 4B , Figure 4—source data 1; false discovery rate ( FDR ) < 0 . 1 ) . Most of the differences in gene expression following SD are not present in HC conditions . Clustering of gene expression patterns for all genes differentially expressed between Shank3ΔC and WT mice reveals 3 groups of genes ( Figure 4C ) . Cluster 1 contains genes that are downregulated in mutants versus WT mice under HC conditions . Genes in this cluster are also downregulated in response to SD in Shank3ΔC mice but are generally unaffected by SD in WT mice . SD seems to exacerbate the difference between genotypes , which explains why differential expression of some of the genes in cluster 1 only reach statistical significance after SD . This cluster contains the majority of the genes that are differentially expressed between genotypes . Cluster 2 contains genes that are upregulated in mutants versus WT mice under HC conditions that are also downregulated by SD in WT mice . Cluster 3 contains genes that are normally upregulated by SD . For both clusters 2 and 3 , the Shank3ΔC mutation seems to dampen the response to SD . To better understand the impact of the gene expression at the pathway level , we carried out functional annotation of the transcripts differentially expressed between Shank3ΔC and WT mice ( Table 2 , Table 2—source data 1 ) . Our results reveal that 1 ) MAPK/GnRH signaling and circadian rhythm-associated transcripts are downregulated in Shank3ΔC mice and 2 ) sleep deprivation exacerbates that difference . Circadian transcription factors are particularly affected . For example , while expression of Per3 , Hlf , and Tef already differs under homecage conditions , SD leads to additional genotype-specific differences in expression of Nr1d1 and Bhlhe41 . All of the above mentioned circadian transcription factors belong to cluster 1 on our heat map ( Figure 4C ) showing downregulation in response to SD only in Shank3ΔC mice . The results of our RNA-seq analysis revealed a general downregulation of circadian transcription factors in the mutants . These findings suggested that Shank3ΔC mice might also have abnormal circadian rhythms . To test this possibility , we measured wheel-running activity in Shank3ΔC and WT mice . Wheel-running data were collected for 2 weeks under LD ( LD weeks 2–3 ) , then mice were released into constant darkness ( DD ) for 3 weeks ( DD weeks 1–3; Figure 5 and Figure 5—figure supplement 1 ) . Alpha , the length of the active phase , was calculated as the time between activity onset and offset . Period length was the time elapsed from the start of an active phase to the start of the subsequent active phase . There was no differences between Shank3ΔC and WT mice for alpha or period length during the DD period ( Table 3 ) , but Shank3ΔC mice ran fewer revolutions than WT mice from LD week 3 – DD week 3 ( F ( 1 , 338 ) = 30 . 96; p<0 . 0001; Table 3 , Table 3—source data 1 ) . Some Shank3ΔC mice greatly reduced their running at certain periods of the DD period , which is reflected in the variance for period , alpha , and activity measures ( Figure 5 , Figure 5—figure supplement 1 , and Table 3 ) . Over 5 weeks of continuous LD conditions , however , wheel-running activity for Shank3ΔC mice instead increased ( F ( 1 , 542 ) = 85 . 99 , p<0 . 0001; Supplementary file 1 , Source data 1 ) . Overall , these data indicate that constant darkness impairs wheel-running activity of Shank3ΔC mice .
The present study is the first to establish a role for Shank3 in mammalian sleep . We show that both PMS patients and Shank3 mutant mice have trouble falling asleep ( Figures 1 , 2 and 3 ) . This phenotype is widely observed in the ASD population and until now had not been replicated in animal models . Shank3∆C mice have problems falling asleep after periods of extended wakefulness when sleep pressure is high , such as at the end of the baseline dark period ( Figure 2A ) or following sleep deprivation ( Figure 3D-E ) . They also have blunted NREM delta power ( Figure 2B ) . However , Shank3∆C mice accumulate sleep pressure ( Figure 3A ) and show no gross abnormalities in circadian rhythms ( Figure 5 ) . This suggests that the primary deficit is in sleep onset . Our molecular studies show that sleep deprivation increases differences in gene expression between Shank3∆C and WT mice ( Figure 4 ) . These differences point to the downregulation of circadian transcription factors and genes involved in the MAPK/GnRH pathways in the mutants ( Table 2 ) . Circadian transcription factors affected include Per3 , Hlf , Tef , Nr1d1 ( REV-ERBα ) , and Bhlhe41 ( DEC2 ) . We therefore investigated the effects of our Shank3 mutation in circadian rhythms . Shank3∆C mice do not show a disruption in circadian rhythmicity , but they do exhibit a large reduction in wheel-running activity in response to constant darkness ( Figure 5 , Table 3 ) . Shank3∆C mice were reported to have deficits in motor coordination ( Kouser et al . , 2013 ) ; however , under a 12:12 hr LD schedule their wheel-running activity increases over time ( Supplementary file 1 , Source data 1 ) . Daily rhythms of activity in rodents are linked to circadian oscillations in dopamine release ( Feenstra et al . , 2000; Menon et al . , 2019 ) in the frontal cortex as well as in the striatum , a motor region with high levels of Shank3 expression ( Peça et al . , 2011 ) . Together with a reduction in circadian gene expression , this DD-specific activity deficit suggests that the mutant sleep phenotype involves clock gene functions outside of their central time-keeping role . Interestingly , mutations in some of the circadian transcription factors we identified , Per3 and Bhlhe41 ( DEC2 ) , indeed lead to deficits in sleep regulation ( Archer et al . , 2018; He et al . , 2009; Hirano et al . , 2018 ) . Our results support a role of clock genes in influencing sleep outside of their roles in generating circadian rhythms ( Franken , 2013 ) . An interesting question is how can deletion of exon 21 of Shank3 lead to dysregulation of transcriptional and signaling pathways linked to sleep and sleep loss ? Exon 21 of Shank3 encodes the homer and cortactin interaction domains of the protein . Homer interacts with metabotropic glutamate receptors ( mGluRs ) and SHANK3/homer complexes anchor mGluRs to the synapse . Shank3∆C mice show a marked reduction of the major isoforms of SHANK3 as well as an increase of mGluRs at the synapse ( Kouser et al . , 2013 ) . mGluR signaling activates the MAPK/ERK pathway , a key regulator of activity-dependent transcription and synaptic plasticity in mature neurons ( Thomas and Huganir , 2004 ) . So the role of SHANK3 at the synaptic membrane explains the observed regulation of MAPK pathway genes ( Table 2 ) . However , it is not yet clear how SHANK3 regulates expression of circadian transcription factors within the nucleus . One possibility might include the role of SHANK3 in Wnt signaling . SHANK3 modulates Wnt-mediated transcriptional regulation by regulating internalization of Wnt receptor Frizzled-2 ( Harris et al . , 2016 ) and nuclear translocation of the Wnt ligand beta-catenin ( Qin et al . , 2018 ) . The Wnt pathway kinase GSK3β phosphorylates circadian transcription factors PER2 ( Iitaka et al . , 2005 ) , CRY2 ( Harada et al . , 2005 ) , and REV-ERBα ( Yin et al . , 2006 ) ; however , this mechanism modulates circadian period length which is not altered in Shank3ΔC mice . A second potential mechanism is nuclear translocation of SHANK3 itself . SHANK3 is known to undergo synaptic-nuclear shuttling in response to neuronal activity and interact with nuclear ribonucleoproteins and components of the RNA Pol II mediator complex ( Grabrucker et al . , 2014 ) . Deletion of the C-terminus leads to nuclear accumulation of SHANK3 and alterations in gene expression ( Cochoy et al . , 2015; Grabrucker et al . , 2014 ) . Thus , mutations in exon 21 of Shank3 could lead to deficits in transcriptional regulation in response to sleep deprivation through direct regulation of transcription in the nucleus . Yeast two-hybrid data show that SHANK3 can directly bind the circadian transcription factors REV-ERBα ( encoded by Nr1d1 ) and DEC1 ( encoded by Bhlhe40 ) , a close paralog of DEC2 ( Bhlhe41 ) ( Sakai et al . , 2011 ) . Future studies of the effects of sleep on SHANK3 nuclear translocation will provide new insights into the non-synaptic function of shank proteins and their role in sleep and circadian rhythms .
Sleep questionnaire data for PMS patients was obtained through a data access agreement with the PMSIR . The PMSIR contains demographic , clinical , and genetic data on PMS patients . Parents and caregivers are able to create a profile on behalf of the patient , enter demographic data , complete questionnaires on symptoms and development , and upload files such as genetic test reports and other medical records . Trained genetic counselors curate all genetic data to ensure each patient has as complete and accurate ‘genetic profile’ insofar as possible . Researchers may apply for data exports containing de-identified clinical data , developmental data , and genetic data . Once approved , the appropriate search is performed in the database and the results are provided to the researcher . Data in this study corresponds to a database export performed on 12/1/2016 containing the results of a sleep questionnaire completed by caregivers , as well as biographic and genetic information of PMS individuals . Difficulty falling asleep is defined as needing more than an hour to fall asleep . Multiple night awakenings is defined as more than two awakenings . Reduced sleep time is defined as less than 6 hr per night . Parasomnias are defined as abnormal movements , behaviors , emotions , perceptions , and dreams that occur while falling asleep or sleeping . Presence of sleep apnea is defined as having received a diagnosis of sleep apnea . Only individuals with a genetic counselor-confirmed deletion of the SHANK3 gene were included in the analysis . The final dataset included 176 individuals – 78 males and 98 females ( age range 1–39 years ) . Heterozygous Shank3+/∆C mice were obtained from Dr . Paul Worley at Johns Hopkins University . Shank3+/∆C breeding pairs were established to obtain wild type ( WT ) and Shank3∆C littermates . Mice were housed in standard cages at 24 ± 1°C on a 12:12 hr light:dark cycle ( unless otherwise specified ) with food and water ad libitum . All experimental procedures were approved by the Institutional Animal Care and Use Committee of Washington State University and conducted in accordance with National Research Council guidelines and regulations for experiments in live animals . | Autism spectrum disorder ( ASD ) is the most common neurodevelopmental disorder in the United States . People with ASD tend to have difficulties with communication and social interactions , restricted interests , and may repeat certain behaviors . They also often struggle to fall or stay asleep . Sleep deprivation may exacerbate other symptoms of the disorder . This makes life more difficult for both the person with ASD and their caregivers . Scientists do not yet know what causes sleep difficulties in people with ASD . Unraveling the complex genetics that underlie ASD may help scientists better understand ASD-related sleep difficulties . One possible genetic culprit for sleep difficulties in ASD is a gene called SHANK3 . Patients with an ASD-associated condition called Phelan-McDermid syndrome are often missing the SHANK3 gene . They also often have sleep problems . Now , Ingiosi , Schoch et al . show that both patients with Phelan-McDermid syndrome and mice with a mutation in the Shank3 gene have problems falling asleep . Using a registry that collects genetic and sleep information on people with Phelan-McDermid syndrome , Ingiosi , Schoch et al . found that people who are missing SHANK3 frequently have trouble falling asleep and wake up many times each night . Mice missing part of the Shank3 gene also had difficulty falling asleep , even after they have been deprived of sleep . Mice naturally have a daily pattern of sleep and activity . This 24-hour activity cycle is maintained by an internal circadian clock . In mice missing part of Shank3 , the circadian clock genes are not turned on correctly . These genes were less active in mice missing Shank3 , and this difference worsened with lack of sleep . These mice also ran less on a wheel than typical mice when kept in total darkness , even though the pattern of activity did not change . The experiments suggest that Shank3 controls sleep , likely through its effects on circadian clock genes . Learning more about what causes these sleep problems may help scientists develop ways to improve sleep in people with ASD and Phelan-McDermid syndrome . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Shank3 modulates sleep and expression of circadian transcription factors |
Understanding the relation between genotype and phenotype remains a major challenge . The difficulty of predicting individual mutation effects , and particularly the interactions between them , has prevented the development of a comprehensive theory that links genotypic changes to their phenotypic effects . We show that a general thermodynamic framework for gene regulation , based on a biophysical understanding of protein-DNA binding , accurately predicts the sign of epistasis in a canonical cis-regulatory element consisting of overlapping RNA polymerase and repressor binding sites . Sign and magnitude of individual mutation effects are sufficient to predict the sign of epistasis and its environmental dependence . Thus , the thermodynamic model offers the correct null prediction for epistasis between mutations across DNA-binding sites . Our results indicate that a predictive theory for the effects of cis-regulatory mutations is possible from first principles , as long as the essential molecular mechanisms and the constraints these impose on a biological system are accounted for .
The interaction between individual mutations – epistasis – determines how a genotype maps onto a phenotype ( Wolf et al . , 2000; Phillips , 2008; Breen et al . , 2012 ) . As such , it determines the structure of the fitness landscape ( de Visser and Krug , 2014 ) and plays a crucial role in defining adaptive pathways and evolutionary outcomes of complex genetic systems ( Sackton and Hartl , 2016 ) . For example , epistasis influences the repeatability of evolution ( Weinreich et al . , 2006; Woods et al . , 2011; Szendro et al . , 2013 ) , the benefits of sexual reproduction ( Kondrashov , 1988 ) , and species divergence ( Orr and Turelli , 2001; Dettman et al . , 2007 ) . Studies of epistasis have been limited to empirical statistical descriptions , and mostly focused on interactions between individual mutations in structural proteins and enzymes ( Phillips , 2008; Starr and Thornton , 2016 ) . While identifying a wide range of possible interactions ( Figure 1 ) , these studies have not led to a consensus on whether there is a systematic bias on the sign of epistasis ( Lalić and Elena , 2013; Kussell , 2013; Velenich and Gore , 2013; Kondrashov and Kondrashov , 2015 ) , a critical feature determining the ruggedness of the fitness landscape ( Poelwijk et al . , 2011 ) . Specifically , it is only when mutations are in sign epistasis that the fitness landscape can have multiple fitness peaks - a feature that determines the number of evolutionary paths that are accessible to Darwinian adaptation ( de Visser and Krug , 2014 ) . Furthermore , even a pattern of positive or negative epistasis has consequences for important evolutionary questions such as the maintenance of genetic diversity ( Charlesworth et al . , 1995 ) and the evolution of sex ( Kondrashov , 1988; Otto and Lenormand , 2002 ) . While the absence of such a bias does not reduce the effect of epistasis on the response to selection , it does demonstrate that predicting epistasis remains elusive . 10 . 7554/eLife . 25192 . 003Figure 1 . The different types of epistasis between two point mutations . Two point mutations , A and B ( grey ) , individually increase the measured quantitative phenotype ( gene expression , for example ) compared to the wildtype . In this study , we use the multiplicative expectation of how the phenotypic effects of two mutations contribute to the double mutant phenotype , according to which epistasis = fm12 / ( fm1fm2 ) , where fm12 is the relative fluorescence of a double mutant ( m12 ) , and fm1 and fm2 the relative fluorescence of the two corresponding single mutants ( m1 and m2 ) , respectively . An alternative to the multiplicative assumption would be the additive one , in which the effect of the double mutant in the absence of epistasis is the sum of the effects of single mutants . The multiplicative model is a better assumption for gene expression data , as there is a lower limit on this trait ( Cordell , 2002 ) . In the absence of an interaction between mutations ( ‘no epistasis’ scenario , represented by a grey circle ) , the phenotype of the double mutant is the product of the individual mutation . If the effect of the double mutant is greater or lower than the multiplicative expectation , the two mutations are said to be in positive ( blue ) or negative ( orange ) magnitude epistasis , respectively . Sign epistasis ( dark green ) occurs when one mutation has the opposite effect in the presence of the other ( as for mutation B above ) . Reciprocal sign epistasis ( light green ) indicates a situation when both mutations have the opposite effect when in the presence of the other , compared to when they occur independently on the wildtype background . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 003 Scarcity of predictive models of epistasis comes as no surprise , given that most experimental studies focused on proteins . The inability to predict structure from sequence , due to the prohibitively large sequence space that would need to be experimentally explored in order to understand even just the effects of point mutations ( Maerkl and Quake , 2009; Shultzaberger et al . , 2012 ) , let alone the interactions between them , prevents the development of a predictive theory of epistasis ( Lehner , 2013; de Visser and Krug , 2014 ) . In fact , the only predictive models of epistasis focus on tractable systems where it is possible to connect the effects of mutations to the underlying biophysical and molecular mechanisms of the molecular machinery ( Dean and Thornton , 2007; Lehner , 2011 ) ; namely , RNA sequence-to-shape models ( Schuster , 2006 ) , and models of metabolic networks ( Szathmáry , 1993 ) . Even though these studies have provided accurate predictions of interactions between mutations , applying their findings to address broader evolutionary questions remains challenging . For RNA sequence-to-shape models , the function of a novel phenotype ( new folding structure ) is impossible to determine without experiments . In addition , this approach cannot account for the dependence of epistatic interactions on even simple variations in cellular environments , which are known to affect epistasis ( Flynn et al . , 2013; Caudle et al . , 2014 ) . On the other hand , metabolic network models are limited to examining the effects of large effect mutations , like deletions and knockouts , and lack an explicit reference to genotype . In order to overcome the limitations of existing theoretical approaches to predicting epistasis , we focused on bacterial regulation of gene expression as one of the simplest model systems in which the molecular biology and biophysics of the interacting components are well understood . We analyze the effects of mutations in a prokaryotic cis-regulatory element ( CRE ) – the region upstream of a gene containing DNA-binding sites for RNA polymerase ( RNAP ) and transcription factors ( TFs ) . As such , we study a molecular system where an interaction between multiple components , rather than a single protein , determines the phenotype . Promoters that are regulated by competitive exclusion of RNAP by a repressor are particularly good candidates for developing a systematic approach to understanding epistasis as , in contrast to coding regions as well as more complex CREs and activatable promoters ( Garcia et al . , 2012 ) , the phenotypic effects of mutations in binding sites of RNAP and repressor are tractable due to their short length and the well-understood biophysical properties of protein-DNA interactions ( Bintu et al . , 2005b; Saiz and Vilar , 2008; Vilar , 2010 ) . Understanding the effects of point mutations in the cis-element on the binding properties of RNAP and TFs allows for the construction of a realistic model of transcription initiation ( Bintu et al . , 2005a; Kinney et al . , 2010 ) , while providing a measurable and relevant phenotype - gene expression level - for the analysis of epistasis .
Throughout we assume a multiplicative model of epistasis , which defines epistasis as a deviation of the observed double mutant expression level ( relative to the wildtype ) from the product of the relative single mutant expression levels ( Phillips , 2008 ) . It should be noted that there is no a priori expectation for the sign of epistasis , even if most mutations are deleterious: epistasis denotes only deviations from the expected phenotype of the double mutant , and can be either positive or negative ( Figure 1 ) . First , we measured expression levels in the absence of CI ( Figure 3—figure supplement 1a , Figure 3—figure supplement 2a ) . We observe that the majority of double mutants are in negative epistasis ( Figure 3a ) — the observed double mutant expression level is lower than the multiplicative expectation based on single mutant expression levels ( Pearson’s χ21 , 112=43 . 82 , p<0 . 0001 ) . Specifically , we observe negative epistasis in 83% of 113 mutants that display statistically significant epistasis , while 28 double mutants do not display significant epistasis ( Figure 3a , Figure 3—source data 1 ) . 10 . 7554/eLife . 25192 . 005Figure 3 . Epistasis in the absence and in the presence of CI . Points show log10 of expected versus log10 of observed double mutant effects ( each relative to wildtype fluorescence ) for all 141 double mutants , in the ( a ) absence; and ( b ) presence of the CI repressor . The solid line represents no epistasis ( expected equal to the observed double mutant expression ) . Six replicates of each mutant were measured . Bar charts show total number of double mutants exhibiting positive ( orange ) and negative ( blue ) epistasis , while the darker areas represent the number that are significantly different from the null expectation of the model ( no epistasis ) . The data presented in this figure can be found in Figure 3—figure supplement 1 , Figure 3—figure supplement 2 , and Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 00510 . 7554/eLife . 25192 . 006Figure 3—source data 1 . Fluorescence measurements of single and double mutants , and the calculated values for epistasis for the random mutant library . Multiplicative epistasis , both in the absence and in the presence of the repressor CI , for each double mutant from the random mutant library is provided along with the standard deviation for the measurement , the t-test value ( 5 degrees of freedom ) , and the FDR-corrected p value . Double mutants that do not exhibit a significant epistatic interaction are marked in green . Wildtype normalized fluorescence measurement of each single and double mutant , from which the epistasis values were calculated , is also provided for both environments . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 00610 . 7554/eLife . 25192 . 007Figure 3—figure supplement 1 . Relative fluorescence of single mutants . Bars are mean fluorescence relative to wildtype in the a ) absence and b ) presence of the repressor CI . Mean fluorescence shown in ascending order . The dotted line shows the wildtype fluorescence . Error bars are standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 00710 . 7554/eLife . 25192 . 008Figure 3—figure supplement 2 . Relative fluorescence of double mutants . Bars are mean fluorescence relative to wildtype in the a ) absence and b ) presence of the repressor CI . Mean fluorescence shown in ascending order . The dotted line shows the wildtype fluorescence . Error bars are standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 008 Next , we estimated epistasis at high CI concentration , when gene expression depends on the competitive binding between RNAP and CI ( Figure 3b , Figure 3—figure supplement 1b , Figure 3—figure supplement 2b , Figure 3—source data 1 ) . In a repressible promoter , the effects of mutations on the binding of the two proteins have opposite effects on gene expression — a reduction in RNAP binding leads to a decrease in gene expression , while a reduction in CI binding leads to higher expression levels . By comparing epistasis between two environments – absence of CI and high CI concentration – we find that the 141 tested random double mutants show a strong dependence on the environment ( ANOVA testing for a GxGxE interaction: F1 , 280 = 21 . 77; p<0 . 0001 ) , in line with previous observations in another bacterial regulatory system ( Lagator et al . , 2016 ) . Interestingly , 58% of double mutants display a change in the sign of epistasis between the two environments ( Figure 4 ) . Especially prevalent is a switch from negative epistasis in the absence of CI , to positive epistasis in its presence ( Figure 4 ) . Strikingly , the proportion of double mutants exhibiting reciprocal sign epistasis ( when the sign of the effect of each mutation changes in the presence of the other mutation ) is greater in the presence ( 66% ) than in the absence ( 8% ) of CI ( Supplementary file 2 ) . This difference likely arises from the molecular architecture of a repressible strong promoter . Mutations affect the binding of both DNA-binding proteins , but in the presence of CI the effect on the binding of RNAP is only unmasked when CI does not fully bind , a scenario that is more likely in the presence of two mutations . 10 . 7554/eLife . 25192 . 009Figure 4 . Sign of epistasis changes with the environment for most double mutants . Points show the log10 value of epistasis in the absence of repressor , and the difference in the log10 value of epistasis in the presence and the absence of repressor: log10 ( εCI ) – log10 ( εnoCI ) , for all 141 double mutants . Points above the solid diagonal line exhibit positive , while points below exhibit negative epistasis in the presence of the CI repressor . Most mutants have a different sign of epistasis between the two environments ( gray area ) . Bar chart shows total number of double mutants that are always in positive ( orange ) or in negative ( blue ) epistasis , and the total number that changes sign between the two environments ( gray ) . The darker areas in the bars represent the number that are significantly different from the null expectation of the model ( no epistasis ) in both environments . Six replicates of each mutant were measured . The data presented in this figure is calculated from Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 009 In order to understand these observations , we created a model of gene regulation that relies on statistical thermodynamical assumptions to model the initiation of transcription , originally developed to describe gene regulation by the lambda bacteriophage repressor CI ( Ackers et al . , 1982 ) . Importantly , our model is generic , as it does not consider the details of any specific transcription factors involved in regulation . Instead , we model competitive binding between two generic transcription factors that share a single binding site ( Figure 5a ) . The binding of one of these TFs leads to an increase in the gene expression level , in a manner similar to the function of a typical RNAP or an activator . The other is a repressor molecule , the binding of which has a negative effect on gene expression level , achieved by blocking access of the activator to its cognate binding site . In order to draw a parallel to our experimental system , we refer to these two TFs in the generic model as ‘RNAP’ and ‘repressor’ , without actually relying on any specific properties of the two molecules , such as CI dimerization , or cooperative binding of CI dimers to multiple operator sites . 10 . 7554/eLife . 25192 . 010Figure 5 . Overview of the generic model . The theoretical approach used in this study , originally developed to describe gene regulation by the lambda bacteriophage repressor CI ( Ackers et al . , 1982 ) , relies on statistical thermodynamics assumptions to model initiation of transcription . ( a ) In this framework , each DNA-binding protein is assigned a binding energy ( Ei ) to an arbitrary stretch of DNA . Given a set of DNA-binding proteins ( a generic RNAP-like and a generic repressor-like TF , in this case ) and a promoter sequence , a Boltzmann weight can be assigned to any configuration of these TFs on the promoter . By assigning a Boltzmann weight to all configurations , one can calculate the probability of finding the system in a configuration that leads to the initiation of transcription . ( b ) When considering only the binding of a single protein to DNA ( for example ‘RNAP’ only ) , if mutations have a negative effect on protein-DNA binding , the model predicts negative epistasis between them in terms of expression . This prediction arises from the non-linear relationship between binding energy and gene expression pon ( dotted line ) . In this illustration , we show a relative change in binding energy compared to the sequence with highest possible binding , in kT . ( c ) By generalizing the properties of the relationship between binding and gene expression , we conclude that the sign of epistasis depends only on the sign of individual mutation effects ( p1 and p2 ) upon binding . When both ‘RNAP’ and ‘repressor’ are present in the system , epistasis depends on the ‘repressor’ concentration and the magnitude of single mutation effects on ‘RNAP’ and ‘repressor’ binding ( d , e , f , g ) . ( d ) One point mutation negatively affects only ‘RNAP’ binding , while the other only ‘repressor’ binding . ( e ) Under such circumstances , the system shows no epistasis at low ‘repressor’ concentrations , but is in positive epistasis when ‘repressor’ concentration increases . Finally , at very high repressor concentrations , epistasis approaches 0 . ( f ) Point mutations negatively affect both ‘RNAP’ and ‘repressor’ binding . ( g ) Under such conditions , epistasis changes the sign from negative to positive as repressor concentration increases . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 010 In the thermodynamic model of transcription , each DNA-binding protein is assigned a binding energy ( Ei ) to an arbitrary stretch of DNA . In our formulation , we assume that each position along the single DNA-binding site under consideration contributes additively to the global free binding energy – an assumption found to be accurate at least for a few mutations away from a reference sequence ( Vilar , 2010 ) . These energy contributions can be determined experimentally ( Kinney et al . , 2010 ) and are typically represented in the form of an energy matrix . Given a set of DNA-binding proteins ( specifically , their energy matrices ) and a promoter sequence , a Boltzmann weight can be assigned to any configuration of these proteins on the promoter . The Boltzmann weight is proportional to the probability of finding the system in each of the possible configurations . By assigning a Boltzmann weight to all configurations , one can calculate the probability of finding the system in a particular state ( a set of configurations sharing a common property ) . Specifically , one can calculate the probability of finding the system in a configuration that leads to the initiation of transcription ( Figure 5a ) . In our generic model , we consider only a single binding site to which ‘repressor’ and ‘RNAP’ compete for binding . Note that the model does not make any assumptions about the identity of the TFs that are binding DNA and hence does not utilize any specific energy matrix . The model is , therefore , general in nature , relying only on the physical and mechanistic properties of protein-DNA binding . In such a system , three basic configurations are possible: no proteins bound to DNA , only ‘RNAP’ bound , or only ‘repressor’ bound ( Figure 5a ) . Each of these states is assigned a Boltzmann weight ( Z ) based on its free binding energy Ei: 1; [P]e−βEP; and [R]e−βER , respectively , where β is 1/kBT; subscript P refers to ‘RNAP’ , subscript R to the ‘repressor’; [P] and [R] to the exponential of the chemical potential for the ‘RNAP’ and the ‘repressor’ which for simplicity we equate to the concentrations of the two molecules; and Ei corresponds to the change in Gibbs free energy of the reaction of the binding between protein and DNA . Assuming that the system is in thermodynamic equilibrium , we can calculate the probability of finding the system in a configuration leading to transcription ( pON ) – when RNAP is bound:pON=[P]e−βEP1+[P]e−βEP+[R]e−βER The phenotype of a mutant is obtained by calculating pON for a free energy Ei′=Ei+Δ , where Δ represents the effect of the mutation on the binding of the protein to the sequence . The energies of single mutants and double mutants are EPm1=EP+p1 and ERm1=ER+p1; and EPm2=EP+p2 and ERm2=ER+p2; and EPm12=EP+p1+p2 and ERm12=ER+p1+p2 , respectively , where pi stands for the effect of mutation i on the binding of ‘RNAP’ and ri for the effect on ‘repressor’ binding . From these measures of the mutational effects , we calculated epistasis against a multiplicative model , in the same manner as done for the experimental measurements:pONm12=ϵpONWTpONm1pONWTpONm2pONWT With the generic model , we ask only about the sign of epistasis and say that it is positive when ε >1 and negative when ε <1 . The generic model cannot predict the magnitude of epistasis in any particular biological system without accounting for the underlying energy matrices and intracellular concentrations of relevant TFs . As the model does not account for the details of any specific regulatory system , it considers only the direct , primary effects of a mutation on binding affinity ( Bintu et al . , 2005a ) , and does not consider any potential interactions arising from secondary effects , namely the effects of a mutation on the structure of DNA ( Rajkumar et al . , 2013 ) , accessibility to the binding sites ( Levo and Segal , 2014 ) , protein cooperativity ( Todeschini et al . , 2014 ) , looping ( Levine et al . , 2014 ) , or any other potential regulatory structures . Using the generic model , we first studied the effects of mutations only on ‘RNAP’ binding ( in the absence of ‘repressor’ ) , and found that epistasis depends only on the sign of individual mutation effects ( Figure 5 ) . Our model predicts that if mutations have the same sign , they are always in negative epistasis . This prediction arises from the non-linear relationship between binding energy and expression pon ( Figure 5b ) . Namely , when repressor concentration goes to zero , epistasis is negative only when e−p1+e−p2<e−p1−p2 - a condition satisfied only when p1 and p2 have the same sign . Conversely , when the two mutations have a different sign , they will always be in positive epistasis . In general , the physical properties of the relationship between binding and gene expression indicate that the sign of epistasis for any given TF depends only on the sign of individual mutation effects ( p1 and p2 ) upon binding ( Figure 5c ) . Experimental observations do not significantly differ from these predictions for the sign of epistasis ( χ21 , 112=3 . 64 , p=0 . 056 ) , as 96 of the 113 double mutants ( 85% ) that are in significant epistasis in the absence of CI conform to model predictions . Experimental deviations from the generic model predictions ( i . e . displaying positive epistasis when both mutations have the same sign ) could be due to the secondary effects of mutations , as they could affect the general context of RNAP binding ( Rajkumar et al . , 2013 ) , or the ability of CI to bind cooperatively ( Stayrook et al . , 2008 ) . The model also describes patterns of epistasis in the presence of a repressor . By assuming that every point mutation affects the binding of both ‘RNAP’ and ‘repressor’ , we find that the environmentally dependent change in the sign of epistasis depends on the concentrations of ‘RNAP’ and ‘repressor’ , as well as the sign and relative magnitude of individual mutation effects ( Table 1—source data 1 ) . At high ‘repressor’ concentrations , effects of mutations on ‘repressor’ binding dominate over their effects on ‘RNAP binding’ . In these environments , the sign of epistasis depends only on the sign of individual mutation effects on ‘repressor’ binding . 10 . 7554/eLife . 25192 . 011Table 1 . Sign of epistasis in a simple CRE depends on the environment and the sign of individual mutation effects . We consider two environments , one without repressor when mutations affect only RNAP binding , and the other with high repressor concentration . In the first environment , sign of epistasis is determined only by the sign of individual mutation effects on RNAP binding , while in the second environment it is the sign of individual mutation effect on the repressor that matters . For each mutation , the signs ( ‘+’ and ‘-‘ ) represent the sign of its effect on the binding of RNAP ( p ) and repressor ( r ) , respectively . ‘neg -> pos’ and ‘pos -> neg’ represent combinations that display transitions from negative to positive , or positive to negative epistasis , respectively . Certain combinations of mutations are always in negative or always in positive epistasis . The extended version of this table , which does not assume a constant ‘RNAP’ concentration in the cell , is provided in Table 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 01110 . 7554/eLife . 25192 . 012Table 1—source data 1 . General conditions for the sign of epistasis in two environments . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 012p1r1 p2r2−−+−−+++++pos→neg neg→posneg→pospos→negneg→pos −+pos→negalways positivealways negative+−pos→negalways negative−−neg→pos In general , assuming that ‘RNAP’ concentration stays relatively constant ( Raser and O'Shea , 2005 ) allows us to derive how the sign of epistasis depends on repressor concentration ( Table 1 ) . When one point mutation negatively affects only ‘RNAP’ binding , while the other only ‘repressor’ binding ( Figure 5d ) , the system does not exhibit any epistasis when ‘repressor’ concentration is very low , as only one of the mutations affects ‘RNAP’ binding ( Figure 5e ) . As ‘repressor’ concentration increases , the system is in positive epistasis . Finally , at very high ‘repressor’ concentrations , which are probably not biologically relevant , epistasis approaches 0 as the ‘repressor’ binds too strongly . When point mutations negatively affect both ‘RNAP’ and ‘repressor’ binding ( Figure 5f ) , epistasis changes the sign from negative to positive as ‘repressor’ concentration increases ( Figure 5g ) . To intuit this finding , consider two mutations that reduce binding of both ‘RNAP’ and ‘repressor’ . In the absence of ‘repressor’ , when only ‘RNAP’ is present , epistasis will be negative because of the negative curvature of the relationship between expression and binding energy ( Figure 5b ) . But , in the presence of ‘repressor’ , it is the relative magnitude of individual mutation effects that will determine the sign of epistasis . This is because mutations that weaken ‘repressor’ binding increase expression . If the mutation effects are larger on ‘RNAP’ , then the negative epistasis on expression arising from ‘RNAP’ will dominate . When the mutations have a greater effect on ‘repressor’ binding , then negative epistasis on ‘repressor’ binding will dominate and lead to positive epistasis on expression , and hence to a dependence on the environment . At high ‘repressor’ concentration , only the sign of the effects of mutations on ‘repressor’ binding will determine the sign of epistasis . As most experimentally tested mutations reduce both RNAP and CI binding , our model explains the observation that most double mutants change the sign of epistasis between the two environments ( Figure 4 ) . The experimental data from the random mutant library ( Figures 3 and 4 ) shows that the patterns of epistasis between two environments follow the generic model predictions , specifically that epistasis switches sign between environments in many mutants . However , our experimental design , where we only measure gene expression levels , does not allow us to identify the effects of a mutation on CI binding alone . For example , if a mutation decreases gene expression level in the presence of CI , we cannot know if it decreases RNAP binding , increases CI binding , or both . This prevents a more thorough verification of the generic model . In order to independently experimentally validate the generic model predictions ( Table 1 ) , it is necessary to know the effects of CRE mutations on RNAP and CI . To obtain this information , we used the experimentally determined energy matrices for RNAP ( Kinney et al . , 2010 ) and CI ( Sarai and Takeda , 1989 ) , and utilized it to create five random double mutants for each possible combination of single mutation effects shown in Table 1 . Due to the high specificity of binding of both RNAP and CI , we could not identify point mutations that simultaneously improved the binding of both ( Supplementary file 3 ) . Therefore , we validate the model by measuring epistasis in 30 double mutants ( five for each of the six possible combinations of single mutant effects ) in the two environments . We find no difference between the predicted and experimental estimates of the sign of epistasis and its dependence on the two experimental environments ( Pearson’s χ22 , 30=0 . 68; p=0 . 72 ) ( Figure 6 ) . As such , the predictions about the sign of epistasis that arise from the generic model ( Table 1 ) hold true in our experimental system . 10 . 7554/eLife . 25192 . 013Figure 6 . The thermodynamic model accurately predicts sign of epistasis and its environment-dependence . In order to conduct an independent test of the assumptions of the generic model , we expanded the generic model to include specific information about the two TFs relevant to the experimental system – namely , the energy matrices for RNAP ( Kinney et al . , 2010 ) and CI ( Sarai and Takeda , 1989 ) . We could not use the 141 random mutants to validate the model , as most of them contained mutations that were in the regions of the CRE that were poorly characterized by the energy matrices . Therefore , using the energy matrices , we had to create a new library consisting of five random double mutants for each category from Table 1 . As we could not identify any single point mutations that simultaneously improved the binding of both RNAP and repressor , we tested if empirical measurements of epistasis conformed to model predictions in 30 mutants . The model predictions of the sign of epistasis and its environment dependence were based only on the sign of individual mutation effects on RNAP and repressor binding . The location of points corresponds to the experimental measurement of epistasis for each mutant , while the color indicates the model prediction: ( i ) blue - double mutants predicted to be in negative epistasis both in the absence and in the presence of the repressor CI; ( ii ) orange - double mutants that are always in positive epistasis; ( iii ) grey - double mutants predicted to change the sign of epistasis in the two environments . The color intensity indicates significance – lighter shades represent non-significant , darker shades represent significant epistasis in both environments ( see ‘Empirical verification of the thermodynamic model’ section in Online Methods ) . Six replicates of each mutant were measured . The data underlying this figure is presented in Figure 6—source data 1 . The quantitative test of how well the thermodynamic model predicts the magnitude of epistasis in this dataset is presented in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 01310 . 7554/eLife . 25192 . 014Figure 6—source data 1 . Fluorescence measurements of single and double mutants , and the calculated values for epistasis for the validation mutant library . Multiplicative epistasis , both in the absence and in the presence of the repressor CI , for each double mutant from the 30-mutant validation library , is provided along with the standard deviation for the measurement , the t-test value ( 5 degrees of freedom ) , and the FDR-corrected p value . Double mutants that do not exhibit a significant epistatic interaction are marked in green . Wildtype normalized fluorescence measurement of each single and double mutant , from which the epistasis values were calculated , is also provided for both environments . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 01410 . 7554/eLife . 25192 . 015Figure 6—figure supplement 1 . The thermodynamic model predicts the magnitude of epistasis . By incorporating specific information about the biological system studied , in the form of energy matrices for RNAP ( Kinney et al . , 2010 ) and CI ( Sarai and Takeda , 1989 ) , we could test if the model predicts not only the sign , but also the magnitude of epistasis . Linear regression between empirical measurements and the model predictions of epistasis is shown ( dashed line ) for all mutants in Figure 5 that exhibited significant epistasis . Epistasis was estimated in the ( a ) absence; and ( b ) presence of CI . Grey lines show no epistasis ( epistasis value of 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25192 . 015 Furthermore , we tested if a simple thermodynamic model that incorporates the two energy matrices ( Sarai and Takeda , 1989; Kinney et al . , 2010 ) can predict not only the sign , but also the magnitude of epistasis in the two environments . Because such a model depends on the concentrations of RNAP and CI , we estimated the values for these parameters so as to maximize the correlation between model predictions and empirical values of epistasis . When we excluded those double mutants which did not empirically exhibit significant epistasis , we found a significant fit between experimental measurements and model predictions of the magnitude of epistasis in the absence ( F1 , 15 = 9 . 86; p<0 . 01 ) and in the presence of CI ( F1 , 15 = 4 . 59; p<0 . 05 ) ( Figure 6—figure supplement 1 ) . As such , the model predicts not only the general patterns of epistasis ( sign ) , but is also reasonably accurate at predicting its magnitude , which is remarkable since the model does not consider detailed molecular aspects of the experimental system , such as CI dimerization or cooperativity .
The theory we present here , which is based on mechanistic properties of protein-DNA binding without accounting for any details of the molecular system studied , provides an accurate prediction of the sign of epistasis and its environmental dependence for a repressible promoter system - the most common form of gene regulation in E . coli ( ~40% of all regulated genes [Salgado et al . , 2013] ) . Furthermore , the fact that we use a generic model with no reference to any particular empirical measures means that our results are derived from first principles . As such , the presented results should hold as long as the effects of mutations on gene expression are mainly driven by their direct impact on TF-DNA binding , as represented by the energy matrix for a given TF . Under such conditions , the thermodynamic model , rather than the multiplicative ( or additive ) expectation , provides a meaningful null model for the sign of epistasis in CREs . The sign of the deviations from a multiplicative expectation can have important evolutionary consequences , such as for the evolution of sex ( Otto and Lenormand , 2002 ) or the maintenance of genetic variation ( Charlesworth et al . , 1995 ) . A particularly important pattern of epistasis is sign epistasis , where the sign of the effect of a particular substitution depends on the genetic background . Sign epistasis can lead to the existence of multiple optima ( local peaks ) . In the system we analyze here , sign epistasis cannot exist in the absence of a repressor , since there is an optimum binding site sequence and the effects of mutations have a definite sign toward this optimal sequence . In the presence of a repressor , however , sign epistasis is possible ( Poelwijk et al . , 2011 ) . Furthermore , we show that the sign of epistasis very often reverses between environments . This phenomenon , previously observed in a different system ( de Vos et al . , 2013; Lagator et al . , 2016 ) , could alleviate constraints coming from the existence of multiple peaks in a particular environment . The thermodynamic model provides a mechanistic basis for this observation: RNAP and repressor have opposite effects on gene expression and this , when combined with the specific shape of response induced by the thermodynamic model , can lead to the environmental dependence of the sign of epistasis . Our results concern the combined effect of mutations ( epistasis ) on phenotype , as opposed to fitness . Phenotypes logically precede fitness and even though it could be argued that fitness is ‘what matters’ for evolution , since mutations spread in part based on their fitness effects , determining the fitness effects of mutations depends on the environment which may or may not be representative of ‘natural’ conditions . Moreover , knowledge about one environment is hardly informative about the fitness patterns in a novel environment . Our results allow for the prediction of patterns of phenotypic epistasis across different environmental conditions , independent of the selection pressures applied to this phenotype . The evolutionary consequences of these patterns of epistasis can then be inferred from the knowledge ( or assumptions ) of how selection is acting on this phenotype , or in other words , how the phenotype maps onto fitness . In order to predict the sign of epistasis in a regulatory system , the thermodynamic model accounts for the underlying physical mechanisms that impose constraints on the genotype-phenotype map under consideration . Incorporating details of physical and molecular mechanisms into models of more complex regulatory elements , as well as coding sequences ( Dean and Thornton , 2007; Li et al . , 2016 ) , can elucidate how epistasis impacts genotype-phenotype maps and their dynamic properties across environments , helping us to understand the environmental dependence of fitness landscapes .
We developed a system based on the right regulatory element of the lambda phage ( PR ) , in which we decoupled the cis- and trans-regulatory elements ( Figure 2 ) ( Johnson et al . , 1981 ) . A Venus-yfp gene ( Nagai et al . , 2002 ) is placed under the control of the cis-regulatory region containing the PR promoter with two lambda repressor CI-binding sites ( OR1 and OR2 ) . The transcription factor CI represses the PR promoter by direct binding-site competition with RNAP . Separated by 500 random base pairs and on the opposite DNA strand , we placed the cI repressor gene under the control of a PTET promoter ( Lutz and Bujard , 1997 ) , followed by a TL17 terminator sequence . Thus , concentration of CI transcription factor in the cell was under external control , achieved by addition of the inducer anhydrotetracycline ( aTc ) . The entire cassette was inserted into the low-copy number plasmid pZS* carrying kanamycin resistance gene ( Lutz and Bujard , 1997 ) . We created a library of random single and double mutants in the 43 bp cis-regulatory element ( consisting of the RNAP binding site and the two CI operator sites OR1 and OR2 ) using the GeneMorph IITM random mutagenesis kit ( Agilent Technologies , Santa Clara , US ) . PCR products of mutagenesis reactions were ligated into the wildtype plasmid and inserted into a modified Escherichia coli K12 strain MG1655 chromosomally expressing tetR gene from a PN25 promoter . We sequenced ~500 colonies in order to create a library of 141 double mutants for which both corresponding single mutants were also identified ( Supplementary file 1 ) . We identified , in total , 89 mutants carrying only a single point mutation . Four single and four double mutants from the library were randomly selected and the whole plasmid sequenced to confirm that during library construction no mutations were found outside the target regulatory region . We measured fluorescence for each single and double mutant , as well as the wildtype PR promoter system , both in the presence and in the absence of the inducer aTc . Six replicates of each mutant in the library were grown overnight in M9 media , supplemented with 0 . 1% casamino acids , 0 . 2% glucose , 30 μg/ml kanamycin , either without or with 15 ng/ml aTc . Presence or absence of aTc determined the two experimental environments . Overnight cultures were diluted 1000X , grown to OD600 of approximately 0 . 05 , and their fluorescence measured in Bio-Tek Synergy H1 platereader . The measured fluorescence was first corrected for the autofluorescence of the media , and then normalized by the wildtype fluorescence . All replicate measurements were randomized across multiple 96-well plates . All replicates were biological , having been kept separate from each other from the moment that the mutant was cloned and identified through sequencing . Six replicates of each mutant were measured as prior experience with similar datasets in the lab has shown it sufficient to detect meaningful differences between mutants . By using a multiplicative model of epistasis , we calculated epistasis relative to the wildtype as ϵ=fm12/ ( fm1fm2 ) , where fm12 is the relative fluorescence of a double mutant ( m12 ) , and fm1 and fm2 the relative fluorescence of the two corresponding single mutants ( m1 and m2 ) , respectively . In order to determine statistically which double mutants exhibit epistasis ( i . e . ε not equal 1 ) , we conducted a series of FDR-corrected t-tests . The errors were calculated based on six replicates , using error propagation to account for the variance due to normalization by the wildtype . Variance is not significantly different between measured mutants ( Figure 3—figure supplement 1; Figure 3—figure supplement 2 ) . We performed a Pearson’s chi-squared test to determine if double mutants had a tendency toward negative epistasis . We asked whether epistasis depended on the environment ( defined as presence or absence of the repressor ) by testing for a genotype x genotype x environment ( GxGxE ) interaction using ANOVA . We also tested if the experimental observations of the sign of epistasis in the absence of CI repressor corresponded to model predictions . To do that , we used the experimental measurements of the sign of single mutation effects to predict the sign of epistasis ( if both mutations had the same sign then epistasis was predicted to be negative , if they differed in sign , it was predicted as positive ) . Then we compared the predicted distribution of the sign of epistasis to the experimental estimates using a chi-squared test , limiting the test to only those double mutants that experimentally exhibited significant epistasis . For all tests , data met the assumptions , and variance between groups was not significantly different . The model is based on previous thermodynamic approaches ( Bintu et al . , 2005a , 2005b; Hermsen et al . , 2006 ) . These models consider all possible promoter occupancy states and assign a Boltzmann weight to each state . The probability of any microstate ( promoter configurations ) is given by Boltzmann weights Wi=e−β ( Ei−Nμ ) , where Ei is the Gibbs free energy of the configuration , N is the number of TF molecules , β is 1/KBT , and μ represents the chemical potential . Pon can then be calculated as the normalized sum of all configurations conducive to the initiation of transcription:pON=∑i∈⊕wi∑iwi where the first summation is over the all configurations conducive to transcription , whereas the second is over all configurations . In our model , we consider a scenario in which an activator ( such as RNAP ) competes with a repressor for access to its binding site . We consider only three possible promoter configurations: the one where neither of the two proteins is bound , the one in which a ‘repressor’ prevents ‘RNAP’ from accessing its binding site , and the one in which’ RNAP’ is bound to its binding site , thereby able to initiate transcription . Under these assumptions , the probability of initiation of transcription is:pON=[P]e−βEP1+[P]e−βEP+[R]e−βER where [P] and [R] represent the exponential of the chemical potential for the ‘RNAP’ and the ‘repressor’ , respectively; and subscripts P and R represent ‘RNAP’ and ‘repressor’ , respectively . Throughout , we measure free energies in natural units such that β = 1 . We assume that mutations simultaneously affect the binding of both ‘RNAP’ and ‘repressor’ to the DNA binding site . We denote the free energies of both ‘RNAP’ and ‘repressor’ binding to DNA by EP and ER , respectively . We model the effect of mutations by perturbing these energies by an additive factor . The energies of single mutants and double mutants are then EPm1=EP+p1 and ERm1=ER+p1; and EPm2=EP+p2 and ERm2=ER+p2; and EPm12=EP+p1+p2 and ERm12=ER+p1+p2 , respectively , We calculate epistasis against a multiplicative model for the effect of mutations on pON:pONm12=εpONWTpONm1pONWTpONm2pONWT and so epistasis is measured by:ε=pONWTpONm1pONm12pONm2= ( 1+Ae−r1+Be−p1 ) ( 1+Ae−r2+Be−p2 ) ( 1+A+B ) ( 1+Ae−r1−r2+Be−p1−p2 ) where A=[R]e−ER and A=[P]e−EP . We say that epistasis is positive when ε >1 and negative when ε <1 . We then find the conditions for which epistasis is positive in the presence ( A > 0 ) or absence ( A = 0 ) of repressor . In order to empirically test the predictions of the generic model on the relationship between sign of individual mutations and the sign of epistasis in two environments , we aimed to select five random double mutants from each category from Table 1 . Effects of mutations on RNAP and on CI were obtained from the experimentally determined energy matrices of RNAP ( Kinney et al . , 2010 ) and CI ( Sarai and Takeda , 1989 ) binding . We could not validate the model from the random mutant library , as the majority of mutants fell in regions that are poorly described by the energy matrices . For this reason , we aimed to create this new library . As the PR promoter is very strong , finding double mutants where both mutations improved expression was not possible . Hence , we selected five double mutants from six categories ( Supplementary file 3 ) , and synthesized them , as well as their corresponding single mutants , using annealed oligonucleotide overlap cloning . We measured fluorescence of these mutants and calculated epistasis in the same manner as described for the random mutant library , and we asked if the epistasis for each double mutant was different from the null-expectation in the manner described in section ‘Statistical analyses’ . We used Pearson’s chi-square test to determine if the environmental-dependence of the sign of epistasis in the experimental measurements differs from model predictions . In order to test whether the thermodynamic model can also predict the magnitude of epistasis , we incorporated the energy matrices for RNAP ( Kinney et al . , 2010 ) and CI ( Sarai and Takeda , 1989 ) into the generic model . As the energy matrix for RNAP contained one additional position in the spacer region between −10 and −35 sites compared to the experimental PR system , we eliminated one position in that region that had lowest impact on overall RNAP binding . In the manner described above , we modeled epistasis in those mutants from the 30-mutant validation library that exhibited significant epistasis . As the thermodynamic model depends on the concentrations of RNAP and CI , we estimated the values for these parameters so as to maximize the correlation between model predictions and empirical values of epistasis . In order to estimate how well the model predicted the magnitude of epistasis , we fitted a linear regression between experimental measurements of epistasis and the model predictions , both in the absence and in the presence of CI . | Mutations are changes to DNA that provide the raw material upon which evolution can act . Therefore , to understand evolution , we need to know the effects of mutations , and how those mutations interact with each other ( a phenomenon referred to as epistasis ) . So far , few mathematical models allow scientists to predict the effects of mutations , and even fewer are able to predict epistasis . Biological systems are complex and consist of many proteins and other molecules . Genes are the sections of DNA that provide the instructions needed to produce these molecules , and some genes encode proteins that can bind to DNA to control whether other genes are switched on or off . Lagator , Paixão et al . have now used mathematical models and experiments to understand how the environment inside the cells of a bacterium known as E . coli , specifically the amount of particular proteins , affects epistasis . These mathematical models are able to predict interactions between mutations in the most abundant class of DNA-binding sites in proteins . This approach found that the nature of the interaction between mutations can be explained through biophysical laws , combined with the basic knowledge of the logic of how genes regulate each other’s activities . Furthermore , the models allow Lagator , Paixão et al . to predict interactions between mutations in several different environments , such as the presence of a new food source or a toxin , defined by the amounts of relevant DNA-binding proteins in cells . By providing new ways of understanding how genes are regulated in bacteria , and how gene regulation is affected by mutations , these findings contribute to our understanding of how organisms evolve . In addition , this work may help us to build artificial networks of genes that interact with each other to produce a desired response , such as more efficient production of fuel from ethanol or the break down of hazardous chemicals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2017 | On the mechanistic nature of epistasis in a canonical cis-regulatory element |
When a human catches a ball , they estimate future target location based on the current trajectory . How animals , small and large , encode such predictive processes at the single neuron level is unknown . Here we describe small target-selective neurons in predatory dragonflies that exhibit localized enhanced sensitivity for targets displaced to new locations just ahead of the prior path , with suppression elsewhere in the surround . This focused region of gain modulation is driven by predictive mechanisms , with the direction tuning shifting selectively to match the target’s prior path . It involves a large local increase in contrast gain which spreads forward after a delay ( e . g . an occlusion ) and can even transfer between brain hemispheres , predicting trajectories moved towards the visual midline from the other eye . The tractable nature of dragonflies for physiological experiments makes this a useful model for studying the neuronal mechanisms underlying the brain’s remarkable ability to anticipate moving stimuli .
A diverse range of animals are capable of visually detecting moving objects within cluttered environments . This discrimination is a complex task , particularly in response to a small target generating very weak contrast as it moves against a highly textured background . The neural processing underlying this behavior must enhance a localized , weak and variable signal , which may only stimulate one or two photoreceptors in turn . Rather than simply respond reactively , some animals even anticipate a target’s path by predicting its future location . In the vertebrate retina , high initial gain combined with neuronal adaptation and sensitization allows responses from a network of overlapping ganglion cells to ‘keep up’ with the current target location and account for sluggish neuronal delays ( Berry et al . , 1999; Kastner and Baccus , 2013 ) . This encoding anticipates targets moving in a straight line , with trajectory reversals eliciting a synchronous burst of activity from a population of ganglion cells ( Schwartz et al . , 2007; Chen et al . , 2014 ) . However , this anticipation does not use the recent trajectory to extrapolate likely target locations at future times . Rather , the last observed location remains sensitized after the target disappears . This differs from studies of human observers , where a temporarily occluded target results in improved sensitivity at the extrapolated forward location ( Watamaniuk and McKee , 1995 ) . This predictive encoding of future target locations indicates the presence of additional processing mechanisms beyond the retina . Like human ball players , dragonflies also estimate target location , capturing single prey in visual clutter , even amidst a swarm of potential alternatives ( Corbet , 1999 ) . We recently described a ‘winner-takes-all’ neuron in the dragonfly likely to subserve such competitive selection of an individual target , whilst ignoring a distracter ( Wiederman and O'Carroll , 2013 ) . In other animal models , inhibitory circuits drive the selection of salient stimuli ( Mysore and Knudsen , 2013 ) and the direction of attention towards targets is evidenced by modulation of contrast gain ( Moran and Desimone , 1985; Reynolds et al . , 2000 ) . How prediction relates to the selection of salient stimuli is unknown ( Zirnsak et al . , 2014 ) and how selection , prediction and attention are encoded at the neuronal level is an intense topic of scientific investigation . Here we used intact , in vivo , recordings from the system of small target-selective neurons in predatory dragonflies to reveal local changes in sensitivity elicited during target tracking . We show that this involves a large increase in contrast gain just ahead of the target’s most recent location , with suppression in the surround . We investigated the spatial extent , temporal persistence and direction tuning within this region of enhancement . Our data shows that a local increase in gain spreads forward after a delay , even anticipating the path of primers presented to the contralateral eye and moved towards the visual midline . Moreover , the direction tuning shifts to match the prior path . Such response attributes differentiate this neuronal processing from typical models of direction selectivity and are ideally suited for a dragonfly’s predictive pursuit of prey ( Mischiati et al . , 2015 ) .
‘Small target motion detector’ ( STMD ) neurons in the dragonfly , Hemicordulia tau , are tuned to target size and velocity and are highly sensitive to contrast ( O’Carroll 1993 , Wiederman et al . , 2008 , 2013 ) . One identified STMD , CSTMD1 , responds selectively to a small , moving target , even when embedded within natural scenes ( Wiederman and O'Carroll , 2011 ) . CSTMD1 also exhibits a sophisticated form of selective attention . The neuronal response to the presentation of two simultaneously moving targets does not simply result in either neuronal summation or inhibition . Instead , CSTMD1 responds in a winner-takes-all manner , selecting a single target as if the distracter does not even exist ( Wiederman and O'Carroll , 2013 ) . We mapped CSTMD1’s receptive field by measuring spiking activity in response to a single , black , square target ( 1 . 5°x1 . 5° ) moving along trajectories at varying spatial locations in the visual field . In one region , a gridded array ( 10 × 10 ) of short , vertical , target trajectories evoke weak neuronal responses ( Figure 1A ) . For each short 200 ms trajectory , we plot mean spike rate over a 100 ms analysis window ( from 50 to 150 ms ) . This colormap represents the spiking activity in response to short trajectories for each of the 100 corresponding spatial locations . In comparison , we present an identical square target moved along long , vertical trajectories ( Figure 1B , Video 1 ) and segment responses at the same corresponding spatial locations as the short paths in Figure 1A ( mean spike rate over 100 ms bins ) . This reveals higher overall spiking activity in response to the long , continuous target trajectories . Here we investigate this effect of stimulus history by separating trajectories into components; a primer and a probe . Each elicit responses when presented alone , however , the probe’s initial response is affected by the gain induced by a preceding primer ( Figure 1C ) . We note that neuronal responses build slowly over hundreds of milliseconds - a property we have previously termed facilitation ( Nordström et al . , 2011 ) . For the primer & probe condition ( where a primer always precedes the probe stimulus ) responses to the probe are facilitated ( green region , cf . black with blue time courses ) . This facilitatory effect is not simply due to slow kinetics , as both responses have a rapid decay time course when the stimulus ends ( Dunbier et al . , 2012 ) . 10 . 7554/eLife . 26478 . 003Figure 1 . CSTMD1’s receptive field mapped with drifting targets . ( A ) Small targets ( black squares , 1 . 5°x1 . 5° ) move along short trajectories ( 200 ms ) that are both vertically and horizontally offset on a 10 × 10 grid . Pictograms are illustrative and not to scale . The colormap reveals CSTMD1 responses to these stimuli producing an ‘unfacilitated’ receptive field ( 50–150 ms analysis window ) . ( B ) Horizontally offset targets are drifted vertically up the monitor display along long , continuous trajectories eliciting strong , facilitated responses ( 100 ms bins to corresponding spatial locations in A ) . ( C ) Separating long paths into two components ( primer followed by probe ) , allows us to examine the facilitatory effects within a short analysis window ( before the probe self-primes , green region ) . In a single neuron , we examined response time courses ( mean of 140 replicates over two hours ) to repeated probe alone ( blue line ) and primer & probe ( black line ) conditions ( D ) We have previously described facilitated receptive fields in response to targets drifted across the entire visual display . Targets moving rightwards ( vertically offset ) reveal inhibition in one eye’s visual field ( in response to motion from the periphery towards the frontal area ) and excitation in the other ( from frontal to periphery ) . ( E ) The facilitated receptive field mapped with upwards moving targets ( hot colors ) is stronger than the weaker , though similarly shaped , unfacilitated receptive field in A . Targets moved upward in the other visual hemifield inhibit responses to below spontaneous levels ( data trace ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 00310 . 7554/eLife . 26478 . 004Video 1 . Visual stimulus for Figure 1 . The receptive field of CSTMD1 is mapped with a series of targets drifted on short paths ( Figure 1A ) , or a single target drifting across the same location on a long path ( Figure 1B ) . Separating a long target path into two components ( a primer and a probe ) allows us to quantify the facilitation induced by a primer ( Figure 1C ) . All trials were presented in a randomised order . In this video trials are presented without rest periods , whilst in experiments trials were separated by at least 7 s to minimize habituation . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 004 Previously , we have reported receptive fields in their facilitated state ( Dunbier et al . , 2012 ) , mapped using targets moving along either long horizontal ( Figure 1D ) or vertical trajectories ( Figure 1E ) . These reveal CSTMD1’s excitatory receptive field which extends from the dorsal , visual midline to the periphery . Spatial inhomogeneity within this receptive field ( interpolated to reduce binning artefacts ) likely results from underlying dendritic integration and local spatiotemporal tuning differences . In the other visual hemifield ( midline at 0° azimuth ) , a drifting target generates inhibition ( Figure 1D , E ) , with activity suppressed to below spontaneous levels ( 0 spikes/s from a spontaneous activity of 11 ± 4 spikes/s , mean ± std . ) . What is the effect on a 2D array of ‘probe’ responses ( short paths in Figure 1A ) when a long primer is presented along a single , constrained , trajectory immediately preceding each probe ? Such an experiment would provide us with a snapshot of the effect of stimulus history ( the primer target ) on the current receptive field . Figure 2 provides examples of individual , neuronal responses to short target trajectories ( probes , blue arrows ) , both with and without a preceding 1 s duration target trajectory ( primer , black arrow ) . For each probe location ( in a 10 × 10 grid ) , we measured the spike rate within a 100 ms time window ( the green shaded regions in Figure 2 ) . The effect of priming was calculated as the difference ( ∆ spike rate ) between the probe response when preceded by the primer ( ‘primer & probe’ , black and blue arrows ) and the probe alone ( blue arrow ) conditions . In this paradigm , we changed the spatial offset ( jumps ) between primer and probe without any delay ( Figure 2A , B ) or following a 300 ms pause ( Figure 2C , D ) . We also tested a condition where the primer target drifted toward the dragonfly’s midline , through the visual field of the other eye ( Figure 2E , F ) . By ensuring primers did not enter the region of binocular overlap , any changes elicited in the probe locations ( in the opposing eye ) must have traversed brain hemispheres . 10 . 7554/eLife . 26478 . 005Figure 2 . A primer target changes probe responses . ( A ) Example traces of CSTMD1’s response to a probe target alone ( blue arrow ) or following a primer target ( black arrow ) . The effect of the primer is measured as the difference ( ∆ spike rate ) in response activity ( primer & probe – probe alone ) in the corresponding 100 ms window ( green shaded region , with enlarged view on right ) . ( B ) With the primer spatially constrained , we repeat primer & probe and probe alone trials in a gridded array of 100 locations ( 200 trials in total , randomly interleaved ) . ( C , D ) A pause of 300 ms is inserted between the conditions where the primer disappears before probe onset ( i . e . simulating a target occlusion ) . ( E , F ) A primer placed in the visual field of the other eye and moved toward the visual midline tests for information traversing the brain hemispheres . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 005 Figure 3A shows the complete two-dimensional map of primer-induced gain modulation , averaged across repeated intracellular recordings from CSTMD1 in different animals . Receptive fields are perspective-corrected from the dragonfly’s point of view to a dragonfly eye map ( mirrored along the vertical midline ) and smoothed using bicubic interpolation to remove binning artefacts . The contour lines in Figure 3A–C indicate the average unfacilitated responses to the probe alone condition . In primer & probe trials , the primer target moved upwards ( Figure 3A , B , Video 2 ) or rightwards ( Figure 3C ) along different paths in each trial but constrained within a region 5° wide ( indicated by the white outlined box ) . To the CSTMD1 we recorded from ( with its excitatory inputs located in the right mid-brain ) , upward and rightward moving targets represent progressive stimuli ( i . e . moving from front-to-back ) . The small variation in primer path decreased local habituation from a repeating primer running over the exact same trajectory . Probe alone and primer & probe trials were randomly interleaved . The color map reveals the average change in neuronal activity ( ∆ spike rate ) elicited by the spatially constrained primer for each probe location ( primer & probe – probe alone ) . Figure 3A reveals a pronounced ‘focus’ of increased sensitivity just ahead of the final location of the priming target and an extensive region of suppression in surrounding locations ( mean , n = 9 dragonflies ) . Thus , what we have previously referred to as facilitation is a more complex phenomenon - local enhancement with spike rate suppression elicited by probes jumped into the surround . Here we use the term ‘focus’ to refer to both the local enhancement and widespread concomitant inhibition . Such neuronal processing may be indicative of an attentional mechanism , rather than a global arousal or sensitization ( Slagter et al . , 2016 ) . In Figure 3A , we observed a large mean change in spike rate – over 50% increase within the focus center ( p=0 . 0007 , n = 9 ) and up to 50% decrease in surrounding locations ( p=0 . 005 , n = 9 ) . 10 . 7554/eLife . 26478 . 006Figure 3 . A predictive focus facilitates responses to a moving target . ( A ) The probe receptive field in response to short , vertical trajectories is indicated by contour lines ( mean , n = 9 dragonflies ) . The color map shows change in spike rate ( for each location ) due to the immediately preceding primer trajectory that is presented within the white outlined box . The change in spiking activity in the corresponding analysis window reveals >50% enhancement in front of the moving target ( red ) , but suppression in the surround ( blue ) . ( B ) With a 300 ms delay introduced after the primer , the focus spreads forward ( color map , n = 7 dragonflies ) , estimating the theoretical future target location ( white crosshairs ) . ( C ) The primer moves toward the midline in the other eye’s visual field , whilst avoiding binocular overlap . The focus transfers between brain hemispheres , with a spatially-localized enhancement in front of the target and suppression at higher elevations ( color map , n = 7 dragonflies ) . ( D ) We examined the statistical significance of all three mappings ( Figure A-C ) by calculating the effect size at each spatial location ( Cohen’s d ) . We see values within the range ±1 . 8 , well above those considered as large effect sizes ( >0 . 5 ) . For spatial points of interest ( + ) , we calculate the corresponding statistical significance ( P value ) between the primer & probe and probe alone versions ( E ) There is a forward shift in the focus region ( mean ± SEM , p=0 . 03 , n = 12 dragonflies ) following an occlusion ( cf . 100 ms pause , yellow line with 300 ms pause , green line ) . The expected target locations following occlusions are indicated with color crosshairs ( 3° for 100 ms and 9° for 300 ms ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 006 If the primer disappears for 300 ms before each probe , a similarly intense focus is still evident ( Figure 3B , Video 2 ) , but now spread forward in spatial extent ( p=0 . 005 , n = 7 dragonflies ) . The focus seems to account for the expected target location had it continued on its original trajectory ( to a position as indicated by the white cross-hair in Figure 3B , C ) , albeit with an increased uncertainty given its broader spatial extent ( mean , n = 7 dragonflies ) . Moreover , if we move a horizontal primer toward the visual midline in the contralateral eye before it disappears for 300 ms , the focus transfers across the brain to the ipsilateral hemisphere ( Figure 3C ) . We then observed enhancement ( red ) localized to a broad region ahead of the primer trajectory ( p=0 . 004 , n = 7 dragonflies ) , but strong suppression ( blue ) at higher elevations ( p=0 . 02 , n = 7 dragonflies ) . Dragonflies have a small area of binocular overlap between the two eyes corresponding to the frontal/dorsal visual field ( Horridge , 1978 ) . Our stimulus was carefully designed to avoid this region , disappearing just before entering the area of overlap . Therefore , our result cannot be explained by facilitation being regenerated in the ipsilateral eye . Rather it must involve a localized , inter-hemispheric transfer of information . Furthermore , a localized and spatially segregated combination of enhancement and suppression ( red and blue regions in Figure 3C ) cannot be explained by a simple global mechanism , such as , a post-inhibitory rebound following a strong inhibitory stimulus ( Bolzon et al . , 2009 ) . This transfer of a predictive focus between brain hemispheres is likely to play a crucial role in the prediction of target location during pursuit flights where the pursuer attempts to fixate the target frontally ( Mischiati et al . , 2015 ) . Prolonged pursuit flights of conspecifics involve highly convoluted paths in which the target may readily cross from one visual field to the other ( Land and Collett , 1974 ) . In Figure 3D , we show the effect sizes of the three maps ( Figure 1A–C ) at all spatial locations . These Cohen’s d values are the mean differences between primer & probe and probe alone ( ∆ spike rate ) , divided by the standard deviation of these differences . Cohen’s d values over 0 . 5 are considered large effect sizes , thus our values of up to 1 . 8 in both excitatory and inhibitory directions are considerable . For particular points in these maps ( Figure 3D , +’s ) we calculate the paired , two-tailed P-values , highlighting the statistically significant effect of the primer . 10 . 7554/eLife . 26478 . 007Video 2 . Visual stimulus for Figure 3 . The unfacilitated receptive field is mapped by a 10 × 10 grid of probes moving on short paths ( Figure 3A , contour lines , only five trials shown in this video ) . A primer drifts on a long trajectory towards the center of the screen , before repeating presentation of each probe . Identical trials are replicated with a 300 ms pause separating the primer and probe ( Figure 3B ) . All trials were presented in a randomised order , separated by rest periods of at least 7 s . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 007 Figure 3E shows data from an additional 12 dragonflies , where we mapped the forward-spreading focus following a delay of either 100 ms or 300 ms along a single dimension . These data ( mean ± SEM ) show that a small ( 6° ) jump backwards precisely over the previously primed path already resets the response magnitude to that of the unfacilitated response ( dashed line ) , whilst larger jumps backwards ( 12° ) reveal potent suppression . Considering that the largest jump in this case is stimulating a part of the receptive field that last saw the target up to 700 ms earlier , the profound inhibition seen for this stimulus suggests that the prior primer target exerts long-lasting effects on the surrounding receptive field . Targets that jump forward after a delay reveal a shift in facilitation , spreading further forward after 300 ms ( green line ) , compared to 100 ms ( yellow line ) . Examining the mean difference combined across all forward jumps ( 6° , 12° and 18° ) reveals a statistically significant difference between 100 ms and 300 ms pauses ( p=0 . 03 , Cohen’s d = 0 . 7 ) . Here the probe target followed directly ‘on path’ to the priming stimulus , without the small horizontal offsets ( up to 5° , Figure 3A–C white priming region ) used previously to limit local habituation . In another eight dragonflies , instead of constraining the position of our primer , we instead tested responses to probes that always landed at the same location following different primers . This stereotyped probe followed either a jump in space , a pause in time , or a combination of both tests . Figure 4A–D show normalized response time-courses from individual CSTMD1 examples . The small 4° instantaneous jump ahead of the primer leads to a response time course with a very rapid rise to a level similar to the fully facilitated state ( Figure 4A ) . However , a 12° instantaneous jump elicits a similar ( slower ) response time course to the unfacilitated probe ( grey line ) , confirming the limited extent to which facilitation initially extends ahead of the target path . A large 20° jump ahead ( Figure 4A ) bypassing the focus-region entirely , again reveals surround suppression , with a much slower response time course than the control . Instantaneous backwards jumps ( Figure 4B ) also reveal potent suppression . 10 . 7554/eLife . 26478 . 008Figure 4 . Spatial jumps and temporal pauses in target trajectories . ( A ) CSTMD1’s normalized response to a short probe trajectory builds over several hundred milliseconds ( grey line ) and is changed by the position and timing of a 500 ms priming target . Probes jumped forward immediately following the primer , reveal kfacilitated responses ( 4° ahead ) , unfacilitated responses ( 12° ahead ) and suppression ( 20° ahead ) , indicative of the focus-region in Figure 3 . ( B ) A jump immediately back over the primer path exhibits unfacilitated ( 4° behind ) or strongly inhibited ( 12° or 20° behind ) responses . ( C ) Inserting a temporal pause between primer and probe shows that weaker facilitation persists at the primed location for over 500 ms , diminishing as the pause duration increases . ( D ) Combining a short pause with a jump reveals a forward spread of facilitation that could account for an occlusion . ( A-D , n = 9 technical replicates from one dragonfly ) ( E ) At the target’s last seen position ( jump size 0° ) , probe responses decrease at times following the primer’s disappearance ( p=0 . 0005 ) . In comparison , responses to probes jumped 12° and 20° ahead increase when matched to their corresponding occlusion durations of 300 ms ( p=0 . 008 ) and 500 ms ( p=0 . 008 ) . Asterisks indicate significance , n = 8 dragonflies . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 008 Pauses without a jump ( 0° ) , show that facilitation strength slowly decays over time at the last seen location of the target ( Figure 4C , E , Cohen’s d = 4 . 48 ) . With no pause ( 0 ms ) , the strongest responses occur 4° in front of the moving target ( Figure 4E ) as observed in the 2D receptive fields ( Figure 3A ) . Given that the target moves at 40°/s , it would have traversed 4° , 12° and 20° during pause ‘occlusions’ of 100 ms , 300 ms and 500 ms respectively . When larger jumps are matched to their respective pauses as might be expected during trajectory occlusions ( 12° and 300 ms; 20° and 500 ms ) there is a statistically significant increase in the resultant spiking activity ( Figure 4E , Cohen’s d = 2 . 0 and 2 . 32 respectively ) . The data presented so far make a strong case for a complex predictive mechanism working to boost responses in a region where a target seen in the recent past is likely to move to in the near future . In primates , one known effect of such attentional or expectation effects is an upregulation of local contrast sensitivity ( gain control ) ( Reynolds et al . , 2000; Carrasco et al . , 2000 ) . To quantify changes in gain , we measured responses to varying contrast probes , preceded by either a low or high contrast primer ( Figure 5A , B , Video 3 ) . Both primers induced a large increase in response , with a larger output range ( increased maximum response ) and a greater than 5-fold increase in contrast sensitivity ( Figure 5C , contrast threshold reduced from 0 . 071 to 0 . 013 for near threshold stimuli , C50 from 0 . 36 to 0 . 13 , n = 9 dragonflies ) . Lower contrast primers themselves induce less overall activity during the priming stimulus ( Figure 5D , Cohen’s d = 0 . 97 ) , yet their effect on subsequent responses to stimuli presented at the expected location is remarkably similar to high contrast primers ( cf . pink with red lines in Figure 5C ) . This suggests that the gain modulation is not elicited solely by the stimulus contrast or the neuronal activity induced by the primer per se , but rather by target presence . This may indicate a ‘switch’ process , such as that suggested for neural circuits in the auditory brain stem of the barn owl ( Mysore et al . , 2011 ) , rather than a simpler , activity-dependent gain control mechanism . Another interesting feature of the facilitated contrast sensitivity is that the boost of response gain is largest at mid-contrast , with softer saturation at high contrasts , extending the range of contrasts over which the response is modulated by a full order of magnitude . Both observations make sense considering the natural context for target detection . During pursuit flights , resources could thus be directed to the expected target location independent of its varying contrast as it moves across a cluttered background . Moreover , the reduction in slope of the contrast sensitivity function would reduce overall response variance to changes in the contrast of the selected target , a phenomenon also observed in humans ( Avidan et al . , 2002 ) . 10 . 7554/eLife . 26478 . 009Figure 5 . Low or high contrast primers increase probe contrast sensitivity . ( A ) Either a low or high contrast primer is presented before varying contrast probes ( contrast sensitivity function ) . These either continue the path trajectory or jump to a distant location . ( B ) Example data traces of responses to either low ( grey ) or high ( black ) contrast primers that are presented before a series of varying contrast probes ( light , medium and dark blue ) ( C ) CSTMD1’s sensitivity to varying contrast probes exhibits a sigmoidal function ( grey ) , with the dashed line indicating a detection threshold above spontaneous levels . Following either a nearby low contrast ( pink ) or high contrast ( red ) primer , contrast sensitivity is substantially increased ( n = 9 dragonflies , p<0 . 0001 ) . A distant primer ( yellow ) does not elicit facilitation , even though spiking activity during low and high contrast primers ( final 100 ms ) is significantly different ( inset , n = 9 dragonflies , p=0 . 02 ) . ( D ) In response to an excitatory stimulus ( e . g . high contrast stimulation ) , the underlying membrane potential is hyperpolarized , a form of motion-after-effect ( MAE ) . ( E ) The hyperpolarizing motion-after-effect is related to the strength ( e . g . target contrast ) of the stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 00910 . 7554/eLife . 26478 . 010Video 3 . Visual stimulus for Figure 5 . Probes of varying contrast drift on short paths to determine unfacilitated contrast sensitivity ( Figure 5C , only 1 contrast shown in video ) . Probes are then preceded by a high contrast primer that drifts at the same horizontal path as the probe - High Contrast Primed ( local ) , or on a different horizontal path - High Contrast Primed ( distant ) . Primed trials are also repeated with lower contrast primers ( Low Contrast Primed ) . All trials were presented in a randomised order , separated by rest periods of at least 7 s . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 010 Our results also show that the increased contrast sensitivity is localized to the focus-region evident in Figure 3 . A more distant primer displaced 20° to the side of the probe does not evoke facilitation of the contrast sensitivity function ( Figure 5C ) . Instead , the contrast sensitivity function reveals a weaker effect of the surround suppression observed in the 2-D receptive fields ( Figure 3 ) . These contrast experiments used a shorter primer duration ( 600 ms vs . 1 s ) , suggesting that suppression could result in part from CSTMD1’s global activity , rather than presynaptic processing . Facilitating stimuli certainly increase the firing rate against a steadily hyperpolarizing membrane potential ( Figure 5D ) . Following a high contrast primer , this hyperpolarizing motion-after-effect ( MAE ) reaches almost 4 mV and suppresses subsequent spiking activity for several hundred milliseconds ( Figure 5D , E ) , an attenuation that may compete with spatially-localized facilitation . Interaction between the facilitation time course and longer-term suppression with slow kinetics may be analogous to the ‘inhibition of return’ observed in human reaction time experiments ( Posner and Cohen , 1984 ) . The facilitated response of CSTMD1 appears to be only weakly direction-selective when stimulated with targets moving along prolonged paths . Within each hemifield , CSTMD1 has a weak preference for progressive motion upwards and away from the midline ( rightwards for the neurons recorded here ) ( Nordström et al . , 2011 ) . To test whether the focus also anticipates the direction of a moving target , we presented a primer moving along one of four cardinal directions , followed by a probe that moves in eight possible directions ( Figure 6A , B , Video 4 ) . Probe responses alone are both weak and weakly direction selective ( Figure 6C , grey dots ) . But all four primers facilitate responses maximally in the direction of the primer’s path , shifting the direction tuning to match that of the primer ( Figure 6C ) . The b1/b0 ratio is a measure of the strength of directionality which is similar for each of the conditions ( Figure 6D ) . However , the magnitude of facilitation ( Figure 6E ) is considerably larger in CSTMD1’s weak preferred , direction ( upwards and rightwards for this hemisphere’s CSTMD1 ) . Such targets would be those moving away from the dragonfly’s own heading ( Olberg , 1986 ) with the mirror-symmetric CSTMD1 expected to exhibit directional preference to progressive targets moving upwards and to the left . This suggests that the preference of both the underlying tuning and the recruitment of facilitation may be linked to a control role in downstream processing of target trajectories for pursuit . Following a reversal of the target trajectory ( Figure 6F , blue and purple lines ) , CSTMD1’s response is strongly inhibited compared to the corresponding probe alone response ( grey lines ) . This contrasts with findings in the vertebrate retina , where a subset of ganglion cells respond strongly and synchronously to motion reversals ( Schwartz et al . , 2007 ) . 10 . 7554/eLife . 26478 . 011Figure 6 . Primer direction establishes probe direction selectivity . ( A ) Primers of four possible directions ( right , upward , left , downward ) preceded probe responses in each of eight possible directions . ( B ) Examples of individual traces to a subset of the experiment conditions . The analysis period is indicated in green . ( C ) Probe responses are weak ( grey points ) until following a primer ( in one of four cardinal directions ) and are most facilitated in the primer’s direction ( mean ± SEM , n = 9 dragonflies ) . ( D ) The b1/b0 is an index showing the strength of directionality . ( E ) Polar plot vector magnitude and direction ( mean ±95% CI ) , shows that probe direction selectivity generally aligns with the primer direction . ( F ) Either upward or downward probe alone ( grey lines ) evoke robust responses . However , ‘reversals’ ( probes opposite in direction to a preceding primer ) generate strong and long-lasting inhibition ( mean time course , n = 9 dragonflies ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 01110 . 7554/eLife . 26478 . 012Video 4 . Visual stimulus for Figure 6 . Probes drift in 8 unique directions to determine the unfacilitated direction tuning of CSTMD1 ( Figure 6C , only 4 directions shown in the video ) . The same 8 probes are preceded by primers moving in each of 4 cardinal directions ( only upwards primer shown ) . All trials were presented in a randomised order , separated by rest periods of at least 7 s . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 012 Does the recruitment of enhanced responses in the direction of travel represent an alteration of the direction selectivity in underlying local motion detectors , or does it result from the offset position of the focus of gain modulation located just ahead of the most recent target location ( Figure 3A ) ? We tested this by jumping the probe stimulus 4° forward into the predicted center of the focus region ( Figure 7A , Video 5 ) . This stimulus induced much weaker direction selectivity ( Figure 7B ) than those that radiate away from the end of the same priming path ( b1/b0 ratio of 0 . 32 vs . 0 . 50 , p=0 . 04 ) . Probes that reverse direction relative to the primer are not facilitated , except when the probe jumps 4° into the focus center ( Figure 7C , Cohen’s d = 3 . 90 ) . Thus , the predictive focus of gain modulation is a spatial phenomenon , established by the past trajectory . This suggests that the apparent direction selectivity induced by primers is not due to any change in the local bias of underlying motion detectors to any one stimulus direction , but rather from the overall displacement of the focus ahead of the target location . Over a target’s developing trajectory , direction selectivity ( quantified here as vector magnitude ) is established even more rapidly than the gain in the facilitated response ( Figure 7D ) . The emergence of directional tuning raises the intriguing possibility that the modulation assists anticipation of target trajectories - promoting the expectation of a continued path . How such tuning matches closed-loop pursuits of the hawking dragonfly with its prey or conspecifics is not yet known . 10 . 7554/eLife . 26478 . 013Figure 7 . Direction selectivity is a result of spatial facilitation . ( A ) The direction experiment is repeated , now with a 4° jump forward into the spotlight . ( B ) Responses are facilitated for all directions ( mean ± SEM , n = 5 dragonflies ) with decreased direction selectivity ( b1/b0 ) . ( C ) Probes in the opposite direction to their corresponding primer reveal no facilitation or inhibition , except when jumped [J] into the spotlight ( p=0 . 03 ) . ( D ) The magnitude of direction selectivity builds on a faster timescale than the response onset . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 01310 . 7554/eLife . 26478 . 014Video 5 . Visual stimulus for Figure 7 . Probes are presented in an identical manner to Figure 6C . However , here probes are preceded by a vertical primer that terminates 4° below the probe start location ( Figure 7B ) . All trials were presented in a randomised order , separated by rest periods of at least 7 s . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 014 CSTMD1 is a higher-order neuron with inputs in the anterior optic tubercle , a midbrain output destination of optic lobe interneurons . We also tested for the facilitatory component of the predictive gain modulation in likely pre-cursor neurons: small-field ( SF ) STMDs located at an earlier stage of visual processing ( Barnett et al . , 2007 ) . Retinotopically organized SF-STMDs have inputs in the outer lobula , a region akin to mammalian primary visual cortex ( Okamura and Strausfeld , 2007; O’Carroll 1993 ) . They have properties similar to end-stopped ( hypercomplex ) cells ( Nordström and O'Carroll , 2009 ) , which are modulated by contextual stimuli presented outside their classical receptive field ( Polat et al . , 1998 ) . We presented primers outside SF-STMD receptive fields , that themselves induce no activity above spontaneous levels ( Figure 8A , B , Video 6 , n = 13 dragonflies ) , with probe stimuli that are limited to the classical ( excitatory ) receptive field . Primers moving toward the receptive field facilitate the probe responses by over 40% , whilst those heading away elicit no facilitation . This predictive gain modulation may be inherited and improved downstream , since we also observe facilitation in other large-field STMD neurons , with an average gain of over 80% ( Figure 8C , Cohen’s d = 1 . 02 ) . Individual responses of both small and large field STMDs vary in facilitation strength , as well as overall activity . The retinotopic organization ( Figure 8D ) and facilitation observed in SF-STMD’s make them ideal candidates for mediating an interhemispheric transfer of localized predictive gain modulation . Supporting this hypothesis , at least one identified ( dye-filled ) SF-STMD axon traverses the brain with an output arborization located within a limited area of the contralateral lobula ( Figure 8D ) . Neurons such as this are thus perfectly suited for the spatially localized inter-hemispheric modulation , both excitatory and inhibitory shown in Figure 3C . 10 . 7554/eLife . 26478 . 015Figure 8 . SF-STMDs are facilitated by a primer that moves toward the receptive field . ( A ) Primers move either toward ( red ) or away ( yellow ) from the classical receptive field ( RF ) , preceding a probe target within the RF ( mean , n = 13 dragonflies ) . ( B ) Outside the receptive field , primer responses do not significantly differ from spontaneous activity . Primers that move towards the receptive field increase probe responses by over 40% ( p=0 . 0004 , n = 13 dragonflies ) . ( C ) Individual STMDs , with either small or large receptive fields , exhibit varying degrees of facilitation ( blue ) . Mean facilitation ( black ) increase responses by over 40% in small-field ( n = 13 dragonflies ) , 80% in large-field STMDs ( n = 11 dragonflies ) and 50% in CSTMD1 ( data not shown ) . ( D ) Six small-field STMD receptive fields ( RF ) are predominantly fronto-dorsal and exhibit variation in overall size and spatial locations . Contour lines represent 25 spikes/s . The SF-STMD with light purple contours is the same neuron in E , with inputs in the binocular region of the dragonfly’s right visual field , whilst input dendrites are in the left hemisphere ( E ) An SF-STMD’s axon traverses the brain , potentially underlying transfer of local predictive gain modulation . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 01510 . 7554/eLife . 26478 . 016Video 6 . Visual stimulus for Figure 8 . Probes are presented within the receptive field of a Small-Field STMD neuron . Probes are preceded by primers that either drift towards the receptive field , or away from the receptive field ( Figure 8A ) . All trials were presented in a randomised order , separated by rest periods of at least 7 s . DOI: http://dx . doi . org/10 . 7554/eLife . 26478 . 016
Neuronal receptive fields are defined by their excitatory and inhibitory responses to stimulation . Populations of such responses elucidate network function , for example , as control systems in insect flight behavior ( Gonzalez-Bellido et al . , 2013; Maisak et al . , 2013 ) . However , our results show that in addition to stimulus selectivity ( contrast , size , velocity ) , a neuron’s receptive field is also a dynamic representation of the spatial ( Wiederman and O'Carroll , 2013 ) and temporal context . Here modulation of the dynamic receptive field represents anticipatory coding , a more complex influence of history than simple neuronal adaptation , sensitization , habituation or fatigue . Indeed , such complexity in processing is also evident in the ‘omitted stimulus response’ in the vertebrate retina , where an omitted component of a periodic pattern predictively elicits robust neuronal activity ( Schwartz et al . , 2007 ) . These examples highlight that the brain is a ‘predictive machine’ ( Rao & Ballard 1999 ) . However , instead of encoding novelty or the unexpected , STMD neurons predict consistency of a selected target’s trajectory , all whilst suppressing distracters . Direction selectivity is , in effect , a simple form of prediction . For example , the Hassenstein-Reichardt correlator provides a nonlinear , facilitated response when an adjacent point is stimulated within a future period ( Hassenstein and Reichardt , 1956 ) . However , such direction selective models cannot explain the observation of a traveling gain modulation that spreads further forward , the longer the occlusion . Neither can these models account for changes in preferred direction , determined by the target’s previous direction of travel . Such models do not result in a contrast invariant ‘switch’ establishing the focus strength , nor the presence of a large suppressive surround . Furthermore , the effects described here are on larger scales either spatially ( tens of degrees ) or temporally ( hundreds of milliseconds ) compared with local motion detection processes , such as optic flow analysis ( tens of milliseconds , Guo and Reichardt , 1987 ) . Finally , our results show a local , predictive focus of facilitation that traverses across brain hemispheres , which is an attribute more reminiscent of higher order attentional networks , rather than local motion encoding circuitry . Our findings of over a 400% increase in contrast sensitivity is consistent with studies that cue spatial attention in vertebrates , albeit with a significantly larger increase . For example , the contrast gain of human observers is increased by approximately 40% for stimuli presented at an attended location ( Carrasco et al . , 2000 ) , with concurrent decreased contrast gain for stimuli presented elsewhere ( Pestilli and Carrasco , 2005 ) . Similar results are also observed in single unit recordings from macaque V4 , where gratings presented at attended locations elicit responses equivalent to a 51% increase in stimulus contrast ( Reynolds et al . , 2000 ) . Whilst there is ongoing debate over whether attention modulates contrast gain ( Reynolds et al . , 2000 ) or response gain ( Lee and Maunsell , 2010 ) , the facilitation in CSTMD1 reveals a combination of both ( Figure 5C ) . CSTMD1’s gain modulation could be an inherent component of the prediction mechanism , or the result of the priming target acting as a cue directing attention to the targets predicted location . We have previously reported that CSTMD1 selectively attends to one target when presented with a pair of competing stimuli , completely ignoring the distracter ( Wiederman and O'Carroll , 2013 ) . In repeated trials , the selected target was not always the same and even occasionally switched mid-trial . This raises the intriguing possibility that the predictive focus ‘locks on’ to a single target , suppressing distracters . The anticipatory gain control measured here provides a possible explanation for this behavior – a positive feedback that allows the neuron to lock onto a single object while other mechanisms , including global inhibition , may help suppress competing objects . Future experiments will address the parameters of the stimuli ( e . g . timing , salience ) that permit the predictive focus to switch between alternative targets . Furthermore , we are currently investigating whether the predictive focus and competitive selection is elicited bottom-up by the stimuli ( exogenous ) or includes a top-down component ( endogenous ) . That is , for a dragonfly feeding in a swarm , are target saliency attributes driving pursuit selection , or is the dragonfly choosing its prey from more complex internal workings ? For decades , scientists studied the neuronal basis of ‘elementary motion detection’ in true flies ( Diptera ) . With morphological ( Takemura et al . , 2013 ) and physiological ( Maisak et al . , 2013 ) experiments making significant progress at elucidating this circuitry , increasing attention is now shifting towards other visual tasks such as feature discrimination ( Aptekar et al . , 2015; Keleş and Frye , 2017 ) . Until now , there has been a divide between such ‘simple’ visual operations and higher-order processing observed in mammals . Our results reveal the dragonfly as a surprisingly sophisticated , yet tractable model , permitting investigation of fundamental physiological and morphological principles underlying neuronal prediction and selective attention .
We recorded from a total of 63 , wild caught male dragonflies , Hemicordulia . Animals were immobilized with a wax-rosin mixture ( 1:1 ) and fixed to an articulating magnetic stand . The head was tilted forward and a small hole dissected in the posterior surface , exposing the left optic lobe . We pulled Aluminium silicate electrodes on a Sutter Instruments P-97 electrode puller , and backfilled them with 2M KCl solution . Electrodes were inserted through the neural sheath into the proximal lobula complex using a piezo-electric stepper ( Marzhauser-Wetzlar PM-10 ) , with typical resistance between 50–150 MΩ . Intracellular responses were digitized at 5 kHz with a 16-bit A/D converter ( National Instruments ) for off-line analysis with MATLAB . Freshly penetrated cells were presented with small targets , bars and wide-field gratings for classification . Neurons were classed as STMDs when responding robustly to visual stimuli composed of small , moving targets and not responsive to bars or gratings . CSTMD1 was identified by its characteristic receptive field , response tuning and action potentials . STMD neurons were categorized into small or large-field by mapping their receptive fields with drifting targets ( a half-width either less than , or greater than 25° ) . We presented stimuli on high definition LCD monitors ( 120 Hz and above ) . The animal was placed 20 cm away and centered on the visual midline . Contrast stimuli were presented at screen center to minimize off-axis artefacts . Stimulus scripts ( https://github . com/swiederm/predictive-gain ) were written using MATLAB’s Psychtoolbox and integrated into the data acquisition system ( Wiederman et al . , 2017; a copy is archived at https://github . com/elifesciences-publications/predictive-gain ) . Unless stated otherwise all targets were 1 . 5°x1 . 5° black squares drifted at 40°/s . A minimum of 7 s rest between trials was implemented to avoid habituation or facilitation from prior trials . Data were only ever excluded due to pathological damage of the neuron or extensive habituation ( experiment cessation ) . All means are calculated from biological replicates ( i . e . repeated measurements from identified neurons in different animals ) . Each biological replicate represents the mean of between 1 and 10 technical replicates . The morphology of an SF-STMD neuron was visualized by intracellular labelling with Lucifer Yellow . Iontophoresis was achieved by passing 3nA negative current through electrodes tip-filled with 4% Lucifer Yellow solution in 0 . 1M LiCl . Brains were then carefully dissected , fixed overnight in 4% paraformaldehyde at 4°C , dehydrated in ethanol series ( 70% , 90% , 100% , 100% ) , cleared in methyl salicylate and mounted on a cavity slide for fluorescence imaging . Data obtained is managed per the ARC/NHMRC Australian Code for the Responsible Conduct of Research . Raw data from experimental testing and numerical simulation is stored on a locally managed server . Processed experimental and numerical data is available on the data management server for The University of Adelaide ( https://adelaide . figshare . com ) | Catching a ball requires a person to track the speed and direction of a small moving target often against a cluttered and varying background . Predatory insects , like dragonflies , face a similar challenge when they pursue their prey through the air . The task is made a little easier , however , by the fact that most moving targets tend to follow predictable trajectories . Indeed , animals are also better at tracking targets that follow smooth continuous trajectories , suggesting that brains have evolved to exploit the normal behavior of visual stimuli to reduce their workload To find out how this process works , Wiederman , Fabian et al . studied the brains of dragonflies as they watched a black square intended to mimic prey . Brain cells called Small Target Motion Detectors ( or STMD neurons for short ) became more active in response to the target . But rather than simply following the target , the STMD neurons instead predicted its future location . In fact , individual neurons were more sensitive to movements occurring just ahead of the target’s current position , and less sensitive to movements occurring elsewhere . If the target abruptly disappeared , the point in space where the neurons were most sensitive to movement continued to gradually move forward over time . Given that real-life targets typically disappear when they move behind other objects , this suggests that the brain is predicting where the target is most likely to reappear . The STMD neurons became more sensitive to movement by increasing their ability to detect differences in brightness between the target and its background . In some cases , the neurons increased their sensitivity more than five-fold . Insects and mammals last shared a common ancestor more than 500 million years ago , and , in many respects , mammalian brains are substantially more complex than insect brains . Nevertheless , the results of Wiederman , Fabian et al . show that the insect brain can perform visual tasks that were previously associated only with mammals . Neuroscientists and engineers have used the insect brain for decades to study the circuits that support biological processes . In the coming years , insects such as the dragonfly may enable us to explore how visual regions of the brain predict future events . This knowledge could ultimately be applied to artificial vision systems , such as those in self-driving cars . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | A predictive focus of gain modulation encodes target trajectories in insect vision |
Efficient detection and reaction to negative signals in the environment is essential for survival . In social situations , these signals are often ambiguous and can imply different levels of threat for the observer , thereby making their recognition susceptible to contextual cues – such as gaze direction when judging facial displays of emotion . However , the mechanisms underlying such contextual effects remain poorly understood . By computational modeling of human behavior and electrical brain activity , we demonstrate that gaze direction enhances the perceptual sensitivity to threat-signaling emotions – anger paired with direct gaze , and fear paired with averted gaze . This effect arises simultaneously in ventral face-selective and dorsal motor cortices at 200 ms following face presentation , dissociates across individuals as a function of anxiety , and does not reflect increased attention to threat-signaling emotions . These findings reveal that threat tunes neural processing in fast , selective , yet attention-independent fashion in sensory and motor systems , for different adaptive purposes .
Perceptual decisions rely on the combination of weak and/or ambiguous samples of sensory evidence . The accuracy of this decision process is particularly important for the interpretation of negative signals , which require rapid and adaptive responses . In the social domain , identifying the emotional state of a conspecific – e . g . , is he/she angry or afraid ? – rarely depends solely on facial features , which are usually ambiguous and can imply different levels of threat for the observer . Surrounding cues , such as gaze direction and body posture , are known to act as contextual information during emotion recognition ( Righart and de Gelder , 2008; Barrett and Kensinger , 2010; Aviezer et al . , 2011 ) . Specifically , the detection of anger represents an immediate threat for the observer when paired with a direct gaze; by contrast , it is when paired with an averted gaze that fear marks the presence ( and possibly the localization ) of a threat in the environment ( Sander et al . , 2007 ) . These threat-signaling combinations of gaze direction and emotion have been shown to be better recognized and rated as more intense than other combinations ( Adams and Kleck , 2003 , 2005; Graham and LaBar , 2007; Sander et al . , 2007; Bindemann et al . , 2008; N’Diaye et al . , 2009 ) , and this as a function of anxiety level of the individuals ( Ewbank et al . , 2010 ) . However , the computational mechanisms underlying the prioritization of threat-signaling information remain unspecified . Classical decision theory distinguishes two classes of mechanisms by which contextual information such as gaze direction could influence the recognition of negative emotions . Gaze direction could bias the interpretation of negative facial expressions in favor of the emotion signaling higher threat in this context – anger for direct gaze , fear for averted gaze . In signal detection theoretical terms ( Green and Swets , 1966; Macmillan and Creelman , 2004 ) , this effect would correspond to an additive shift of the decision criterion as a function of gaze direction . However , gaze direction could also increase the perceptual sensitivity to the facial features diagnostic of the emotion signaling higher threat . In contrast to the first account , this effect would correspond to a multiplicative boost of threat-signaling cues in the decision process . While the two accounts predict similar effects of gaze direction on the recognition of threat-signaling emotions , a bias effect would be maximal for neutral ( emotionless ) expressions , whereas a sensitivity effect would be maximal at low emotion strengths ( Figure 1 ) . 10 . 7554/eLife . 10274 . 003Figure 1 . Model predictions for the effect of gaze direction on emotion categorization . Left panel: prediction of an effect of gaze direction on decision bias . Upper left panel: if gaze direction biases the interpretation of negative facial expressions in favor of the emotion signaling higher threat , direct gaze would additively bias the choice selection toward anger . Lower left panel: the predicted psychometric function would accordingly be shifted to the left for direct gaze , as participants will be biased toward interpreting faces displaying a direct gaze as angry . Maximal effects would appear for neutral ( emotionless ) expressions as highlighted through the filled grey area on the emotion axis that represents the difference between the two psychometric functions for direct and averted gaze . Right panel: prediction of an effect of gaze direction on perceptual sensitivity . Upper right panel: if gaze direction increases the sensitivity to the facial features diagnostic of the emotion signaling higher threat , direct gaze would now multiplicatively boost the processing of an angry expression displaying a direct gaze . Lower right panel: the predicted psychometric function would now show an increased slope for threat-signaling emotions , with maximal effects at low emotion strengths ( as shown in the filled grey area on the emotion axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 003 Here we arbitrated between these two possible accounts by recording human electroencephalogram ( EEG ) signals while participants categorized facial expressions as displaying anger or fear . We manipulated emotion strength by presenting ‘morphed’ facial expressions ranging from neutral to intense anger or fear , and contextual information by pairing facial expressions with direct or averted gaze . The parametric control over emotion strength afforded fitting decision theoretical models to the behavioral and neural data to arbitrate between bias and sensitivity accounts of threat-dependent effects on emotion recognition . At the neural level , previous studies have reported interactions between emotion and gaze direction from 200 ms following face presentation ( Sato et al . , 2004; N’Diaye et al . , 2009; Adams et al . , 2012; Conty et al . , 2012 ) , but failed to characterize the computational mechanism responsible for these effects . Here , we applied model-guided regressions of single-trial EEG signals to determine whether the neural ‘encoding’ of threat-signaling emotions is enhanced in ventral face-selective and/or dorsal motor regions ( El Zein et al . , 2015 ) , and whether this enhancement is mediated by increased top-down attention to threat-signaling facial features . As high-anxious individuals show increased sensitivity to threats , but also negative signals in general ( Bishop , 2007; Cisler and Koster , 2010 ) , we further assessed the neural mechanisms by which anxiety influences the detection of and reaction to social threats .
Participants were presented at each trial with a face expressing fear or anger of varying emotion strength ( 7 levels of emotion strength for each emotion ) and had to categorize the displayed emotion ( Figure 2 ) . Crucially , direction of gaze ( direct or averted ) was manipulated independently of the displayed emotion in a completely implicit fashion , as it was never mentioned to the subjects nor relevant to the emotion categorization task . Nevertheless , in addition to an expected increase in categorization performance with emotion strength ( F 6 , 138 = 187 . 3 , p<0 . 001 ) , gaze direction strongly interacted with the displayed emotion on performance ( F1 , 23 = 21 . 2 , p<0 . 001 ) . Facial displays of anger were better categorized when paired with a direct gaze ( t23 = 4 . 3 , p<0 . 001 ) , whereas expressions of fear were better categorized when paired with an averted gaze ( t23 = -3 . 4 , p<0 . 01; Figure 3a ) . These combinations of gaze and emotion , anger paired with a direct gaze and fear paired with an averted gaze , are associated with higher threat for the observer ( Sander et al . , 2007 ) , albeit of different natures . In the case of anger , gaze direction indicates the target of the threat , while in the case of fear gaze direction signals its source . Nevertheless , just as the combination of anger with a direct gaze is more threatening/relevant than with an averted gaze , fear is more threatening when paired with an averted gaze than with a direct gaze . These two combinations , anger direct and fear averted , will thus be labeled as THREAT+ combinations as opposed to THREAT− combinations ( i . e . , anger paired with averted gaze , and fear paired with direct gaze ) . 10 . 7554/eLife . 10274 . 004Figure 2 . Stimuli and experimental procedure . ( a ) Examples of morphed expressions for one identity: morphs from neutral to intense fearful/angry expressions providing evidence for one or the other emotion . Stimuli displayed either an averted or a direct gaze . THREAT+ conditions ( in orange ) correspond to combinations of gaze and emotion that signal higher threat for the observer as compared to THREAT− conditions ( in green ) . ( b ) Following fixation , a facial expression appeared for 250 ms , after which the participant had to indicate whether the face expressed anger or fear within 2 seconds . No feedback was provided after response . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 00410 . 7554/eLife . 10274 . 005Figure 3 . Enhanced recognition accuracy and perceptual sensitivity to threat-signaling emotions . ( a ) Proportion of correct responses for ( from left to right ) averted/anger , direct/anger , averted/fear and direct/fear . THREAT+ combinations of gaze and emotion ( in orange ) were associated with increased recognition accuracy . ( b ) Psychometric function representing the proportion of ‘anger’ responses as a function of the evidence for anger ( proportion morph , 0 = neutral , negative towards fear , and positive towards anger ) for THREAT+ ( orange ) and THREAT− ( green ) combinations of gaze and emotion . Dots and attached error bars indicate the human data ( mean ± s . e . m . ) . Lines and shaded error bars indicate the predictions of the best-fitting model . ( c ) Parameter estimate for the slope of the psychometric curve ( corresponding to emotion sensitivity ) for THREAT+ and THREAT− combinations . **p < 0 . 01 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 005 Moreover , a significant emotion by gaze by emotion strength interaction was observed ( F6 , 138 = 4 . 3 , p<0 . 01 ) , explained by a stronger influence of gaze on emotion categorization at weak emotion strengths ( gaze by emotion interaction for levels 1 to 4 , F1 , 23 = 23 . 8 , p<0 . 001 ) than at high emotion strengths ( gaze by emotion interaction for levels 5 to 7 , F1 , 23 = 5 . 1 , p<0 . 05 ) . Reaction time ( RT ) analyses revealed a decrease of correct RTs with emotion strength ( repeated-measures ANOVA , F6 , 138 = 54 . 5 , p<0 . 001 ) , faster responses to angry as compared fearful faces ( F1 , 23 = 12 , p<0 . 01 ) , and faster responses to direct as compared to averted gaze ( F1 , 23 = 7 . 7 , p<0 . 05 ) . Furthermore , an emotion by gaze interaction was observed ( F1 , 23 = 8 , p<0 . 01 ) , corresponding to faster reaction times for direct as compared to averted gaze in the anger condition only ( t23 = -3 . 9 , p<0 . 001 ) . To characterize the mechanism underlying the improved recognition of threat-signaling emotions , we fitted participants’ behavior using a family of nested models of choice which hypothesize that decisions are formed on the basis of a noisy comparison between the displayed emotion and a criterion , under the following formulation ( see Materials and methods for details ) : P ( anger ) =Φ[w·x+b]· ( 1-ε ) +0 . 5·ε where P ( anger ) corresponds to the probability of judging the face as angry , Ф[ . ] to the cumulative normal function , w to the perceptual sensitivity to the displayed emotion , x to the evidence ( emotion strength ) in favor of anger or fear in each trial ( from -7 for an intense expression of fear to +7 for an intense expression of anger ) , b to an additive stimulus-independent bias in favor one of the two responses/emotions , and ε to the proportion of lapses ( random guesses ) across trials . We compared a ‘null’ model which did not allow for contextual influences of gaze direction on the decision process , to two additional models which instantiate two different mechanisms which could account for the observed increase in recognition accuracy for THREAT+ combinations of gaze and emotion: 1 . a first variant in which gaze direction biases the decision criterion in favor of the emotion signaling higher threat , and 2 . a second variant in which gaze direction enhances the sensitivity to the emotion signaling higher threat . Bayesian model selection revealed that a sensitivity enhancement for THREAT+ combinations explained substantially better the behavioral data than a criterion shift ( Bayes Factor ≈ 108 , exceedance probability pexc > 0 . 74 ) . Maximum-likelihood estimates of the perceptual sensitivity parameter w extracted from the winning model were significantly increased for THREAT+ combinations of gaze and emotion ( t23 = 3 . 9 , p<0 . 001; Figure 3b , c ) . The proportion of lapses did not vary between THREAT+ and THREAT- combinations ( t23 = 0 . 4 , p>0 . 5 ) . To validate the finding of enhanced sensitivity to threat-signaling emotions , and to identify its neural substrates , we then investigated how facial expressions modulated scalp EEG activity recorded during the emotion categorization task . Instead of computing event-related averages , we relied on a parametric regression-based approach consisting in regressing single-trial EEG signals against the strength of the displayed emotion at each electrode and time point following the presentation of the face ( Wyart et al . , 2012a , 2015 ) . A general linear regression model ( GLM ) was fit to the EEG data where emotion strength ( from 0 for a neutral/emotionless expression to 7 for an intense fear/anger expression ) was introduced as a trial-per-trial predictor of broadband EEG signals at each electrode and time point following stimulus onset ( from 0 . 2 s before to 1 . 0 s following stimulus onset ) . The resulting time course at each electrode represents the degree to which EEG activity ‘encodes’ ( co-varies with ) the emotion strength provided by morphed facial features . Parameter estimates of the regression slope revealed significant correlations between emotion strength and EEG activity peaking initially around 280 ms following face presentation at temporal ( t-test against zero , t23 = -12 . 7 , p<0 . 001 ) and frontal electrodes ( t23 = 8 . 7 , p<0 . 001 ) , and then around 500 ms and at response time at centro-parietal ( t23 = 10 . 2 , p<0 . 001 ) and frontal electrodes ( t23 = -7 . 9 , p<0 . 001 ) ( Figure 4a–c ) . Time points and electrodes where parameter estimates diverge significantly from zero indicate neural encoding of emotion information . The strength of this neural encoding – indexed by the amplitude of the parameter estimate – provides a measure of the neural sensitivity to emotion information . 10 . 7554/eLife . 10274 . 006Figure 4 . Enhanced neural encoding of threat-signaling emotions . ( a ) Middle panel: scalp topography of neural encoding at 280 ms , corresponding to its first peak of the encoding of emotion strength averaged across conditions ( peak of deviation from zero ) , and expressed as mean parameter estimates in arbitrary units ( a . u . ) . Dots indicate electrodes of interest where neural encoding was maximal . Left and right panels: encoding time course for THREAT+ and THREAT− conditions at electrodes of interest . Shaded error bars indicate s . e . m . Thick orange and green lines indicate significance against zero at a cluster-corrected p-value of 0 . 05 . Shaded grey areas indicate significant differences between THREAT+ and THREAT− conditions at p < 0 . 05 . ( b ) Same conventions as ( a ) at the second neural encoding peak at 500 ms . ( c ) Same conventions as ( a ) at the third neural encoding peak at response time . ( d ) Estimated cortical sources of the encoding difference between THREAT+ and THREAT− conditions at the time of significant difference between conditions at 170 ms . ( e ) Same as ( d ) at 500 ms . ( f ) Same as ( d ) at response time . FG: fusiform gyrus , pSTS: posterior superior temporal sulcus , SMG: supramarginal gyrus , ANG: angular gyrus , STG: superior temporal gyrus , MTG: middle temporal gyrus , OCG: occipital gyrus , aINS: anterior insula , IFS: inferior frontal sulcus , TP: temporal pole , OFC: orbitofrontal cortex , OP: occipital pole , TPJ: temporo-parietal junction , dlPFC: dorsolateral prefrontal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 006 To test for a neural signature of the increased sensitivity to threat-signaling emotions , we compared parameter estimates extracted separately for THREAT+ ( anger direct and fear averted ) and THREAT− ( anger averted and fear direct ) combinations of gaze and emotion . This contrast revealed increased parameter estimates for THREAT+ combinations first at 170 ms at temporal ( paired t-test , t23 = -2 . 5 , p<0 . 05 ) and frontal electrodes ( t23 = 2 . 2 , p<0 . 05 ) , and then later at 500 ms and at response time at centro-parietal ( t23= 2 . 2 , p<0 . 05 ) and frontal electrodes ( t23 = -2 . 4 , p<0 . 05 ) ( Figure 4a–c ) . This finding indicates that the neural gain of emotion encoding was enhanced at these time points and electrodes for threat-signaling emotions . This threat-dependent enhancement remained significant when considering only correct responses ( temporal: t23 = -2 . 1 , p<0 . 05; centro-parietal t23 = 4 . 2 , p<0 . 001 ) . Interestingly , THREAT+ combinations were not associated with increased event-related averages at classical peak latencies ( P1 , N170 , P2 , P3: all t23 < 1 . 95 , p>0 . 07 ) . To assess which brain regions generated the scalp-recorded EEG signals , we computed the cortical sources of this enhanced encoding of threat-signaling emotions by performing the same regression approach to minimum-norm current estimates distributed across the cortical surface . Parameter estimates at time points of interest ( where differences between THREAT+ and THREAT− combinations were observed ) were then contrasted between the two conditions ( see Materials and methods ) . Increases in regression slopes for THREAT+ combinations shifted from ventral visual areas selective to facial expressions of emotion ( fusiform gyrus and superior temporal sulcus ) around 170 ms , to associative brain regions encompassing parietal , temporal and frontal cortices ( superior and middle temporal , temporal pole , and orbitofrontal cortices ) at 500 ms , and then to sensorimotor regions around response onset ( dorsal central , parietal and frontal regions ) ( Figure 4d–f ) . These neural effects converge with behavioral modeling in favor of a sustained enhancement of perceptual sensitivity to threat-signaling emotions , starting 170 ms following face presentation and lasting until after response onset . Additional evidence supports our hypothesis that enhancements in neural sensitivity to THREAT+ combinations are specifically linked to an increase in implied threat for these combinations of gaze and emotion . A separate group of participants rated the identities used in the emotion categorization task in terms of perceived threat and trustworthiness ( see Materials and methods ) , and the group-level ratings for each identity were regressed against single-trial EEG signals as additional regressors . This regression showed that perceived threat , but not trustworthiness , correlated significantly with temporal and centro-parietal EEG activity at 500 ms following face presentation , in the same direction as the contrast between THREAT+ and THREAT− combinations ( threat: t23 > 3 . 6 , p<0 . 01; trustworthiness: t23 < 0 . 7 , p>0 . 48 ) . Analyses of the neural data have so far confirmed the hypothesis that contextual gaze information affects emotion categorization by increasing the perceptual sensitivity to threat-signaling emotions . Such an effect could be mediated by increased top-down attention to threat-signaling emotions – i . e . , THREAT+ combinations ( anger direct and fear averted ) . To test this possibility , we explored whether residual fluctuations in single-trial EEG signals unexplained by variations in emotion strength ( measured by the previous regressions ) modulated the accuracy of the subsequent categorical decision – i . e . , the perceptual sensitivity to the displayed emotion . This approach is reminiscent of ‘choice probability’ measures applied in electrophysiology to measure correlations between neural activity and choice behavior ( Britten et al . , 1996; Shadlen et al . , 1996; Parker and Newsome , 1998 ) – by estimating how much fluctuations in recorded neural signals are ‘read out’ by the subsequent decision ( Wyart et al . , 2012a , 2015 ) . Stimulus-independent improvements in neural-choice correlations have been classically interpreted as increases in ‘read-out’ weights – i . e . , increased top-down attention to these neural signals ( Nienborg and Cumming , 2009 , 2010 ) . Here , an increased neural modulation of choice for THREAT+ conditions could indicate an increase in top-down attention to threat-signaling emotions , which could in turn explain the observed increase in perceptual and neural sensitivity to these combinations of gaze and emotion . To test this hypothesis , we entered EEG residuals from the previous regression against emotion strength as an additional ‘mediation’ predictor of choice – as means to test whether these neural signals co-vary with perceptual sensitivity ( see Materials and methods for details ) . In practice , we estimated the parameters bmod and wmod of these neural modulation terms at each time point following face presentation via an EEG-informed regression of choice for which the trial-by-trial neural residuals e from the regression against emotion strength were entered either alone ( additive influence , parameter bmod ) or as their interaction with emotion strength ( multiplicative influence , parameter wmod ) as additional predictors of the subsequent choice: P ( anger ) =ϕ[ ( w+wmod· e ) x+b+bmod·e] The time course and spatial distribution of this neural modulation of perceptual sensitivity ( wmod ) followed qualitatively the neural encoding of emotion strength ( Figure 5a–c ) , with a negative temporal component peaking at 270 ms ( t23 = -4 . 2 , p<0 . 001 ) , followed by a positive centro-parietal one peaking around 600 ms ( t23 = 8 . 0 , p<0 . 001 ) and then at response time ( t23 = 7 . 6 , p<0 . 001 ) . We used Bayesian model selection to confirm that EEG residuals co-varied multiplicatively with the perceptual sensitivity ( wmod ) of the subsequent decision , not additively as a bias ( bmod ) in emotion strength , both at temporal ( Bayes factor ≈ 103 . 4 , pexc = 0 . 79 ) and centro-parietal electrodes ( Bayes factor ≈ 108 . 9 , pexc = 0 . 99 ) . Critically , no difference in modulation strength was observed between THREAT+ ( anger direct and fear averted ) and THREAT− ( anger averted and fear direct ) combinations ( temporal: t23 = -0 . 4 , p>0 . 5; centro-parietal: t23 = 0 . 1 , p>0 . 5 ) . To determine whether this absence of significant difference is due to a genuine absence of effect ( rather than a lack of statistical sensitivity ) , we computed Bayes factors under the same parametric assumptions as conventional statistics ( see Materials and methods ) . We obtained Bayes factors lower than 10–4 at temporal and centro-parietal electrodes , indicative of no increase in ‘read-out’ weights for THREAT+ conditions . This null effect suggests that the observed enhancement in perceptual and neural sensitivity to these threat-signaling combinations of gaze and emotion is not triggered indirectly by an increase in top-down attention in these conditions . 10 . 7554/eLife . 10274 . 007Figure 5 . Absence of threat-dependent enhancement of neural-choice correlations . ( a ) Middle panel: scalp topography of neural-choice correlations , expressed as the modulation of perceptual sensitivity by EEG encoding residuals at 280 ms , same time point shown in Figure 4a . Electrodes of interest indicated with dots are the same as in Figure 4a . Left and right panels , time course of the modulation of perceptual sensitivity by EEG encoding residuals expressed in arbitrary units ( a . u . ) . Same conventions as in Figure 4a . ( b ) Same conventions as ( a ) at 500 ms . ( c ) Same conventions as ( a ) at response time . The variation of the modulation strength over time is consistent with the variation of the encoding parameter estimate . No difference between THREAT+ and THREAT− is observed . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 007 We reasoned that threat could impact not only the neural representation of the displayed emotion in visual and associative cortices , but also the preparation of the upcoming response in effector-selective structures ( Conty et al . , 2012 ) . To measure response-preparatory signals in the neural data , we computed spectral power in the mu and beta frequency bands ( 8–32 Hz ) ( Donner et al . , 2009; de Lange et al . , 2013 ) . Limb movement execution and preparation coincide with suppression of low-frequency ( 8–32 Hz ) activity that is stronger in the motor cortex contralateral as compared to ipsilateral to the movement . Thus , subtracting the contralateral from ipsilateral motor cortex activity is expected to result in a positive measure of motor preparation . The contrast between left-handed and right-handed responses at response time identified lateral central electrodes , associated with focal sources in motor cortex ( Figure 6a ) . Subtracting contralateral from ipsilateral signals relative to the hand assigned to the ‘fear’ response ( counterbalanced across participants ) provided a motor lateralization index whose sign predicts significantly the upcoming choice ( anger or fear ) from 360 ms before response onset ( paired t-test , t23 = 4 . 6 , p<0 . 001; Figure 6b ) . 10 . 7554/eLife . 10274 . 008Figure 6 . Encoding of threat-signaling emotions in motor response lateralization measures . ( a ) Top panel , scalp topography before response of the time frequency power in the 8–32 Hz band in the last 100 ms before response , for the trials where subjects responded with their left hand minus the trials where they responded with their right hand . Dots correspond to the selected electrodes , where the effect was maximal . Bottom panel: corresponding neural sources . ( b ) Time course of response lateralization ( time frequency power activity from the contralateral electrodes minus ipsilateral electrodes to the hand used to respond ‘fear’ ) towards anger and fear when the choice was anger ( red ) or fear ( blue ) . Shaded error bars indicate s . e . m . The shaded gray area indicates a significant difference in motor lateralization between Anger and Fear responses . ( c ) Encoding of emotion strength in response lateralization index for THREAT+ ( orange ) and THREAT− ( green ) conditions . Differences between conditions are observed at 200 ms after stimulus onset ( stimulus-locked , upper panel ) and at response time ( response-locked , lower panel ) . Conventions are the same as in Figure 4 . ( d ) Time course of neural-choice correlations , expressed as the modulation of additive bias by motor lateralization encoding residuals in arbitrary units ( a . u . ) stimulus-locked ( upper panel ) and response locked ( lower panel ) . Conventions are the same as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 008 We applied the previous neural encoding approach by regressing this motor lateralization index against the signed emotion strength ( from 0 for a neutral expression , to ±7 for an intense anger/fear expression ) on a trial-by-trial basis . Parameter estimates of the regression slope diverged significantly from zero from 400 ms after stimulus onset ( t-test against zero , t23 = 5 . 1 , p<0 . 001 ) and at response time ( t23 = 5 . 2 , p<0 . 001 ) – reflecting stronger response preparation to stronger ( i . e . , more diagnostic ) emotions . Computing regression slopes separately for THREAT+ ( anger direct and fear averted ) and THREAT- ( anger averted and fear direct ) combinations revealed that THREAT+ combinations produced a stronger encoding of emotion strength in motor preparation late at response onset ( t23 = 2 . 9 , p<0 . 01 ) , but also early around 200 ms following face presentation ( t23 = 3 . 2 , p<0 . 01 ) . This early threat-dependent motor enhancement remained significant when considering only correct responses ( t23 = 3 . 0 , p<0 . 01 ) . While THREAT− combinations of gaze and emotion were not associated with significant neural encoding in motor preparation until 440 ms following face presentation ( t23 < 0 . 8 , p>0 . 4 ) , THREAT+ combinations resulted in significant neural encoding between 100 and 320 ms , peaking at 200 ms ( t23 = 3 . 2 , p<0 . 01; Figure 6c ) . To determine whether this early neural encoding of threat-signaling emotions in motor preparation influences the speed of subsequent responses , we recomputed and compared regression parameters estimated separately for fast and slow responses to THREAT+ combinations ( anger direct and fear averted ) , on the basis of a median split of response times informed by emotion strength . This comparison revealed a single , gradual neural encoding of emotion strength in motor preparation preceding fast , but not slow responses , arising as early as 150 ms ( at a threshold p-value of 0 . 05 ) following the presentation of the face ( difference in encoding onset between fast and slow responses , jackknifed ( Kiesel et al . , 2008 , see Materials and methods ) t23 = 5 . 2 , p<0 . 001; Figure 7a ) . This effect indicates that the early neural encoding of THREAT+ combinations in motor preparation is characteristic of efficient ( fast ) responses . We verified that this latency shift in neural encoding was selective of motor preparation signals , by performing the same comparison on the neural encoding of emotion strength at centro-parietal electrodes . This contrast revealed only a difference in peak amplitude , not onset latency , between fast and slow responses ( peak amplitude: t23 = 5 . 1 , p<0 . 001; onset latency: jackknifed t23 = -1 . 3 , p>0 . 2; Figure 7b ) . 10 . 7554/eLife . 10274 . 009Figure 7 . Encoding of emotion strength as a function of reaction times ( RT ) in motor and parietal structures . ( a ) Neural encoding of emotion strength for THREAT+ conditions in motor lateralization for fast and slow reaction times ( RT ) : when RTs were fast , the encoding of emotion strength became significant at 150 ms and rose gradually until response; by contrast , when RTs were slow , the encoding of emotion strength became significant later at 540 ms . Shaded error bars indicate s . e . m . Thick dark and light grey lines indicate significance against zero at a cluster-corrected p-value of 0 . 05 . Shaded grey bars indicate significant differences between fast and slow responses . Encoding latency is significantly different between fast and slow RTs , ***: p<0 . 001 ( b ) Emotion strength encoding in parietal electrodes . Convention are the same than ( a ) . Fast responses are associated with a stronger neural encoding of emotion strength , but without any change in encoding latency . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 009 Finally , we performed neural-choice correlations analyses to assess whether the early neural encoding of threat-signaling emotions in motor preparation influences not only the speed , but also the content ( anger or fear ) of subsequent responses . Across conditions , the neural ‘mediation’ analysis described above revealed that stimulus-independent fluctuations in motor lateralization index co-vary as an additive choice bias in the upcoming response from 400 ms following face presentation ( t23 = 2 . 9 , p<0 . 01 ) . Indeed , in contrast to fluctuations in temporal and centro-parietal activity , the impact of variability in motor lateralization on emotion categorization was better described as an additive choice bias rather than a change in perceptual sensitivity ( Bayes factor ≈ 1036 . 4 , pexc = 0 . 98 ) – consistent with its hypothesized role as a motor representation of the decision variable ( Donner et al . , 2009; de Lange et al . , 2013 ) . No difference in modulation strength was observed between THREAT+ ( anger direct and fear averted ) and THREAT− ( anger averted and fear direct ) combinations ( t23 < 1 . 6 , p>0 . 1; Figure 6d ) . Critically , even when considering combinations alone , residual variability in motor lateralization measured between 100 and 320 ms ( where the neural encoding of threat-signaling emotions was significant ) did not bias significantly the upcoming choice ( t23 < 1 . 4 , p>0 . 17 ) . This null effect was supported by Bayesian model selection that identified a genuine absence of neural-choice correlation as the most likely account of the data ( Bayes factor ≈ 102 . 3 , pexc = 0 . 96 ) . This finding indicates that the early neural encoding of threat-signaling emotions in motor preparation occurs earlier than the formation of the upcoming choice . In the general population , anxiety has been classically associated with an oversensitivity to threat signals in social conditions ( Bishop , 2007; Cisler and Koster , 2010 ) . Here , we assessed whether the enhanced neural processing of threat-signaling emotions in temporal and motor regions co-varied with the level of anxiety in our participants . For this purpose , we measured anxiety at the beginning of the experimental session , before data collection , using the Spielberger State-Trait Anxiety Inventory ( STAI ) ( Spielberger et al . 1983 ) . This self-questionnaire provides a measure of vulnerability for anxiety disorders ( Grupe et al . 2013 ) . Participants’ state anxiety scores ranged from 20 to 45 ( mean = 30 . 5 , SD = 6 . 8 ) . Trait anxiety scores ranged from 22 to 52 ( mean = 38 . 2 , SD = 7 . 8 ) . These scores are comparable to the original published norms for this age group ( Spielberger , 1983 ) and to those from French normative data ( Bruchon-Schweitzer and Paulhan , 1993 ) . We analyzed the effect of anxiety on the behavioral and neural data in two complementary ways: 1 . by splitting the participants in two equally-sized groups based on their measured anxiety , and 2 . by correlating neural encoding parameters estimated at the level of individual participants with their measured anxiety . Surprisingly , we found no effect of anxiety on overall measures of performance ( t11 < 0 . 05 , p>0 . 9 ) , nor on the difference between THREAT+ ( anger direct and fear averted ) and THREAT- ( anger averted and fear direct ) combinations of gaze and emotion ( F1 , 22 < 0 . 4 , p>0 . 5 ) . Nevertheless , the absence of effect of anxiety at the behavioral level was accompanied by a compensatory double dissociation in the neural data . Indeed , state anxiety influenced significantly the neural encoding of emotion strength at temporal electrodes at the peak of neural encoding , 280 ms following face presentation ( median split , interaction: F1 , 22 = 7 . 3 , p=0 . 01; Figure 8a ) : high-anxious observers showed no difference in neural encoding between THREAT+ and THREAT- combinations ( THREAT+ : t11 = -5 . 8 , p<0 . 001; THREAT-: t11 = -6 . 1 , p<0 . 001 , difference: t22 = 0 . 84 , p=0 . 4 ) , whereas low-anxious observers encoded exclusively THREAT+ at the same latency ( THREAT+: t11 = -6 . 5 , p<0 . 001; THREAT-: t11 = -1 . 8 , p=0 . 08; difference: t22 = -3 . 0 , p=0 . 01 ) . A parametric assessment of the relationship between state anxiety and the difference in neural encoding between THREAT+ and THREAT- combinations proved to be significant ( Pearson correlation coefficient r = 0 . 51 , d . f . = 22 , p=0 . 01; Figure 8a ) . In other words , high anxiety was associated with a significant and indifferent neural encoding of negative emotions , whether threat-signaling or not , in ventral face-selective regions . 10 . 7554/eLife . 10274 . 010Figure 8 . Modulation of threat encoding by individual anxiety . ( a ) Left panel: correlation ( Pearson ) between state anxiety and the difference of the encoding parameter estimates between THREAT+ and THREAT− conditions in temporal electrodes at 280 ms . Right panel: encoding parameter estimates in temporal electrodes split into high and low anxious individuals for both THREAT+ and THREAT− conditions at 280 ms . T+: THREAT+ , T-: THREAT- . ( b ) Left , correlation ( Pearson ) between state anxiety and the encoding parameter estimates in motor lateralization signals for THREAT+ condition at 200 ms . Right , encoding parameter estimates in motor lateralization signals split into high and low anxious individuals for both THREAT+ and THREAT− conditions at 200 ms . ***: p<0 . 001 , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 10274 . 010 Interestingly , at the early time window ( peak of the encoding at 200 ms ) where only THREAT+ combinations ( anger direct and fear averted ) were encoded in motor signals , a reverse pattern was observed: only high anxious individuals showed a significant encoding at this latency ( interaction between between-subject state anxiety and gaze pairing F1 , 22 = 4 , p=0 . 05; Figure 8b ) . The more the individuals were anxious , the more they encoded observer-relevant threat signals in motor systems ( correlation between parameter estimates for THREAT+ conditions and state anxiety Pearson coefficient r = 0 . 52 , d . f . = 22 , p<0 . 01; Figure 8b ) . Moreover , the neural encoding of THREAT+ emotions in motor signals correlated with behavioral sensitivity to THREAT+ emotions for high-anxious individuals ( Pearson correlation coefficient r = 0 . 66 , d . f . = 10 , p=0 . 01 ) , whereas it was not the case for low-anxious individuals ( Pearson correlation coefficient r = −0 . 42 , d . f . = 10 , p>0 . 16 , difference between coefficients , p<0 . 01 ) . To sum up , while high anxious individuals process all threat signals equivalently in face selective regions , they selectively encode threat signals that are relevant to them in motor specific systems , and this encoding reflects their behavioral sensitivity to threat-signaling emotions .
Accurate decoding of emotions in others , especially negative ones , conveys adaptive advantages in social environments . Although typical social interactions do not require an explicit categorization of the emotion expressed by others , a precise understanding of the neural mechanisms involved in emotion recognition provides important information regarding how the human brain processes socially meaningful signals . And while past work has uncovered the neural correlates of perceptual decisions ( Gold and Shadlen , 2007; Heekeren et al . , 2008 ) , only few studies have addressed the issue of how such decisions are formed on the basis of socially relevant stimuli such as facial displays of emotion . As in most perceptual categorization tasks , we manipulated the ambiguity of sensory evidence – here , using controlled morphs between angry or fearful expressions and neutral ones . But owing to the social nature of our stimuli , we could simultaneously and implicitly manipulate the contextual significance of the displayed emotion in terms of implied threat for the observer , using gaze direction , and apply a model-guided approach to characterize the neural prioritization of threat-signaling information in electrical brain signals . Gaze direction , which acts as a contextual cue in our emotion categorization task , differs from contextual cues found in perceptual decision-making studies which are typically provided hundreds of milliseconds before the decision-relevant stimulus ( Rahnev et al . , 2011; Kok et al . , 2012; Wyart et al . , 2012b; de Lange et al . , 2013 ) . Here , as in many social situations , contextual cues can co-occur with the decision-relevant stimulus – a property which strongly constrains their impact on stimulus processing . Moreover , the meaning of contextual cues ( e . g . , attention or expectation cues ) used in perceptual decision-making studies is usually instructed explicitly , and thus processed explicitly by the participants during task execution ( Kok et al . , 2012; Wyart et al . , 2012b ) . Here , by contrast , gaze direction is irrelevant for the emotion categorization task , and thus does not need to be processed explicitly . Despite these two differences with other contextual cues , we show that gaze direction tunes the neural processing of emotion information from 200 ms following stimulus onset until response in sensory , associative and motor circuits of the human brain . Previous observations of increased subjective ratings and improved recognition of angry expressions paired with a direct gaze and fearful expressions paired with an averted gaze have been interpreted in terms of a contextual evaluation of the displayed emotion during its processing ( Adams and Kleck , 2003; Sander et al . , 2007; Adams et al . , 2012 ) . In particular , ‘appraisal’ theories ( Sander et al . , 2007 ) emphasize that an angry expression paired with a direct gaze can be interpreted as behaviorally ‘relevant’ to the observer as being the target of a verbal or physical assault , whereas a fearful expression looking aside from the observer might signal a source of danger in the immediate vicinity of the observer . However , the mechanisms which instantiate the proposed contextual evaluation of emotions as a function of their implied threat for the observer have remained unclear . Gaze direction could either bias the perceived emotion towards its most relevant ( threat-signaling ) interpretation – i . e . , anger when paired with direct gaze , or fear when paired with averted gaze , or increase the sensitivity to the most relevant emotion . The present study answers directly this issue by showing , both behaviorally ( by comparing quantitative fits of the two effects to the behavioral data ) and neurally ( by regressing brain signals against emotion strength ) , that the improved recognition accuracy for threat-signaling emotions corresponds to a selective neural enhancement of perceptual sensitivity to these combinations of gaze and emotion . Emotion information modulated EEG signals at centro-parietal electrodes from 500 ms following face presentation until response execution , a finding in accordance with the ‘supramodal’ signature of perceptual integration reported in previous studies ( O’Connell et al . , 2012; Wyart et al . , 2012a ) . This centro-parietal positivity has been proposed to encode a ‘domain-general’ decision variable , as it varies with the strength of sensory evidence for both visual and auditory decisions , independently from the associated response ( O’Connell et al . , 2012 ) . Here , the same centro-parietal positivity was found to increase with the emotion strength of facial expressions – which indexes the decision variable in our emotion categorization task . Importantly , the strength of this relationship was enhanced for threat-signaling emotions . This improved neural representation of threatening combinations of gaze and emotion cannot be explained by increased attentional or surprise responses , since the centro-parietal ‘P3’ potential , previously reported to vary as a function of attentional resources ( Johnson , 1988 ) and surprise ( Mars et al . , 2008 ) , was not increased in response to threat-signaling emotions . Moreover , we could also rule out the possibility that this enhanced neural encoding is triggered indirectly by an increase in selective attention , which should have been associated with an improved ‘decoding’ of participants’ decisions from their underlying neural signals ( Nienborg and Cumming , 2009 , 2010; Wyart et al . , 2015 ) . We therefore hypothesize that the enhanced neural processing of threat-signaling emotions proceeds in an attention-independent , bottom-up fashion . Earlier contextual modulations of emotion processing were also observed in ventral face-selective areas from 170 ms following face presentation . While these findings contradict a ‘two-stage’ view according to which emotion and gaze information would be processed independently during the first hundreds of millisecond ( Pourtois et al . , 2010 ) before being integrated as a function of their significance to the observer ( Klucharev and Sams , 2004 ) , they are in agreement with recent findings ( Conty et al . , 2012; El Zein et al . , 2015 ) of early interactions between emotion and gaze information on N170 and P200 components . At these early latencies , only threat-signaling emotions were encoded by face-selective neural signals , reflecting a faster processing of emotions signaling an immediate threat to the observer as a function of their associated gaze . More strikingly , gaze direction also modulated the encoding of emotional expressions in effector-selective regions , in parallel with the effects observed in ventral face-selective areas: only threat-signaling emotions were encoded in response preparation signals overlying human motor cortex at 200 ms following face presentation . Recent work sheds light on the adaptive function of this early representation of threat signals in motor cortex . Disrupting this motor representation using TMS impairs the facial recognition of negative ( i . e . , potentially threatening ) emotions , not positive ones ( Balconi and Bortolotti , 2012; 2013 ) . Moreover , the perception of natural scenes engages the motor cortex at very early latencies only when the emotional valence of the scene is negative ( Borgomaneri et al . , 2014 ) . Taken together , these findings support a strong connection between emotion and motor circuits ( Grèzes et al . , 2014 ) enabling the brain to react swiftly and efficiently to threat signals ( Ohman and Mineka , 2001; Frijda , 2009 ) . Our findings build on these earlier observations by showing that the brain encodes parametrically the strength of threat signals in motor cortex in parallel to their representation in face-selective , sensory regions . Finally , our data reveal a clear functional dissociation between face- and effector-selective regions as a function of individual anxiety . The enhanced sensitivity to threat-signaling emotions in face-selective temporal cortex is driven by low-anxious observers , whereas the early enhancement measured in motor cortex is only found in high-anxious observers . The observation that high-anxious individuals encode all negative emotions as equally ( and strongly ) salient in face-selective regions is consistent with earlier reports of a ‘hyper-vigilance’ to potentially threatening signals in these individuals ( Bishop , 2007; Cisler and Koster , 2010 ) , and with their tendency to interpret ambiguous stimuli as threatening ( Beck et al . , 1985 ) – both associated with amygdala hyperactivity ( Bishop , 2007; Etkin and Wager , 2007 ) . Nevertheless , our findings reveal that high-anxious individuals are capable of encoding threat signals in a selective fashion in motor cortex . Consistent with the idea of a compensatory mechanism , the distinct neural enhancements of temporal and motor activity found in low- and high-anxious individuals lead to similar behavioral improvements in terms of perceptual sensitivity to threat signals . Together , this pattern of findings suggests that anxiety increases the relative contribution of the motor pathway during the processing of negative social signals , in accordance with the adaptive function of anxiety in detecting efficiently and reacting swiftly to threats in the environment ( Bateson et al . , 2011 ) . It is worth noting that the present study only involved participants with anxiety scores within the range of the healthy adult population ( Spielberger , 1983 ) , leaving open the question as to whether clinically anxious individuals would similarly recruit their motor cortex in response to threatening social stimuli . Moreover , further research should assess the specificity of these anxiety-dependent effects , in light of the growing evidence in favor of comorbidity between anxiety and depressive disorders . By applying theoretical models of decision-making to socially-relevant stimuli , we were able to characterize the neural and computational mechanisms underlying the integration and interpretation of facial cues in the implicit context of threat . Evolutionary pressure might have shaped the human brain to prioritize threat signals in parallel in sensory and motor systems ( Darwin , 1872; LeDoux , 2012 ) . Such prioritization – found to proceed in a fast , selective , yet attention-independent fashion – could increase perceptual sensitivity to other features of the sensory environment ( Phelps et al . , 2006 ) to enable rapid and adaptive responses in complex , multidimensional situations of danger .
Twenty-four healthy subjects ( 12 females; mean age , 22 . 7 ± 0 . 7 years ) participated in the EEG experiment . All participants were right-handed , with a normal vision and had no neurological or psychiatric history . They provided written informed consent according to institutional guidelines of the local research ethics committee ( Declaration of Helsinki ) and were paid for their participation . Stimuli consisted of 36 identities ( 18 females ) adapted from the Radboud Faces Database ( Langner et al . , 2010 ) that varied in emotion ( neutral , angry or fearful expressions ) and gaze direction ( direct toward the participant or averted 45° to the left or right ) . Using Adobe Photoshop CS5 . 1 ( Adobe Systems , San Jose CA ) , faces were modified to remove any visible hair , resized and repositioned so that eyes , nose and mouth appeared within the same circumference . All images were converted to greyscale and cropped into a 280 x 406 pixel oval centered within a 628 x 429 pixel black rectangle . To vary the intensity of emotional expressions , faces were morphed from neutral to angry expressions and from neutral to fearful expression using FantaMorph ( Abrosoft http://www . fantamorph . com/ ) . At first , we created 7 levels of morphs from neutral to angry expressions and from neutral to fearful expressions ( separately for direct and averted gaze stimuli ) using a simple linear morphing transformation . This resulted in 30 conditions for each identity: 7 levels of morphs * 2 emotions * 2 gaze directions = 28 and 2 neutral stimuli with direct and averted gaze . We then calibrated the morphing between angry and fearful expressions by performing an intensity rating pre-test of the emotional expressions and adjusting the morphs based on the results . 19 subjects ( 9 females , mean age , 24 . 7 ± 0 . 9 years ) were presented with the facial expressions for 250 ms and rated the emotional intensity perceived on a continuous scale from “not at all intense” to “very intense” using a mouse device ( with a maximum of 3 seconds to respond ) . We adjusted for differences between emotions by linearizing the mean curves of judged intensities and creating corresponding morphs that were validated on 10 new subjects ( 4 females , mean age 24 . 1 ± 1 . 9 ) . To summarize , the stimuli comprise of 36 identities with an Averted gaze condition and a Direct gaze condition , each with 7 levels of Anger and 7 levels of Fear equalized in perceived emotional intensities and a neutral condition , resulting in a total of 1080 items ( see Figure 2a for examples of stimuli ) . Using the Psychophysics-3 Toolbox ( Brainard , 1997; Pelli , 1997 ) , stimuli were projected on a black screen . Each trial was initiated with a white oval delimiting the faces that was kept during all the trial . The white oval appeared for approximately 500 ms , followed by a white fixation point presented at the level of the eyes for approximately 1000 ms ( to keep the fixation to the upcoming faces natural and avoid eye movements from the center of the oval to eye regions ) , than the stimuli appeared for 250 ms . Participants’ task was to decide whether the faces expressed Anger or Fear by pressing one of the two buttons localized on two external devices held in their right and left hands , with their right or left index correspondingly ( Figure 2b ) . An Anger/Fear mapping was used ( e . g Anger: Left hand , Fear: Right hand ) kept constant for each subject , counterbalanced over all subjects . All stimuli were presented once , resulting in a total of 1080 trials . The experiment was divided in 9 experimental blocks , each consisting of 120 trials , balanced in the number of emotions , directions of gaze , gender and levels of morphs . After each block , the percentage of correct responses was shown to the participants to keep them motivated . Repeated-measures ANOVA was performed on the percentage of correct responses and average reaction times , with gaze direction ( direct/averted ) , emotion ( anger/fear ) , and intensity ( 7 levels of morphs ) as within-subjects factors . We performed model-guided analyses of the behavioural data to characterize the observed increase in recognition accuracy for THREAT+ combinations of gaze and emotion . We used Bayesian model selection based on the model evidence ( estimated by a 10-fold cross-validation estimation of model log-likelihood , which penalizes implicitly for model complexity without relying on particular approximations such as the Bayesian Information Criterion or the Akaike Information Criterion ) . We applied both fixed-effects and random-effects statistics previously described in the literature . The fixed-effects comparison assumes all participants to have used the same underlying model to generate their behavior , such that the overall model evidence for a given model is proportional to the product of model evidence for the model for all participants . Based on this model evidence , we compared different models by computing their Bayes factor as the ratio of model evidence of the compared model ( Jeffreys , 1961; Kass and Raftery , 1995 ) . The random-effects comparison is more conservative in allowing different participants to use different models to generate their behavior , and aims at inferring the distribution over models that participants draw from ( Penny et al . , 2010 ) . For this comparison , we computed support for the winning model by the exceedance probability ( pexc ) , which is the probability that participants were more likely to choose this model to generate behavior over any alternative model . We started with the simplest model ( model 0 ) that could account for each subject’s decisions using a noisy , ‘signal detection’-like psychometric model to which we included a lapse rate , thereby considering that subjects guessed randomly on a certain proportion of trials: P ( anger ) =ϕ[w*x+b]* ( 1-ε ) +0 . 5*ε where P ( anger ) corresponds to the probability of judging the face as angry , Ф[ . ] to the cumulative normal function , w to the perceptual sensitivity to the displayed emotion , x to a trial-wise array of evidence values in favor of anger or fear ( emotion strength , from −7 for an intense expression of fear to +7 for an intense expression of anger ) , b to an additive , stimulus-independent bias toward one of the two emotions , and ε to the proportion of random guesses among choices . We compared a ‘null’ model which did not allow for contextual influences of gaze direction on the decision process , to two additional models which instantiate two different mechanisms which could account for the observed increase in recognition accuracy for THREAT+ combinations of gaze and emotion . A first possibility ( model 1 ) would be that gaze direction biases emotion recognition in favor of the interpretation signaling higher threat ( anger for a direct gaze , fear for an averted gaze ) . Alternatively ( model 2 ) , gaze direction might selectively increase sensitivity to emotions signaling higher threat in this context ( modeled by a different sensitivity to emotions in THREAT+ vs . THREAT− conditions ) . An EEG cap of 63 sintered Ag/AgCl ring electrodes ( Easycap ) was used to record EEG activity . EEG activity was recorded at a sampling rate of 1000 Hz using a BRAINAMP amplifier ( Brain Products , BRAINAMP MR PLUS ) and low pass filtered online at 250 Hz . The reference channel was placed on their nose and a forehead ground was used . Impedances were kept under a threshold of 10 kΩ . The raw EEG data was recalculated to average reference , down-sampled to 500 Hz , low-pass filtered at 32 Hz , and epoched from 1 s before to 4 s after the face stimulus onset using EEGLAB ( Delorme and Makeig , 2004 ) . First , EEG epoched data was visually inspected to remove muscle artifacts and to identify noisy electrodes that were interpolated to the average of adjacent electrodes . Second , independent component analysis ( ICA ) that excluded interpolated electrodes was performed on the epoched data and ICA components capturing eye blink artifacts were manually rejected . A last , visual inspection was done on the resulting single epochs to exclude any remaining trials with artifacts . After trial rejections , an average of 999 ± 10 trials per subject remained . Time frequency analysis was performed using the Fieldtrip toolbox for MATLAB ( Oostenveld et al . , 2011 ) . We were particularly interested in motor mu-bands ( 8–32 Hz ) and thus estimated the spectral power of mu-beta band EEG oscillations using ‘multitapering’ time frequency transform ( Slepian tapers , frequency range 8–32 Hz , five cycles , three tapers per window ) . The purpose of this multitapering approach is to obtain more precise power estimates by smoothing across frequencies . Note that this time–frequency transform uses a constant number of cycles per window across frequencies , hence a time window whose duration decreases inversely with increasing frequency . | Facial expressions can communicate important social signals , and understanding these signals can be essential for surviving threatening situations . Past studies have identified changes to brain activity and behavior in response to particular social threats , but it is not clear how the brain processes information from the facial expressions of others to identify these threats . Here , El Zein , Wyart and Grèzes aimed to identify how signals of threat are represented in the human brain . The experiment used a technique called electroencephalography to record brain activity in healthy human volunteers as they examined angry and fearful facial expressions . El Zein , Wyart and Grèzes found that emotions that signaled a threat to the observer are better represented in particular regions of the brain – including those that control action – within a fraction of a second after the facial expression was shown to the volunteer . Moreover , the response of the brain regions that control action was greater in volunteers with higher levels of anxiety , which highlights the role of anxiety in reacting rapidly to social threats in the environment . El Zein , Wyart and Grèzes’ findings show that social threats can alter brain activity very rapidly , and in a more selective manner than previously believed . A future challenge is to find out whether other aspects in threatening environments can stimulate similar increases in brain activity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Anxiety dissociates the adaptive functions of sensory and motor response enhancements to social threats |
Excitatory and inhibitory synapses are the brain’s most abundant synapse types . However , little is known about their formation during critical periods of motor skill learning , when sensory experience defines a motor target that animals strive to imitate . In songbirds , we find that exposure to tutor song leads to elimination of excitatory synapses in HVC ( used here as a proper name ) , a key song generating brain area . A similar pruning is associated with song maturation , because juvenile birds have fewer excitatory synapses , the better their song imitations . In contrast , tutoring is associated with rapid insertion of inhibitory synapses , but the tutoring-induced structural imbalance between excitation and inhibition is eliminated during subsequent song maturation . Our work suggests that sensory exposure triggers the developmental onset of goal-specific motor circuits by increasing the relative strength of inhibition and it suggests a synapse-elimination model of song memorization .
During critical period learning early in life , brain structure is regulated not only by intrinsic factors such as age , but also by extrinsic factors including sensory experience ( Andersen , 2003; Hubel and Wiesel , 1970; Kirkwood et al . , 1995 ) . During this period , neuronal connections are highly susceptible to experience-dependent modifications ( Hensch , 2004 ) , often manifest as competitive synapse elimination ( Lichtman and Colman , 2000 ) . Both excitatory and inhibitory transmissions have important roles in shaping the structural and functional outcomes of sensory experience ( Bear et al . , 1990; Hata and Stryker , 1994 ) . However , the requirements on excitatory and inhibitory synapses during critical period learning have remained elusive . Excitatory and inhibitory neurotransmitter systems are involved in severe mental disorders ( Lee et al . , 2015; Rubenstein and Merzenich , 2003; Yizhar et al . , 2011 ) , which is why a better understanding of these systems during developmental learning may be of therapeutic relevance . We study the influence of sensory experience and of subsequent motor maturation on the structural balance between excitation and inhibition . We use the songbird as a model system because of its highly specialized song system . Similar to human speech development ( Neville and Bavelier , 2002 ) , the normal development of birdsong requires early sensory experience ( Barrington , 1773; Beecher and Brenowitz , 2005 ) . Adult birdsong is generated and temporally controlled by interactions among inhibitory and excitatory neurons in HVC ( Aronov et al . , 2008; Kosche et al . , 2015; Long and Fee , 2008; Mooney and Prather , 2005; Nottebohm et al . , 1976 ) . An important role of excitation is to participate in successful acquisition of a template of tutor song , because blockage of NMDA-mediated ( excitatory ) synaptic transmission in HVC during tutor exposure abolishes normal song development ( Roberts et al . , 2012 ) . One important role of HVC inhibition may be to protect HVC neurons after tutoring from further sensory influence , as the strength of HVC inhibition correlates with the learned portions of song but not with age ( Vallentin et al . , 2016 ) . To obtain structural insights into the dynamics of HVC excitation and inhibition , we study the HVC synaptic organization and the changes associated with tutor exposure and with aging .
Using electron microscopy ( EM ) , we investigated the effects of tutor song exposure on HVC synapses . We compared synapse densities in zebra finches that were either exposed to an adult tutor for one day ( SHORT birds ) or that were never exposed to a tutor ( ISO birds ) . To compare with densities in normally reared birds , we included a third group of birds that were tutored for 24 days ( LONG birds ) . In this first experiment , we minimized the influence of age by sacrificing all birds mid-development at 59 days post hatch ( dph ) , which is near the end of the critical sensory song learning period , Table 1 , Figure 1A . We found that one day of tutor exposure led to a 26 ± 5% decrease in the HVC asymmetric ( excitatory ) synapse density ( p = 3*10−9 , linear mixed effect ( LME ) model with bird group as fixed effect and bird identity as random effect , df = 1152 , n = 4 SHORT birds and n = 4 ISO birds , see Materials and methods ) , Figure 1B , C . In contrast , the HVC symmetric ( inhibitory ) synapse density increased by about 20 ± 10% during one day of tutor exposure ( p = 0 . 05 , LME model , df = 1152 , n=4 SHORT and n=4 ISO birds ) , Figure 1B , D . In combination , upon first tutoring and within 24 hr , the percent inhibitory synapses in HVC rose by 42% from on average 19% ( ISO group ) to 27% ( SHORT group , p = 1 . 7*10−6 , LME model with 32 observations , two fixed effect coefficients , and eight random effect coefficients , see Materials and methods ) , Figure 1E . Extended tutoring ( LONG ) had the net effect of removing both excitatory and inhibitory synapses . Compared with the ISO group , we found that 24 days of tutoring led to a decrease in excitatory synapse density of 37 ± 1% ( p = 4 . 5*10−18 , LME model , n = 4 LONG and n = 4 ISO birds ) and to a decrease in inhibitory synapse density of 30 ± 2% ( p = 0 . 003 , LME model , n = 4 LONG and n = 4 ISO birds ) . Altogether , tutoring induced a synaptic bias towards inhibition that was transient , because the percent inhibitory synapses in LONG birds was barely larger than in ISO birds ( 21% vs 20% , p = 0 . 48 , LME model , n = 4 LONG and n = 4 ISO birds ) . To assess whether the observed changes in synapse densities were aligned ( or anti-aligned ) with normal developmental trends , in a second experiment , we studied HVC synapse turnover and the excitatory-inhibitory balance as a function of age . We measured HVC synapse densities at the onset of the sensory learning phase at 30 dph in untutored birds ( ISO30 birds ) and at the onset of adulthood at 90 dph , both in extensively tutored birds ( LONG90 birds ) and in untutored birds ( ISO90 birds ) . We included two additional groups of extensively tutored and untutored birds sacrificed mid-development ( 60 dph , LONG60 and ISO60 birds ) , Table 1 , Figure 2A . We provided tutored ( LONG60 and LONG90 ) birds in this second experiment with just 90 min of tutor exposure per day instead of the unrestricted exposure we had provided to LONG birds in the first experiment . The motivation for this change in tutoring paradigm was to induce better song copying . Song learning in juvenile zebra finches is inversely related to tutor song abundance , and more playbacks of a given tutor song lead to less complete song imitations ( Tchernichovski et al . , 1999 ) . LONG birds in the first experiment produced rather poor song imitations , at 59 dph their average similarity score with tutor song was only about 27% . As expected , LONG60 birds tutored in 24 sessions of 90 min produced more accurate song copies: their average similarity score with tutor song exceeded that of LONG birds by 21% on average ( p = 0 . 006 , nonzero fixed effect of 0 . 21 in LME model comparing n = 4 LONG60 with n = 4 LONG birds , t = 2 . 41 , df = 158 ) , Figure 2B . Extended tutoring was again observed together with synapse removal: at 60 dph , both excitatory and inhibitory synapse densities were lower in tutored than in untutored birds ( p = 10−30 for excitatory density and p = 3*10−6 for inhibitory density , LME model , n = 4 LONG60 and n = 4 ISO60 birds ) . At 90 dph , densities remained lower in tutored birds compared to their age-matched untutored controls ( p = 0 . 02 for excitatory density and p = 3*10−5 for inhibitory density , LME model , n = 4 LONG90 and n = 4 ISO90 birds ) , Figure 2C , D . Sparse tutoring was associated with lower excitatory synapse densities than dense tutoring ( p = 3*10−6 , LME model , n = 4 LONG60 and n = 4 LONG birds ) , suggesting that pruning of excitatory synapses is related to song maturation ( i . e . , increases in song similarity scores ) . Indeed , there was a negative correlation between song similarity scores and excitatory synapse densities ( r = −0 . 77 , p = 0 . 026 , Pearson correlation coefficient , n = 4 LONG and n = 4 LONG60 birds ) , Figure 2E . Sparse tutoring did not appear to alter inhibitory synapse densities at 60 dph ( p = 0 . 2 , LME model , n = 4 LONG60 and n = 4 LONG birds ) . In tutored birds , as a function of age , excitatory synapse densities followed a trough-increase trend ( Figure 2C ) , whereas inhibitory synapse densities followed a mirrored peak-decline trend ( Figure 2D ) . In tutored birds from 60 to 90 dph , excitatory synapse densities increased ( p = 9*10−6 , LME model , n = 4 LONG60 and n = 4 LONG90 birds ) whereas inhibitory synapse densities decreased ( p = 0 . 03 , LME model , n = 4 LONG60 and n = 4 LONG90 birds ) . As a consequence , over the course of song development , the percentage of inhibitory synapses in tutored birds ranged from 7% at the onset of song development ( 30 dph ) , to roughly 23% at 60 dph , down to 14% at 90 dph . Thus , the developmental trajectory of the synaptic balance is highly dynamic in normally tutored birds . In untutored birds , synaptic densities did not stay constant , either . The age-related peak-decline trend of inhibitory synapse densities occurred irrespective of sensory experience ( it was seen in both tutored and untutored birds , Figure 2D ) By contrast , in untutored birds , excitatory synapse densities did not decrease until past closure of the critical period , Figure 2C . At the young age of 30 dph , birds normally do not sing yet , except that they occasionally may produce highly unstructured subsongs that do not require HVC ( Aronov et al . , 2008 ) . These observations suggest that abundance of HVC excitatory synapses in young birds is not a requirement for song production , but that instead it signals readiness for sensory memory formation . Moreover , in untutored birds from 60 to 90 dph , both excitatory and inhibitory synapse densities decreased ( p = 0 . 009 for excitatory and p = 3*10−6 for inhibitory synapses , LME model , n = 4 ISO60 and n = 4 ISO90 birds ) , suggesting that high synapse densities are costly and therefore not maintained beyond the critical period . We combined the data from both experiments to perform a group-level analysis on the ( near ) 60 dph animals . Combining the ISO and ISO60 birds , we found that one day of tutoring was associated with a 22% increase in inhibitory synapse densities ( p = 0 . 016 , n = 8 ISO and n = 4 SHORT birds , Wilcoxon ranksum test ) and with a 33% decrease in excitatory synapse density ( p = 0 . 008 , n = 8 ISO and n = 4 SHORT birds , Wilcoxon ranksum test ) . In combination , one day of tutoring was associated with an increase in the percentage of inhibitory synapses from 20% to 29% ( p = 0 . 004 , n = 4 SHORT and n = 8 ISO birds , Wilcoxon ranksum test ) . Combining also the LONG and LONG60 birds , we found that extended tutoring was associated with a decrease in inhibitory and excitatory synapse densities of 38% and 50% , respectively ( both tests p = 0 . 00016 , n = 8 LONG and n = 8 ISO birds , Wilcoxon ranksum test ) , Figure 3A , B . Tutoring has no major influence on HVC size ( Bottjer et al . , 1985; Herrmann and Bischof , 1986; Nordeen and Nordeen , 1988 ) . We therefore expected our findings on synapse densities to translate more or less directly into findings on synapse numbers . Indeed , we found that HVC volumes were variable across animals but did not strongly depend on age or treatment , with the exception that the average HVC volume in 30 dph-old birds was about 30% smaller than in the other bird groups ( see Figure 1—figure supplement 1 ) . Despite this moderate growth in HVC volume from 30 to 60 dph , LONG60 birds tended to have fewer excitatory HVC synapses than did ISO30 birds , evidencing in sparse tutored birds from early to mid-development a process of net excitatory synapse elimination . Synapse insertions and deletions are extreme cases of subtler processes of synapse size changes . To inspect the latter , we reconstructed HVC synapses in brains from the first experiment , imaged with focused ion beam electron microscopy ( FIBSEM , see Materials and methods ) , Figure 4A–C . We found that synapse sizes ( their physical volumes ) were well approximated by log-normal distributions , Figure 4—figure supplement 1 . Log mean synapse sizes displayed large differences across animals . Nevertheless , we found an effect of 40% larger excitatory synapses in SHORT birds compared to ISO birds ( LME , p = 0 . 05 ) , Figure 4D . Extended tutoring was not associated with significant size changes of excitatory synapses ( LME , ISO vs LONG , p = 0 . 42 ) . Sizes of inhibitory synapses were not affected by either short or long tutoring ( LME , p > 0 . 05 ) , Figure 4E . Also , we found no effect of tutoring on Feret diameters ( directional measure of object size ) of either excitatory or inhibitory synapses ( LME , p > 0 . 05 ) .
Our findings suggest that insertion and elimination of HVC synapses are age and experience dependent . There was a strong decoupling among synapse types in that we observed a more than three-fold variation across age and experience in the relative density of inhibitory synapses . After extended tutor exposure , the density , ratio , and average sizes of synapses returned to levels close to those of age-matched untutored controls , which hints at a structural homeostasis ( Turrigiano and Nelson , 2004 ) that is independent of learning experience . Synapse density has been commonly observed to follow a peak-decline trend from youth to adulthood ( Cragg , 1975; Herrmann and Arnold , 1991; Murphy and Magness , 1984; Winfield , 1981a ) . The peak is thought to represent overproduction of neuronal connections , reflecting elevated structural plasticity at the onset of behavioral development . In HVC of normally raised birds , we find an early peak-decline trend for excitatory synapses , and a late such trend in inhibitory synapses . Excitatory synapses were formed very early and in untutored birds they were regulated at a constant rate throughout the sensory learning period from 30 to 60 dph . Inhibitory synapses were formed later , reminiscent of delayed development of inhibition in visual ( Winfield , 1981b ) and auditory cortices ( Dorrn et al . , 2010 ) . Given that increases in inhibition tends to cause the end of experience-related plasticity ( Chen et al . , 2011; Fagiolini and Hensch , 2000; Hensch , 2005; Iwai et al . , 2003 ) , the age- and tutoring dependent increases in inhibitory synapse densities we find raise the possibility that increased inhibition causes not only the closure of the sensory learning period but also accelerates the ending of structural plasticity right after tutoring . Processes of synapse formation and elimination are associated with sprouting and retraction of dendritic spines ( Holtmaat et al . , 2006; Trachtenberg et al . , 2002 ) . Because tutoring is associated with a rapid and monotonic decline in HVC spine turnover ( Roberts et al . , 2010 ) , there seems to be diminished requirement for structural plasticity right after tutoring , in line with our observed increase in inhibitory synapse density that supposedly closes critical period learning . Synapse pruning tends to be an activity-dependent process , for example mediated by microglia ( Schafer et al . , 2012 ) . Tutor exposure is associated with decreases in excitatory synapse densities , and so naively one would expect higher excitatory synapse densities in sparsely tutored birds compared to densely tutored birds . However , we find the opposite , and therefore the better explanation for fewer excitatory synapses in sparsely tutored birds is their higher song maturity ( better song imitation ) . The extent to which higher maturity stems either from better memorization of tutor song or from more targeted song practice remains to be investigated . Neither explanation can be currently ruled out , although our findings align mostly with the former explanation , as detailed in the following two paragraphs . On the one hand , excitatory connections have been proposed to implement a sequence-generating chain network ( Fee et al . , 2004; Kornfeld et al . , 2017 ) . Elimination of HVC excitatory synapses during song maturation agrees with the sparse firing of excitatory motor-projecting neurons in adulthood ( Kozhevnikov and Fee , 2007; Hahnloser et al . , 2002 ) . Over the course of song maturation , HVC projection neurons fire first densely during each syllable and then sparsely during only a single syllable type ( Okubo et al . , 2015 ) . Such sparsening of firing has been suggested to result from a synaptic chain splitting mechanism that underlies the transformation of immature subsongs into more stereotyped plastic songs . Our findings of progressive elimination of excitatory synapses up to 60 dph supports such chain splitting , whereas the subsequent insertion of excitatory synapses ( LONG90 vs LONG60 ) could reflect a process of chain strengthening . On the other hand , excitatory synapses and their dendritic spines can be formed by repetitive motor training ( Xu et al . , 2009; Fu et al . , 2012 ) , which aligns with our observation of increasing excitatory synapse density towards adulthood . Given that optical ablation of task-specific spines in motor cortex can selectively disrupt newly acquired motor skills ( Hayashi-Takagi et al . , 2015 ) , it is likely that the late-formed excitatory synapses between 60 and 90 dph contribute to increased motor performance in adults . Accordingly , our findings support the view that sensory memorization is associated with loss of excitatory synapses , whereas increases in motor performance are associated with insertion of excitatory synapses . Paradoxically , we find a net removal of inhibitory synapses ( LONG vs ISO ) during a developmental phase in which the strength of inhibition onto motor-projecting HVC neurons increases ( Vallentin et al . , 2016 ) . Likely , these contrasting findings can be reconciled by maturation processes such as insertion of AMPA receptors at excitatory synapses ( Hensch , 2005 ) , for which we find indirect evidence in terms of increase in excitatory synapse size ( Nusser et al . , 1998 ) . Additionally , strengthened inhibition with fewer inhibitory synapses can arise when synapse elimination primarily affects competing neuron pools linked by inhibition ( Kornfeld et al . , 2017 ) , that is when eliminated synapses disinhibit the local excitatory-inhibitory pools ( Markowitz et al . , 2015 ) that are associated with learned song syllables ( Vallentin et al . , 2016 ) , Figure 5 . Removal of excitatory inputs is in line with developmental sharpening of excitatory inputs seen in auditory cortex ( Sun et al . , 2010 ) . However , it is unclear how the structural preponderance of excitatory versus inhibitory synapses relates to the functional balance between excitatory and inhibitory synaptic inputs . Functional balances in individual neurons can be measured using whole-cell recordings , but it will be excessively difficult to obtain such information from singing birds , given that such recordings have not been feasible thus far . Overall , our work provides a structural foundation of developmental learning mechanisms and the roles of neurotransmitter systems . Given that the pathophysiology and treatment of many severe mood disorders are linked with excitatory ( glutamatergic ) and inhibitory ( GABAergic ) neurotransmitters , imbalances in these systems either associated with depression or induced by antidepressants ( Sanacora et al . , 2003 ) can interfere with requirements dictated by age and experience . For example , depression during pregnancy extends the critical period for infant speech discrimination , whereas treatment of depression with selective serotonin reuptake inhibitors ( SSRIs ) shortens this critical period ( Weikum et al . , 2012 ) . SSRI-induced shortening of critical periods might be caused by physiological deficits such as premature increases in inhibition ( Sanacora et al . , 2002 ) , premature decreases in excitation , or a combination of both mechanisms , given their known interdependence ( Nakayama et al . , 2012 ) . The roles of amino acid neurotransmitters in depression and mood disorders appear diverse because there have been mixed reports of SSRIs causing either increases or decreases in cortical GABA and glutamate concentration ( Maya Vetencourt et al . , 2008; Sanacora et al . , 2002; Sanacora et al . , 2003 ) . Part of the conundrum might be the alternating experience-dependent dynamics in inhibitory and excitatory synapses we find . Our work suggests that exposure therapies and the pharmacological treatment of mood disorders might benefit from targeted manipulation of both excitation and inhibition , taking into consideration their rich experience-dependent dynamics . One expected immediate effect of a tutoring-induced bias towards inhibition is to reduce neural activity in postsynaptic targets . However , more exciting is the idea that our observed structural changes are associated with holding a memory of the tutor song . On the one hand , a memory could be held by inhibitory synapses inserted right after tutoring . Support for this idea comes from barrel cortex , where single whisker stimulation triggers formation of inhibitory synapses within 24 hr in the corresponding barrel but not in other barrels ( Knott et al . , 2002 ) . On the other hand , a song memory could be formed by irreversible removal of excitatory synapses right after tutoring . Accordingly , the tutor template would be stored by pruning unused excitatory connections , Figure 5 . In this model , the initially formed excitatory connections would represent the entire hypothetical space of notes and note transitions , consistent with zebra finches’ vocal repertoire . The eliminated excitatory connections would represent the notes and note transitions that are not matched by tutor song , whereas the strengthened synapses would be the ones required for tutor song imitation .
34 male juvenile zebra finches ( <90 days ) were raised in our facility at the University of Zurich , Switzerland . At 15 days post hatching ( dph ) , the young birds were transferred together with their mothers to a sound-isolation chamber where they could not hear male songs . Between 25 and 35 dph , sex determination was performed based on feather appearance and on genotyping . At 35 dph , selected male birds were separated from siblings and transferred to individual sound-isolation chambers ( inner volume 60 × 60 × 60 cm3 ) where their songs were recorded . We randomly assigned birds to treatment groups with one exception: to minimize the influence of genetic background on results , siblings from the same nest were never put in the same group . Birds were housed in two cages with inner dimensions 39 × 23 × 39 cm3 ( length ×width × height ) , joined together by the doors , resulting in a movement range of approximately 39 × 46 × 39 cm3 . Each juvenile bird was housed together in the same cage with an adult bird , inside the recording chamber: either a male serving as live tutor or a female serving as companion ( female zebra finches do not sing ) . Singing of juveniles and tutors was recorded continuously and monitored on a daily basis . We aborted the experiment in two juveniles because the tutor failed to sing at least 20 song motifs per day . Birds were maintained on a 14:10 hr light:dark cycle with food and water provided ad libitum . To avoid human bias , we coded all birds with numeric identifiers and performed all analyses in a manner that was blind to bird identity and treatment group . All experimental procedures were in accordance with the Veterinary Office of the Canton of Zurich . Experiment I was designed to investigate whether experience of the tutor song alters excitatory and inhibitory synaptic connectivity in HVC . We divided birds into the following three groups ( Figure 1A ) , each composed of four animals: At 59 dph , all juveniles were deeply anesthetized and perfused . Experiment II was designed to explore changes in HVC excitatory and inhibitory synapses as a function of age . To separately explore age dependence in tutored and untutored birds , we divided birds into the following five groups , each composed of four animals: All vocalizations produced by the juveniles and their companions were recorded with a wall-mounted microphone ( Audio Technica PRO 42 ) , amplified with a microphone preamplifier ( RME Quadmic ) , and digitally sampled at 32 kHz ( PCI card , National Instruments ) . Songs were detected and saved using custom written software ( Labview , National Instruments ) . To evaluate the similarity between the juveniles’ songs and tutors’ songs one day before sacrifice , we randomly selected from the last day of recording 20 song motifs from the pupil and 20 motifs from the tutor . The comparison of the 20 motif pairings was performed using Sound Analysis Pro ( SAP ) ( Tchernichovski et al . , 2000 ) . Birds were sacrificed with 50 µL sodium pentobarbital ( Streuli Pharma AG Esconarkon ) injected intramuscularly . They were perfused with 5 mL 0 . 9% NaCl , followed by 300 ml freshly prepared fixative solution at body temperature ( 4% paraformaldehyde and 0 . 1% glutaraldehyde in 0 . 1 M pH 7 . 4 phosphate buffer ( PB ) ) . Our procedure was adapted from Knott et al . ( 2011 ) , instead we used a slightly reduced glutaraldehyde concentration . After perfusion , brains were dissected from the skull and briefly washed in PB . Prior to sectioning , to enhance cutting stability , we separated the two hemispheres along the midline and embedded each in 3% agar . Parasagittal brain slices of 100 µm thickness were cut on a vibratome ( Thermo Scientific , Microm HM 650V ) and collected at 4°C in 0 . 1 M PB . The sections were first washed in cacodylate buffer ( 0 . 1 M , pH 7 . 4 ) and then incubated for 40 min in 1 . 5% potassium ferrocyanide and 1% OsO4 in cacodylate buffer , following another 40 min incubation in only 1% OsO4 . Thereafter , the sections were incubated for another 40 min in 1% uranyl acetate in double distilled water ( ddH2O ) . After these heavy metal stainings , the sections were dehydrated by 10 min incubations each in a gradient of ethanol solutions with increasing concentrations ( 50% , 70% , 90% , and 95% ) . The sections were further dehydrated by twice incubating in 100% ethanol following twice incubating in propylene oxide , each for 15 min . During the dehydration , small amounts of liquid from the preceding steps remained inside the vial to prevent the tissue from drying and cracking . After dehydration , the sections were immersed into a freshly made epoxy resin for at least 12 hr to reach complete tissue infiltration . The epoxy resin ( Durcupan ACM , Sigma-Aldrich Fine Chemicals , Buchs , Switzerland ) consisted of 10 g: 10 g: 0 . 3 g: 0 . 2 g of component A/M , B , C , and D , respectively . After incubation , the sections were placed between two Aclar films ( ACLAR sheets , Agar Scientific Ltd . , Stansted , UK ) that were sandwiched between two glass slides . A small weight ( ~40 g ) was placed on top of the upper glass slide . The sections were then cured for 48 hr at 52°C . Before proceeding to ultramicrotomy , the ROI in HVC was located by comparing the LM images of a section before and after embedding . With a scalpel blade ( cat . # 10050–00; Fine Science Tools GmbH . , Heidelberg , Germany ) , a small piece of tissue containing HVC was dissected . The small tissue piece was then glued on top of a resin block with the brain tissue side facing down ( remaining Aclar film facing up ) . HVC was located by reference to the LM images and trimmed in a trapezoidal shape with a diamond trimming knife ( trimtool 20 , DiATOME Ltd . , Nidau , Switzerland ) . The Aclar film on top was trimmed away and the surface of the brain tissue block was polished . Serial 70-nm-thick ultrathin sections were cut with an ultramicrotome ( Leica FC6 ) and a sharp diamond knife ( Histo Jumbo , DiATOME Ltd ) . The sections formed a ribbon of tissue that were floating on the surface of the water bath . The section thickness was monitored based on the reflection index ( golden:>100 nm; silver-gray: 50–100 nm ) . Sections that were too thick were discarded from collection . When the desired ribbon of serial ultrathin sections was produced , it was detached from the diamond blade and moved on the water surface with a human eyelash that was fixed to a toothpick . A custom-made silicon wafer ( Si-Mat Silicon Materials , Kaufering , Germany ) was first deionized with a charging generator ( EN SL , Haug Biel AG , Biel , Switzerland ) and then slowly dipped into the water bath . The floating ribbon of ultra-thin sections was flattened and attached to the silicon wafer and both were withdrawn from the water bath . Prior to the ssSEM imaging of the dried ultrathin sections , the silicon wafer was fixed with clips onto a SEM sample holder ( cat . # 16112–20 , Plano GmbH , Wetzlar , Germany ) . HVC volume grows significantly from 20 to 40 dph , after which it reaches 91% of its adult size ( Bottjer et al . , 1985;Herrmann and Bischof , 1986;Nordeen and Nordeen , 1988 ) . The size of HVC is mainly genetically regulated ( Airey and DeVoogd , 2000a ) and manipulations of song experience have little influence ( Brenowitz et al . , 1995; Burek et al . , 1991 ) . However , although HVC volume and neuron number remain constant in adults ( Wang et al . , 2002 ) , both HVC volume and HVC neuron number positively correlate with the number of song syllables copied from a tutor ( Airey et al . , 2000b; Ward et al . , 1998 ) . To measure HVC size in Experiment II , wet brain sections and Nissl-stained brain sections were imaged under a bright field light microscope ( Olympus BX61 ) . Wet brain sections were first mounted onto glass slides and kept moist with PB during imaging . Light microscopy images of HVC were taken at different magnifications ( 1 . 25x , 4x , and 10x ) . To compensate for fluctuations in environmental illumination , microscope parameters were set prior to imaging such that histograms of live images appeared normalized . In order to measure HVC size defined as the physical volume of HVC , we adopted the procedure described by Airey and colleagues ( Airey et al . , 2000c ) . Basically , all images of HVC in a given hemisphere were grouped together and imported as an image stack into a single TrakEM2 project . The images were then aligned using translational and rotational transformations . The HVC outline on each image was hand-drawn and the number n of pixels inside HVC was calculated . With known pixel size p and section thickness d , the size Vsec of HVC in a given brain section was estimated as the product of these three terms: Vsec=npd . The total size VHVC was calculated as the sum over all sections: VHVC=∑sec Vsec and the final HVC size for a given animal was calculated as the average of the right and left HVC sizes , Figure 1—figure supplement 1 . To test for tutoring or aging-related changes in HVC size , we modeled HVC sizes using a linear mixed effects ( LME ) model with bird group as fixed effect and bird identity as random effects ( see Section on Linear Mixed Effects Models ) . We found the following peak-decline trend: at 30 dph , HVC size was smaller than at 60 dph ( p = 0 . 0004 , LME model , ISO30 and ISO60 ) ; and at 60 dph , HVC size was larger than at 90 dph ( p = 0 . 02 , LME model , ISO60 and ISO90 ) . Furthermore , we found that extended tutoring lead to HVC shrinkage at 60 dph ( p = 0 . 03 , ISO60 vs LONG60 ) , whereas no effect of tutoring on HVC size was observed at 90 dph ( p > 0 . 05 , LME model ISO90 vs TUT90 ) , which agrees with previous reports ( Brenowitz et al . , 1995; Burek et al . , 1991 ) . For synapse counting , we performed serial section electron microscopy ( ssSEM ) on thin tissue volumes of dimensions ~ 80 µm×80 µm×140 nm located near the center of HVC . We performed EM imaging with a high-throughput scanning electron microscope ( Merlin , Carl Zeiss Microscopy GmbH , Oberkochen , Germany ) . The section holder was fixed onto the sample stage and loaded into the microscope chamber . The approximate locations of the serial sections were first determined with 5 kV of extra-high tension ( EHT , the voltage applied to the electron gun ) and an in-lens detector . The subsequent working distance ( the distance between the sample surface and the electron gun ) was 3 . 5 mm . The EHT was then reduced to 1 . 6 ± 0 . 1 kV , and the view mode was switched to the energy selective backscattered electron detector ( EsB ) , with the EsB Grid voltage set to 550 V and the detector probe current set to 550 pA . All high-resolution ssSEM datasets were acquired with the EsB detector in the pixel-averaged noise-reducing scanning mode . After fine adjustments of the imaging parameters , the focus and astigmatism correction were optimized . All of these adjustments were set manually with the SmartSEM software ( Carl Zeiss Microscopy GmbH ) . We acquired images with a pixel size of 4 nm and an electron beam dwell time of 7 µs . Image acquisition was performed automatically using the ATLAS 3D software ( Fibics Incorporated , Ottawa , Canada ) . We identified excitatory and inhibitory synapses in EM imagery based on morphology ( Gray , 1969; Klemann and Roubos , 2011 ) . Excitatory ( asymmetric ) synapses typically display a pronounced post-synaptic density ( PSD ) . In contrast , the PSD in inhibitory ( symmetric ) synapses looks similar to the presynaptic membrane , showing no obvious differences in membrane specialization . In addition , asymmetric synapses have wider synaptic clefts and are always associated with larger ( ~40 nm ) and rounder synaptic vesicles when compared with symmetric synapses whose synaptic vesicles are smaller ( ~20 nm ) and of irregular oval shape , Figures 1B and 4B . In both Experiments I and II , physical disector volumes were carefully calibrated before estimating synapse densities . To quantify tissue deformations that can occur during EM staining , embedding , and ultramicrotomy , we took LM and EM images at diverse tissue preparation stages . Deformations resulting from tissue embedding were estimated by comparing LM images of wet brain sections taken prior to resin embedding with LM images of the same sections after embedding . In such image pairs , we identified landmarks such as blood vessels , ventricles , and sharp sample borders . We then estimated the distances S between three pairs of such identified landmarks using Fiji . We assumed that tissue deformations caused by EM staining and embedding were isotropic . We calculated the relative 3D physical volume deformation Re ( shrinkage or dilation ) from the ratio of measured 1-D line distances as Re= Sembedded sectionSwet section3 . For each animal , we estimated the relative 3D physical volume deformation Re ( shrinkage or dilation ) caused by tissue embedding , Figure 1—figure supplement 2 . We found that embedding caused a dilation of 2 . 2 ± 0 . 2 % in each spatial direction . We similarly estimated tissue deformations caused by ultramicrotomy by comparing LM images of trimmed sample block faces with SEM images of ultrathin sections cut from the same sample block , Figure 1—figure supplement 3 . Physical cutting was always along the same direction and thus introduced non-isotropic tissue deformations . Therefore , we separately measured the deformations along the three orthogonal axes X , Y , and Z . We defined the X-axis to be perpendicular to the cutting direction and the Y-axis to be parallel to the cutting direction ( Figure 1—figure supplement 3 , red and blue lines ) . We calculated the deformation RX along the X axis as the deformation ratio: RX= S ( ultrathin section ) S ( embedded section ) , the deformation RY along the y axis was computed analogously . In summary , ultramicrotomy caused on average a 1 . 4 ± 0 . 3 % tissue dilation along the X-axis and a 17 . 9 ± 0 . 3 % shrinkage along the Y-axis ( cutting direction ) . Thicknesses of ultrathin sections along the Z-axis were estimated using a cylindrical diameter method adopted from Fiala and Harris ( Fiala and Harris , 2001 ) . This method provides an estimation of the average separation of a given set of consecutive 2D images based on precise measurements of the diameters of cylindrical objects such as mitochondria , Figure 1—figure supplement 4 . For a mitochondrion i that was longitudinally dissected in the images , we measured the diameter di and counted the number si of sections it spanned . By averaging the resulting ratio disi over all N inspected mitochondria ( N≃20 in each bird ) , we obtained the following estimate t- of mean section thickness: t−= 1N ∑i disi . The average ultrathin section thickness was 1 . 5 ± 0 . 7 % thinner than the target thickness ( 70 nm ) set in the ultra-microtome . The deformation RZ along the Z-axis was estimated as RZ=t−70 nm , where 70 nm represents the advancement of the resin block between two consecutive ultrathin section cuts . The deformations RZ of all samples are depicted as green bars in Figure 1—figure supplement 5; as expected , the tissue deformations caused by ultramicrotomy were non-isotropic , Figure 1—figure supplement 5 . Along the cutting direction , the ultrathin sections shrunk to roughly 80% of their original size ( Y-axis ) , whereas along the perpendicular X-axis there was essentially no deformation . The cumulative tissue deformation Rcum from wet brain sections to embedded ultrathin sections was estimated as Rcum=Re * RX * RY* RZ , Figure 1—figure supplement 6 . This number represents the overall change in tissue volume from the wet state to the ultrathin section state . We calibrated density estimates for each sample with the value of Rcum . The synapse density estimates in Experiments I and II therefore reflected the density in the wet state , which was close to the in-vivo state . On average , physical volumes in ultrathin sections were 16 . 1 ± 0 . 7 % smaller than in wet brain sections . To estimate synapse densities , we used disector counting in ssSEM section pairs , which is a standard stereology method , Figure 6 . Disector methods for synapse counting do not rely on the counting of every synapse in the sample . Disectors provide reliable estimates of object numbers ( errors smaller than 6% ) when the separating distance between consecutive disectors is no larger than twice the mean size of objects . When the separating distance is larger than three times the mean object size , density errors rapidly increase to 27% ( Merchán-Pérez et al . , 2009 ) . Density estimates converge to the true density when more than a hundred disectors are inspected ( Merchán-Pérez et al . , 2009 ) . In each animal , we calibrated disector physical volumes . In experiment I , we inspected on average 96 disectors per animal ( range 68–111 disectors , n = 12 birds ) . Therein , we counted on average 188 dissected synapses per bird ( range 120–245 , n = 12 birds ) , among which on average 42 were symmetric ( range 26–68 , n = 12 birds ) . In experiment II , we inspected on average 102 disectors per animal ( range 77–124 disectors , n = 20 birds ) . Therein , we counted on average 175 synapses per bird ( range 114–250 , n = 20 birds ) , among which on average 29 were symmetric ( range 5–49 , n = 20 birds ) . We found HVC synapse densities in the range 5–8 × 108/mm3 , in consistency with previous EM studies of HVC ( Herrmann and Bischof , 1986; Peng et al . , 2012 ) . Synapse density estimates from ssSEM imagery could in principle be biased because of missed synapses located between two consecutive sections ( e . g . synapses smaller than 70 nm oriented parallel to the imaging plane ) . However , based on our measured synapse sizes of around 200 nm , we expect that our density estimates are not severely affected by such biases ( see Section on Synapse sizes ) . To calculate the percentage of symmetric synapses , we computed in each disector i the total number Nis of dissected symmetric synapses and the total number Ni of dissected synapses . The percentage q of symmetric synapses can be estimated by dividing the sums of these two numbers , q = 100 ∑inNis ∑inNi , where n is the number of disectors . To estimate the variance of the percentage of symmetric synapses , we used a jackknife resampling procedure . Accordingly , the i th estimate qi of symmetric synapse percentage was defined as:qi = 100 ∑k≠inNis ∑k≠inNi . The jackknife estimate q¯ of mean percentage and the estimate σq2 of its variance was then calculated as q¯= 1 n ∑inqi=q and σq2= n-1 n ∑i=1n ( qi-q¯ ) 2 , see ( Efron and Stein , 1981; Shao and Wu , 1989 ) . The mean and variance of symmetric synapse percentage are shown for each bird in Figure 1E . We assessed statistical differences in the percentage of inhibitory synapses among the diverse bird groups in experiment 1 using two independent procedures . Our first approach was to use the method reported in the Results Section based on dividing the disectors in each animal into four separate groups and using an LME model on the 4 independent estimates of percent inhibitory synapses per animal . Our second approach was to use a basic bootstrapping procedure , in which we simulated n=106 experiments with 4 ISO and 4 SHORT birds with Gaussian-distributed percent symmetric synapses , taking the per-animal statistics reported in Figure 1E . Using this latter bootstrapping procedure , we obtained a p-value of p = 5*10−5 , where the p-value represents the fraction of the 106 simulated experiments in which SHORT birds had a lower percentage of inhibitory synapses than ISO birds . Clearly , both approaches yielded the nearly identical statistical result that SHORT birds had a higher percentage of inhibitory synapses than ISO birds . We imaged carbon-coated brain sample blocks with a FIBSEM microscope ( Auriga 40 , Carl Zeiss Microscopy GmbH ) . We first set the working distance to at least 10 mm and obtained live images of the block surface under an EHT of 5 kV and using the secondary electron detector . Thereafter , the working distance was reduced to 4 . 8 mm , and the EHT was reduced to 1 . 7 ± 0 . 2 kV . We ensured that the ROI on the sample surface was located precisely at the coincident point of the cross-beam system of the instrument . We switched the imaging modes back and forth between the SEM and FIB views . The Z-axis was moved only in the FIB view , while the beam shift in the X-Y axis was only used in the SEM view . Once the SEM and FIB views were well aligned , we chose a 40 × 40 µm ROI near the center of HVC while avoiding large synapse-free structures , such as blood vessels and cell bodies . We then exposed the cross-sectional surface by milling a coarse trench into the tissue with a 16-nA ion beam . A fine polish was then performed with a 600 pA ion current . We fine-tuned the electron beam and imaged the exposed cross-section using the SmartSEM software and the EsB grid voltage set to 1 . 3 kV . We applied an angle correction to the acquired images to compensate for the tilting of the sample surface . We set the pixel size to 5 nm and the electron dwell time to 13 µs for all acquired FIBSEM imagery . 10-nm-thick layers were then serially milled away with a 600 pA Ga ion current , and the exposed block surfaces were serially imaged with a 8 × 8 µm square-shaped ROI window defined in the ATLAS 4D software ( Fibics Incorporated ) . The auto-focus and auto-stigmator were used every 90 min to counter the milling-induced focus drift along the Z-axis . We estimated the sizes of HVC synapses in FIBSEM imagery using the following segmentation procedure . The FIBSEM imagery was first aligned using TrakEM2 ( Cardona et al . , 2012 ) and then exported in tiff format . Because of the 2D alignment , the exported images contained margin areas with zero pixel values . We converted the values of these margin pixels to 255 ( from black to white ) using a custom macro implemented in Fiji . We then exported a subset of 200 consecutive images from each dataset to serve as training set for subsequent synapse segmentation . We performed synapse segmentation using Ilastik , an interactive machine-learning toolkit for 3D image segmentation ( Sommer et al . , 2011 ) . The FIBSEM training set and the entire dataset were first imported into Ilastik as single 3D volumes . We performed a semi-automatic pixel classification procedure followed by an object classification procedure ( Kreshuk et al . , 2011 ) . Ilastik uses a random forest classifier ( Breiman , 2001 ) to classify pixels or objects into different classes based on manually labeled objects defined as ground truth . In each training set , we labelled 5 categories of objects: synapses , mitochondria , membrane compartments , vesicles , and areas containing none of these features ( ‘remainder’ ) , Figure 4—figure supplement 2a–b . Large black objects were observed to interfere with synapse segmentation , which is why in a pre-processing step we changed the appearance of the image margins from black to white and why we categorized myelin sheaths and extracellular space as ‘remainder’ . The initial labeling was performed on 20 slices that were evenly distributed across the training set . After each iteration of the supervised pixel classifier , we visualized the resulting pixel prediction map ( Figure 4—figure supplement 2c ) and then we corrected the ground truth until there were no major mistakes left ( i . e . no missing synapses ) . The smoothed pixel prediction map was then subjected to object classification . The object classifier only examined voxels with predicted synapse probabilities above 50% . It also excluded small objects containing less than 1000 interconnected voxels , which were too small to be synapses . The quality of the segmentation was visually verified after object classification . After training and verification , the classifier was applied to the larger dataset that contained all of the images from a given bird . After inspection of the obtained synapse segmentations , the segmentations were exported as a binary image stack in multipage tiff files . The image stack had the same dimensions as the original input stack , its pixel values indicated whether a pixel belonged to a synapse ( value 1 ) or not ( value 0 ) . The FIBSEM datasets were of size 2–4 GB each . During synapse segmentation , Ilastik created several intermediate datasets that were times the size of the original FIBSEM stacks . Therefore , adequate RAM and disk space were required to perform this analysis . We performed all analysis on a 2 . 8 GHz 6-Core Intel Xeon work-station with 24 GB RAM running Windows 7 . We wrote several MATLAB scripts to visualize , process , and analyze the segmentation results . Binary synapse segmentations were colored in red and superimposed on the original FIBSEM stack , Figure 4—figure supplement 2d . The superimposed image stack was visualized in MATLAB with a custom 3D image viewer . Each segmented synapse was recognized as an object composed of connected pixels and assigned an identification number . False positives ( not synapses ) and partial segmentations ( that overlapped with the margin area ) were automatically discarded . Falsely split or merged synapses were automatically identified and corrected . False negatives ( missing synapses ) were counted to ensure that no more than 3% of synapses were missed by our segmentation procedure , otherwise synapse segmentation was performed anew . The corrected and recolored ( blue: symmetric , red: asymmetric , Figure 3b ) synapses were then subjected to volumetric and geometric measurements . The size of a synapse was defined as its physical volume containing the pre- and post-synaptic membranes , their associated synaptic densities , and the synaptic cleft . The precise physical size of a single FIBSEM voxel was estimated using the procedure described in the Section on Physical Section Deformation . For the synapse size and Feret diameter measurements , on average 320 synapses were fully segmented and pooled in each bird ( range 202–406 synapses per bird , n = 12 birds ) . Synapses were classified into asymmetric and symmetric subtypes and their sizes and diameters were measured . Histograms of logarithmically transformed synapse sizes were fitted with Gaussians , Figure 4—figure supplement 1 . To determine how well synapse sizes could be fit with log-normal distributions , we performed goodness-of-fit tests ( Merchán-Pérez et al . , 2014 ) . In all 12 birds , asymmetric synapses were log-normally distributed ( p < 0 . 002 in every animal , Kolmogorov-Smirnov test ) . In 6/12 animals , symmetric synapses were log-normally distributed ( p < 0 . 05 , Kolmogorov-Smirnov test ) , whereas in the remaining six there was a trend for log-normal distribution ( p < 0 . 3 in all six cases ) . Based on these findings , we decided to examine synapse sizes using log-transformed data . For a given synapse i with measured physical volume Vi , the log-transformed size Wi is given by: Wi=log ( Vi ) . Let μ=1N ∑iNWi be the mean of Wi and σ2 its variance , where N is the total number of segmented synapses . The average synapse size V¯ in a given bird is then given by V¯=eμ . The standard error SE of logarithmic synapse size is given by SE=varN , where var=exp ( 2μ+σ2 ) ( exp ( σ2 ) −1 ) , see ( Mood et al . , 1973 ) . Both V¯ and SE are displayed for each animal in Figure 2B , C . We calculated the Feret diameters of synapses , defined as the diameter of the minimum-bounding sphere for each synapse . To improve the efficiency of this calculation , we implemented a linear time-randomized algorithm ( Welzl , 1991 ) . Average Feret diameters of both excitatory and inhibitory HVC synapses types were around 200 nm , which is much larger than the ssSEM-disector distance of 70 nm ( we included an intermediate section in between the reference and look-up sections , the latter were separated by 140 nm , Figure 6B ) . Symmetric synapses had thinner synaptic clefts in the range 20 to 40 nm . Note that flat-shaped synapses parallel to the imaging plane with synaptic cleft size smaller than one third of the disector distance ( 70 nm / 3 ≈ 23 . 3 nm ) could be missed by our disector counting . Therefore , it is possible that the disector counting systematically missed more symmetric synapses than asymmetric synapses . However , given the observed distribution of synapse sizes , we estimate to have missed at most 2 . 2 ± 0 . 7% of symmetric synapses ( assuming random synapse orientations ) . We obtained similar findings for synapse Feret diameters that were log-normally distributed in each animal for asymmetric synapses ( p < 0 . 002 , Kolmogorov-Smirnov test , n = 12 birds ) and that were log-normally distributed in 7/12 animals for symmetric synapses ( p < 0 . 05 ) , with a trend for log-normal distribution in the remaining five animals ( p < 0 . 31 in all five cases ) . Based on these findings , we performed across-group analysis of synapse sizes and synapse Feret diameters using log-transformed data that we fitted with linear mixed effect models . We fitted linear mixed-effect ( LME ) models ( Gelman and Hill , 2007 ) to our synapse data . We interpreted each measurement as a linear mixture of the following terms: a baseline value ( usually representative of the ISO group ) , a fixed effect of tutoring or age , a random individual effect of the bird , and a zero-mean Gaussian observational error of fixed variance . We then evaluated the probability p ( p value ) that the fixed effect of tutoring or age significantly deviated from zero . More specifically , for bird min the ISO group , the synapse density yimISO in disector iwas modeled as the sum of the fixed group mean βISO , the random individual-bird effect bmISO , and the observation errorϵiISO:yimISO=βISO+ bmISO+ϵiISO . The synapse densities yimSHORT of birds in the SHORT group were modeled asyimSHORT=βISO+βSHORT+ bmSHORT+ϵiSHORT , and the synapse densities yimLONG of birds in the LONG group asyimLONG=βISO+βLONG+ bmLONG+ϵiLONG , where βSHORT is the fixed effect of short-term tutoring on synapse densities and βLONG the fixed effect of long-term tutoring . In the Result Section , we report the fixed effect βSHORT and βLONG and their standard errors as obtained with the function fitlmematrix in Matlab ( Mathworks Inc . ) . Tutoring had a significant effect on SHORT densities when the 95% confidence interval for βSHORT did not contain the solution βSHORT=0 ( same for βLONG ) . Reported p values correspond to the confidence of βSHORT≠0 ( same for βLONG ) . Linear mixed-effect modeling of data from the second experiment was performed analogously . | A wide range of species use complex sounds to communicate , including humans and songbirds like zebra finches . During a critical period of learning , infants and young animals learn how to remember and discriminate this ‘language’ from other sounds . However , the changes that happen in the brain during this learning period are not well understood . The process of learning forms new connections between neurons in the brain and prunes away old connections . These connections , known as synapses , come in different types . Signals sent across excitatory synapses increase the activity of the receiving neuron , while signals sent across inhibitory synapses reduce neuron activity . What happens to the synapses in the brain during the critical period ? To find out , Huang et al . used electron microscopy to examine the brains of young zebra finches that either had never heard birdsong , or had just heard birdsong for the first time . A single day of hearing song dramatically shifted the balance of excitatory and inhibitory synapses in the main vocal control area of the young birds’ brains . The number of excitatory synapses decreased , and the number of inhibitory synapses increased . The balance between excitation and inhibition is important for the brain to work correctly . Therefore , as well as helping us to understand how infants learn their first language , the results presented by Huang et al . could also help us to improve treatments for conditions where this balance goes wrong , such as mood disorders . For example , tailoring the time point of medication intake in combination with sensory exposure therapies could improve how effectively either one works . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Excitatory and inhibitory synapse reorganization immediately after critical sensory experience in a vocal learner |
Maintenance of muscle function requires assembly of contractile proteins into highly organized sarcomeres . Mutations in Kelch-like protein 41 ( KLHL41 ) cause nemaline myopathy , a fatal muscle disorder associated with sarcomere disarray . We generated KLHL41 mutant mice , which display lethal disruption of sarcomeres and aberrant expression of muscle structural and contractile proteins , mimicking the hallmarks of the human disease . We show that KLHL41 is poly-ubiquitinated and acts , at least in part , by preventing aggregation and degradation of Nebulin , an essential component of the sarcomere . Furthermore , inhibition of KLHL41 poly-ubiquitination prevents its stabilization of nebulin , suggesting a unique role for ubiquitination in protein stabilization . These findings provide new insights into the molecular etiology of nemaline myopathy and reveal a mechanism whereby KLHL41 stabilizes sarcomeres and maintains muscle function by acting as a molecular chaperone . Similar mechanisms for protein stabilization likely contribute to the actions of other Kelch proteins .
Nemaline myopathy ( NM ) is one of the most severe forms of congenital myopathy , affecting ~1 in 50 , 000 births ( Romero et al . , 2013; Wallgren-Pettersson et al . , 2011 ) . NM encompasses a set of genetically heterogeneous diseases defined by the presence of rod-like structures , called nemaline bodies , in skeletal muscle fibers . Nemaline bodies are formed by the abnormal aggregation of proteins within muscle thin filaments and are associated with myofibril disorganization , reduced contractile force , mitochondrial dysfunction , and a spectrum of abnormalities ranging from mild muscle weakness to complete akinesia . There is no effective treatment for NM and mechanistic insight into the molecular basis of the disorder is lacking . To date , mutations in 11 different genes have been linked to NM , most of which encode components of the sarcomeric thin filament , including nebulin ( NEB ) , actin-1 , tropomyosins-2 and -3 , troponin-1 and leiomodin-3 ( LMOD3 ) ( Agrawal et al . , 2004; Cenik et al . , 2015; Donner et al . , 2002; Johnston et al . , 2000; Lehtokari et al . , 2006; Wattanasirichaigoon et al . , 2002; Yuen et al . , 2014 ) . Interestingly , 3 genes associated with NM , KLHL40 , KLHL41 and KBTBD13 , encode Kelch proteins , characterized by the presence of a Kelch-repeat domain , a BTB/POZ domain involved in protein-protein interaction , and a BACK domain that binds E3 ubiquitin ligases ( Geyer et al . , 2003; Gupta and Beggs , 2014 ) . More than 60 different Kelch proteins have been identified ( Dhanoa et al . , 2013 ) , many of which function as substrate-specific adaptors for Cullin-3 ( CUL3 ) E3 ubiquitin ligase , a component of the ubiquitin-proteasome system ( Genschik et al . , 2013 ) . Modification of proteins by the covalent attachment of ubiquitin to lysine residues via CUL3 serves as a signal for protein degradation . However , there are also Kelch proteins that function independently of proteolysis ( Werner et al . , 2015 ) and much remains to be learned about their functions . Recently , we and others showed that loss of function of the muscle-specific Kelch protein KLHL40 in mice causes NM similar to that seen in human patients with KLHL40 mutations ( Garg et al . , 2014; Ravenscroft et al . , 2013 ) . Unlike other Kelch proteins that promote protein degradation , KLHL40 is required for stabilization of LMOD3 , an actin nucleator , and NEB , a molecular ruler that controls myofibrillogenesis ( Cenik et al . , 2015; Garg et al . , 2014 ) . The absence of LMOD3 or NEB causes lethal NM and severe disruption of skeletal muscle sarcomeric structure and function in mice and humans , confirming the essential roles of these proteins in maintenance of sarcomere integrity . KLHL40 shares 52% identity with KLHL41 , which is also expressed in a muscle-specific manner ( Taylor et al . , 1998 ) . Similarly , KLHL41 mutations in humans have been associated with NM , and morpholino knockdown of KLHL41 in zebrafish causes NM-like abnormalities with aberrant myofibril formation ( Gupta et al . , 2013 ) . However , the molecular functions of KLHL41 and the mechanistic basis of these abnormalities have not been determined . Here , we describe a mouse model of severe NM caused by a loss of function mutation in the Klhl41 gene . Although KLHL40 and 41 share high homology and muscle-specific expression , we show that their mechanisms of action are distinct . KLHL41 preferentially stabilizes NEB rather than LMOD3 , and this activity is dependent on poly-ubiquitination sensed through the BTB domain of KLHL41 . In the absence of KLHL41 , NEB and other sarcomeric components fail to accumulate , resulting in early neonatal lethality . These findings provide new insight into the molecular etiology of NM and also reveal a previously unrecognized role for Kelch proteins in protein stabilization and chaperone activity .
Klhl41 is expressed in a muscle-specific manner with highest levels in adult skeletal muscle relative to the heart ( Figure 1A–B ) . During embryogenesis , Klhl41 is highly expressed in somite myotomes at embryonic day ( E ) 10 . 5 and in skeletal muscles throughout the body at later stages , as detected by in situ hybridization ( Figure 1C ) . To investigate the function of Klhl41 in mice , we obtained embryonic stem cells that contained a LacZ-Neo cassette in intron 1 of the Klhl41 locus from KOMP ( Figure 1—figure supplement 1 ) . In this allele , exon 1 of the Klhl41 gene is predicted to be spliced to LacZ , preventing expression of a functional Klhl41 transcript . Mice heterozygous for the mutant allele ( Klhl41+/− ) were normal and were intercrossed to obtain Klhl41−/−knockout ( KO ) mice . LacZ expression was not detected in Klhl41+/− mice , suggesting that the LacZ cassette was spliced out . Quantitative RT-PCR ( qRT-PCR ) confirmed the complete loss of Klhl41 transcript in KO mice ( Figure 2A ) and western blot analysis revealed loss of KLHL41 protein in muscle from KO mice ( Figure 2B ) . KO mice were born at Mendelian ratios from heterozygous intercrosses and were indistinguishable from WT littermates at birth ( Figure 2C ) . However , KO pups failed to thrive and showed progressive lethality from birth to postnatal day ( P ) 12 , after which no surviving KO mice were observed ( Figure 2D ) . In order to ascertain that the failure to thrive was not due to difficulty in suckling or breathing , we confirmed that KO mice had milk spots comparable to those of their WT littermates . KO mice that survived the early neonatal period displayed severe runting at P3 and P10 ( Figure 2C ) , and their body weight failed to increase with age ( Figure 2E ) , even when other littermates were removed from their mothers , indicating an intrinsic abnormality rather than simply an inability to compete with stronger siblings for nursing . Histological analysis of skeletal muscles of KO mice at various time points revealed occasional ragged fibers ( fibers with discontinuities in their staining ) in the diaphragm and the hindlimbs at P0 and P10 ( Figure 3A and Figure 3—figure supplement 1A ) . Further abnormalities in muscle histology were observed by Gomori’s trichrome staining ( Figure 3B ) . KO myofibers presented abundant depositions that were absent in WT muscle . Desmin staining showed additional cytoskeletal disarray ( Figure 3C ) . In WT muscle at P3 , desmin was evenly expressed throughout transverse sections of myofibers . However , in KO mice , desmin protein aggregates were distributed across myofibers , indicating abnormal sarcomere structure ( Figure 3C ) . We observed a similar general disarray in sarcomeric α-actinin staining in KO muscle ( Figure 3—figure supplement 1B ) . Electron microscopy of diaphragm ( Figure 3D–E ) and hindlimb muscle ( Figure 3—figure supplement 2 ) at P3 also showed sarcomere disarray and Z-line streaming , as well as electron dense inclusions , corresponding to nemaline bodies . Hearts from WT and KO mice were indistinguishable ( Figure 3—figure supplement 1A ) . To further define the muscle abnormalities of Klhl41 KO mice , we compared the protein compositions of WT and Klhl41 KO hindlimb muscle at P0 by unbiased quantitative proteomics . These studies revealed a total of 389 proteins that were up- or down-regulated in muscle from the KO mice , with KLHL41 being the most down-regulated protein ( Figure 4A and Figure 4—source data 1 ) . Analysis of enriched biological pathways using the Database for Annotation , Visualization and Integrated Discovery ( DAVID ) ( Jiao et al . , 2012 ) revealed that ‘sarcomere organization’ and ‘regulation of muscle contraction’ proteins were aberrantly down-regulated in KO mice ( Figure 4—figure supplement 1A ) . Remarkably , NEB was the second most down-regulated protein in the muscle of the KO mice ( Figure 4A ) . Many other down-regulated proteins in the KO mice were essential components of the sarcomere ( Figure 4B ) , including slow skeletal muscle troponin T , myosin light chain-3 , myozenin-3 , and β-tropomyosin ( also associated with NM ) . As revealed by deep sequencing of RNA transcripts ( Figure 4—source data 2 ) , the mRNAs encoding these proteins were unchanged in KO muscle . Therefore , the changes in accumulation of these proteins likely reflect post-translational mechanisms . Notably , western blot analysis showed down-regulation of NEB and only a slight decrease in LMOD3 in Klhl41 KO mice ( Figure 2B ) , while both NEB and LMOD3 were markedly down-regulated in Klhl40 KO mice ( Garg et al . , 2014 ) . The mRNA levels of both proteins did not change in KO muscle , as assessed by RNA-seq and qRT-PCR ( Figure 4—source data 2 and Figure 4—figure supplement 2C ) . We therefore reasoned that KLHL41 could mainly stabilize NEB instead of LMOD3 , further underscoring the possibility that these two Kelch proteins act , at least partially , through different mechanisms . Nebulin-related anchoring protein ( NRAP ) , another member of the nebulin family ( Pappas et al . , 2011 ) , was up-regulated both at protein and mRNA levels ( probably as a compensatory consequence of sarcomere disarray ) ( Figure 4A and Figure 4—source data 2 ) , suggesting that the presence of nebulin repeats is not sufficient for KLHL41 to recognize and stabilize its partners . In contrast to LMOD3 , LMOD2 , another member of the LMOD family involved in thin filament shortening and cardiomyopathy ( Pappas et al . , 2015 ) , was also up-regulated in Klhl41 KO mice both at protein and mRNA levels . Among the pathways identified in the up-regulated proteins in KO mice ( Figure 4—figure supplement 1B ) , we found ‘ubiquitin-dependent catabolic processes’ . Indeed , a remarkable number of up-regulated proteins in the KO mice were involved in ubiquitination , including E3 ubiquitin ligases HERC2 , TRIM63 and TTC3 ( Figure 4B and Figure 4—source data 1 ) . Among these , TRIM63/MuRF1 has been associated with atrophy and degradation of sarcomeric proteins ( Bodine et al . , 2001 ) . Up-regulation of ubiquitination regulators occurred at both protein and mRNA levels ( Figure 4B and Figure 4—source data 2 ) , suggesting that accumulation of sarcomeric proteins within nemaline bodies might activate compensatory protein degradation pathways . Overall , these results indicate that KLHL41 is required to maintain normal levels of sarcomeric proteins in vivo . In contrast to other Kelch proteins , KLHL40 stabilizes its two main binding partners ( NEB and LMOD3 ) instead of degrading them ( Garg et al . , 2014 ) . Due to the high similarity between KLHL40 and 41 ( Figure 4—figure supplement 3A ) , we tested the stabilization of NEB and LMOD3 by KLHL41 in transfected COS-7 cells ( Figure 4C ) . Because of the large size of NEB ( >800 KDa ) , which prohibits efficient expression of the full-length protein , we used a fragment of the NEB protein ( NEBfrag ) previously found to associate with KLHL40 in yeast two-hybrid assays ( Garg et al . , 2014 ) . NEBfrag alone could not be detected by western blot , but NEBfrag protein levels were stabilized when either KLHL40 or KLHL41 were co-expressed ( Figure 4C ) . In contrast , LMOD3 levels only increased modestly in the presence of KLHL41 , whereas KLHL40 overexpression was sufficient to dramatically increase LMOD3 stability ( Figure 4C ) . These results indicate that KLHL41 preferentially stabilizes NEBfrag over LMOD3 , confirming the functional distinctions between KLHL40 and KLHL41 . To explore the molecular functions of KLHL41 , we performed tandem affinity purification ( TAP ) of 3xFLAG-HA tagged KHL41 in C2C12 myotubes and identified KHL41 binding partners by mass spectrometry ( Figure 4D ) . Structural components of the sarcomere , such as NEB , NRAP and filamin-C ( FLNC ) were identified . Notably , mutations in NEB , NRAP , and FLNC have been associated with NM and other myopathies ( D'Avila et al . , 2016; Duff et al . , 2011 ) . KLHL40 was also identified as a KLHL41 binding partner , suggesting a heterodimeric function of these proteins . We further validated the interaction between KLHL41 and both NEBfrag and FLNC by co-immunoprecipitation in COS-7 cells ( Figure 4E and Figure 4—figure supplement 3B ) . We did not observe changes in either NRAP or FLNC protein levels when KLHL41 was co-expressed , indicating that the stabilizing activity of KLHL41 may be partner specific . Overall , these results suggest that KLHL41 interacts with components of the sarcomere and loss of these interactions leads to sarcomeric disarray and NM . LMOD3 was not detected by TAP using 3xFLAG-HA-KLHL41 , nor did we detect an interaction between these proteins in co-immunoprecipitation experiments ( Figure 4—figure supplement 3B ) . These results indicate that despite their similarity in structure , KLHL41 and KLHL40 have both common and distinct partners that may contribute to NM through different mechanisms . Kelch proteins are known to associate with each other and to form complexes with CUL3 ( Dhanoa et al . , 2013 ) . We reasoned that the overlapping partners between KLHL41 and KLHL40 could reflect their association with a common complex . Indeed , in co-immunoprecipitation assays , we found that KLHL41 self-associated and interacted with KLHL40 ( Figure 5A ) , as well as with CUL3 ( Figure 5B ) . To define the domains required for these interactions , we created deletion mutants lacking each of the three annotated domains of KLHL41 ( BTB , BACK and Kelch repeats ) . By co-immunoprecipitation , we observed that deletion of the BTB domain was sufficient to abolish homodimerization of KLHL41 ( Figure 5C ) and its association with CUL3 ( Figure 5D ) . Similarly , deletion of the BTB domain of KLHL40 greatly impaired its association with KLHL41 ( Figure 5E ) . Overall , these experiments indicate that the BTB domain of KLHL41 constitutes a critical region for interaction with other components of the CUL3 complex . Most Kelch proteins act as adaptors that confer substrate specificity to E3 ubiquitin ligase complexes ( Gupta and Beggs , 2014 ) . To assess whether poly-ubiquitination could regulate NEBfrag accumulation , we tested the effect of HA-tagged ubiquitin mutants on NEBfrag protein levels . The formation of poly-ubiquitination chains typically occurs via the covalent attachment of the 76-amino acid ubiquitin peptide to lysine residues of proteins that are targeted for degradation ( Komander and Rape , 2012 ) . The HA-tagged Ubiquitin-K0 ( lysine zero ) ( Ub-K0 ) is a mutant protein that contains all its lysines mutated to arginines , thus preventing poly-ubiquitination in a dominant negative manner ( Kulathu and Komander , 2012 ) . Surprisingly , overexpression of Ub-K0 was sufficient to prevent the accumulation of NEBfrag in the presence of KLHL41 ( Figure 6A ) . In contrast , overexpression of Ub-K0 did not affect LMOD3 stabilization by KLHL40 ( Figure 6—figure supplement 1 ) . These findings suggest that poly-ubiquitination was unexpectedly required for the stabilization of NEBfrag by KLHL41 . Next , we sought to identify the protein target of poly-ubiquitination . Overexpression of Ub-K0 collapsed high molecular bands of KLHL41 corresponding to a potentially poly-ubiquitinated pool ( Figure 6B ) , suggesting that KLHL41 was the target of poly-ubiquitination . Co-immunoprecipitation experiments showed that KLHL41 was poly-ubiquitinated and overexpression of Ub-K0 greatly reduced KLHL41 poly-ubiquitination ( Figure 6C ) . Poly-ubiquitination can take place through any of the 7 lysine residues of ubiquitin or the free amino group from the first methionine . To understand which type of ubiquitination regulated KLHL41 , we used HA-tagged ubiquitin mutants in which individual lysines were mutated to arginine ( K6R , K11R , K27R , K29R , K33R , K48R and K63R ) . We found that overexpression of K48R reduced NEBfrag stabilization by KLHL41 , whereas other mutants did not affect NEBfrag levels ( Figure 6D ) . To identify the region of KLHL41 that is sensitive to poly-ubiquitination , we co-expressed KLHL41 deletion mutants with Ub-WT or Ub-K0 and observed that deletion of the BTB domain was sufficient to strongly impair KLHL41 poly-ubiquitination ( Figure 6C ) . The loss of NEBfrag stability in the presence of Ub-K0 ( Figure 6A ) suggested that reduced poly-ubiquitination could inhibit KLHL41 activity . Based on the preferential poly-ubiquitination of the BTB domain , we tested whether deletion of any functional domain of KLHL41 could rescue NEBfrag levels even in the presence of Ub-K0 . As previously reported for KLHL40 , deletion of the Kelch repeats of KLHL41 only decreased NEBfrag protein levels slightly when Ub-WT was overexpressed ( Garg et al . , 2014 ) . However , when poly-ubiquitination was inhibited by Ub-K0 , ∆BTB-KLHL41 could still stabilize NEBfrag ( Figure 6E ) . While components of the E3 ligase complex are usually ubiquitinated ( Fang et al . , 2000; Nuber et al . , 1998 ) , the requirement of the BTB domain for poly-ubiquitin sensitivity suggests that ubiquitination plays a direct role in the regulation of KLHL41 activity . Therefore , we conclude that poly-ubiquitination of the BTB domain of KLHL41 mediates NEBfrag stabilization . Because mutations in NEB are the most common cause of NM ( Romero et al . , 2013 ) , we investigated the mechanism by which KLHL41 stabilizes NEB . Cycloheximide chase experiments showed that NEBfrag had a short half-life of ~6 hr ( Figure 7—figure supplement 1A ) . We hypothesized that KLHL41 could either prevent degradation of NEB by another E3 ligase or act as a chaperone . Inhibition of proteasomal activity or autophagy by MG132 or chloroquine , respectively , was not sufficient to rescue NEBfrag protein levels in the absence of KLHL41 ( Figure 7A–B ) , suggesting that KLHL41 did not evoke its stabilizing effect on NEBfrag by degrading another E3 ligase . Next , we sought to identify a potential degron in NEBfrag that could target it for degradation . We therefore generated a series of NEBfrag mutants each containing a 25-residue deletion spanning the entire protein ( Figure 7—figure supplement 1B–C ) , and assessed their protein levels by western blot in the presence or absence of KLHL41 . Strikingly , we observed that in the absence of KLHL41 , and in contrast to the full length NEBfrag , most deletion mutants could be detected even in the absence of KLHL41 , albeit at lower levels than with co-expression of KLHL41 and the full length NEBfrag ( Figure 7C , up ) . Furthermore , in the presence of KLHL41 , all NEBfrag deletion mutants exhibited increased protein levels compared to the full length NEBfrag and they interacted with KLHL41 by co-immunoprecipitation ( d5 and d6 had weaker interactions than others ) ( Figure 7C , bottom ) . NEBfrag contains multiple nebulin repeats enriched in basic and aromatic amino acids ( Figure 7—figure supplement 1B–C ) . There are no apparent sequence differences among deletion mutants that could explain the different activities of d5 and d6 from other mutants ( Figure 7—figure supplement 1C ) . The fact that multiple deletion mutants could interact with KLHL41 suggests that the nebulin repeats present in NEBfrag may be sufficient for the interaction . These findings indicate that NEBfrag is unlikely to be regulated by a degron recognized by an unknown E3 ligase but rather general conformational changes promote NEBfrag stability . To assess whether KLHL41 prevented NEBfrag aggregation , we performed protein extraction of the insoluble fraction of transfected COS-7 cells with high detergent concentration . Surprisingly , we found that when NEBfrag was expressed alone , it could be detected in the insoluble fraction , whereas co-expression of KLHL41 was sufficient to shift a portion of the NEBfrag pool to the soluble fraction ( Figure 7D ) . We further validated these results by immunofluorescence ( Figure 7E and Figure 7—figure supplement 1D ) . NEBfrag alone formed aggregates localized predominantly in the nucleus , and co-expression of KLHL41 or KLHL40 resulted in homogenous cytosolic staining . Thus , these results suggest that KLHL41 and KLHL40 can act as chaperones and prevent NEBfrag aggregation .
Protein ubiquitination involves the covalent attachment of the 76-amino acid ubiquitin peptide to the epsilon amino group of target lysine residues . Poly-ubiquitin chains can be formed by sequential addition of ubiquitin molecules through any of its 7 lysines ( K6 , K11 , K27 , K29 , K33 , K48 and K63 ) or the free amino group of the initial methionine ( Yau and Rape , 2016 ) . K48-linked poly-ubiquitination has been extensively characterized as the canonical signal for proteasomal degradation and most Kelch proteins whose functions have been explored mediate protein degradation via E3 ligase-dependent poly-ubiquitination ( Furukawa and Xiong , 2005; Liu et al . , 2016; Shibata et al . , 2013; Xu et al . , 2003 ) . However , new nonproteolytic roles for Kelch proteins are also being reported . For example , KLHL20 controls trafficking of the actin stabilizing protein Coronin 7 via K33-linked poly-ubiquitination ( Yuan et al . , 2014 ) , while mono-ubiquitination NOLC1 and TCOLF by KBTBD8 regulates translation and cell fate specification ( Werner et al . , 2015 ) . Our findings highlight a unique protein stabilizing function of KLHL41 . In contrast to other Kelch proteins , KLHL41 stabilizes its partner , NEB , instead of marking it for degradation . Inhibition of the two major degradation pathways ( the ubiquitin-proteasome and the lysosome ) was not sufficient to rescue NEBfrag levels in the absence of KLHL41 , suggesting that KLHL41 does not regulate another E3 ligase responsible for NEB degradation . Intriguingly , inhibition of poly-ubiquitination abolished the ability of KLHL41 to stabilize NEBfrag . Additionally , we could not identify a degron responsible for NEBfrag instability in the absence of KLHL41 but rather , multiple mutants were stable when short sequences of the protein were deleted . These results suggest that KLHL41 stabilizes NEBfrag by regulating its folding rather than by degrading an unknown protein that might decrease NEBfrag levels . Specific inhibition of K48-linked poly-ubiquitination decreased NEBfrag stabilization by KLHL41 . Besides its canonical role in degradation , K48-linked poly-ubiquitination has been implicated in substrate stabilization ( Flick et al . , 2006 ) and segregation of ubiquitinated proteins from their partners ( Ramadan et al . , 2007; Rape et al . , 2001 ) . In contrast to those studies , however , KLHL41 itself is ubiquitinated instead of its partner . Components of the E3 ubiquitin ligase complex can be self-ubiquitinated during the transfer of ubiquitin to their substrates or as a negative feedback loop to downregulate their protein levels ( Fang et al . , 2000; Nuber et al . , 1998 ) , but the requirement of Kelch protein poly-ubiquitination for protein stabilization constitutes a novel regulatory mechanism . We speculate that K48-linked poly-ubiquitination of KLHL41 could regulate protein-protein interactions between the BTB domain and other partners . K48-linked poly-ubiquitination can act as a recognition signal for CDC48 , a chaperone required to extract misfolded proteins from the ER during ER-associated protein degradation ( Jentsch and Rumpf , 2007 ) . Additional work will be required to understand how KLHL41 activity is regulated in a poly-ubiquitin dependent manner . Although KLHL40 and 41 share extensive homology , and loss of either gene causes NM , our results indicate that these two proteins possess both overlapping and distinct functions . Loss of either KLHL40 or KLHL41 leads to severe NM in humans , whereas Klhl41 KO mice present earlier onset of muscle dysfunction than Klhl40 KO mice . While both proteins stabilize NEB , KLHL41 stabilizes LMOD3 at lower levels than KLHL40 . Considering that the loss of function phenotype of KLHL41 is stronger than that of KLHL40 , it is likely KLHL41 is more critical for sarcomere integrity by interacting with additional partners . The stabilization of LMOD3 by KLHL40 occurs through a mechanism distinct from the stabilization of NEB , as inhibition of poly-ubiquitination does not decrease LMOD3 protein levels in the presence of KLHL40 ( Garg et al . , 2014 ) . Furthermore , in the absence of KLHL40 , LMOD3 levels are increased by proteasome inhibition . KLHL40 and KLHL41 are highly similar in their BTB and BACK domains , which could explain why both are sensitive to poly-ubiquitination . However , they present differences within the Kelch repeats , which likely enables them to discriminate between different substrates . It is currently unknown whether other Kelch proteins , especially those closely related to KLHL41 , act as chaperones or can stabilize their partners in a poly-ubiquitin dependent manner . Of special interest is KBTBD13 , the only other Kelch protein associated with NM . Further work is needed to understand if KBTBD13 acts through a similar mechanism to that of KLHL40 and KLHL41 . NM and other protein aggregation myopathies are characterized by formation of pathogenic protein inclusions ( Goebel and Blaschek , 2011 ) . Although the most common causes for protein aggregation in myofibers are mutations in sarcomeric genes , mutations in the MuRF1 and MuRF3 E3 ubiquitin ligases have also been reported ( Olivé et al . , 2015 ) . Nemaline bodies can be detected in the cytoplasm and the nucleus of patient biopsies , and the presence of intranuclear rods has been associated with more severe clinical phenotypes ( Ryan et al . , 2003 ) . Chaperones have become interesting therapeutic targets in protein aggregation diseases ( Smith et al . , 2014; Winter et al . , 2014 ) . Indeed , during myofibrillogenesis , chaperones are required for proper assembly of the sarcomere and loss of their activity has been associated with a broad array of muscle disorders ( Sarparanta et al . , 2012; Selcen et al . , 2009 ) . Our results reveal an unexpected chaperone activity for KLHL40 and KLHL41 , suggesting that modulation of chaperone activity could represent an approach to treat NM . There are currently no effective therapies for NM . Thus , elucidation of the precise mechanisms of action of KLHL40 and KLHL41 may ultimately allow new interventions into the pathogenic processes associated with this disorder .
All experimental procedures involving animals in this study were reviewed and approved by the University of Texas Southwestern Medical Center’s Institutional Animal Care and Use Committee . Mice were generated using a targeted Klhl41 embryonic stem cell clone ( Klhl41tm1a ( KOMP ) Wtsi ) obtained from the KOMP Repository ( http://www . KOMP . org ) as previously described ( Millay et al . , 2013 ) . Klhl41+/– mice were intercrossed to generate KO mice . KO mice were maintained in a pure C57BL/6 background . Klhl40 KO mice have been previously described ( Garg et al . , 2014 ) . Klhl41 and Klhl40 genotypes were determined based on the presence or absence of WT and KO alleles using two genotyping reactions . The primers for Klhl41 span intron 1 ( WT ) or the targeted allele plus intron 1 ( KO ) . The primers for Klhl40 span exon 1 and a region removed in the targeted allele ( WT ) or the LacZ cassette and a region outside of the KO allele ( KO ) . The following primers were used: PrimerSequence ( 5’ to 3’ ) KLHL41-WT-FAGAAAGTAAGTGCCAAAATGAATCCKLHL41-WT-RAGGCTGACTGTGCTCCTAGGTGCTGTTCKLHL41-KO-FGAGATGGCGCAACGCAATTAATKLHL41-KO-RCAGTTTCTCGTTCAGTTCTTCTCTGKLHL40-WT-FTATATATAGCCCAGGGCAGACAGKLHL40-WT-RCGCACTACACAGTCTAGGAACTTGKLHL40-KO-FTCGAGAGACCTTCCAGTTCKLHL40-KO-RGTCTGTCCTAGCTTCCTCACTG Radioisotopic in situ hybridization ( ISH ) on E10 . 5 , E12 . 5 and E15 . 5 embryonic sections was performed as previously described ( Shelton et al . , 2000 ) . The following sequence was used to detect Klhl41 expression: AGCTGGAATCTAAAGAGTTTGCACCCACTGAAGTCAATGACATATGGAAGTATGAAGATGATAAAAAAGAATGGGCTGGGATGCTGAAGGAAATCCGTTACGCTTCGGGAGCTAGTTGCCTAGCAACGCGCTTAAATCTGTTTAAACTGTCTAAACTATAAAGGAGGTGACAAAGACACAGTTTGAGAGGTGGCTTGTTGGGACAAGAGGCTTTAATTTATTGTCATTCTTTAAGCCTATACAATGATTCACATAGGGTTACAGGGATTCACGCAGTTTCTCTGGGGTAAAACAGTGTAACCGAATCCCAGAGATTTTCAGTGTGCCAAGTATAAAACCATTTGCTAGGAAGTTTAGTATTCAGTTGAACAATATATTTTTTTTTTTTTTTTTTTTGGTTTTTTGAGACAGGGTTTCTCTGTATAGC The DNA was cloned into PCRII-Topo vector ( Life Technologies ) as per manufacturer’s instructions using the following primer set and skeletal muscle cDNA as a template: Kl41-IS-F: 5’- cactgaagtcaatgacatatggaag-3’ , Kl41-IS-R: 5’- gctatacagagaaaccctgtctcaa-3’ . Following cloning , probe sequence was cut out from the PCRII-Topo backbone and transcribed using the MAXIscript in vitro transcription kit ( Life Technologies ) by the T7 and SP6 promoters to generate the anti-sense and sense probes , respectively . Tissues were harvested from 8 week old C57BL/6 mice and flash-frozen in liquid nitrogen . Total RNA was extracted from tissue with TRIZOL reagent ( Invitrogen ) according to manufacturer instructions . The in-situ hybridization probe sequence was used to generate the Northern blot probe by labeling with [α-32P]dCTP using the RadPrime DNA Labelling System ( Invitrogen ) , as per manufacturer’s instructions . As a loading control , 28S and 18S rRNAs from gel run were visualized . RNA levels were measured by in situ hybridization , Northern blot analysis , RNA-seq , and qRT-PCR , as previously described ( Cenik et al . , 2016 ) . RNA-seq ( n = 3 mice per genotype ) was performed by the UT Southwestern Genomic and Microarray Core Facility . For analysis of multiple groups , Holm-Sidak correction for multiple comparisons was utilized with a false discovery rate of 0 . 05 . Data are available from Gene Expression Omnibus ( GSE95543 ) . For qRT-PCR , total RNA was extracted from tissue with TRIZOL reagent ( Invitrogen ) according to manufacturer instructions . cDNA was synthesized using Superscript III reverse transcriptase with random hexamer primers ( Invitrogen ) , as per manufacturer instructions . Gene expression was analyzed by qRT-PCR using KAPA SYBR FAST ( Kapa Biosystems ) . The following primers were used: GeneNameSequence ( 5’ to 3’ ) Klhl41RT1_Klhl41-E1 . 2-FCTGTATGTGGACGAAGAAAATAAGGKlhl41RT1_Klhl41-E1 . 2-RCCACCACATAGATTTTGTCATCTACTKlhl41RT1_Klhl41-E3 . 4-FTTTTTCCAGCTTGATAACGTAACATKlhl41RT1_Klhl41-E3 . 4-RAGATTTTTCACTTCACTCCACTTTGKlhl41RT2_Klhl41-E3 . 4-FCCAGCTTGATAACGTAACATCTGAKlhl41RT2_Klhl41-E3 . 4-RAGATTTTTCACTTCACTCCACTTTGKlhl41RT1_Klhl41-E5 . 6-FGAAGATGGTCTTTCAGCTTCAGTTKlhl41RT1_Klhl41-E5 . 6-RAGTGGGTGCAAACTCTTTAGATTCKlhl40RT_Klhl40-FCCCAAGAACCATGTCAGTCTGGTGACKlhl40RT_Klhl40-RTCAGAGTCCAAGTGGTCAAACTGCAGLmod3RT_Lmod3-FCCGCTGGTGGAAATCACTCCCLmod3RT_Lmod3-RACTCCAGCTCCTTTGGCAGTTGCNebRT_Neb-FTGACTTGAGAAGTGATGCCATTCNebRT_Neb-RCTCTAGCGCCAATGTGGTGAC Skeletal muscle tissues were flash-frozen in a cryoprotective 3:1 mixture of tissue-freezing medium ( Triangle BioSciences International ) and gum tragacanth ( Sigma-Aldrich ) as previously described , followed by sectioning on a cryostat ( Liu et al . , 2014 ) . Hearts and diaphragms were fixed in 4% paraformaldehyde , followed by paraffin embedding and sectioning . Routine hematoxylin and eosin staining was performed on both paraffin-embedded tissue and cryosections as previously described ( Cenik et al . , 2015 ) . Desmin ( M0760 , clone D33; Dako ) , laminin ( L9393; Sigma-Aldrich ) and sarcomeric α-actinin ( A7811 , Sigma-Aldrich ) staining was performed on cryosections of skeletal muscle ( Millay et al . , 2013 ) . Gomori’s trichrome staining was performed as previously described ( Frey et al . , 2004 ) . For electron microscopy , processing of muscle tissues was performed as previously described ( Nelson et al . , 2013 ) . Images were acquired using a FEI Tecnai G2 Spirit Biotwin transmission electron microscope . Tagged KLHL40 , LMOD3 and NEBfrag plasmids were cloned as previously described ( Garg et al . , 2014 ) . A fragment of FLNC ( amino acids 2133–2725 ) cloned into pcDNA3 . 1-FLAG was used from previous studies ( Frey et al . , 2000 ) . A fragment of NRAP ( NCBI reference sequence NM_008733 . 4 , nucleotides 162–1106 ) was cloned into pcDNA3 . 1-myc by conventional PCR using a synthetic gBlock ( Integrated DNA Technologies ) as template . KLHL41 was cloned from mouse quadriceps cDNA using Phusion High-Fidelity DNA Polymerase ( NEB ) and the following primers: KLHL41-F: gatgaattcGATTCCCAGCGGGAGCTTGCAGA KLHL41-R: tgctcgagTTATAGTTTAGACAGTTTAAACAGATTTAAGCGCG KLHL41 was then subcloned into pcDNA3 . 1-FLAG , pcDNA3 . 1-myc and pCS2-3xFLAG-HA . For tandem affinity purification , N-terminal tagged pCS2-3xFLAG-HA-KLHL41 was subcloned into pBx vector . Domain deletions of KLHL41 were generated by conventional PCR and cloned into pcDNA3 . 1-FLAG . The deleted regions were ( numbers correspond to nucleotides in NCBI reference sequence NM_028202 . 3 ) : ΔBTB ( 78-467 ) , ΔBACK ( 518-788 ) and ΔKR ( 789-1898 ) . For generation of ΔBACK , 2 PCR products were used for triple ligation . Domain deletion mutants were subcloned into pCS2-3xFLAG-HA . NEBfrag deletions were generated by conventional PCR and triple ligation into pcDNA3 . 1 ( + ) . pRK5-HA-Ub-WT and pRK5-HA-Ub-K0 were a gift from Ted Watson ( Addgene plasmids #17608 and #17603 respectively ) ( Lim et al . , 2005 ) . The remaining HA-Ubiquitin mutants were cloned by conventional mutagenesis ( QuikChange Lightning Site-Directed Mutagenesis Kit , Agilent ) using HA-Ub-WT as template . CUL3 was cloned from mouse quadriceps cDNA into C-terminal myc pcDNA3 . 1 ( Invitrogen ) using Phusion High-Fidelity DNA Polymerase ( NEB ) and the following primers: CUL3-F: TAAGCAGGTACCCGCCACCATGTCGAATCTGAGCAAAGGC CUL3-R: TGCTTACTCGAGTGCTACATATGTGTATACTTTGCGATC Flash-frozen muscle tissues were dissociated in IP-A lysis buffer ( 50 mM Tris , 150 mM NaCl , 1 mM EDTA , 1% Triton , supplemented with cOmplete mini EDTA protease inhibitor cocktail ( Sigma ) and PhosSTOP phosphatase inhibitor cocktail ( Sigma ) ) using a mini pestle and mechanical disruption with a 25G 5/8 needle . Tissue lysate was centrifuged at 20 , 817 x g for 15 min at 4°C and the supernatant containing protein was transferred to a new tube . Protein was mixed with one volume of 2X Laemmli buffer with 5% β-mercaptoethanol and SDS-PAGE electrophoresis was performed following 5 min of sample incubation at 100°C . All protein was transferred to Immobilon-P PVDF membrane ( EMD Millipore ) using a Mini Trans-Blot cell ( Bio-Rad ) . NEB western blot was performed as previously described ( Zhang et al . , 2017 ) . All antibodies were diluted in 5% non-fat milk in TBS/0 . 1% Tween-20 . Primary antibodies: anti-KLHL41 ( ab66605 Abcam 1:2 , 000 ) , anti-LMOD3 ( 14948–1-AP , Proteintech . 1:500 ) , anti-NEB ( 19706–1-AP , Proteintech , 1:200 ) and anti-GAPDH ( MAB374 , EMD Millipore , 1:10 , 000 ) . All primary antibodies were incubated with blots overnight at 4°C , while secondary antibodies were incubated for 1 hr at room temperature . Tandem affinity purification was performed as previously described ( Garg et al . , 2014 ) . Briefly , 80% confluent Platinum E cells ( Cell Biolabs ) on 15 cm plates were transfected with pBx-3xFLAG-HA-GFP or pBx-3xFLAG-HA-KLHL41 using Fugene 6 ( Promega ) at a 3:1 DNA to Fugene 6 ratio according to manufacturer’s directions . Viral media were collected 24 , 36 and 48 hr after transfection and filtered through a 0 . 45 µm syringe filter ( Corning ) . Polybrene was added to the media at a final concentration of 6 µg/mL . Media were added to 80% confluent C2C12 myoblasts ( American Type Culture Collection ) in growth media ( DMEM with 10% fetal bovine serum and 1% antibiotic-antimycotic ) ( Life Technologies ) . C2C12 cells were then washed with PBS and differentiated for 5 days with differentiation media ( DMEM with 2% horse serum and 1% antibiotic-antimycotic ) ( Life Technologies ) . Protein was collected from C2C12 myotubes by washing cells and then collecting them into 5 mL of PBS by scraping cells with a sterile cell lifter . Protein extraction was done in IP-A buffer according to standard procedures . Anti-FLAG M2 affinity gel ( A2220 , Sigma ) and EZView Red anti-HA affinity gel ( E6779 , Sigma ) were used for pull-down after equilibration in IP-A buffer . 3xFLAG peptide ( F4799 , Sigma ) and HA peptide ( I2149 , Sigma ) were used for elution in IP-A buffer as per manufacturer’s instructions . IP-B buffer ( 50 mM Tris , 700 mM NaCl , 1 mM EDTA , 1% Triton ) was used to wash beads before elution . Silver staining and peptide identification were performed as previously described ( Garg et al . , 2014 ) . C2C12 and PE cells were tested for mycoplasma contamination with Universal Mycoplasma Detection Kit ( 30–1012K , ATCC ) and they were negative for mycoplasma . COS-7 cells ( American Type Culture Collection ) in 60 mm dishes were transfected at 75% confluency using Fugene 6 ( Promega ) as per manufacturer’s directions . A total of 5 ug of DNA was used for each transfection . 48 hr after transfection , cells were washed with PBS and lysed in IP-A for 15 min . Lysates were rotated for 1 hr at 4°C and centrifuged at 20 , 817 x g for 15 min . The pellet was discarded . Standard SDS-PAGE was performed . FLAG M2 ( F1804 , Sigma , 1:5 , 000 ) , mouse c-myc ( R950-25 , ThermoFisher , 1:5 , 000 ) , rabbit c-myc ( sc-789 , Santa Cruz , 1:500 ) , HA ( 32–6700 , ThermoFisher , 1:5 , 000 ) , GAPDH ( MAB374 , EMD Millipore 1:20 , 000 ) , LC3 ( NB100-2220 , Novus Biologicals , 1:2 , 000 ) and fibrillarin ( ab5821 , Abcam , 1:1 , 000 ) antibodies were used for blotting . For solubility assays , the remaining pellet was solubilized in high detergent lysis buffer ( 50 mM Tris , 150 mM NaCl , 1% NP-40 , 0 . 1% SDS ) , boiled and incubated for 24 hr at 4°C under constant mixing . For ubiquitination assays , cells were lysed in IP-A buffer supplemented with 10 mM N-ethylmaleimide ( E3876 , Sigma ) , a deubiquitinase inhibitor . For HA-Ub co-immunoprecipitation , cells were lysed in high detergent lysis buffer with 10 mM N-ethylmaleimide and the final volume was diluted 1:2 with IP-A buffer . For chemical treatments , MG132 ( C2211 , Sigma ) or chloroquine ( C6628 , Sigma ) was dissolved in DMSO and added to COS-7 at a final concentration of 10 µM for 24 hr and 12 hr , respectively . For cycloheximide pulse experiments , cycloheximide ( C7698 , Sigma ) was resuspended in ethanol and added to COS-7 cells at a final concentration of 100 µg/ml . COS-7 cells were tested for mycoplasma contamination with Universal Mycoplasma Detection Kit ( 30–1012K , ATCC ) and they were negative for mycoplasma . COS-7 cells were plated in µ-Slide 8 Well ibiTreat ( 80826 , ibidi ) and transfected with the indicated plasmids . 30 hr later , cells were fixed in 4% paraformaldehyde for 15 min , permeabilized in 0 . 2% Triton X-100 for 20 min and blocked in 5% BSA for 1 hr . Primary antibody incubation was performed for 1 hr with antibodies against c-myc ( R950-25 , ThermoFisher ) , FLAG ( F7425 , Sigma ) and fibrillarin ( ab5821 , Abcam ) . Primary and secondary Alexa-Fluor antibodies were used at 1:200 dilution . Hindlimb muscle from P0 mice ( n = 3 mice per genotype ) was collected for quantitative proteomic analysis by 10-fraction LC/LC-MS/MS by Proteomics and Metabolomics Shared Resource at Duke University . For sample preparation , each sample was added 8 M urea in 50 mM ammonium bicarbonate , pH 8 . 0 at a constant 10 µL per mg of tissue . Samples were then subjected to mechanical disruption using a Tissue Tearer followed by 2 rounds of probe sonication on ice at 30% power . Samples were spun to remove insoluble material and 4 µL was removed and subjected to Bradford assay to determine protein quantity . From each sample , 100 µg of total protein was removed and concentrations were normalized . Samples were then diluted in 1 . 6M urea with 50 mM ammonium bicarbonate . All samples were reduced for 20 min at 80°C with 10 mM dithiothreitol and alkylated for 40 min at room temperature with 25 mM iodoacetamide . Trypsin was added to a 1:50 ratio ( enzyme to total protein ) and allowed to proceed for 18 hr at 37°C . Samples were then acidified with 0 . 2% TFA ( pH 2 . 5 ) and subjected to C18 SPE cleanup ( Sep-Pak 50 mg bed ) . Following elution , all samples were frozen and lyophilized to dryness . For TMT labeling , each sample was resuspended in 100 µL 200 mM triethylamonium bicarbonate , pH 8 . 0 . Fresh TMT reagents ( 0 . 8 mg for each 6-plex reagent ) were resuspended in 100 µL acetonitrile . 50 µL of each TMT tag was added to a specific sample and incubated for 4 hr at room temperature . Afterwards , 8 µL 5% hydroxylamine was added to quench the reaction . 20% of each sample were combined at 1:1:1:1:1:1 ratio and was then lyophilized to dryness prior to LC/LC-MS/MS analysis . Quantitative two-dimensional liquid chromatography-tandem mass spectrometry ( LC/LC-MS/MS ) was performed on approximately 5 µg of protein digest per sample . The method uses two-dimensional liquid chromatography in a high-low pH reverse phase/reverse phase configuration on a nanoAcquity UPLC system ( Waters Corp . ) coupled to a Thermo QExactive Plus high resolution accurate mass tandem mass spectrometer with nanoelectrospray ionization . Peptides were first trapped at 2 ul/min at 97/3 v/v water/MeCN in 20 mM ammonium formate ( pH 10 ) on a 5 µm XBridge BEH130 C18 300 µm x 50 mm column ( Waters Corp . ) . A series of step-elutions of MeCN at 2 µl/min was used to elute peptides from the 1st dimension column . Ten steps of 14 . 0% , 16 . 0% , 17 . 3% , 18 . 5% , 20 . 3% , 22 . 0% , 23 . 5% , 25 . 0% , 30 . 0% and 50 . 0% MeCN were utilized for the analyses; these percentages were optimized for delivery of an approximately equal load to the 2nd dimension column for each fraction . For 2nd dimension separation , the elution of the 1st dimension was first diluted 10-fold online with 99 . 8/0 . 1/0 . 1 v/v/v water/MeCN/formic acid and trapped on a 5µ Symmetry C18 180 µm x 20 mm trapping column ( Waters Corp . ) . The 2nd dimension separations were performed using a 1 . 7 µm Acquity BEH130 C18 75mmx250mm column ( Waters Corp . ) using a 90 min gradient of 3% to 25% acetonitrile with 0 . 1% formic acid at a flow rate of 400 nL/min with a column temperature of 55°C . Data collection on the QExactive Plus mass spectrometer was performed in a data-dependent acquisition mode of acquisition with a r = 70 , 000 ( at m/z 200 ) full MS scan from m/z 375–1600 with a target AGC value of 1e6 ions followed by 20 MS/MS scans at r = 17 , 500 ( at m/s 200 ) at a target AGC value of 5e4 ions . A 30 s dynamic exclusion was employed to increase depth of coverage . The total analysis cycle time for each sample injection was approximately 2 hr . Following the 10 LC-MS/MS analyses , raw data were processed by Protein Discoverer to create MGF files . These MS/MS data were searched against a SwissProt_Mouse database within Mascot Server ( Matrix Science ) that also contained a reversed-sequence ‘decoy’ entry for each protein for false positive rate determination . Because mouse nebulin ( NEB ) is not a reviewed entry in SwisProt_Mouse , the unreviewed entry E9Q1W3_MOUSE was manually included in the analysis . Search tolerances were 5ppm precursor and 0 . 02 Da product ions with full trypsin protease rules and up to two missed cleavages . Search results were imported to Scaffold Q + S v4 . 4 . 6 ( Proteome Software ) and data was annotated at a Protein False Discovery Rate of 1 . 0% . The overall dataset yielded identifications for 23 , 910 TMT labeled peptides corresponding to 4 , 418 TMT labeled proteins . Only peptides uniquely identified to a protein were considered . To normalize the six different channels to account for differences in labeling efficiencies and mixing percentages , the summed intensity for each channel was calculated and then normalized to the 126 channel . Then , protein level intensities were generated by summing all of the unique peptide intensities to that protein . A list of proteins differentially changed between WT and KO is presented in Figure 4—source data 1 . For representation , proteins more than 30% up- or down- regulated were selected as input for Morpheus ( https://software . broadinstitute . org/morpheus/ ) . Z-score was used for heat map scale . For pathway analysis , a list of significantly up- or down-regulated proteins was used as input for DAVID analysis of enriched GO Terms related to biological pathways ( https://david . ncifcrf . gov ) ( Jiao et al . , 2012 ) . Data were presented as mean ±SEM . Differences between two groups were tested for statistical significance using the unpaired two-tailed Student’s t test . p<0 . 05 was considered significant . For analysis of multiple groups , we utilized the Holm-Sidak correction for multiple comparisons with a false discovery rate of 0 . 05 . | Together with the tendon and joints , muscles move our bones by contracting and relaxing . Muscles are formed of bundles of lengthy cells , which are made up of small units called sarcomeres . To contract , the proteins in the sarcomere need to be able to slide past each other . In healthy muscle cells , the proteins in the sarcomeres are evenly distributed in an organized pattern to make sure that the muscle can contract . However , when some of the proteins in the sarcomere become faulty , it can lead to diseases that affect the muscles and movement . For example , in a genetic muscle disease called nemaline myopathy , the proteins in the sarcomere are no longer organized , which leads to a build-up of proteins in the muscle fiber . These protein masses form rod-like structures that are lodged in between sarcomeres , which makes it more difficult for the muscles to contract . This can cause muscle weakness , difficulties eating or breathing , and eventually death . A common cause of nemaline myopathy is mutations in the gene that encodes the nebulin protein , which serves as a scaffold for the sarcomere assembly . Other proteins , including proteins named KLHL40 and KLHL41 , have also been linked to the disease . These two proteins are both ‘Kelch’ proteins , most of which help to degrade specific proteins . However , a recent study has shown that KLHL40 actually stabilizes nebulin . As the two Kelch proteins KLHL40 and 41 are very similar in structure , scientists wanted to find out if KLHL41 plays a similar role to KLHL40 . Now , Ramirez-Martinez et al . have created genetically modified mice that lacked KLHL41 . These mice showed symptoms similar to people with nemaline myopathy , in which the sarcomeres were disorganized and could not form properly . Further experiments with cells grown in the laboratory showed that KLHL41 stabilized nebulin by using a specific chemical process that usually helps to degrade proteins . These results suggest that Kelch proteins have additional roles beyond degrading proteins and that some proteins linked to nemaline myopathy may actively prevent others from accumulating . A next step will be to find drugs that can compensate for the lack of KLHL41 . A better understanding of the causes of this fatal disease will contribute towards developing better treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2017 | KLHL41 stabilizes skeletal muscle sarcomeres by nonproteolytic ubiquitination |
Filopodial dynamics are thought to control growth cone guidance , but the types and roles of growth cone dynamics underlying neural circuit assembly in a living brain are largely unknown . To address this issue , we have developed long-term , continuous , fast and high-resolution imaging of growth cone dynamics from axon growth to synapse formation in cultured Drosophila brains . Using R7 photoreceptor neurons as a model we show that >90% of the growth cone filopodia exhibit fast , stochastic dynamics that persist despite ongoing stepwise layer formation . Correspondingly , R7 growth cones stabilize early and change their final position by passive dislocation . N-Cadherin controls both fast filopodial dynamics and growth cone stabilization . Surprisingly , loss of N-Cadherin causes no primary targeting defects , but destabilizes R7 growth cones to jump between correct and incorrect layers . Hence , growth cone dynamics can influence wiring specificity without a direct role in target recognition and implement simple rules during circuit assembly .
Live dynamics data in intact nervous systems are critical to understand developmental processes and mutant phenotypes during the establishment of synaptic connectivity . However , most analyses of molecular perturbation experiments are based on fixed tissue and most live data are obtained in cell culture . Dynamics measurements have been difficult to obtain in intact developing brains at the resolution of growth cone filopodia , especially over long developmental time periods , in any organism ( Mason and Erskine , 2000; Langen , et al . , 2015 ) . Growth cone filopodia have been shown to follow guidance cue gradients ( Gallo and Letourneau , 2004; Zheng , et al . , 1996 ) and provide physical support for growth cone migration ( Heidemann , 1990; Chan and Odde , 2008 ) . They have also been associated with dendritic spine formation ( Sekino et al . , 2007 ) . However , filopodia may exhibit very different and changing roles during the lifetime of a growth cone ( Mason and Erskine , 2000; Kolodkin and Tessier-Lavigne , 2011; Mason and Wang , 1997 ) . The types and roles of filopodial dynamics that control specific growth cone behaviors during neural circuit assembly in developing brains are largely unknown . Amongst genetic model organisms with a complex brain , Drosophila provides a unique combination of small size , rapid development and the ability to culture developing eye-brain complexes ( Gibbs and Truman , 1998; Ayaz , et al . , 2008 ) . The fly visual system provides a well-studied model for axon outgrowth , targeting , layer formation , and quantitative synapse formation ( Hadjieconomou et al . , 2011; Clandinin and Feldheim , 2009; Feller and Sun , 2011 ) . The fly’s compound eye is an assembly of ~750 ommatidia . Each ommatidium contains six outer photoreceptors ( R1–R6 ) that terminate in the first optic neuropil , the lamina; the axons of the central photoreceptors R7 and R8 establish a retinotopic array of terminals in two separate layers of the second optic neuropil , the medulla . In particular , the development of the deepest projecting photoreceptor neuron , subtype R7 , has been analyzed in great detail from axon outgrowth to layer-specific targeting and synapse formation ( Hadjieconomou et al . , 2011; Feller and Sun , 2011; Ting , et al . , 2007; Ting , 2005 ) . However , to our knowledge , these steps have not yet been shown in the living , developing brain and the underlying types and roles of growth cone dynamics are unknown . We have recently performed a slow time-lapse intravital imaging study of photoreceptor R1-R6 growth cone dynamics in intact pupae ( Langen et al . , 2015 ) . However , intravital imaging has reduced resolution in deeper brain regions and is limited to early pupal stages . Previous imaging in cultured brains established high-resolution imaging in short developmental time windows ( Medioni et al . , 2015; Zschätzsch , et al . , 2014 ) and over long periods at low resolution and with slow time lapse ( Rabinovich et al . , 2015 ) , thus preventing in depth analysis of the role of filopodial dynamics during an entire neural circuit assembly process . Here we present the development of an ex vivo imaging method for Drosophila eye-brain development using 2-photon microscopy that allows widely applicable continuous , fast , high-resolution 4D live imaging anywhere in the fly brain throughout pupal development . We present R7 growth cone imaging at single filopodium resolution for both long periods ( up to 24 hr per session ) and at high temporal resolution ( <1 min ) , without deleterious effects on normal development . Our measurements show that R7 growth cones do not actively extend after initial target recognition . Concurrently , the vast majority of R7 filopodia are motile and function in growth cone stabilization during layer formation . Loss of the cell adhesion molecule N-Cadherin ( Ting , 2005;Lee , et al . , 2001; Nern , et al . , 2008 ) reduces filopodial dynamics and causes destabilization of R7 growth cones , resulting in active growth cone ‘jumping’ between layers even days after targeting has been concluded in wild type . These findings reveal an unexpected role for growth cone filopodia during layer formation and highlight the importance of assessing subcellular dynamics in relation to long-term neuronal development during brain wiring .
In preparation for fast and high-resolution imaging of growth cone dynamics through different phases of brain development ( Figure 1a ) we systematically tested and adapted culture methods of developing Drosophila brains ( Gibbs and Truman , 1998; Ayaz , et al . , 2008 ) in an imaging chamber ( Figure 1b; detailed description in Online Methods and Figure 1—figure supplement 1 ) . Pupal eye-brain complexes dissected at P + 24% exhibited only minor overall shape changes after 24 hr culture in the imaging chamber ( Figure 1c , d ) . Eye pigmentation became apparent at the end of this time period , indicating developmental progress . In contrast , a parallel control in which the brain developed normally in the pupae during the same time period exhibited a more pronounced expansion of the eye discs , but not yet any obvious eye pigmentation ( Figure 1e ) . Similarly , eye-brain complexes dissected at P + 50% and cultured for 24 hr exhibited increasing eye pigmentation , but less overall shape changes than a brain developed inside the fly ( Figure 1f–h ) . 10 . 7554/eLife . 10721 . 003Figure 1 . Development of Drosophila pupal brains in an imaging chamber . ( a ) Timeline of photoreceptor circuit formation during brain development and the periods accessible by live imaging . ( b ) Ex vivo imaging chamber , top ( left ) and side ( right ) views ( see Figure 1—figure supplement 1 for step-by-step assembly ) . ( c-h ) Changes in brain morphology during development ex vivo v . in vivo . ( c , f ) Pupal brains dissected at P + 24% and P + 50% . ( d , g ) The same brains after 24 hr of development ex vivo . ( e , h ) Brains that were dissected from pupae collected at P + 24% and P + 50% and aged in parallel to the ex vivo brains . See Figure 1—figure supplement 2 for comparison with free-floating cultures . ( i-p ) Optic lobe development ex vivo v . in vivo ( i’-p’ ) magnified details of ( i-p ) . All photoreceptors express CD4-tdGFP . Initial layer separation ( P + 24% + 19 hr ) occurs ex vivo ( i’ , j’ ) similarly to the in vivo controls ( k’ , l’ ) aged in parallel ( blue arrows: R8 , green arrows: R7 ) . Lamina rotation ( red arrows ) observed in vivo ( k , l ) is defective ex vivo ( i , j ) . Final layer formation and lamina expansion ( P + 40% + 18 hr ) occurs similarly ex vivo and in vivo , ( m’-n’ ) v . ( o’-p’ ) ( arrows ) and ( m-n ) v . ( o-p ) ( between arrowheads ) , respectively . Note that for the ex vivo brains , images of the same specimens were taken at different time points , while for the in vivo controls different brains had to be fixed and imaged for the different time points . Scale bars , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 00310 . 7554/eLife . 10721 . 004Figure 1—figure supplement 1 . Culture imaging chamber . ( a ) Step-by-step construction of the imaging chamber . ( i ) Spacers are placed on the Sylgard layer in a triangle formation . ( ii ) A drop of diluted dialyzed agarose is pipetted onto the Sylgard . ( iii ) Dissected eye-brain complex is placed into the agarose drop . ( iv ) The mix is covered with a coverslip . ( v ) After the agarose polymerization , remaining space under coverslip is filled with the culture media; ( vi ) and sealed completely with rubber cement . The schematic of the final chamber ( b ) from the side and ( c ) the top . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 00410 . 7554/eLife . 10721 . 005Figure 1—figure supplement 2 . Brain development in imaging chamber compared to liquid media . Changes in brain morphology during development ex vivo in chamber vs . ex vivo in liquid media ( free floating ) vs . in vivo; from brains dissected at P + 24% ( a-d ) , P + 50% ( f-i ) and cultured for 24 hr . Parallel developed in vivo controls ( e , j ) were dissected at the end of cultures . At early stages , brains that developed in the imaging chamber ( b ) are more similar in morphology to the in vivo controls ( e ) than the brains that developed in fully liquid media ( d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 005 To analyze development at the level of axon targeting , we expressed a membrane-tagged GFP ( CD4-tdGFP ) ( Han et al . , 2011 ) in all photoreceptor neurons . We again compared the development of a brain in culture ( from now on referred to as ex vivo ) , at different time points with brains that developed normally inside the fly ( from now on referred to as in vivo ) . As shown in Figure 1i , l and Video 1 , the distance between the terminals of R7 and R8 axons increase significantly after 19 hr ex vivo and in vivo ( blue and green arrows ) . In contrast , a prominent rearrangement of neuropils , where the lamina repositions itself around the medulla , appeared to occur only partially ex vivo ( red arrows Figure 1k–l ) . We next compared 18 hr ex vivo and in vivo development of photoreceptor axon projections starting at P + 40% . We observed increases of both lamina and distal medulla thickness ( between arrowheads ) that occurred both ex vivo and in vivo , suggesting similar developmental progress ( Figure 1m–p and Video 1 ) . These observations suggest that large-scale morphological rearrangement may require factors outside of the eye-brain complex but may have no obvious effects on axon targeting inside the brain . 10 . 7554/eLife . 10721 . 006Video 1 . Ex vivo imaging of Drosophila brain development in culture . All photoreceptors are labeled with CD4-tdGFP . Two live imaging sessions ( 30 min intervals ) starting at P + 24% ( 19 hr ) and at P + 40% ( 18 hr ) are shown . Four developmental processes ( i ) lamina rotation ( ii ) lamina column expansion ( iii ) first-stage separation of R7 and R8 terminals and iv ) Final layer formation of R7 and R8 terminals , are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 006 Next , we sought to determine ( 1 ) to what extent the rate of development is affected in culture and ( 2 ) the effect of continuous laser scanning during development . To measure developmental speed we compared stage-matched brains ex vivo and in vivo . To assess the effect of laser scanning , we compared a continuously imaged optic lobe ( scanned every 30 min for approximately 15 min ) with an unscanned control optic lobe of the same brain ex vivo . For brains dissected at P + 22% and cultured for 20 hr we found no difference in distal medulla expansion between the optic lobes subjected to continuous laser scanning and the control optic lobes within the same brains ( Figure 2a , b ) . However , this expansion in the ex vivo brains was more rapid than in brains kept in vivo; the latter required 10 hr extra to achieve the same distal medulla size ( Figure 2a , b ) . Importantly , the final thickness of the distal medulla was identical and did not increase further in both cases . 10 . 7554/eLife . 10721 . 007Figure 2 . Effects of culture conditions and laser scanning on the optic lobe development ex vivo . Two-photon imaging of the medulla was performed with brains cultured at P + 22% for 20 hr ( a ) and P + 41% for 19 hr ( d ) all photoreceptors express CD4-tdGFP . For each experiment one image stack was acquired containing both optic lobes of a brain . Next , only one of the lobes was scanned every 30 min . Finally , another stack was acquired with both lobes . Different brains aged in parallel in pupae have been dissected as in vivo controls . ( b ) Quantification of the layer distance increase in P + 22% cultures . The distance between R8 ( green rectangles in a , d ) and R7 ( blue rectangles ) layers increase identically in scanned and unscanned ex vivo lobes , but higher than the in vivo control ( p = 0 . 0036 , n = 3 ) . ( c ) Quantification of the change in the angle between the planes of posterior lamina and the anterior medulla . Ex vivo lobes rotate similarly but slower than in vivo controls ( p <0 . 0001 , n = 3 ) . ( e ) Quantification of the layer distance increase in P + 41% cultures . All groups show a similar increase in the distance between R8 temporary layer and R7 terminals . Error bars depict SEM . ( f ) Calibration of the developmental speed in culture to in vivo development , based on distal medulla expansion . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 00710 . 7554/eLife . 10721 . 008Figure 2—figure supplement 1 . Lamina rotation is incomplete ex vivo . Two-photon imaging of the medulla was performed with brains cultured at P + 22% for 20 hr , all photoreceptors express CD4-tdGFP . Continuously scanned ex vivo culture , unscanned control optic lobe and in vivo ( fixed ) control experiments were done as described in Figure 2 . The angles ( blue arches ) between the planes of posterior lamina and anterior medulla have been measured for the start and end points of each culture as well as the corresponding in vivo controls; and plotted in Figure 2c . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 00810 . 7554/eLife . 10721 . 009Figure 2—figure supplement 2 . Effects of 20-Hydroxyecdysone ( 20-HE ) and type of microscope on imaging in the culture chamber . ( a-h ) 20-HE is required for early but detrimental to late pupal development in the optic lobe . ( a-d ) All photoreceptors were labeled with CD4-tdGFP . Cultures were set-up at P + 22% ( a ) , with ( b ) or without ( c ) 20-HE in the culture media . Parallel developed pupae were dissected and imaged at the end of cultures as in vivo controls ( d ) . R7-R8 layer separation in the medulla was impaired in cultures without 20-HE compared to in vivo controls or cultures with 20-HE . Scale bars , 10 μm . ( e-h ) All photoreceptors were labeled with td-Tomato and R7 cells were sparsely labeled with CD4-tdGFP using GMR-FLP through MARCM . Cultures were set-up at P + 22% ( e ) , with ( f ) or without ( g ) 20-HE in the culture media . Parallel developed pupae were dissected and imaged at the end of cultures as in vivo controls ( h ) . R7 axons that developed in the presence of 20-HE showed excessive filopodial formations on their terminals compared in vivo controls or the cultures without 20-HE . Scale bars , 4 μm . ( i ) Comparison of resonant confocal and 2-photon microscopy signal stengths in the imaging culture chambes . R7 cells were sparsely labeled with CD4-tdGFP using GMR-FLP through MARCM . Individual R7 growth cones were imaged in the culture chamber at P + 30% . Images were acquired with a Leica TCS SP5 confocal microscope with a resonant scanner or a Zeiss LSM 780 multiphoton microscope at various depths from the coverslip . The confocal signal reduces below 60 μm compared to the 2-photon signal . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 009 Next , we analyzed the development of brains dissected at P + 41% and cultured for 19 hr . Similar to the earlier time point , we found no differences in the levels of distal medulla expansion between the continuously scanned and unscanned optic lobes . In addition , we found no quantitative differences with in vivo controls for medulla expansion at this later stage ( Figure 2d–f ) . For the effects of application times of the molting hormone 20-Hydroxyecdysone and usage of a resonant confocal microscope vs a 2-photon microscope see Figure 2—figure supplement 2 . In sum , at this resolution the developmental outcomes appear normal in culture , and are not affected by continuous 2-photon imaging . We next set out to image R7 growth cone dynamics at high resolution . To visualize individual growth cones we sparsely labeled ~10% of R7 cells , the deepest projecting photoreceptor axons in the Drosophila brain ( Fischbach and Dittrich , 1989 ) through MARCM ( Lee and Luo , 1999 ) using GMR-FLP . The development of R7 axons , particularly with regards to their layer specificity ( Ting , 2005; Lee , et al . , 2001; Clandinin , et al . , 2001 ) and columnar restriction ( Ting et al . , 2007; Ting , et al . , 2014 ) has been studied extensively in fixed preparations . As before , we compared brains ex vivo and in vivo , which were staged in parallel . We compared cultures starting at P + 20% ( Figure 3a ) , P + 40% ( Figure 3b ) and P + 55% ( Figure 3c ) that were imaged continuously for up to 20 hr each ( Video 2 ) ; together these time intervals cover the development from layer selection to synapse formation over a 50 hr time period . As expected , live imaging deep in the brain leads to significant loss of fine structure; it was difficult to ascertain many faintly labeled filopodia and as a result we consistently counted only about half as many filopodia ex vivo compared to in vivo fixed controls ( Figure 3a–d ) . However , this undercount affected filopodia of different lengths equally , resulting in the same mean lengths ( Figure 3e ) . Amongst live preparations , a P + 20% preparation after additional 20 hr in culture looked qualitatively and quantitatively identical to fresh preparation at P + 40%; similarly a P + 40% preparation after additional 15 hr in culture looked like a fresh preparation at P + 55% . We conclude that changes in the R7 growth cone structure occur similarly ex vivo and in vivo at this resolution . 10 . 7554/eLife . 10721 . 010Figure 3 . Different filopodial signatures accompany separate circuit formation steps . Slow ( 30 min interval ) time-lapse imaging of pupal brains dissected at P + 20% ( a ) , P + 40% ( b ) and P + 55% ( c ) in comparison with in vivo fixed controls at the same stages . The same growth cones were analyzed for all live imaging experiments while different samples from parallel aged pupae had to be dissected for the in vivo controls . All photoreceptors were labeled with myr-mRFP and R7 cells were sparsely labeled with CD4-tdGFP using GMR-FLP through MARCM . ( a ) As the R7 and R8 layers go through their initial separation ( upper panel ) , R7 terminals have numerous filopodia that invade neighboring columns ( lower panel ) , which are pruned around P + 40% both ex vivo and in vivo . ( b ) As the layers start to reach their final configuration , R7 terminals form a bipartite structure around P + 50% . Filopodia numbers remain low . Around P + 55% , more ( shorter ) filopodia are observed again as R7 axon assumes a brush-like look . ( c ) After P + 55% shorter filopodia are pruned and R7 growth cones form new , longer filopodia that are fewer in number and have bulbous tips ( arrows ) . Quantifications of ( d ) total number of filopodia per growth cone and e , mean length of filopodia through the ex vivo experiments ( a-c ) and respective in vivo controls . Error bars depict SEM . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01010 . 7554/eLife . 10721 . 011Figure 3—figure supplement 1 . Filopodial dynamics are restricted to the growth cone and axon shaft inside the medulla neuropil . ( a ) representative R7 terminal structures inside the medulla neuropil ( grey background ) reveal the transition of a more classical growth cone to a branched axonal structure . ( b ) 3D visualization of individual R7 axons ( green ) on the background of all photoreceptors ( magenta ) at P + 70% . ( c ) analysis of R7 axons and extended growth cones/axon shafts in the medulla reveals that filopodia only occur within the medulla neuropil . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01110 . 7554/eLife . 10721 . 012Video 2 . Long-term ex vivo imaging of R7 photoreceptor growth cone filopodial dynamics . All photoreceptors are labeled with myr-tdTomato and R7 photoreceptors are sparsely labeled with CD4-tdGFP using GMR-FLP . Three live imaging sessions ( 30 min intervals ) starting at P + 22% ( 21 hr ) , P + 42% ( 19 hr ) and P + 55% ( 15 hr ) are shown . The development of the filopodial structure of R7 growth cones are shown throughout layer and synapse formation . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 012 The high-resolution structure of R7 growth cones revealed two distinct filopodial ‘signatures’ before and after P + 50% . Prior to P + 50% , we observed that each R7 growth cone had numerous filopodia that invaded several neighboring columns . During this period R7 growth cones slowly became restricted to their individual columns ( Figure 3a , lower panel; Figure 3—figure supplement 1a ) . Around P + 40% R7 growth cones underwent extensive loss of filopodia ( Figure 3d–e; Figure 3—figure supplement 1a ) . Initial high filopodial activity coincides with the beginning of layer separation ( Figure 3a , upper panel ) as lamina neuron axons intercalate between R7 and R8 terminals ( Ting , 2005 ) . Afterwards , R7 terminals have significantly fewer and shorter filopodia during the remainder of R7/R8 layer selection ( P + 45–55%; Figure 3b; Figure 3—figure supplement 1a ) . Surprisingly , longer filopodia reemerge after P + 55% ( Figure 3c , e ) . These are fewer in number per growth cone compared to the early-stage filopodia . In addition , these late-stage filopodia often develop bulbous-like tips ( Figure 3c , arrows ) unlike any of the earlier filopodia . These observations suggest previously undescribed filopodial dynamics that start after P + 55% . During and after this time the R7 terminal undergoes a transition from a morphologically distinct growth cone to an elongated structure with a branched axon shaft , reminiscent of previously observed axonal filopodia in spinal cord culture ( Gallo and Baas , 2011; Spillane , et al . , 2012 ) . However , we only observed filopodia within the medulla neuropil where active layer formation and synapse formation occurs , and not on the main axon leading to the medulla neuropil ( Figure 3—figure supplement 1 ) . In summary , distinct growth cone structures accompany separate developmental events and suggest different roles of filopodia during columnar restriction , layer separation and synaptogenesis . To correlate fast filopodial dynamics with developmental events that are hours apart , we applied an imaging protocol that alternated between slow time-lapse imaging of the overall structure and high-resolution fast time-lapse imaging ( every 1 min ) for 1 hr periods . We focused on critical periods of three major developmental events: the first stage of layer separation until P + 40% , the separation or R7/R8 terminals in what will become the M3 and M6 medulla layers in the adult , and the onset of synaptogenesis . We found distinct signatures of filopodial dynamics for each of these three processes ( Figure 4a ) . Specifically , at P + 28% ( during the first stage of layer separation ) , many transient filopodia ( <8 min lifetime ) as well as stable filopodia ( lifetime >60 min ) are apparent ( Figure 4b , c; Video 3 ) . In contrast , at P + 50% , a reduced number of transient filopodia ( with the same kinetic characteristics ) are present , but no stable filopodia ( Videos 3 , 4 ) . At P + 60% ( onset of synaptogenesis ) , transient filopodia are dramatically reduced and a new type of stable filopodia has emerged that are less active than those during the first stage of layer separation ( Video 4 ) . The culture and imaging conditions had little or no deleterious effects , as we observed very similar dynamics for different growth cones that had been in culture for different times at the same developmental timepoints ( Figure 4a; Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 10721 . 013Figure 4 . Distinct classes of transient and stable filopodia underlie different developmental events . Fast ( 1 min interval ) time-lapse imaging was performed at multiple points of three ex vivo experiments . ( a ) Three time points are shown; during the first-stage ( P + 28% ) and second-stage ( P + 50% ) layer formation , and synaptogenesis ( P + 60% ) . 3D graphs ( upper panel ) show the dynamics of individual filopodia observed in a one hour period . In the heat maps on blue background , individual filopodia are shown as verticals lines . The filopodia were sorted by their initial orientation angle ( x-axis ) . The length of the vertical lines represents the life time of the filopodia ( time on the y-axis ) . The color map indicates the length ( μm ) for each filopodium through time . Representative images of the growth cones at the above time-points ( lower panel ) . See Figure 4—figure supplement 1 for heat maps and representative images at all time points . ( b ) Numbers of filopodia per growth cone for the time-points shown in a; for filopodia with lifetime <8 min ( transient ) and lifetime >1 hr ( stable ) . ( c ) Numbers of filopodia relative to the numbers at P + 28% for all time-points imaged . Fitted curves: y = 28 . 17 + 4 . 597x-0 . 075x2 ( transient ) and y = 583 . 1–24 . 53x + 0 . 26x2 ( stable ) . ( d ) Mean length ( μm ) ( e ) Average speed ( μm/min ) and ( f ) , Inactivity ( ratio of intervals with no significant extension or retraction ) for transient and stable filopodia at all time-points . Stable filopodia observed after P + 50% have significantly higher inactivity than those observed before ( Means: 0 . 3002 v . 0 . 4346 , p = 0 . 0002 , n = 14 for each ) . See Figure 4—figure supplement 2 for these parameters as a function of filopodia lifetime on the same growth cone . Error bars depict SD . Scale bars , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01310 . 7554/eLife . 10721 . 014Figure 4—figure supplement 1 . Fast filopodial dynamics throughout pupal development . Dynamics data from all 6 growth cones ( 2 independent growth cones for each time point ) that were used in Figure 4 . The heat maps on blue background show individual filopodia as verticals lines . The filopodia were sorted by their initial orientation angle ( x-axis ) . The length of the vertical lines represents the life time of the filopodia ( time on the y-axis ) . The color map indicates the length ( μm ) for each filopodium through time . a , starting at P + 28% , after 9 hr in culture and after 19 hr in culture . b , starting at P + 40% , after 8 hr in culture and after 21 hr in culture . ( c ) , starting at P + 52% and 8 hr in culture . Scale bars , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01410 . 7554/eLife . 10721 . 015Figure 4—figure supplement 2 . Filopodial dynamics as a function of lifetime . ( a-i ) For the same growth cones depicted in Figure 4 , every filopodia observed in a 1 hr period were binned into different lifetime classes: <1 min , 2–3 min , 4–7 min , 8–15 min , 16–31 min , 32–59 min or >60 min . Mean length ( a-c ) , speed ( d-f ) and inactivity ( g-i ) were plotted for each group of the three growth cones . The boxes cover the entire range and horizontal lines show the mean . Scale bars , 2 μm . ( j-k ) , Mean inactivity ( ratio of intervals with no significant extension or retraction ) for transient ( <8 min ) and stable ( >60 min ) filopodia at all time-points . ( j ) , Inactivity: due to the high ratio of filopodia with ‘zero’ inactivity in transient filopodia ( g-i ) , average inactivity appear much lower for transient filopodia . ( k ) Inactivity of filopodia with at least one inactive time point: after the exclusion of ‘zero inactivity’ filopodia , inactivity of transient and early stage stable filopodia are identical . Error bars depict SD . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01510 . 7554/eLife . 10721 . 016Video 3 . Ex vivo imaging of fast filopodial dynamics-1 . All photoreceptors are labeled with myr-mRFP and R7 photoreceptors are sparsely labeled with CD4-tdGFP using GMR-FLP . Live imaging started P + 28% and continued for 20 hr . We used an alternating slow ( 30min intervals ) imaging of the general structure and fast ( 1min interval ) imaging of two growth cones at higher resolution at three different time points . Fast filopodial dynamics of the same two growth cones at P + 28% , P + 40% and P + 55% are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01610 . 7554/eLife . 10721 . 017Video 4 . Ex vivo imaging of fast filopodial dynamics-2 . All photoreceptors are labeled with myr-mRFP and R7 photoreceptors are sparsely labeled with CD4-tdGFP using GMR-FLP . Two live imaging sessions starting at P + 40% ( 22 hr ) and P + 52% ( 9 hr ) are shown . We used an alternating slow ( 30 min intervals ) imaging of the general structure and fast ( 1 min interval ) imaging of two growth cones at higher resolution at different time points . Fast filopodial dynamics of the same two growth cones at P + 40% , P + 50% and P + 60% and fast filopodial dynamics of another three growth cones at P + 52% and P + 62% are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 017 What is the role of transient and stable filopodia during development ? Transient filopodia constitute more than 90% of all filopodia at P + 28% and more than 70% of all filopodia at P + 60% ( Figure 4b ) . Remarkably , these filopodia exhibit indistinguishable dynamics throughout pupation . We measured no significant changes in their mean lengths ( Figure 4d ) , average speed of extension and retraction ( Figure 4e ) , levels of inactivity ( Figure 4f ) and the variance of these measurements ( blue traces in Figure 4d–f ) . Hence , these dynamics do not correlate with any particular developmental time period or event . Instead , the transient filopodial extensions suggest a continuous function for the even spatial distribution of the columnar and layer structure throughout early developmental stages . The number of transient filopodia reduces increasingly with time ( Figure 4a ) while more cellular processes solidify the adult anatomy . In contrast to transient filopodia , stable filopodia exhibit a bimodal distribution . A first type of stable filopodia exists up to P + 35% and then rapidly vanishes before P + 50% . These filopodia are significantly longer than transient filopodia , with some exploring up to two columns ( Figure 4b , d ) . Surprisingly , their average speed and inactivity ( i . e . intervals with no significant extension or retraction ) are not significantly different from transient filopodia at this stage ( Figure 4e , f; Figure 4—figure supplement 2 ) . This indicates that at early stages some filopodia appear stable only because they are longer . Unlike the earlier stable filopodia , those that emerge after P + 55% exhibit a significantly greater inactivity compared to both the transient filopodia at this stage as well as the filopodia in earlier stages ( Figure 4f; Figure 4—figure supplement 2 ) . Combined with their peculiar bulbous tips ( Figure 3c ) , they define a distinct class of filopodia in both structure and dynamics . The two types of stable filopodia correlate with different developmental subprograms: The first type accompanies columnar stabilization and restriction of the growth cone , while more layers are being formed . In contrast , the second type of stable filopodia emerges after layers are defined and interactions with presumptive synaptic partners commence . The time period around P + 50% where stable filopodia are absent matches precisely the moment when the final R7 and R8 layers are defined . In summary , measurements of fast filopodial dynamics reveal that the majority of R7 filopodia are transient and may function during continuous column and layer stabilization; in contrast , distinct classes of stable filopodia may be substrates for the specific types of neurite interactions underlying the developmental events they accompany . The temporary absence of stable filopodia around P + 50% marks a critical developmental period when the final R7 and R8 layers are determined ( Ting , 2005 ) . In this well-studied model for layer formation the R8 growth cone is known to actively extend ( Timofeev et al . , 2012 ) . In contrast , how R7 reaches its final target layer is less clear . R7 ends up in the deepest layer of the distal medulla through one of two processes: it may extend away from a temporary layer to a new , more proximal layer ( active model ) ( Hadjieconomou et al . , 2011; Feller and Sun , 2011; Mast , et al . , 2006 ) alternatively , more layers are intercalated by other neurons while R7 remains in the same layer throughout ( Ting , 2005 ) ( passive model ) ( Figure 5a ) . The recent finding that R7 is in close proximity with its Dm8 target dendrites as early as P + 17% ( Ting et al . , 2014 ) supports the passive model ( Ting , 2005 ) ; however , it remains unclear whether R7 growth cones actively participate in any part of the layer formation process . Live imaging of the entire process of layer formation can provide an unequivocal answer to the question whether R7 growth cones exhibit any extension activity by following the same growth cones over time . 10 . 7554/eLife . 10721 . 018Figure 5 . R7 growth cones do not actively extend in the medulla . ( a ) R7 may reach its final target layer through active extension or passive displacement and intercalation . ( b ) Live imaging starting at P + 30% . All photoreceptors were labeled with myr-mRFP and R7 cells were sparsely labeled with CD4-tdGFP using GMR-FLP through MARCM . R7 growth cone ( triangle ) initially has a cone structure . As the layer formation progresses , a new varicosity ( arrow ) is formed from the axon shaft . This structure expands further and by P + 50% the entire terminal thickens . See Figure 5—figure supplement 1 for all time points . ( N = 31 ) . ( c ) Live imaging starting at P + 42% . Both R7 and R8 cells were sparsely labeled with CD4-tdGFP using hsFLP . R7 axon has already formed its distal varicosity ( arrowhead ) ; the R8 axon has extended a single filopodia proximally ( arrow ) . Later , this filopodia reaches to the R8 final layer and forms the new terminal . R7 terminal shows no active extension activity . ( N = 17 for R7 and 15 for R8 ) . ( d ) Model of layer formation in the distal medulla . After their arrival to the medulla R7 and R8 terminals are initially separated by intercalation of lamina cell ( LM ) axons . After P + 40% , R8 growth cones actively extend to new layer while R7s remain in their arrival layer throughout . Scale bars , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 01810 . 7554/eLife . 10721 . 019Figure 5—figure supplement 1 . Single growth cone tracking demonstrates R7 terminals remain passive throughout layer formation without a stationary landmark . Live imaging starting at P + 30% . All photoreceptors were labeled with myr-mRFP and R7 cells were sparsely labeled with CD4-tdGFP using GMR-FLP through MARCM . R7 terminal ( red arrow ) can be followed throughout 17 . 5 hr based on its specific filopodial morphology and dynamics . A new varicosity ( yellow arrow ) was formed from the axon shaft and expands over the course of 15 hr , pushing the terminal distally . No directed activity was observed at the growth cone tip throughout the imaging period . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 019 We used a time-lapse interval of 30 min to track individual growth cones and their shape changes ( Video 5 ) . At P + 30% the R7 growth cone exhibits a cone-shape that expands towards the terminal ending from its thin axonal process ( triangle in Figure 5b = 0 ) . Over the next 18 hr we observed a gradual change of this shape , but no extension away from it ( Figure 5b ) . How does the R7 axon accommodate new layer formation in the expanding distal medulla without extension ? As shown in Figure 5b new varicosity emerges distally on the axon shaft of the cone-shaped growth cone ( t = 5 hr , arrow ) . This varicosity expands to give the entire terminal a bipartite structure ( t = 5–14 hr ) ; these observations suggest the intercalation of a new layer and support the passive model . Importantly , the continuous observation of the same growth cone and its dynamics ( Video 5 , Figure 5—figure supplement 1 ) unequivocally reveals the lack of active extension without the need for a stationary landmark that is necessary in fixed images to verify movement . 10 . 7554/eLife . 10721 . 020Video 5 . Second stage layer targeting of R7 and R8 . Two live imaging experiments are shown . ( 1 ) All photoreceptors are labeled with myr-mRFP and R7 photoreceptors are sparsely labeled with CD4-tdGFP using GMR-FLP . Imaging started at P + 30% and continued for 18 hr , with 30 min intervals . Two R7 growth cone tips ( red arrow ) were followed . At the 2 . 5 hr mark a varicosity starts to develop from the axon shaft and expands over the next 15 hr , contributing to the elongation of the R7 axon . Note that being able to follow the same growth cone tip based on its unique filopodial structure allows us to verify lack of active extension without a stationary landmark . ( 2 ) R7 and R8 photoreceptors were sparsely labeled with CD4-tdGP using hs-FLP . Imaging started at P42% and continued for 21 hr . We used an alternating slow ( 30 min intervals ) imaging of the general structure and fast ( 1 min interval ) imaging of two neighboring R7 and R8 growth cones at higher resolution at different time points . R8 axon relocates to its final layer by sending a single filopodia proximally , which is initially very dynamic but later stabilizes and expands in the new layer , forming the new R8 terminal . In contrast , R7 terminal elongates along the axon shaft , but no directed extension activity is observed on the growth cone . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 020 In contrast to R7 , the growth cones of R8 extend a single filopodium towards their final layer ( Figure 5c , arrow ) ; this filopodium is initially highly dynamic and exhibits almost complete retractions and re-extensions . It is finally stabilized in a deeper layer and gradually becomes thicker to form the new R8 terminal in the same layer as the distal end of the intercalated R7 varicosity ( Figure 5c , arrowhead ) . The formation of the bipartite R7 growth cone and its intercalating varicosity precedes the stabilization of the dynamically extending and retracting R8 process . This observation suggests that some other cell type first defines the layer where first the R7 growth cone forms its expanding varicosity and finally R8 targets . In summary , our live data demonstrate that R7 terminals do not actively extend after P + 30% . Instead , R7 growth cones arrive directly to their final layer and are only passively dislocated by the intercalation of other axons and dendrites ( Figure 5d ) . This process requires their continuous stabilization . The idea of passive retention implies that R7 growth cones do not engage in any active targeting process after P + 30% and is consistent with the continuous transient filopodial dynamics shown above ( Figure 4 ) . However , previous mutant analyses described R7 targeting defects into layers that form after P + 30% . For example , the homophilic adhesion molecule N-Cadherin ( CadN ) ( Hatta et al . , 1985 ) has emerged as a major regulator of synaptic layer specificity and its loss of function causes R7 mistargeting to the R8 layer ( Ting , 2005; Lee , et al . , 2001 ) . Previous studies focused on structure-function analyses of CadN ( Nern et al . , 2005 ) and its molecular interactions with other proteins ( Prakash et al . , 2009 ) and found that the penetrance of the mistargeting defect increased over time , suggesting retractions ( Ting , 2005 ) . However , how loss of CadN causes mistargeting or retractions is still unclear . In particular , it is unknown what changes in the growth cone dynamics cause this phenotype . To investigate these aspects we performed live imaging of CadN mutant R7 axons ( positively labeled using MARCM ) in an otherwise wild-type brain ( Video 6 ) . We observed that almost all R7 terminals arborized correctly in a layer right below R8 terminals upon arrival at the medulla prior to P + 20% . At P + 23% , some of the ‘oldest’ mutant terminals that first arrived at the medulla were mislocalized ( 17% of all R7 terminals , n = 54 ) . As predicted ( Ting , 2005 ) , this increase is due to the retraction of R7 terminals which were initially in the correct position ( Figure 6a , d ) . These retractions were always preceded by a gradual collapse of their filopodial structure that could predict the remobilization of the growth cone at least 2 and up to 10 hr prior to retraction . 10 . 7554/eLife . 10721 . 021Figure 6 . N-Cadherin is required for the stabilization but not the layer specific targeting of R7 growth cones . All photoreceptors were labeled with myr-mRFP . CadN405 R7 cells were generated with MARCM , using GMR-FLP and positively labeled with CD4-tdGFP . ( a ) Live imaging started at P + 24% shows a mutant R7 growth cone ( arrow ) that retracts from its target layer over the course of 5 hr . ( b ) , Live imaging started at P + 53% shows a mutant R7 growth cone ( arrow ) that retracts from its target layer over the course of 10 hr . Some mutant axons retract completely from the medulla ( Figure 6—figure supplement 1 ) ( c ) Live imaging started at P + 48% shows an R7 axon ( arrow ) that has been retracted to the edge of distal medulla but re-extends and attempts to re-innervate both wrong ( 5 . 5 hr ) and the right ( 7 hr ) layers . ( d ) Schematics of observed retraction and re-extensions events . Left and middle: Full Retraction leads to complete loss of the R7 axons from the medulla ( left ) , while partial retraction ( middle ) leads to R7 terminals in an incorrect layer . Number of mislocalized terminals: 33% ( n = 85 ) at P + 40% and 56% ( n = 62 ) at P + 52% . Right: Previously retracted R7 axons can re-extend , even days after they would have been stabilized in wild type . 52% ( n = 23 ) of retracted axons at P + 40% re-extended before P + 50%Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 02110 . 7554/eLife . 10721 . 022Figure 6—figure supplement 1 . CadN mutant R7 axons may retract completely from the medulla . All photoreceptors were labeled with myr-mRFP . CadN 405 R7 cells were generated with MARCM , using GMR-FLP and positively labeled with CD4-tdGFP . Live imaging starting at P + 35% demonstrates an R7 axon which retracts from its target layer in the first 2 hr . During the remaining 7 hr , the axon retracts below the R8 temporary layer ( upper red layer ) and leaves the distal medulla completely . Scale bar , 4 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 02210 . 7554/eLife . 10721 . 023Video 6 . N-Cadherin functions in growth cone stabilization . All photoreceptors are labeled with myr-mRFP and approximately 10% of R7 photoreceptors were made mutant for CadN and labeled with CD4-tdGFP using GMR-FLP through MARCM . Three live imaging sessions are shown . ( 1 ) Starting at P + 24% ( 17 hr ) . R7 axons arrive correctly to their target layer but they gradually retract from it , preceded by growth cone collapse . ( 2 ) Starting at P + 53% ( 20 hr ) . Retractions continue despite the wild-type photoreceptors reached their final layer configurations . ( 3 ) Starting at P + 42% ( 11 hr ) . Some of the R7 axons that retracted at the earlier stages re-extend back into the distal medulla . Note that the growth cones are streamlined during active movement but show expansion while the axons attempt to re-innervate various medulla layers . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 023 The fraction of mislocalized terminals increased to 33% ( n = 85 ) by P + 40% and 56% ( n = 62 ) by P + 52% . In addition , these numbers are underestimates since some of the mutant terminals retract completely from the medulla ( Figure 6—figure supplement 1 ) . Retractions continued even after the wild-type neurons formed their final layers ( Figure 6b , d ) , resulting in the previously observed penetrance of 70% in adult brains ( Lee et al . , 2001 ) . These late retractions could be the consequence of dying back axons , or , alternatively , CadN deficiency is sufficient for R7 axons to regain active mobility days after their targeting is concluded . Live observations of growth cone dynamics provided a clear distinction of these two possibilities: we observed that 52% ( n = 23 ) of retracted axons at P + 40% APF actually re-extended towards more proximal layers within the next 8 hr . These axons often re-arborize in both correct and incorrect layers ( Figure 6c , d ) , but again fail to stabilize those arborizations ( Video 6 ) . This phenotype was previously impossible to recognize in fixed preparations and masked by the overall increase in mistargeting penetrance . These data show that CadN mutant axons regain motility for days after their targeting should have been concluded . We conclude that CadN is not required for targeting per se , but for the stabilization of R7 growth cones after initial targeting . What is the role of filopodia in growth cone stabilization ? Our R7 filopodial dynamics measurements revealed that >90% of all filopodia were transient and exhibit continuous , stochastic extension/retraction dynamics that did not correlate with any specific developmental processes ( Figure 4 ) . These dynamics are consistent with continuous stabilization of the passively retained R7 growth cones throughout development ( Figure 5 ) . If filopodia control growth cone stabilization , then CadN growth cones should exhibit reduced filopodial dynamics . CadN R7 growth cones do not appear obviously disrupted as long as they remain in their initial , correct arrival layer ( Figure 6 ) and filopodia numbers are not significantly affected at P + 28% ( Figure 7a ) . However , both transient and stable filopodia of mutant growth cones exhibit reduced average speed of extension/retraction ( Figure 7b , Video 7 ) . As a consequence , both types of filopodia are on average also significantly shorter than wild-type ( Figure 7c , d ) . These findings point to a general slow-down of the filopodial dynamics in CadN growth cones and suggest that N-Cadherin mediated adhesion ( Hatta et al . , 1985 ) is important for the stabilization of R7 growth cones through filopodial interactions at the target layer . 10 . 7554/eLife . 10721 . 024Figure 7 . N-Cadherin is required for fast filopodial dynamics . CadN405 R7 cells were generated with MARCM , using GMR-FLP and positively labeled with CD4-tdGFP . Fast ( 1 min interval ) time-lapse imaging was performed at P + 28% . ( a ) , The average numbers of filopodia per growth cone are not significantly different between wt and CadN405 . ( b ) Mutant filopodia are slower , ( for transient , means wt: 1 . 303 ( n = 143 ) , CadN405: 0 . 791 ( n = 169 ) , p<0 . 0001; for stable , means wt: 0 . 898 ( n = 10 ) , CadN405: 0 . 636 ( n = 5 ) p = 0 . 0199 ) and ( c ) shorter ( for transient , means wt: 1 . 542 ( n = 143 ) , CadN405: 0 . 939 ( n = 169 ) , p <0 . 0001; for stable: means wt: 3 . 707 ( n = 10 ) , CadN405: 2 . 275 ( n = 5 ) , p = 0 . 1257 ) . ( d ) CadN405 R7 growth cones at the correct layer . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 02410 . 7554/eLife . 10721 . 025Video 7 . Loss of N-Cadherin leads to reduced filopodial dynamics . Representative wild type and cadN mutant R7 growth cones are shown at P + 28% . Extraction of individual filopodia reveals reduced dynamics over the same time period ( 1 hr with 1 min time lapse ) as shown in the quantifications in Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 10721 . 025 In summary , we find that filopodial dynamics predict growth cone stabilization in a specific layer . This attachment of the growth cone in a specific layer is a continued requirement long after initial targeting is completed and it is further reflected in the majority of transient filopodial extension/retraction dynamics . Loss of CadN reduces these dynamics and increases the likelihood of layer destabilization even days after targeting is concluded .
Our ex vivo brain development system in a closed imaging chamber allows continuous laser scanning during development for at least 20 hr per session . For longer imaging periods , the system can be modified to a semi-open state with perfusion ( Williamson and Hiesinger , 2010 ) . However , the ease of the closed chamber outweighed the advantage of longer imaging periods in our hands . Our key goal was to follow subcellular dynamics at the resolution limit of conventional light microscopy with fast enough time-lapse to quantitatively describe subcellular dynamic properties in developing brains over many hr . Important advances in Drosophila ex vivo brain imaging have recently established high-resolution imaging in short developmental time windows ( Medioni et al . , 2015; Zschätzsch , et al . , 2014 ) and over long periods at low resolution and with slow time lapse ( Rabinovich et al . , 2015 ) . We identify phototoxicity and drift as key problems to obtain high spatial and temporal resolution 3D dynamics data over long developmental time periods , for which the imaging system presented here provides a successful approach . We have tested our system for developmental processes ranging from L3 brain development ( cell migration , data not shown ) and throughout pupal development ( growth cone dynamics ) . We further provide the calibration of developmental progress in this culture system under imaging conditions . For example , morphological changes of the eye and ‘lamina rotation’ occur only incompletely outside of the fly’s head ( Figure 2c and Figure 2—figure supplement 1 ) . In contrast , early layer formation of photoreceptor axonal projections are accelerated with normal outcome; development after P + 40% occurs with identical speed in our ex vivo system and in vivo . These findings indicate that layer and synapse formation are not directly dependent on distal tissue morphogenesis . However , different developmental processes must be calibrated for their ex vivo progress compared to in vivo development in the fly . Based on our quantitative analyses of layer formation in the distal medulla , we anticipate that the developmental progress of more proximal brain regions will be similar to the calibrated optic lobe development . We show that conventional 2-photon microscopy can safely be used over long periods with virtually no drift and at high resolution in our imaging chamber when following a simple ‘no bleaching’ rule . In some cases we even observed mild photobleaching ( e . g . Figure 2d ) without adverse effects on developmental progress . We conclude that as long as there is no significant decrease in the signal intensity over time , 2-photon imaging per se does not negatively affect the development . In addition , ex vivo imaging has the advantage that the culture media allow pharmacological manipulations which are not easily possible in vivo or with intravital imaging . In summary , the ex vivo imaging system and conditions developed here allow to observe live the formation of neural circuits anywhere in the Drosophila brain . Importantly , imaging at different spatial and temporal scales allows relating fast , high-resolution filopodial dynamics to much slower , long-term developmental processes . Growth cone behavior is highly dynamic and context-dependent ( Mason and Erskine , 2000 ) . Understanding the role of growth cone dynamics as part of a longer developmental process requires observation in their normal environment . Growth cone filopodia have traditionally been interpreted as probes that detect guidance cue gradients ( Gallo and Letourneau , 2004; Zheng , et al . , 1996 ) or as ‘sticky fingers’ that provide the traction required for growth cone migration ( Heidemann , 1990; Chan and Odde , 2008 ) . Our characterization of the R7 growth cones revealed a different role for the vast majority of its filopodia during layer formation in the distal medulla: Surprisingly , more than 90% of R7 filopodia exhibit apparently stochastic extension/retraction dynamics that do not correlate with any major structural change during layer formation in the distal medulla . Instead , these movements are fast , transient and only slowly reduce over the period of days during brain development , while new neurons innervate and new layers form . What is the role of these filopodia ? Our imaging data revealed that R7 growth cones do not actively extend after their initial target recognition , in contrast to some of the earlier models ( Hadjieconomou , et al . , 2011; Clandinin and Feldheim , 2009; Ting and Lee , 2007 ) . Instead , other axons and dendrites intercalate while R7 growth cones define the most proximal boundary of the distal medulla . Hence , R7 must stably maintain their position while active intercalation of other neurons , e . g . R8 extension , pushes the R7 layer proximally ( Figure 5d ) . This stabilization is consistent with continued filopodial extension/retraction dynamics that are decreasingly required as the final adult column and layer organization solidifies . However , we note that our imaging data do not establish a causal relationship between the observed dynamic behaviors . The stabilizing function is reminiscent of zebrafish retinotectal axons which display a broadened structure while resting but are more streamlined during extension ( Kaethner and Stuermer , 1992 ) . Stabilization through filopodial dynamics is further supported by the observation of ‘jumping’ growth cones in CadN mutants , as discussed below . Finally , we also observed a previously undescribed kind of filopodia that emerge at later stages . These are more stable , appear to coincide with the timing of synaptogenesis and are reminiscent of densities observed in hippocampal cultures using VAMP-GFP ( Smith et al . , 2000 ) . However , adult R7 synapses are restricted to the main axonal trunk of the R7 terminal and the precise roles of these late , stable filopodia away from the main axonal trunk remain to be investigated . Our observation of growth cones from initial arrival through layer formation and finally synaptogenesis reveals a remarkable transitioning of the R7 terminal shape from a more classical growth cone to an elongated structure with branched axon shaft ( Figure 3—figure supplement 1a ) . Filopodia on this extended R7 axon are reminiscent of axonal filopodia observed in spinal cord culture ( Gallo and Baas , 2011; Spillane , et al . , 2012 ) , but restricted to the axon shaft inside the medulla neuropil where layer formation and synaptogenesis occur ( Figure 3—figure supplement 1b , c ) . Before P + 40% , we only observe filopodia at the axon tips as in classical growth cones . Therefore , we suggest these structures are still growth cones although they appear to use filopodia as a means to stabilize rather than as a substrate for migration . We think that the subsequent transition morphologies could reasonably be interpreted as an extended growth cone or a distinct and short part of the proximal axon that got recruited to new active functions during layer formation ( Figure 3—figure supplement 1a ) . Our findings support a model in which continuous stabilization of R7 growth cones in a column/layer grid depends on the levels of N-Cadherin ( CadN ) . The observation of mistargeted photoreceptor axons ( Lee et al . , 2001 ) as well as its classical role in axon guidance ( Matsunaga et al . , 1988 ) have previously led to the interpretation of CadN as a guidance cue . The interpretation as a part of a specificity code is complicated by the observation that CadN is expressed in all presynaptic and postsynaptic neurons during distal medulla development; however , temporal regulation ( Petrovic and Hummel , 2008 ) as well fine-tuning of expression levels ( Schwabe et al . , 2014 ) have been proposed as solutions . Our live imaging data reveal that CadN-deficient R7 axons have no initial targeting defects and CadN does not function as a target layer-specific cue . Instead , growth cones fail to stabilize and engage in an aberrant process of ‘jumping’ between incorrect and correct layers . Remarkably , CadN-deficient R7 growth cones retain the ability to jump between distal medulla layers for days after their normal targeting should have been concluded and stabilized , presumably through the filopodial dynamics described here . CadN has been shown to localize to the filopodia of R1-6 photoreceptor axons in the lamina neurpil ( Schwabe et al . , 2013 ) ; we speculate that R7 growth cones could use the surface area of their filopodia to form stabilizing adhesions through CadN . Consistent with this interpretation , loss of CadN reduces filopodial extension/retraction dynamics and ‘jumps’ between medulla layers are preceded by a slow , several hours-long filopodial retraction process . CadN mediated adhesive interactions were shown to be essential for growth cone migration in primary neuronal cultures ( Bard et al . , 2008 ) . We observed decreased filopodial lengths in CadN growth cones , consistent with the longer neurite lengths observed with increased CadN mediated adhesions . While such interactions were previously correlated with growth cone velocity , our observations provide a link to their stability in a different context . CadN-mediated adhesion with other medulla cells is required for R7 terminals to end up in the correct layer independent of its signaling function ( Yonekura et al . , 2007 ) . This finding is consistent with a growth cone stabilization function via interactions with many different medulla cells , independent of the correct synaptic partners . This idea is further consistent with the widespread expression of CadN in many cell types . In sum , our data together with previous observations support a ‘non-guidance cue’ function for CadN in stabilizing the positions and contacts of neurites once the targeting is complete . Indeed , several cell adhesion molecules previously thought to function as guidance cues have recently been shown to exert ‘non-cue’ functions cell-autonomously ( Petrovic and Schmucker , 2015 ) and through implementing simple developmental rules ( Hassan and Hiesinger , 2015 ) . Initial R7 targeting to the correct layer could be achieved by other molecules or by a developmental rule such as 'stop at the first target layer encountered past the pioneer R8 axon , and then stabilize' . It is possible that such a rule could result in the correct initial targeting of R7 and L-cell axons simply by their arrival order ( Figure 5d ) , requiring no layer-specific molecular code ( Hassan and Hiesinger , 2015 ) .
Pan-photoreceptor labeling was done by GMR-Gal4 expressing the membrane tethered CD4-tdGFP ( Han et al . , 2011 ) . Sparse R7 labeling as well as the generation of CadN mutant R7 neurons were achieved through MARCM ( Mosaic analysis with a repressible cell marker ) ( Lee and Luo , 1999 ) using GMRFLP . GMR-myr-RFP or GMR-myr-tdTomato was used to label all photoreceptors in the background . Fly stocks: i ) ;; GMR-Gal4 , UAS-CD4-tdGFP ii ) GMR-FLP; GMR-Gal4; FRT80B , UAS-CD4-tdGFP iii ) GMR-myr-RFP;; FRT80B , tub-Gal80 iv ) GMRFLP; FRT40A , tub-Gal80; GMR-Gal4 , UAS-CD4-tdGFP , GMR-myr-RFP v ) ; FRT40A , CadN405; vi ) hs-FLP; GMR-FRT-w + -FRT-Gal4; UAS-CD4-tdGFP vii ) GMR-myr-tdTomato; FRT80B , tub-Gal80 . Eye-brain complexes were dissected in PBS , fixed in 3 . 7% paraformaldehyde ( PFA ) in PBS for 40 min , washed in PBST ( 0 . 4% Triton-X ) and mounted in Vectashield ( Vector Laboratories , CA ) . Images were collected using a Leica TCS SP5 confocal microscope with a 63X glycerol objective ( NA = 1 . 3 ) . The culture chambers were built inside 60x15 mm petri dish lids with a layer of Sylgard 184 ( Dow Corning ) at the center ( 2 cm in diameter ) . 200 μm thick X-ray films cut in 1x1 mm pieces were used as spacers to prevent the coverslip from crushing the tissue . 2% low melting point agarose was prepared in water and dialyzed in pure water for 48 hr with changing the water every 12 hr at room temperature , then stored at 4oC . The culture media was modified from a previous recipe ( Ayaz et al . , 2008 ) . It was prepared with 1:10 fetal bovine serum ( FBS ) , 10 μg/ml human insulin recombinant zinc ( Stock: 4 mg/ml ) , 1:100 Penicillin/streptomycin ( Stock: 10000 IU/ml penicillin , 10 mg/ml streptomycin ) , 1 μg/ml 20-Hydroxyecdysone ( Stock: 1 mg/ml in ethanol ) in Schneider’s Drosophila Medium . All were acquired from Life Technologies . Brains were dissected in chilled Schneider’s Medium and mounted in 0 . 4% dialyzed low-melting agarose diluted in the culture media . Step-by-step chamber assembly ( Figure 1—figure supplement 1 ) : The final imaging chamber ( Figure 1—figure supplement 1b , c ) provides sufficient oxygen and nutrients through diffusion for at least 24 hr . 20-Hydroxyecdysone is excluded from the cultures that start after 50% APF . This is due to previously measured physiological titers ( Paul Bainbridge and Bownes , 1988 ) as well as our experimental data ( Figure 2—figure supplement 2a–h ) . Live imaging was performed at room temperature using a Zeiss LSM 780 multiphoton microscope with a 40X LD water objective ( NA = 1 . 1 ) or a Leica SP8 MP microcope with a 40X IRAPO water objective ( NA = 1 . 1 ) with a Chameleon Ti:Sapphire laser ( Coherent ) . For single-channel CD4-tdGFP imaging , excitation was done at 900 nm . For double-channel CD4-tdGFP and myr-RFP imaging , excitation was done at 800 nm . Our chamber can be imaged in conjunction with both water and glycerol objectives with both upright and inverted microscopes . We compared the images of R7 growth cones of a P + 27% brain acquired by above setup with those acquired by a conventional Leica TCS SP5 confocal microscope with a resonant scanner , using a 63X Glycerol objective ( NA = 1 . 3 ) . A resonant scanner provides superior scan speeds compared to standard two-photon systems . However , we observed a decrease in signal intensity and quality with the confocal microscope at tissue deeper than 60 μm from the coverslip ( Figure 2—figure supplement 2i ) . Even at moderately deep tissue , the laser power required on the confocal system to acquire images with comparable quality to the two-photon system is too high to take advantage of the higher scan speeds for extended periods ( data not shown ) . Nevertheless , the resonant scanner would still be the preferred option for imaging at low depths when speed is the most important factor . Imaging data were analyzed and processed with Imaris ( Bitplane ) . Deconvolved data were used in Figures 4 , 5 and 6 and supplementary Videos . 3D deconvolution was performed with Autoquant X3 using adaptive PSF ( blind ) ( Hiesinger et al . , 2001 ) . For all datasets , 10 iterations were performed at medium noise level ( noise value: 20 ) with recommended settings . Distance from the coverslip was set to 40 μm . Filopodial analysis was done with the Filament module of Imaris . Each filopodium was manually segmented and tracked across time points . 'Automatic placement' option was used while drawing to ensure that we measured the actual 3D length of each filopodium . We exported the 'length over time data for all of the filopodia of a growth cone to an Excel sheet and performed further analysis with MATLAB . We used a custom MATLAB code to calculate the number of extension and retraction events , mean extraction and retraction speeds , mean lengths and lifetimes for each TrackID . Heat maps of lengths versus time for all filopodia in a growth cone were also generated . In those , filopodia were sorted by the angle of their orientation at the time of their initial formation . We did not find any overall , significant difference between the average speeds of extension and retraction on any growth cone; so they were combined to calculate a single average speed for all further analysis . We considered any changes in length less than 0 . 3 μm between consecutive time points as zero movement or 'static' periods because manual segmentation cannot be precise enough to reliably account for such a small retraction or extension . Average speeds were therefore calculated only from the points that had a change in length greater than 0 . 3 μm . We used the ratio of static time points to the lifetime of a filopodium to calculate 'inactivity' Source code 1 . Further analysis , i . e . classification into transient and stable filopodia , statistical analysis , and the generation of graphs were done with GraphPad Prism . Where needed statistical differences were calculated with unpaired , parametric t-tests . Filopodia number percentages over time in Figure 4c were fitted with second order polynomials to generate curves . For inactivity measurements , we generated two different graphs . Due to their short lifetimes , many transient filopodia have zero inactivity by definition; resulting in drastically lower average inactivity for transient filopodia compared to other filopodia ( Figure 4—figure supplement 2g–i ) . This may unfairly imply an intrinsic difference of dynamics between transient and stable filopodia ( Figure 4—figure supplement 2j ) . Indeed , when the filopodia with ‘zero inactivity’ are excluded , their average inactivity is statistically identical with the early-stage stable filopodia ( Figure 4—figure supplement 2k ) . We therefore used these graphs in Figures 4 and 6 . We provide exemplary datasets in Zenodo ( https://zenodo . org/record/33141 ) . One file is a Zeiss ( . lsm ) file ( raw data ) and the other is the deconvolved version of this dataset in the Imaris ( . ims ) format , including the segmented filopodia as filament objects . This dataset belongs to the first fast imaging session of P+28% growth cones shown in Video 3 . We would be happy to provide any other datasets upon request . | Genes encode complicated developmental processes , but it is clear that genetic information cannot encode each and every individual connection that forms between the nerve cells in a brain . Instead , the individual cells and nerve endings must make decisions during brain development . Up until now , few examples were known for how these nerve endings move and choose their paths and partners in a living , developing brain . The fruit fly Drosophila provides a useful model to explore the ‘wiring’ of nerve cells in the brain , partly because a fruit fly’s brain develops within a few days . However , most previous studies have relied on identifying mutant flies with disrupted brain wiring and studying them using still images . Now , Özel et al . have developed a new imaging method that has enough resolution and speed over sufficiently long periods to track the growing nerve endings in a developing fly brain . The method was applied to a model nerve cell in the fly’s visual system . This revealed that most of this nerve’s dynamic changes are short-lived and random , and appear to help to stabilize the developing nerve ending , rather than guide it to a target . Özel et al . also found that a protein called N-Cadherin , previously thought to be required for the targeting of developing nerve endings , actually plays a role in their stabilization . These findings uncover the roles of changes in nerve endings during long-term brain development; this was previously largely unknown for any organism . The next stage in this research will involve further analyses of both wild type and mutant flies to try and work out general principles about how the brain develops via the decoding of genetic information . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Filopodial dynamics and growth cone stabilization in Drosophila visual circuit development |
Fibrolamellar carcinoma ( FLC ) is a rare liver cancer . FLCs uniquely produce DNAJ-PKAc , a chimeric enzyme consisting of a chaperonin-binding domain fused to the Cα subunit of protein kinase A . Biochemical analyses of clinical samples reveal that a unique property of this fusion enzyme is the ability to recruit heat shock protein 70 ( Hsp70 ) . This cellular chaperonin is frequently up-regulated in cancers . Gene-editing of mouse hepatocytes generated disease-relevant AML12DNAJ-PKAc cell lines . Further analyses indicate that the proto-oncogene A-kinase anchoring protein-Lbc is up-regulated in FLC and functions to cluster DNAJ-PKAc/Hsp70 sub-complexes with a RAF-MEK-ERK kinase module . Drug screening reveals Hsp70 and MEK inhibitor combinations that selectively block proliferation of AML12DNAJ-PKAc cells . Phosphoproteomic profiling demonstrates that DNAJ-PKAc biases the signaling landscape toward ERK activation and engages downstream kinase cascades . Thus , the oncogenic action of DNAJ-PKAc involves an acquired scaffolding function that permits recruitment of Hsp70 and mobilization of local ERK signaling .
Fibrolamellar carcinoma ( FLC ) is a variant of liver cancer that has distinctive histologic features ( Craig et al . , 1980 ) . This rare cancer afflicts healthy adolescents and young adults between the ages of 15–25 with no history of liver disease . This latter feature can compromise early diagnosis of FLC as patients frequently present with vague symptoms that include abdominal pain , loss of appetite , or a palpable mass . The diagnosis is often made after disease has spread outside the liver , leading to an overall survival of 35% ( Ang et al . , 2013 ) . Unfortunately , FLC frequently recurs , as it is intractable to standard chemotherapies and radiation . Surgical resection is currently the only opportunity for a cure . The search for new therapies for these patients is hindered by the limited availability of clinical samples and a lack of disease relevant cell lines or animal models that faithfully recapitulate the pathogenesis of FLC ( Dinh et al . , 2017; Engelholm et al . , 2017; Kastenhuber et al . , 2017; Oikawa et al . , 2015 ) . Recent transformative advances in our understanding of the molecular basis of FLC offer renewed hope for the development of drug therapies to treat this disease ( Honeyman et al . , 2014 ) . Sequencing tumor genomes of FLCs identified the underlying genetic defect as a heterozygous in-frame deletion of ~400 kb in chromosome 19 ( Honeyman et al . , 2014; Xu et al . , 2015 ) . This genetic lesion leads to translation of a de novo chimeric gene product where the chaperonin-binding domain of heat shock protein 40 ( DNAJ ) is fused to the Cα subunit of PKA ( Cheung et al . , 2015; Honeyman et al . , 2014 ) ( Figure 1A ) . We have recently shown that DNAJ-PKAc is solely expressed in FLCs , is cAMP-responsive , and importantly is incorporated into A-Kinase Anchoring Protein ( AKAP ) signaling complexes ( Riggle et al . , 2016a ) . This latter property provides a mechanism by which this pathological kinase is sequestered within defined subcellular locations and in immediate proximity to a subset of target substrates ( Langeberg and Scott , 2015; Scott and Pawson , 2009; Smith et al . , 2017 ) . While protein kinase A is generally not considered an oncogene , PKAc has been detected in the serum of patients with colon , renal , lungs , or adrenal carcinomas ( Cho et al . , 2000; Cvijic et al . , 2000; Porter et al . , 2001 ) . Whole exome sequencing from independent patient cohorts have identified pathological mutations in PKAc that are linked to Cushing’s syndrome ( Sato et al . , 2014 ) . This disease occurs either as consequence of pituitary tumors that overproduce adrenocorticotropic hormone ( ACTH ) or as a consequence of aberrant signaling events that stimulate excess cortisol release from the adrenal glands ( Beuschlein et al . , 2014; Lacroix et al . , 2015 ) . In the latter instance , amino acid substitution of arginine 205 to lysine in PKAc prevents binding to the regulatory ( R ) subunits of PKA to promote mislocalization of uncontrolled PKA activity ( Cao et al . , 2014 ) . In this report , we define a mechanism of action of DNAJ-PKAc , the fusion kinase exclusively expressed in fibrolamellar carcinoma . We have discovered that this fusion kinase is recruited into AKAP signaling complexes where , by virtue of its DNAJ domain , selectively interacts with the chaperonin heat shock protein 70 ( Hsp70 ) . This cellular chaperonin facilitates protein folding thereby providing an explanation as to why levels of DNAJ-PKAc protein are elevated over wildtype PKA in FLCs . The association of Hsp70 with DNAJ-PKAc also creates a unique therapeutic target for combinations of Hsp70 and kinase inhibitor drugs .
Immunoblot screening of clinical samples with antibodies against PKAc revealed that human FLCs are heterozygous in that they express both wildtype PKA and the DNAJ-PKAc fusion ( Figure 1B , top panel ) . This unique PKA fusion is solely expressed in FLCs , remains responsive to the second messenger cAMP , and importantly is incorporated by A-Kinase Anchoring proteins ( AKAPs ) into signaling complexes ( Riggle et al . , 2016a; Riggle et al . , 2016b; Turnham and Scott , 2016 ) . Immunofluorescent analysis of normal liver and FLC sections illuminated the distinctive morphology of this subtype of hepatocellular carcinoma where liver tumor is infiltrated with fibroid bands interspersed between cancerous hepatocytes ( Craig et al . , 1980 ) . This ‘intratumoral heterogeneity’ is distinct from the undulating sinusoidal pattern of normal liver ( Figure 1C & D ) . Co-localization of PKA catalytic ( green ) and regulatory subunits ( RIIα , red ) was evident in both sections . Counterstaining with DAPI ( blue ) is included to denote nuclei ( Figure 1C & D ) . Additional biochemical characterization of these clinical samples substantiated the elevated expression of the type Iα regulatory subunit of PKA ( RIα ) in FLC tumors as compared normal adjacent tissue ( Figure 1—figure supplement 1A , top panel ) ( Riggle et al . , 2016a ) . Related experiments demonstrate that type II regulatory subunit ( RII ) levels do not fluctuate ( Figure 1—figure supplement 1A , bottom panel ) . The active site of DNAJ-PKAc is identical to that of the native kinase; both PKA forms are inhibited by PKI and are sensitive to the same spectrum of ATP analog inhibitors ( Cheung et al . , 2015; Riggle et al . , 2016a ) . Immunoblot analyses using a phospho-PKA substrates antibody detects a different pattern of PKA phosphorylation in tumors as compared to adjacent liver extracts ( Figure 1—figure supplement 1B ) . In addition , an RII overlay survey of AKAPs reveals a distinct pattern of anchoring proteins in FLC as compared to adjacent liver tissue ( Figure 1—figure supplement 1C ) . These findings infer that introduction of DNAJ-PKAc results in changes in the substrate preference of this kinase or its access to subcellular targets . Yet , it remained important to ascertain whether the substrate specificity of this pathological fusion enzyme is altered in FLC . Phosphoproteomic profiling of human FLC and adjacent normal liver samples by label-free LC-MS/MS analysis identified 7697 phosphopeptides ( Hogrebe et al . , 2018 ) ( Figure 1E; n = 6 technical replicates ) . Of these , 628 phosphopeptides were significantly enriched in FLCs as compared to adjacent normal liver ( Figure 1E; orange ) . Substrate profiling with the NetworKIN platform predicted consensus kinase phosphorylation motifs ( Horn et al . , 2014 ) . Of the phosphosites increased in FLC , 20% were putative PKC targets and 8% were ERK-MAPK sites ( Figure 1F ) . This analysis revealed a systemwide rewiring of several protein kinase networks leading to increases and decreases in phosphorylation of specific substrates ( Figure 1—figure supplement 2 ) . Interestingly , PKA phosphosites were only enriched by 6 . 5% . However , phosphorylation of several key signaling effectors , scaffolding and anchoring proteins were enhanced ( Figure 1F and Figure 1—figure supplement 1D ) . One plausible explanation for this surprisingly modest effect on PKA signaling is that oncogenesis driven by the fusion kinase may not only solely proceed through the kinase domain but also involves the chaperonin-binding site . Thus , DNAJ-PKAc may function to recruit additional elements that underlie the pathology of FLC ( Figure 1G ) . Further immunoprecipitation experiments from clinical samples revealed that DNAJ-PKAc interacts with heat shock protein 70 ( Hsp70; Figure 1H ) , a cellular chaperonin that facilitates protein folding and is frequently up-regulated in cancers ( Calderwood et al . , 2006; Mayer and Bukau , 2005 ) . Proximity ligation ( PLA ) is an in situ technique that amplifies detection of native protein-protein interactions that occur within in a range of 40–60 nm ( Whiting et al . , 2015 ) . This approach was used to identify interaction between endogenous Hsp70 and PKAc in liver sections from FLC patients ( Figure 1I , J & K ) . PLA puncta indicative of native DNAJ-PKAc/Hsp70 sub-complexes were readily detected in regions of tumor ( Figure 1J and Figure 1—figure supplement 3 ) . In contrast , the number of PLA puncta was reduced in adjacent sections of healthy liver ( Figure 1I ) . Quantification is presented in Figure 1K and additional PLA images of tissue sections are included in Figure 1—figure supplement 3 . Recruitment of Hsp70 to DNAJ-PKAc may explain why protein levels of this fusion are frequently elevated compared to native PKA in FLCs ( Figure 1B , top panel ) . FLC research to date has been hampered by the limited availability of patient samples , a paucity of disease-relevant cell-lines , and mouse models exhibiting a 24 month latency to develop hepatic tumors ( Engelholm et al . , 2017; Kastenhuber et al . , 2017; Oikawa et al . , 2015 ) . Additionally , the most rigorously characterized PDX model is missing several key phenotypic traits of FLCs ( Oikawa et al . , 2015 ) . Therefore , we employed CRISPR/Cas9 gene editing of chromosome eight in AML12 non-transformed murine hepatocytes to generate sustainable and homogenous cell lines . A 400 kb region was excised between intron 1 of the gene for Hsp40 ( Dnajb1 ) and intron 1 of the gene for PKAc ( Prkaca; Figure 2A ) . Initial characterization by PCR detected transcripts of intervening genes ( Gipc1 , Ddx39 and Lphn1 ) located at the 5’ end , middle and 3’ end of the non-engineered strand of chromosome 8 ( Figure 2A & B ) . Quantitative PCR measurement of mRNA transcripts for Dnajb1 and Prkaca in wildtype and four gene-edited AML12DNAJ-PKAc cell lines revealed differential expression of both transcripts in each clonal AML12DNAJ-PKAc cell line ( Figure 2C & D , orange ) . Likewise , the Dnajb1-Prkaca fusion transcript was present at different levels in each cell line ( Figure 2E ) . Characterization by nucleotide sequencing and immunoblot analyses confirmed that these AML12DNAJ-PKAc cell lines encode and express a single copy of DNAJ-PKAc ( Figure 2F & G ) . As observed in FLCs , introduction of the DNAJ-PKAc allele promote the up-regulation of RIα expression ( Figure 2—figure supplement 1A ) . Clone 14 was selected for further analyses as these cells express similar levels of DNAJ-PKAc and native PKA as compared to human FLC patients ( Figure 2G ) . Interestingly , these clonal AML12DNAJ-PKAc cells have similar levels of PKA activity and comparable migratory properties to the wildtype cell line ( Figure 2—figure supplement 1B–F ) . We next evaluated the formation of DNAJ-PKAc/Hsp70 complexes in our cell lines . Immunoblot analysis detected DNAJ-PKAc within Hsp70 immune complexes isolated from our AML12DNAJ-PKAc cell line , while PKAc was not present in Hsp70 immune complexes isolated from control AML12 cells ( Figure 2H , top panel ) . Proximity ligation was used to evaluate DNAJ-PKAc/Hsp70 sub-complex formation ( Figure 2I & J ) . In control cells , few puncta were evident when PLA was performed with antibodies against PKAc and Hsp70 ( Figure 2I ) . In contrast , quantitation of PLA puncta ( yellow ) from >200 AML12DNAJ-PKAc cells revealed increased amounts of the DNAJ-PKAc/Hsp70 sub-complexes in our gene-edited cell lines ( Figure 2J & K ) . Counterstaining with antibodies against actin ( green ) and DAPI ( blue ) defined whole-cell and nuclear boundaries , respectively . Additional PLA images from both cell types are included as Figure 2—figure supplement 2 . Thus , our AML12DNAJ-PKAc cell line affords a disease relevant model with sufficient material to explore the mechanism of action of DNAJ-PKAc/Hsp70 assemblies . Accelerated cell proliferation is a hallmark of carcinogenesis ( Hanahan and Weinberg , 2011 ) . Thus , three independent measurements assessed growth of AML12DNAJ-PKAc cells . First , cell proliferation was measured over 72 hr in culture using the MTS assay . Amalgamated data show that AML12DNAJ-PKAc cells proliferate more rapidly than wildtype AML12 cells ( Figure 3A; n = 3 ) . Second , immunostaining for BrdU incorporation showed that DNA synthesis is increased AML12DNAJ-PKAc cells as compared to wildtype AML12 cells ( 82 ± 2% vs 36 ± 5% , Figure 3B & C; n = 3 ) . Third , colony formation assays were performed to reinforce our data that AML12DNAJ-PKAc cells have increased proliferative capacity as compared to their wildtype counterparts ( Figure 3D; n = 3 ) . Quantitation of amalgamated data confirms that AML12DNAJ-PKAc cells proliferate more rapidly than their wildtype counterparts ( Figure 3E ) . These findings lead us to surmise that the oncogenic nature of DNAJ-PKAc may not be simply due to changes in intrinsic kinase activity , but rather from the recruitment of Hsp70 . A logical extension of this premise is to determine whether pharmacologically blocking Hsp70 influences proliferation of AML12DNAJ-PKAc cells . Ver-155008 is an ATP-analog inhibitor of Hsp70 ( IC50 = 0 . 5 μM ) that halts cell proliferation in several cancer models ( Eugênio et al . , 2017; Wen et al . , 2014 ) . However , sole application of this drug over a range of concentrations did not have a differential effect on the viability of AML12DNAJ-PKAc cells compared to wildtype AML12 cells as assessed by MTS assay at 72 hr ( Figure 3F; n = 3 ) . Consequently , we screened drug combinations that target additional elements within DNAJ-PKAc/Hsp70 signaling complexes . Cells were seeded and screened against a panel of 125 FDA-approved anti-cancer compounds in the presence or absence of Ver-155008 ( Pauli et al . , 2017 ) ( Figure 3G–K ) . Cell viability was assessed by CellTiter-Glo assay and plotted against a standard deviation ( Z-score ) derived from collated mean responses ( Figure 3G & H; Pauli et al . , 2017; Toyoshima et al . , 2012 ) . Drug combinations in the lower right quadrant ( Sensitivity ) are more effective at reducing proliferation than drug combinations plotted in the upper right quadrant ( Resistance ) . In wildtype AML12 cells , which lack the fusion enzyme , there was little change in the response to any of the FDA-approved drugs irrespective of whether the Hsp70 inhibitor was present ( Figure 3G ) . Similarly , AML12DNAJ-PKAc cells were refractory to most FDA-approved anti-cancer drugs in the absence of Ver-155008 , but , when screening was repeated in the presence of Ver-155008 ( over a range of concentrations up to 10 μM ) , certain drug combinations preferentially blunted AML12DNAJ-PKAc cell proliferation ( Figure 3H & I ) . Three Hsp70 inhibitor/drug combinations were appreciably more toxic to cells harboring DNAJ-PKAc than to cells only expressing wildtype kinase ( Figure 3G–I , pink dots ) . Deconvolution of our screening data revealed that these compounds were the MEK kinase inhibitors cobimetinib , binimetinib and trametinib . Further validation that these Hsp70/MEK inhibitor cocktails selectively target AML12DNAJ-PKAc cells was obtained when the combination drug screen was repeated using lower doses of Ver-155008 ( 3 μM; Figure 3—figure supplement 1 ) . Dose response curves revealed that wildtype AML12 cells are sensitive to cobimetinib alone ( Figure 3J ) whereas AML12DNAJ-PKAc cells were more resistant to this drug over the same concentration range ( Figure 3K ) . Importantly , in the presence of Ver-155008 the cytotoxic effect of cobimetinib in AML12DNAJ-PKAc cells was enhanced ( Figure 3K ) . Taken together , this screening venture provides two exciting new pieces of information: inhibition of Hsp70 in conjunction with blocking the RAF-MEK-ERK kinase cascade selectively affects the growth of cells expressing a single allele of DNAJ-PKAc , and drug combinations that target DNAJ-PKAc/Hsp70 assemblies offer a therapeutic strategy for FLC that warrants further investigation . A hallmark of FLC is the presence of fibroid bands that are interspersed between cancerous hepatocytes ( Craig et al . , 1980 ) . This morphological feature is indicative of ‘intratumoral heterogeneity’ which promotes microenvironmental diversity in the primary liver cancer ecosystem ( Liu et al . , 2018; Pribluda et al . , 2015 ) . Through a combination of biochemical , imaging and proteomic approaches we show that intratumoral heterogeneity influences ERK signaling within FLCs . Immunoblot analyses of tumor lysates detect a slight reduction in global phospho-ERK signal in patient samples ( Figure 4A , top panel ) . Yet immunofluorescent staining of tumor sections reveals clusters of prominent phospho-ERK signal in the cancerous hepatocytes ( Figure 4B & C , yellow; from patient 3 ) . Such regional detection of phospho-ERK is consistent with heterogeneous activation of the ERK cascade within the tumor . Likewise , the phosphoproteomic screen presented in Figure 1E & F identifies numerous ERK substrates that are elevated in FLC tumor as compared to normal liver ( Figure 4D ) . This includes the protein kinase P90RSK , a well-characterized downstream target of ERK ( Dalby et al . , 1998 ) . Validation of this ERK phosphorylation event is provided in two ways . First , immunoblot detection of pSer 221-P90RSK indicates variable activation of this kinase in the same cohort of FLC samples ( Figure 4E , top panel ) . Second , immunofluorescent detection of phospho-P90RSK in tissue sections of FLC uncovered clusters of cells containing activated kinase ( Figure 4F & G , magenta; from patient 3 ) . Collectively these findings infer that the RAF-MEK-ERK kinase cascade is active in a subset of cells within the heterogeneous intratumoral environment of FLCs . On the basis of our understanding of how local signaling events are organized , we reasoned that AKAPs may be integral components of DNAJ-PKAc complexes ( Smith et al . , 2017 ) . A logical candidate was AKAP-Lbc , a multifunctional anchoring protein and enhancer of ERK signaling ( Smith et al . , 2010 ) that interacts with another scaffolding protein , kinase suppressor of Ras ( KSR ) , to form the core of a signaling network that integrates cAMP regulation of RAF-MEK-ERK signaling ( Figure 5A ) . We found that AKAP-Lbc protein is up-regulated in human FLCs as compared to normal adjacent liver ( Figure 5B , top panel , lane 2 ) and immunoblot analysis detected DNAJ-PKAc in AKAP-Lbc immune complexes isolated from FLCs ( Figure 5C , top panel , lane 2 ) . Parallel experiments show that DNAJ-PKAc/Hsp70 sub-complexes co-fractionate with this anchoring protein in AML12DNAJ-PKAc cells ( Figure 5D , top panel , lane 2 ) . Thus , AKAP-Lbc can sequester Hsp70 and DNAJ-PKAc with an ERK signaling module in AML12DNAJ-PKAc cells and human FLCs . Detection of phospho-ERK1/2 is frequently used as a biochemical readout for activation of the RAF-MEK-ERK kinase cascade ( Rossomando et al . , 1992 ) . Notably , basal levels of phospho-ERK 1/2 were elevated 2 . 8 ± 1 . 5 fold ( n = 4 ) in AML12DNAJ-PKAc cells as compared to wildtype controls ( Figure 5E ) . This finding was confirmed in situ by immunofluorescent detection . Phospho-ERK signal was barely detectable in control AML12 cells ( Figure 5F & G ) , but clearly evident in the cytoplasm of AML12DNAJ-PKAc cells ( Figure 5H & I ) . Actin ( red ) and DAPI ( blue ) were used as cytoskeletal and nuclear markers respectively ( Figure 5G & I ) . We next monitored the efficacy of Hsp70/MEK inhibitors on basal ERK activity in AML12DNAJ-PKAc cells . In wildtype cells , treatment with Ver-155008 ( 3 μM ) alone had no effect on ERK activation ( Figure 5J , top panel , lanes 1 and 2 ) . However , administration of cobimetinib ( 100 nM ) or a combination of both drugs abolished detection of the phospho-ERK signal ( Figure 5J , top panel , lanes 3 and 4 ) . In contrast , basal phospho-ERK levels were high in AML12DNAJ-PKAc cells , treatment with Ver-155008 ( 3 μM ) alone had a modest effect on phospho-ERK signal ( Figure 5J , top panel , lanes 5 and 6 ) . Application of cobimetinib ( 100 nM ) or in combination with Ver-155008 abolished detection of phospho-ERK signals ( Figure 5J , top panel , lanes 7 and 8 ) . Thus , dual inhibition of Hsp70 and elements of the RAF-MEK-ERK cascade impedes mitogenic signals to preferentially block proliferation of AML12DNAJ-PKAc cells . This postulate was confirmed by clonogenic growth assays that monitor colony formation . Crystal violet staining showed that the synergistic effect of cobimetinib ( 100 nM ) and Ver-155008 ( 3 μM ) blocked AML12DNAJ-PKAc cell proliferation more potently than either drug alone ( Figure 5K ) . Qualitatively similar results were obtained when parallel experiments were conducted with the more potent MEK inhibitor trametinib ( 30 nM ) ( Figure 5—figure supplement 1 ) . One intriguing outcome of our study is the question of whether or not interrupting the association between DNAJ-PKAc and Hsp70 impacts activation of the RAF-MEK-ERK cascade . Mutation of a conserved HPD motif that demarks a critical loop in the DNAJ domain abolishes interaction with Hsp70 ( Hennessy et al . , 2005 ) ( Figure 6A ) . Thus , substitution of H33 to Q in the context of DNAJ-PKAc would be expected to prevent association with endogenous Hsp70 in AML12 cells ( Figure 6B ) . Wildtype AML12 cells were transfected with vectors encoding DNAJ-PKAc or DNAJ-PKAc H33Q . Additional co-immunoprecipitation experiments used transiently transfected AKAP-Lbc as the scaffold to isolate DNAJ-PKAc-Hsp70 sub-complexes . Introduction of the H33Q mutation greatly reduces the level of Hsp70 in AKAP-Lbc complexes ( Figure 6C , top panel , lane 3 ) . The simplest explanation of this result is that addition of the J-domain onto the N-terminus of PKAc induces a novel interaction with Hsp70 , thereby permitting the recruitment of this chaperonin to AKAP signaling islands . Immunoblot detection confirmed that basal levels of phospho-ERK were elevated upon introduction of DNAJ-PKAc in wildtype AML12 cells while transfection with the DNAJ-PKAc H33Q mutant diminished ERK activation ( Figure 6D , top panel ) . Densitometry analysis of four independent experiments confirmed this result ( Figure 6D , graph ) . Control immunoblotting monitored total ERK levels as a loading control and confirmed equivalent expression of each DNAJ-PKAc form in transfected cells ( Figure 6D , lower panel ) . Independent support for our hypothesis was provided through a phosphoproteomic screen that identified 2912 unique phosphopeptides in wildtype and AML12DNAJ-PKAc cells ( Figure 6E ) . Of these , 96 phosphopeptides were increased ( orange ) and 76 were reduced in AML12DNAJ-PKAc cells ( black ) . Substrate profiling using the NetworKIN platform revealed that 23% of ERK phosphosites were up-regulated in AML12DNAJ-PKAc cells whereas only 3% of PKA consensus sites were enriched ( Figure 6F ) . Enrichment of PKC ( 7% ) , DAP kinase ( 5% ) and CDK ( 3% ) phosphosites were also evident . This systemwide analysis suggests that DNAJ-PKAc/Hsp70 macromolecular assemblies bias the signaling landscape toward ERK activation and mobilize other downstream kinase networks .
We have discovered that DNAJ-PKAc , a unique fusion protein that is emblematic of fibrolamellar carcinoma ( FLC ) , functions as a scaffolding protein to assemble additional signaling elements that contribute to the pathogenesis of this cancer . More specifically , we show that the chaperonin-binding domain of this fusion enzyme supports recruitment of the co-chaperonin Hsp70 . This creates a unique molecular context in which the DNAJ-PKAc chimera acts in FLCs . Chaperonopathies are a group of diseases caused by genetic lesions or aberrant post-translational modifications of molecular chaperones ( Macario and Conway de Macario , 2007 ) . ‘Chaperonopathies by mistake’ are a sub-group of related disorders , including certain cancers , in which chaperonin activity is normal , but becomes inappropriately assimilated into molecular pathways that enhance disease progression ( Macario and Conway de Macario , 2007 ) . We believe that formation of DNAJ-PKAc/Hsp70 sub-complexes in FLC is an example of this latter category ( Calderwood et al . , 2006; Mayer and Bukau , 2005; Whitesell and Lindquist , 2005 ) . Chaperonins can repair misfolded proteins to reduce cellular stress or , as we believe is the case in FLC , recruitment of Hsp70 through the DNAJ domain preferentially stabilizes the chimeric PKAc fusion protein . This hypothesis is borne out by data in Figures 1B and 2G wherein we demonstrate that protein levels of the DNAJ-PKAc variant are elevated in tumor samples and disease-relevant cell lines as compared to native PKA . In addition , the abnormal pairing of Hsp70 with DNAJ-PKAc creates new and unique drug targets . This rationale provided the impetus to screen a panel of recognized chemotherapeutics in combination with an Hsp70 inhibitor . This new precision pharmacology approach ascertained if certain drug pairings act synergistically to inhibit proliferation of cells harboring macromolecular complexes of this chaperonin and the fusion kinase . Since aberrant kinase activity is known to drive many cancers , we further reasoned that augmented PKA activity could also contribute to the pathobiology of FLC ( Druker et al . , 2001; Turnham and Scott , 2016 ) . However , one confounding factor is that the pathological DNAJ-PKAc fusion and its native kinase counterpart share similar sensitivities to the inhibitor PKI and efficiently bind R subunits to form type I and type II PKA holoenzymes ( Cao et al . , 2019; Cheung et al . , 2015; Riggle et al . , 2016a; Scott et al . , 1985 ) . Although physiochemically similar , notable differences between the PKA holoenzyme subtypes include lack of an autoregulatory phosphorylation site in RI isoforms , different in vitro binding affinities for cAMP and dispersal to distinct subcellular sites via interaction with distinct AKAPs ( Aggarwal et al . , 2019; Burgers et al . , 2012; Feramisco et al . , 1980; Means et al . , 2011; Smith et al . , 2018; Taylor et al . , 2012 ) . Another noteworthy feature is that expression of the DNAJ-PKAc enhances production or stabilization of the RIα subunit . Interestingly this phenomenon occurs in both FLCs and AML12DNAJ-PKAc cells ( Langeberg and Scott , 2015; Riggle et al . , 2016a ) and Figure 1—figure supplement 1A ) . Such increased availability of RIα subunits may be indicative of tumor-specific variation in the ratio of type I to type II PKA activity . Switching of PKA isotypes may be clinically relevant as lesions in RI subunit genes are linked to disease ( Cho-Chung and Nesterova , 2005; Stratakis , 2013 ) . For example , nonsense and insertion mutations reduce levels of RIα in the endocrine neoplasia Carney complex ( Stratakis , 2013 ) . Similarly , mutations in the cAMP binding sites that render RIα less sensitive to cAMP have been linked to the rare skeletal dysplasia syndrome acrodysostosis type I ( Rhayem et al . , 2015 ) . Yet , perhaps the most intriguing example is a single case report of inactivating mutations in RIα that induce sporadic fibrolamellar carcinomas in the absence of classic DNAJ-PKAc ( Graham et al . , 2018 ) . Although the molecular mechanism surrounding this unusual case is not clear , one can postulate that reduced type I PKA activity , loss of anchoring to RI selective AKAPs , or overcompensation by type II PKA holoenzymes contributes to pathogenesis . Thus , marked changes in the quantity , isotype ratio and subcellular distribution of PKA holoenzymes , combined with the availability of DNAJ-PKAc may be factors that contribute to the etiology of FLC . A second postulate is that re-localization of Hsp70 to AKAP complexes by DNAJ-PKAc may be a critical event in transformation in FLC patients . Thus , the chaperonin binding properties of DNAJ-PKAc may as pertinent to oncogenesis as the intrinsic kinase activity of the fusion enzyme . Hence , we propose that the genetic lesion in chromosome 19 that is a hallmark of FLC incorporates a new binding interface that transforms PKA from an essential ‘homeostatic enzyme’ into a dual-function kinase/scaffolding protein with pathological implications . Our combination drug screen implicates mobilization of the ERK signaling cascade as a likely factor in the progression of FLC . Although mitogen-activated protein kinase ( MAPK ) pathways feature prominently in many cancers ( Kolch et al . , 2015 ) , we propose that the impact of RAF-MEK-ERK signaling on FLC is complex and atypical . Three factors contribute to this view . First , whole exome sequencing confirms that FLCs lack activating mutations in Ras or B-RAF , but rather arise from a monogenic lesion in chromosome 19 that produces the DNAJ-PKAc fusion ( Cornella et al . , 2015; Lalazar and Simon , 2018; Simon et al . , 2015; Xu et al . , 2015 ) . Second , our screen of FDA approved anti-cancer compounds in Figure 3H and I reveals that drugs targeting upstream elements of the ERK cascade , including EGF receptor antagonists erlotinib , lapatinib and afatinib and the B-RAF inhibitors dabrafenib and vemurafenib were ineffective , or at best exhibited modest antiproliferative effects when used in combination with Ver-155008 . Interestingly , the ERK inhibitor GDC-0994 had little effect on proliferation in these combination screens . Therefore , we interpret the exquisite sensitivity of AML12DNAJ-PKAc cells to MEK inhibition to suggest that DNAJ-PKAc may be acting downstream of Ras-Raf activation . A third contributing factor seems to be the atypical pattern of ERK activation in FLCs , which impacts downstream phosphorylation events . We base this conclusion on the regional immunofluorescent detection of phospho-ERK and its substrate P90RSK in FLC sections ( Figure 4C & G ) . If these findings are reconciled with immunoblot data indicating that global levels of phospho-ERK and phospho-P90RSK minimally change tumor lysates , it argues for heterogeneous activation of both kinases occurs only in pockets of tumor . Collectively , these observations argue that the distinctive morphological features of FLC where cancerous cells are intermingled with fibrous tissue creates a heterogeneous tumor microenvironment that is prone to irregular activation of the ERK signaling cascade . Although molecular links between ERK and DNAJ-PKAc were not immediately evident , we reasoned that one commonality was the proto-oncogene AKAP-Lbc . Anchored PKA activity has been implicated in the phosphorylation of RAF kinase and KSR-1 in the context of AKAP-Lbc signaling complexes ( Smith et al . , 2017; Smith et al . , 2010; Takahashi et al . , 2017 ) . In addition , AKAP-Lbc is upregulated in human FLCs , and interacts directly with RAF-MEK-ERK kinase signaling scaffolds . In keeping with this molecular mechanism , our phosphoproteomic analysis identifies elevated PKA phosphorylation of serine 838 on KSR in FLCs ( Figure 1—figure supplement 1D ) . This is especially interesting in light of early findings that Hsp90 and certain Hsp70 isoforms are elements of KSR scaffolds ( Stewart et al . , 1999 ) and data in Figure 5B–D showing that the chaperonins including the Hsp70-DNAJ-PKAc subcomplex are selectively recruited to AKAP-Lbc-KSR signaling units in FLC . The relationship between cAMP and ERK signaling in cancer is complex and context dependent ( Dumaz and Marais , 2005 ) . For example , PKA has pleotropic effects on tumor-initiation . Paradoxically , a recent report postulates that PKA activity leads to mesenchymal-to-epithelial transitions that impede oncogenesis; yet DNAJ-PKAc kinase activity is thought to be necessary for tumor initiation ( Kastenhuber et al . , 2017; Pattabiraman et al . , 2016 ) . Therefore , one pertinent and unanswered question is whether or not the kinase activity residing within DNAJ-PKAc is an absolute requirement for FLC progression . This view is further substantiated by the phosphoproteomics data presented in Figure 6E and F showing that DNAJ-PKAc/Hsp70 macromolecular assemblies skew the signaling landscape toward enhanced ERK signaling rather than simply potentiating the action of PKA . Together , these results imply that recruitment of Hsp70 enhances basal ERK signaling in AML12DNAJ-PKAc cells by preferentially stabilizing this oncogenic signaling unit . Indirect support for this notion is presented in Figure 6D showing that abolishing the binding of Hsp70 to DNAJ-PKAc complex decreases ERK signaling . Thus , we postulate that the preferential stabilization of DNAJ-PKAc proceeds through the local action of Hsp70 . Such a mechanism could explain why greater amounts of the DNAJ-PKAc fusion kinase are detected in FLCs and our AML12DNAJ-PKAc cells as compared to wildtype PKA . That said , we do not discount the kinase activity of DNAJ-PKAc as a pathological factor in FLC . Rather , we propose that recruitment of Hsp70 via the DNAJ domain in the chimeric DNAJ-PKAc kinase is an important new element that contributes to the dysregulation of this unique fusion enzyme . One shared objective of the FLC research community and those investigating rare adolescent cancers is to identify and test therapeutic targets in the most efficient possible manner ( Kastenhuber et al . , 2019 ) . One advantage of screening FDA-approved compounds is that the pharmacotoxicity , therapeutic indices , and off-target effects of most components are well documented . Accordingly , each MEK inhibitor identified in our screen has been approved for the treatment of melanoma and other cancers ( Caunt et al . , 2015 ) . Another benefit of the combination screening approach is the potential to identify drug pairings that can be used at lower effective doses; though there is also the possibility that new drug combinations may prove more toxic . This could be an important consideration for Ver-155008 as clinical trials with other Hsp70 inhibitors hold promise for the treatment of cancers ( Goloudina et al . , 2012 ) . Although the utility of Hsp70 and MEK inhibition as combination therapy for FLC is far from clear , our discovery of drug pairs that selectively halt the growth of cells expressing DNAJ-PKAc but not wildtype hepatocytes provides a valuable tool to further the investigation for new treatments of this debilitating disease of adolescents .
Human FLCs with paired normal liver were consented for tissue donation under IRB-approved protocols ( #31281 and #51710 ) . Human FLC and normal adjacent liver was harvested according to above IRB and flash frozen . AML12 and AML12DNAJ-PKAc cells were grown on a 15 cm dish and after rinsing twice with ice cold PBS , cells were harvested in 750 μL of 6M aq . Guanidine hydrochloride ( Gdn*HCl ) containing 100 mM Tris , 5 mM TCEP*HCl , and 10 mM chloroacetamide ( CAM ) , pH 8 . 5 , using a cell scraper . Frozen human FLC specimens of ca . 100 mg wet weight were ground into a fine powder using the CryoGrinder Kit from OPS Diagnostics ( Lebanon , NJ ) and added to the Gdn*HCl buffer described above . Cell lysates were pipetted into 1 . 5 mL microtubes , voretexed briefly and heated to 95C for 5 min . Samples were then sonicated in a Qsonica cup sonicator ( Newton , CT ) at 100 W for 10 min ( 30 s on , 30 s off ) on ice . Protein content was measured using the Pierce 660 nm assay reagent ( Thermo Fisher Scientific , Waltham , MA ) . Aliquots of 300 μg of protein were pipetted into a new tube and diluted 2-fold with 100 mM triethylammonium bicarbonate ( TEAB ) pH = 8 . 5 . 3 μg of sequencing-grade endoproteinase Lys-C ( Wako , Richmond , VA ) were added ( 1:100 ratio ) and the mixture agitated on a thermomixer at 1400 rpm at 37°C for 2 hr . The mixture was diluted another 2-fold with 100 mM TEAB pH = 8 . 5 and 3 μg of trypsin were added . The mixture was agitated on a thermomixer at 1400 rpm at 37°C for overnight , acidified with formic acid ( 1% final ) , and cleared by centrifugation for 10 min at RT and 14 , 000 rcf . Peptides were extracted from the supernatant using Oasis HLB 1cc ( 10 mg ) extraction cartridges ( Waters , Milford , MA ) . Cartridges were activated by passing through 200 μL of methanol followed by 200 μL 80% aq . ACN containing 0 . 1% TFA , equilibrated with 400 μL 1% aq . formic acid . Peptides were loaded and then washed with 400 μL 1% aq . formic acid . Peptides were eluted with 300 μL 80% aq . ACN containing 0 . 1% TFA and directly subjected to the published batch IMAC phosphopeptide enrichment protocol with the following minor modifications ( Golkowski et al . , 2016; Villén and Gygi , 2008 ) . 20 μL of a 50% IMAC bead slurry composed of 1/3 commercial PHOS-select iron affinity gel ( Sigma Aldrich ) , 1/3 in-house made Fe3+-NTA superflow agarose and 1/3 in-house made Ga3+-NTA superflow agarose were used for phosphopeptide enrichment ( Ficarro et al . , 2009 ) . The IMAC slurry was washed three times with 10 bed volumes of 80% aq . ACN containing 0 . 1% TFA and phosphopeptide enrichment was performed in the same buffer . Phosphopeptides were desalted using C18 StageTips according to the published protocol with the following minor modifications; after activation with 50 μL methanol and 50 μL 80% aq . ACN containing 0 . 1% TFA the StageTips were equilibrated with 50 μL 1% aq . formic acid . Then the peptides that were reconstituted in 50 μL 1% aq . formic acid were loaded and washed with 50 μL 1% aq . formic acid . The use of 1% formic acid instead of 5% aq . ACN containing 0 . 1% TFA prevents the loss of highly hydrophilic phosphopeptides . The LC-MS/MS analyses were performed on a Thermo Fisher Scientific Orbitrap Elite instrument ( AML12 cell lines ) or a Thermo Fisher Scientific Orbitrap Fusion ( human FLC specimens ) as described previously with the following minor modifications ( Golkowski et al . , 2017 ) . Peptide samples were separated on a Thermo-Dionex RSLCNano UHPLC instrument ( Sunnyvale , CA ) using 20 cm long fused silica capillary columns ( 100 μm ID ) packed with 3 μm 120 Å reversed phase C18 beads ( Dr . Maisch , Ammerbuch , DE ) . For phosphopeptide samples the LC gradient was 120 min long with 3–30% B at 300 nL/min . LC solvent A was 0 . 1% aq . acetic acid and LC solvent B was 0 . 1% acetic acid , 99 . 9% acetonitrile . Data-dependent analysis was applied using Top15 selection with CID fragmentation . Raw files were analyzed by MaxQuant/Andromeda ( Olsen et al . , 2010 ) version 1 . 5 . 2 . 8 using protein , peptide and site FDRs of 0 . 01 and a score minimum of 40 for modified peptides , 0 for unmodified peptides; delta score minimum of 17 for modified peptides , 0 for unmodified peptides . MS/MS spectra were searched against the UniProt human database ( updated July 22nd , 2015 ) . MaxQuant search parameters: Variable modifications included Oxidation ( M ) and Phospho ( S/T/Y ) . Carbamidomethyl ( C ) was a fixed modification . Max . missed cleavages was 2 , enzyme was Trypsin/P and max . charge was 7 . The MaxQuant ‘match between runs’ feature was enabled . The initial search tolerance for FTMS scans was 20 ppm and 0 . 5 Da for ITMS MS/MS scans . MaxQuant raw data were processed , statistically analyzed and clustered using the Perseus software package v1 . 5 . 6 . 095 ( Tyanova et al . , 2016 ) . Human gene ontology ( GO ) terms ( GOBP , GOCC and GOMF ) were loaded from the Perseus Annotations file downloaded on the 01 . 08 . 2017 . Expression columns ( phosphopeptide MS intensities ) were log2 transformed and normalized by subtracting the median log2 expression value of each column from each expression value of the corresponding column . Potential contaminant , reverse hits and proteins only identified by site were removed . Reproducibility was analyzed by column correlation ( Pearson’s r ) and replicates that showed a variation of >0 . 25 in the r value compared to the mean r-values of all replicates of the same experiment were removed as outliers . Significant differences in phosphopeptide expression between experiments were quantified with a two-tailed two sample t-test with unequal variances and Benjamini-Hochberg correction for multiple comparisons was applied ( FDR = 0 . 05 ) . For human FLC and normal adjacent liver , significantly enriched phosphosites in FLC were input into the NetworKIN platform . For sites significantly enriched in AML12DNAJ-PKAc cells , the conserved phosphosite in human was identified in PhosphoSitePlus and then input into NetworKIN . Minimum score cutoff was 1 . AML12 mouse hepatocytes were used in this study . These cells were developed by the Nelson Fausto lab ( Wu et al . , 1994 ) . The cells from this study came from Dr . KJR and are also available at ATCC ( https://www . atcc . org/Products/All/CRL-2254 . aspx ) . The cells were verified and mycoplasma free before beginning these studies and are currently being re-tested by STR and mycoplasma detection at IDEXX ( Westbrook , ME ) . AML12 cells were cultured in DMEM/F12 supplemented with 10% FBS , 0 . 04 µg/mL dexamethasone , 0 . 1% gentamicin , 1 µg/mL recombinant human insulin , 0 . 55 µg/mL human transferrin , and 0 . 5 ng/mL sodium selenite . All cell lines were maintained in a 5% CO2 incubator at 37°C . For lysates probed with phospho-ERK , cells were serum-starved for 16–24 hr and lysed . Serum-starved medium was prepared as above with the exception of addition of FBS . Cells for cobimetinib and Ver-155008 treatment were serum-starved for 16–24 hr and then incubated with 3 μM Ver-155008 for 30 min , and either DMSO or 100 nM Cobmetinib was incubated for 10 min . AML12 cells for Figure 6B–D were transfected with constructs as indicated with TransIT-LT1 ( Mirus Bio ) . Cells for Figure 6B & C were collected for immunoprecipitation after 24 hr , while cells for Figure 6D cells were switched to serum-free media for 16–24 hr . Guide ( g ) RNAs were designed to target intron 1 of either mouse Dnajb1 ( GCATTCCGGGGATCTAGCGG ) or Prkaca ( GTAGTGCTGAGGAGAGTGGGG ) in order to introduce DNA double-stranded breaks in the regions similar to the deletion seen in FL-HCC . We engineered constructs expressing Cas9 and both guide ( g ) RNAs into SpCas9-2A-Puro V2 . 0 ( Addgene plasmid number 62988 ) ( Ran et al . , 2013 ) and transfected the vector into AML12 cells using lipofectamine LTX with plus ( Thermo Fisher ) according to manufacturer's instructions . Cells were subjected to 2 µg/mL puromycin ( Sigma ) selection 48 hr post-transfection . After 3 days in puromycin-containing media , cells were clonally isolated . After selection , cells were dissociated using 0 . 25% trypsin-EDTA ( Gibco ) and 200 cells were plated into 15 cm2 dish and incubated for 48–96 hr or until single-cell derived colonies were visible . Single-cell derived colonies were hand picked with cloning disks ( 3 . 2 mm diameter , Sigma-Aldrich ) soaked with 0 . 25% trypsin-EDTA and plated into single wells of a 96-well plate . Genomic DNA was extracted ( GeneJET Genomic DNA purification kit , Thermo Fisher ) to screen clonally-isolated cells . Polymerase chain reaction ( PCR ) was performed to determine a heterozygous deletion . Primer sequences are found in the Key Resources Table . Total RNA was extracted from wildtype AML12 and Dnajb1-Prkaca clones using trizol and RNeasy Mini Kits ( Qiagen ) and reverse transcribed using iScript Reverse Transcription Supermix for RT-qPCR ( Bio-Rad ) according to manufacturer’s instructions . The cDNA was subjected to PCR with primers against Dnajb1-Prkaca fusion , and the resulting amplification was subjected to Sanger sequencing . Quantitative PCR was performed on ABI Fast 7500 using PowerUp SYBR Green Master Mix ( Thermo Fisher ) according to manufacturer’s instructions with primers ( see Key Resources Table ) against Dnajb1-Prkaca fusion , wildtype Dnajb1 , wildtype Prkaca . Data are reported as delta delta Ct after normalizing to Gapdh . Dnaj-PKAc cDNA was isolated from clone 14 and cloned into Zero Blunt TOPO PCR Kit ( Thermo Fisher ) and sequenced to verify the in-frame fusion . Cells and human FLCs were lysed in ice-cold RIPA buffer ( 10 mM Tris-HCl , 150 mM NaCl , 1% sodium deoxycholate , 1% Nonidet P-40 , 0 . 1% SDS , 2 mM EDTA , 50 mM sodium fluoride ) with protease inhibitors . Cleared lysate was measured using BCA Protein Assay ( Pierce ) . Lysate was boiled in 1X LDS buffer ( Thermo Fisher ) , separated on 4–12% NuPAGE gradient gels ( Thermo Fisher ) and transferred onto nitrocellulose using standard techniques . Membranes were incubated overnight at 4°C in 5% w/v milk with TBST and the following antibodies: PKAc ( BD Transduction , 610981 ) , Hsp70 ( Proteintech , 10995–1 ) , β-actin ( Sigma-Aldrich , A1978 ) , AKAP-Lbc ( V096 , 1 µg ml−1 ) , phospho p44/42 MAPK ( CST , 9101 ) , p44/42 ( CST , 9102 ) . Membranes were washed in TBST , incubated with HRP-labeled secondary antibodies ( GE Life Sciences ) , washed as before and developed using ECL ( Thermo Fisher ) on an iBright FL1000 . For re-probing , membranes were striped with 1X Re-Blot Plus Strong ( Millipore ) for 15 min and then re-blocked in Blotto before incubation with primary antibodies again . Densitometry for blot quantification was done using ImageJ software ( NIH; http://rsb . info . nih . gov/ij ) . Human tissue and cell lysates were lysed in 0 . 5% or 1% Triton-X buffer ( 50 mM Tris-HCl , 130 mM NaCl , 20 mM NaF , 2 mM EDTA , 0 . 5% or 1% Triton-X with protease inhibitors ) . Lysates were pre-cleared with IgG and protein A/G agarose beads ( Millipore ) then incubated with anti-PKAc , anti-HSP70 , anti-GFP , or anti-AKAP-Lbc antibodies overnight at 4°C . Immunocomplexes were separated by incubation with protein A/G agarose beads for 2 hr at 4C and washed 4 × 1 mL in lysis buffer . For FLAG immunoprecipitation , lysates were incubated with anti-FLAG M2 magnetic beads ( Sigma M8823 ) overnight . Immunocomplexes were washed 4 × 1 mL in lysis buffer . AML12 and AML12DNAJ-PKAc cells were plated on a 96-well plate and subjected to IncuCyte ZOOM 96-Well Scratch Wound Cell Migration and Invasion assay ( Essen Bioscience ) . Matrigel ( Corning ) was used in invasion assays . Data are representative images of n = 3 . Images collected every 45 min for 24 hr ( migration assay ) or 48 hr ( invasion assay ) . SignaTECT cAMP-dependent Protein Kinase ( PKA ) Assay System ( Promega , V7480 ) was used to measure kinase activity . Cells were lysed and PKA activity was measured according to protocol ( ATP , [γ−32P]- 3000 Ci/mmol 10mCi/ml EasyTide; Perkin Elmer , BLU502A001MC ) . Experiments were carried out ±25 μM cAMP to stimulate PKA , ±Kemptide substrate for normalization , and ±50 μM PKI to inhibit PKA . AML12 cells were grown on coverslips and fixed with 4% paraformaldehyde/PBS for 20 min . After several washes in PBS , samples were permeabilized in 0 . 5% Triton X-100/PBS for 10 min and washed extensively in PBS . Cells were then subjected to PLA or immunofluorescence . Human liver tissue for PLA was fresh frozen , cut on a cryostat at 8 µm , and fixed in 4% paraformaldehyde/PBS at RT for 4 min . For PLA , samples were processed according to manufacturer’s instructions with anti-mouse and anti-rabbit reagents ( Sigma ) using PKAc ( BD Transduction , 610981 ) and Hsp70 ( Proteintech , 10995–1 ) . Z-stacks of fluorescent images were collected using a Keyence BZ-X710 using relevant filter cubes . Maximum intensity projections were quantified for puncta number using Fiji/ImageJ . For AML12 cell PLA , images were smoothened and a duplicate image was created for use as a mask . The duplicate file was thresholded to capture as many puncta as possible without significant blending of densely packed signal . The binary mask was then used to measure selected regions from the original image . Total cell number per field of view was counted as DAPI-stained nuclei . For quantification of human liver tissue PLA , unfocused light was removed using the Keyence haze reduction function . Puncta number and fluorescence intensity were measured by automation using Keyence hybrid cell counter set to detect thresholded puncta between 0 and 1 . 0 µm in diameter . Puncta counts were normalized to the total image area . Human liver tissue for immunostaining was formalin fixed and paraffin embedded . Samples were permeabilized in 0 . 5% Triton X-100/PBS for 10 min . Images for immunofluorescence were immunostained with primary antibodies [PKAc ( CST , 5842 ) , PKA RIIα ( BD Transduction , 610626 ) , Erk ( BD 610123 ) , ( phospho p44/42 MAPK ( CST , 9101 ) , or phospho P90RSK ( Thermo Fisher PA5-37829 ) ] overnight at 4C . Cells were washed three times in PBS and incubated with Alexa Fluor conjugated secondary antibodies ( Thermo Fisher ) for 2 hr at room temperature . Nuclei were stained with DAPI and samples were mounted on glass slides using ProLong anti-fade media ( Invitrogen ) or Aqua-Mount ( Thermo Scientific ) . Images were taken on a Leica DMI6000B inverted microscope with a spinning disk confocal head ( Yokagawa ) and a CoolSnap HQ camera ( Photometrics ) controlled by MetaMorph 7 . 6 . 4 ( Molecular Devices ) , or a BZ-X710 microscope ( Keyence ) . Cells were seeded at 3 , 000 cells/well into 96-well plates , allowed to recover for 16–24 hr and either treated with Ver-155008 , DMSO , or left untreated . MTS reagent ( CellTiter 96 Aqueous One Solution , Promega ) was added per the manufacturer’s instructions , and absorbance was read at 490 nm 3 hr later . Wildtype AML12 and AML12DNAJ-PKAc cells were seeded at 20 , 000 cells/well on a 2-well chamber slide ( Lab-Tek ) . Fourty-eight hours after plating , cells were treated with 25 µg/mL BrdU ( Roche Diagnostics ) for 4 hr . Cells were washed twice in ice-cold PBS and fixed with 100% ice-cold methanol . BrdU labeling was then determined by immunohistochemistry by using anti-BrdU antibody ( DAKO ) . For clonogenic growth assays , cells were seeded at 200 cells/well in 12-well dishes . For inhibitor tests , drug was added following day to appropriate concentrations ( 100 nM cobimetinib or 30 nM trametinib; 3 µM Ver-155008 ) in normal growth media . Media/drug was refreshed every 5 days . After two weeks , cells were washed in PBS and fixed for 20 min in 4% paraformaldehyde/PBS . Cells were then stained with 0 . 1% crystal violet in 10% methanol , washed 3x with water and air dried for image capture . Images were quantified in ImageJ using masking and particle analysis to determine well surface area covered by stained cells . Data were further analyzed and plotted in Prism 7 . Drug screening of AML12 and AML12DNAJ-PKAc cells was performed using a drug library assembled by SEngine Precision Medicine ( Seattle , Washington ) that includes FDA-approved chemotherapies as well as drugs in clinical development . The drug screens used a dilution series of the inhibitors that started at 10 µM and decreased in half-log units to a final concentration of ~41 nM . Initial combination screens were performed with 10 µM Ver-155008 , a concentration well above the IC50 ( in vitro IC50 0 . 5 µM-2 . 6 µM ) to assure strong Hsp70 inhibition . Cells were tested in 2D and data evaluated as described ( Pauli et al . , 2017 ) . Statistically significant differences between samples were calculated as indicated in figure legends , using Student’s two-tailed t-test or ANOVA with post-hoc multiple comparisons for groups of 3 or more . All results are presented as the mean ±s . d unless otherwise indicated . Sample size ( n ) indicated the number of independent experiments represented in amalgamated data; total cell numbers used in experiments are indicated . P values of < 0 . 05 were considered statistically significant . Raw mass spectrometry data has been uploaded to MassIVE , an NIH supported MS data repository ( MSV000083167 ) . | Fibrolamellar carcinoma ( or FLC for short ) is a rare type of liver cancer that affects teenagers and young adults . FLC tumors are often resistant to standard radiotherapy or chemotherapy treatments . The only way to treat FLC is to remove tumors by surgery . However , often the tumors come back after initial treatment and spread to other locations . Therefore , there is a genuine need to improve the treatment options available to FLC patients . The tumor cells of FLC patients contain a genetic defect that fuses together two genes , which produce proteins called DNAJ and PKAc . Normally , DNAJ helps other proteins in the cell to fold into their correct shapes , while PKAc is an enzyme that can control how cells communicate . However , it is not clear what the abnormal DNAJ-PKAc fusion protein does , or how it causes FLC . Turnham , Smith et al . have now used gene editing to make mouse liver cells that mimic the human FLC mutation . Biochemical experiments on these cells showed that the DNAJ-PKAc protein brings together unique combinations of enzymes that drive uncontrolled cell growth . Analyzing cells taken from tumors in FLC patients confirmed that these enzymes are also activated in the human disease . Turnham , Smith et al . also found that combinations of drugs that simultaneously target the DNAJ-PKAc protein and the recruited enzymes slowed down the growth of FLC cells . More experiments are now needed to test these drug combinations on human FLC cells or in mice . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cancer",
"biology"
] | 2019 | An acquired scaffolding function of the DNAJ-PKAc fusion contributes to oncogenic signaling in fibrolamellar carcinoma |
Limited understanding of infant pain has led to its lack of recognition in clinical practice . While the network of brain regions that encode the affective and sensory aspects of adult pain are well described , the brain structures involved in infant nociceptive processing are less well known , meaning little can be inferred about the nature of the infant pain experience . Using fMRI we identified the network of brain regions that are active following acute noxious stimulation in newborn infants , and compared the activity to that observed in adults . Significant infant brain activity was observed in 18 of the 20 active adult brain regions but not in the infant amygdala or orbitofrontal cortex . Brain regions that encode sensory and affective components of pain are active in infants , suggesting that the infant pain experience closely resembles that seen in adults . This highlights the importance of developing effective pain management strategies in this vulnerable population .
The network of brain regions that encode both the affective and sensory aspects of the pain experience have been well described in the adult ( Apkarian et al . , 2005; Tracey and Mantyh , 2007 ) . It is not known which cortical and subcortical brain structures are activated following noxious events in infants . Early evidence demonstrated that infants exhibited reflex responses and concluded that pain was not processed at the level of the cortex ( Rodkey and Pillai Riddell , 2013 ) . This , coupled with an infant's inability to describe their pain experience verbally , led to extreme controversy regarding whether an infant has the ability to experience the unpleasant affective components of pain ( Rodkey and Pillai Riddell , 2013 ) . Consequently , infants have received poor pain management , exemplified during the 1980s by surgery being routinely performed using neuromuscular blocks without provision of adequate analgesia ( Anand and Hickey , 1987 ) . More recent research has primarily focussed on behavioural and physiological measures , which has led to the development of a number of infant pain assessment tools ( Duhn and Medves , 2004 ) . However , the lack of sensitivity and specificity of these measures means the trend to undertreat pain remains in clinical practice ( Carbajal et al . , 2008 ) , despite concerted efforts to improve the management of pain in this population ( Anand and International Evidence-Based Group for Neonatal Pain , 2001 ) . For example , it is remarkable that current UK NHS guidelines for ankyloglossia ( tongue tie ) surgery state that ‘in small babies , being cuddled and fed are more important than painkillers’ ( NHS Choices , 2015 ) . Indeed , a recent review of neonatal pain management practice in intensive care highlighted that although infants experience an average of 11 painful procedures per day , 60% of the population did not receive any pharmacological analgesia ( Roofthooft et al . , 2014 ) . Recent studies using EEG and near-infrared spectroscopy have been used to provide reliable evidence that nociceptive information is transmitted to the newborn infant brain ( Slater et al . , 2006 , 2010a , 2010b ) , and have highlighted the limitations of using observational behavioural measures to quantify pain in infants . For example , nociceptive information can be processed in the infant brain without a concomitant behavioural response ( Slater et al . , 2008 ) , and interventions thought to alleviate pain ( i . e . , oral sucrose ) can reduce clinical pain scores without reducing evoked nociceptive brain activity ( Slater et al . , 2010a ) . While these studies confirm that the infant central nervous system can process noxious stimulation , they do not elucidate the nature of the infant experience—in particular , which brain regions are involved , and therefore , whether the sensory , cognitive , and emotional aspects of pain are present in this population . Here , we identify the cortical and subcortical structures activated following acute noxious stimulation in the healthy newborn infant , and compare the activity with that observed in adults . The feasibility of this approach was demonstrated in a foundational pilot study ( Williams et al . , 2015 ) . A case study in a single infant demonstrated that noxious stimulation evoked widespread brain activity ( Williams et al . 2015 ) , which included brain regions previously reported to be involved in adult pain ( Tracey et al . , 2007 ) . Using a reverse inference approach to compare active brain regions in infants with those reported during adult pain , we postulate which aspects of the pain experience are present ( Wager et al . , 2013 ) , providing an opportunity to gain insight into the organisation of nociceptive circuitry in the naïve infant brain . In this study , acute noxious stimulation ( PinPrick Stimulators , MRC Systems ) was applied to the foot in both adults ( n = 10; applied force: 32–512 mN ) and infants ( n = 10; applied force: 32–128 mN; greater force was not applied due to the potential risk of tissue damage ) . Using functional magnetic resonance imaging ( fMRI ) changes in blood oxygen level dependent ( BOLD ) activity in the brain were recorded in response to the stimuli . Adults were asked to verbally report their pain intensity and , using the McGill Pain questionnaire ( Melzack and Torgerson , 1971 ) , to describe the quality of the pain they experienced . As infants are unable to describe their pain , reflex leg withdrawal from the stimuli was visually observed during scanning . Parents were present during the studies and no infants were withdrawn from the study after recruitment .
Adult participants reported increased pain with increasing stimulus intensity ( r = 0 . 48; p < 0 . 0001 ) , and most frequently described the pain as pricking ( n = 8 of 10 ) and sharp ( n = 6 of 10 ) . In infants , application of the stimuli evoked visible withdrawal of the stimulated leg , which could be observed at all stimulus intensities , whereas in adults , reflex withdrawal of the leg or foot was not observed at any stimulus intensity . While low threshold stimuli can also evoke reflex withdrawal in infants ( Cornelissen et al . , 2013 ) , this observation confirms that the stimuli applied in this study were detected by the peripheral nervous system and transmitted to the central nervous system . Although noxious stimulation can elicit reflex limb withdrawal in adults , supraspinal modulation of the input means this activity is often suppressed in experimental studies . In adults , noxious stimulation evoked significant increases in BOLD activity in cortical and subcortical brain regions , including primary somatosensory cortices , anterior cingulate cortex ( ACC ) , bilateral thalamus , and all divisions of the insular cortices ( Figure 1 ) . All brain regions that had a significant increase in BOLD following noxious stimulation are identified in Table 1 , and are consistent with previous literature ( Tracey and Mantyh , 2007 ) . In infants , the increases in BOLD activity evoked by the noxious stimulation were extremely similar to that seen in adults , and all but two of the 20 regions that were active in the adults were active in infants ( Table 1 and Figure 1 ) . While in adults the parietal lobe , pallidum , and precuneus cortex were only active in the brain regions contralateral to the site of stimulation , in infants these brain regions were also active on the ipsilateral side to the stimulus . Additional brain regions that were only active in the infants included the bilateral auditory cortices , hippocampus , and caudate ( Table 1 ) . The increased bilateral activity and greater number of active regions in infants are likely due to the immature cortico-cortical and interhemispheric pathways ( Kostovic and Jovanov-Milosevic , 2006 ) . Major reorganisation of the cortical circuitry occurs after the first postnatal month when there is a striking retraction of exuberant axons in the corpus callosum and there is a cessation of growth of the long cortico-cortical afferent pathways ( Jovanov-Milosevic et al . , 2006; Kostovic and Jovanov-Milosevic , 2006 ) . 10 . 7554/eLife . 06356 . 003Figure 1 . Comparison of nociceptive-evoked brain activity in selected brain regions that are active in both adults and infants . Significantly , active voxels across each stimulus intensity level are presented for ( A ) adult and ( B ) infant participants ( applied force: adults 32–512 mN; infants 32–128 mN ) . Each colour represents activity in a different anatomical brain region . ( A ) Adult activity is overlaid onto a standard T1 weighted MNI template and ( B ) infant activity is overlaid onto a standard T2 weighted neonatal template , corresponding to a 40-week gestation infant . ACC: anterior cingulate cortex; S1: primary somatosensory cortex: PMC: primary motor cortex; SMA: supplementary motor area . DOI: http://dx . doi . org/10 . 7554/eLife . 06356 . 00310 . 7554/eLife . 06356 . 004Table 1 . Identification of all active brain regions in adults and infants following acute noxious stimulation at all stimulus intensities ( applied force: adults 32–512 mN; infants 32–128 mN ) DOI: http://dx . doi . org/10 . 7554/eLife . 06356 . 004AdultsInfantsAnatomical areaRegionPeak Z within clusterMNI coordsRankSlope of regression ( *E-03 ) P val*Peak Z within clusterNeonate template coordsRankSlope of regression ( *E-03 ) P val*xyzxyzActive regions in both adults and infantsIntensity encoding regions ( in adults ) Temporal gyrusContra3 . 9264−342011 . 010 . 00023 . 0532−321212 . 460 . 0083Cingulate gyrusAnterior4 . 11644020 . 650 . 00052 . 58−1126111 . 010 . 3971Opercular cortexContra5 . 604061030 . 630 . 00013 . 3832−131922 . 230 . 0391InsulaContra4 . 183414640 . 610 . 00013 . 0419−222332 . 150 . 0207Supramarginal gyrusContra4 . 3364−382050 . 600 . 00083 . 2925−233991 . 080 . 1749Postcentral gyrusContra4 . 2858−182260 . 600 . 00123 . 8515−2252101 . 010 . 2667Visual cortexContra3 . 6244−62470 . 590 . 00043 . 2521−523461 . 410 . 0814PutamenContra3 . 68226680 . 550 . 00013 . 3017−171881 . 200 . 1656ThalamusContra3 . 5114−14090 . 500 . 00103 . 586−161541 . 910 . 0592InsulaIpsi4 . 67−38−1814100 . 490 . 00012 . 59−26−141451 . 690 . 1015Supplementary motor areaContra3 . 918446110 . 390 . 00083 . 506−184871 . 230 . 2315Non intensity encoding regions ( in adults ) CerebellumIpsi3 . 88−20−66−440 . 350 . 00293 . 53−3−46−63 . 570 . 0164Temporal gyrusIpsi3 . 72−52−56100 . 180 . 54873 . 41−32−22142 . 900 . 0196Supramarginal gyrusIpsi4 . 59−64−28200 . 510 . 00353 . 13−31−24302 . 790 . 0055CerebellumContra3 . 3620−70−500 . 310 . 02463 . 162−44−62 . 720 . 1634Opercular cortexIpsi5 . 23−50−28260 . 500 . 00182 . 69−27−12132 . 230 . 0710Postcentral gyrusIpsi4 . 71−62−18240 . 440 . 03753 . 52−31−15412 . 120 . 0845ThalamusIpsi3 . 52−12−14100 . 420 . 00183 . 48−1−20131 . 670 . 1009Angular gyrusIpsi3 . 59−58−50180 . 530 . 01072 . 98−23−39331 . 560 . 0528Precentral gyrusIpsi4 . 01−580100 . 430 . 05783 . 46−23−17481 . 530 . 1247Frontal gyrusContra3 . 88581200 . 560 . 02123 . 1111−12481 . 420 . 0646Cingulate gyrusPosterior3 . 71−14−28380 . 080 . 24803 . 18−9−23351 . 420 . 1101Angular gyrusContra3 . 7160−46180 . 540 . 00803 . 1222−51351 . 420 . 0407Precuneous cortexContra3 . 6016−68400 . 380 . 07143 . 705−30521 . 190 . 1623Visual cortexIpsi3 . 82−52−7010−0 . 090 . 37582 . 59−7−40111 . 170 . 1657Brainstem3 . 8610−26−80 . 330 . 17102 . 99−3−27−101 . 110 . 4350Parietal lobuleContra3 . 1020−44680 . 610 . 10973 . 1027−24461 . 090 . 1271PutamenIpsi3 . 63−1610−20 . 450 . 00233 . 13−14−14190 . 920 . 2813Supplementary motor areaIpsi3 . 55−64440 . 400 . 02193 . 16−4−10460 . 910 . 3903Precentral GyrusContra4 . 0558480 . 440 . 02763 . 766−20530 . 880 . 2672Frontal gyrusIpsi3 . 57−82232−0 . 240 . 19542 . 79−13−9500 . 700 . 4820PallidumContra3 . 4016−4−40 . 490 . 00712 . 8413−13130 . 640 . 4863Active regions in adults onlyAmygdalaContra3 . 4920−2−140 . 690 . 0160AmygdalaIpsi4 . 28−20−2−120 . 430 . 0860Orbitofrontal cortexIpsi3 . 40−184−160 . 420 . 0157no activityOrbitofrontal cortexContra3 . 573430−20 . 440 . 0460Active regions in infants onlyPrecuneous cortexIpsi3 . 80−1−26521 . 260 . 1699PallidumIpsi3 . 16−8−5140 . 590 . 4787Parietal lobuleIpsi3 . 31−28−23330 . 990 . 2711Auditory cortexContra2 . 8926−14183 . 070 . 0119Auditory cortexIpsi3 . 34−17−29192 . 560 . 0304CaudateContrano activity3 . 6113−17220 . 590 . 5822CaudateIpsi3 . 47−7−8181 . 050 . 3415HippocampusContra2 . 6121−2591 . 840 . 1288HippocampusIpsi2 . 77−15−3191 . 000 . 3326ParahippocampusContra3 . 0211−2301 . 530 . 3740ParahippocampusIpsi2 . 99−7−24−80 . 190 . 9013 Although the infant brain activity was widespread , the specificity of the response was demonstrated , as it was not present across all brain regions . For example , brain regions not commonly associated with the cerebral processing of nociceptive stimulation in the adult , such as the olfactory cortex , cuneus , and fusiform gyrus , were also not active in the infants . 14% of voxels across the whole brain were active following the application of the 128 mN stimuli in infants compared with 9% of voxels following the 512 mN stimuli in adults ( Figure 2 ) . In contrast , the 128 mN stimulus activated less than 1% of voxels in the adult brain . This demonstrates that the coverage and distribution of brain activity evoked by the 128 mN stimulus in infants was most similar to that evoked by the 512 mN stimulus in adults ( Figure 2 ) . This suggests that infants have increased sensitivity to nociceptive stimuli compared with adults , which is supported by previous data that show spinal nociceptive reflex withdrawal activity has greater amplitude and duration in infants compared with adults ( Andrews and Fitzgerald , 1999; Skljarevski and Ramadan , 2002; Cornelissen et al . , 2013 ) . These data strongly imply that the threshold for evoking widespread nociceptive brain activity in infants is substantially lower than in adults . It is , however , not known whether the increased brain activity observed at a lower threshold in the infants is due to increased peripheral drive , for example due to differences in skin thickness between the adult and infant populations , or due to differences in transduction or subsequent central processing of the nociceptive input . 10 . 7554/eLife . 06356 . 005Figure 2 . Noxious-evoked brain activity in response to the maximal presented stimulus in adults ( 512 mN ) and infants ( 128 mN ) . Red-yellow coloured areas represent active brain regions ( threshold z ≥ 2 . 3 with a corrected cluster significance level of p < 0 . 05 ) . An image of a midline sagittal brain slice ( right panel ) identifies the location of each example slice in the horizontal plane . ( A ) Adult activity is overlaid onto a standard T1 weighted MNI template and ( B ) infant activity is overlaid onto a standard T2 weighted neonatal template , corresponding to a 40-week gestation infant . DOI: http://dx . doi . org/10 . 7554/eLife . 06356 . 00510 . 7554/eLife . 06356 . 006Figure 2—figure supplement 1 . Relationship between percentage change in BOLD signal and stimulus intensity ( force ) in four example active brain regions in adult and infant participants . A: Contralateral insula; B: Contralateral primary somatosensory cortex ( S1 ) ; C: Anterior cingulate cortex ( ACC ) ; and D: Ipsilateral cerebellum . The crosses represent activity in individual participants . Red and blue lines are fitted regression lines and dashed lines show 95 % confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 06356 . 006 Noxious stimulation in infants did not evoke activity in the amygdala or orbitofrontal cortex ( OFC ) ( Table 1 ) , and in contrast to the adults , where activity was present across all divisions of bilateral insular cortices , activity in the anterior division was not present ( Figure 1 ) . A recent white matter tractography study of the adult brain shows that the anterior insula has dominant connections with the OFC ( Wiech et al . , 2014 ) . Based on many imaging studies spanning a range of stimuli and tasks , it is thought that activation in the anterior insula reflects the net evaluation of the affective impact of an impending situation . Similarly , the OFC is sensitive to stimuli with an emotional valence , however , it primarily responds to the reward value of the stimulus ( including negative value ) rather than its sensory features . Importantly , the OFC also encodes the anticipation of future outcomes , which makes it well suited for guiding subsequent decisions ( Kahnt et al . , 2010 ) . It is likely that the infants are too immature and inexperienced to evaluate and contextualise the nociceptive stimulus into a coordinated decision and response , which might account for the lack of activity within these regions . Similarly , in adults the amygdala is thought to attach emotional significance to the nociceptive inputs it receives , and to play a role in fear and anxiety ( Simons et al . , 2014 ) , which may reflect affective qualities that the newborn infant does not yet ascribe to the stimulus . In light of these observations , it is plausible that infants do not experience the full range of aversive qualities that adults associate with nociceptive input . Indeed , this hypothesis is supported by evidence from rat pups , which shows that avoidance behaviour in a fear-conditioning paradigm does not manifest until postnatal day 10 , and is associated with the enhancement of neural activity within the amygdala ( Sullivan et al . , 2000; Sullivan , 2001 ) . Nevertheless , the observation that brain structures involved in affective processing , such as the anterior cingulate cortex , are activated following noxious stimulation suggests that infants do have the capacity to experience an emotionally relevant context related to incoming sensory input . Indeed , in adults the modulation of pain-related activity in the anterior cingulate cortex closely parallels a selective change in perceived unpleasantness ( Rainville et al . , 1997 ) . 11 brain regions significantly encoded stimulus intensity in adults , whereas none of the active regions in infants exhibited significant intensity encoding ( Table 1 ) . Although the trend for intensity encoding in infants is clearly evident in some brain regions , these data suggest that infants do not discriminate stimulus intensity as well as adults ( Figure 2—figure supplement 1 ) . As only three stimulus intensities were applied to the infants it is plausible that if the intensity range were increased , significant intensity encoding may be observed . Nevertheless , when considering adult brain regions that did significantly intensity encode , three of the four highest ranked brain regions ( ranked according to the degree of intensity encoding , and identified as the contralateral temporal gyri , opercular cortex , and all divisions of the insular cortex ) , were ranked in the same order within the top three regions in infants , highlighting the remarkable similarity in how the immature infant brain and adult brain encode nociceptive information ( Table 1 ) . Intensity encoding has been reported following low intensity von Frey hair stimulation ( Williams et al . , 2015 ) . Inferences about the subjective experience of pain are highly speculative , whether based on brain imaging data , behavioural responses or other autonomic or physiological observations . In most adults , where the pain experience can be communicated verbally , it is not always necessary to rely on surrogate measures when attempting to quantify an individual's pain experience or when assessing the need for analgesic provision . However , where verbal report is not possible as in the infant population or in those who are cognitively impaired , reliance on surrogate measures is essential when making inferences about pain perception . As cortical activation is a fundamental requirement for an experience to be interpreted as painful , inferences based on patterns of brain activity may provide the most reliable surrogate measure of pain compared with alternative approaches based on behavioural and physiological indicators that may not be reliably linked to central sensory or emotional processing in the brain ( Oberlander et al . , 2002; Ranger et al . , 2007 ) . This does not , however , negate the importance of taking a multidimensional approach to infant pain assessment by considering measures of brain activity in the context of other well-characterised behavioural and physiological indicators . Indeed , some researchers have argued that reverse inference based on brain imaging results should be used merely as a guide to direct further enquiry rather than a direct means to interpret results ( Poldrack , 2008 ) . Nevertheless , it has been shown using multivariate pattern analysis that pain-related brain activity can be classified and discriminated from other psychological states , suggesting a neural state for pain perception that is distinct from other sensory modalities and affective experiences ( Yarkoni et al . , 2011; Wager et al . , 2013 ) . Although we cannot necessarily infer an infant's subjective experience based on a given pattern of brain activity , these results make certain conclusions more likely . The closer the pattern of brain activity mimics activity observed in adults—who can report their subjective experience—the stronger the inference . The patterns of brain activity observed in this study make it likely that the infant experience is similar to that described by adults . Pain is defined as an unpleasant sensory and emotional experience . This study provides the first demonstration that many of the brain regions that encode pain in adults are also active in full-term newborn infants within the first 7 days of life . This strongly supports the hypothesis that infants are able to experience both sensory and affective aspects of pain , and emphasizes the importance of effective clinical pain management .
10 healthy adults ( mean age = 28 . 3 years; range: 23–36 ) and 10 healthy term-born infants ( mean gestational age at time of study = 40 . 6 weeks; range: 38 . 6–42 . 7 ) participated in the study . Adult participants were members of staff or postgraduate students at The University of Oxford , and infants were recruited from the Maternity Unit at the John Radcliffe Hospital , Oxford . At the time of study all infants were less than 7 days old ( mean postnatal age = 3 days; range: 1–6 ) . Infant participants were eligible for inclusion in the study if they were healthy , had no history of neurological problems , born after 37 weeks gestation , self-ventilating in air and clinically stable at the time of study . Informed written consent and consent to publish the results were provided by adult participants or by the infant's parent before the study commenced . The study was approved by the National Research Ethics Service and the University of Oxford Central University Research Ethics Committee . The study conformed to the standards set by the Declaration of Helsinki and Good Clinical Practice guidelines . A member of the research team identified infants who were eligible for inclusion in the study shortly after birth . Prior to obtaining consent for infants to take part , parents were shown the experimental stimulators and given the opportunity to test the stimulators before they were applied to the infants . A full description of the MRI scanning environment was also provided . Parents of 113 infants were approached to take part in the study . 44 parents expressed an interest in the proposed research and 11 infants were recruited to the study . Parents were invited to stay with their infants during the study and in nearly all cases , one or both parents chose to do so . Parents were also informed that if their infant became restless while in the scanner , the study would be stopped . Recruitment success rate was highly dependent on infant availability during the pre-booked MRI scan slots . One study was stopped due to the baby being restless when placed on the MRI bed . In adults , 100% of the subjects ( 10 out of 10 ) who were approached to take part in the study gave their consent for the psychophysical and MRI aspects of the study . Functional magnetic resonance imaging ( fMRI ) of the brain was performed on all participants in response to acute noxious stimulation . On a second test occasion in the adults , the experimental protocol was repeated outside the scanner and the psychophysical data were recorded . During this session , participants were asked to verbally rate pain intensity using a numerical scale ( 0–10 ) and to describe the type of pain they experienced using the McGill pain questionnaire ( Melzack and Torgerson , 1971 ) . | Doctors long believed that infants do not feel pain the way that older children and adults do . Instead , they believed that the infants' responses to discomfort were reflexes . Based on these beliefs , it was a routine practice to perform surgery on infants without suitable pain relief up until the late 1980s . Even now , infants may receive less than ideal pain relief . For example , a review found that although newborns in intensive care units undergo 11 painful procedures per day on average , more than half of the babies received no pain medications . Some guidelines continue to emphasize that for infants cuddling and feeding are more important sources of comfort than pain-relieving drugs . There is growing support for better pain control for infants . Doctors and nurses now routinely observe behaviour and physiological responses—such as heart rate—to assess whether infants are experiencing pain . When an infant shows signs of pain , medical staff may give the infant sugar water or other interventions aimed at reducing their distress . However , recordings of brain activity suggest that infants may experience pain without exhibiting physical signs and that sugar water may reduce the behaviours associated with pain but not the pain itself . More objective measurements of infant pain would be useful , but to create such measurements scientists must first understand how infants experience pain . So Goksan et al . used a technique called functional magnetic resonance imaging ( fMRI ) to compare the brain responses of adults and newborns to the same stimulus—a sharp poke of the foot . The adults were also asked about the pain they experienced , and whether the infants pulled their foot away when poked was documented . The fMRI results revealed that pain increased activity in 20 regions in the adults' brains , and 18 of the same regions in the infants' brains . The brain regions activated in the infants' brains in response to a poke on the foot are involved in processing sensations and emotions . The two regions that did not activate in the infant brains—the amygdala and the orbitofrontal cortex—help individuals interpret the stimuli . Goksan et al . therefore conclude that infants experience pain in similar ways to adults , though they may not experience all the emotions that adults have when they are in pain . It is , therefore , important to give infants suitable pain relief during potentially painful procedures . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2015 | fMRI reveals neural activity overlap between adult and infant pain |
Alveolar rhabdomyosarcoma is a pediatric soft-tissue sarcoma caused by PAX3/7-FOXO1 fusion oncogenes and is characterized by impaired skeletal muscle development . We developed human PAX3-FOXO1 -driven zebrafish models of tumorigenesis and found that PAX3-FOXO1 exhibits discrete cell lineage susceptibility and transformation . Tumors developed by 1 . 6–19 months and were primitive neuroectodermal tumors or rhabdomyosarcoma . We applied this PAX3-FOXO1 transgenic zebrafish model to study how PAX3-FOXO1 leverages early developmental pathways for oncogenesis and found that her3 is a unique target . Ectopic expression of the her3 human ortholog , HES3 , inhibits myogenesis in zebrafish and mammalian cells , recapitulating the arrested muscle development characteristic of rhabdomyosarcoma . In patients , HES3 is overexpressed in fusion-positive versus fusion-negative tumors . Finally , HES3 overexpression is associated with reduced survival in patients in the context of the fusion . Our novel zebrafish rhabdomyosarcoma model identifies a new PAX3-FOXO1 target , her3/HES3 , that contributes to impaired myogenic differentiation and has prognostic significance in human disease .
Rhabdomyosarcoma ( RMS ) presents as a solid tumor that displays characteristics of primitive skeletal muscle ( Parham and Ellison , 2006 ) . There are two major histological subtypes , embryonal ( ERMS ) and alveolar rhabdomyosarcoma ( ARMS ) , with ARMS being more aggressive and particularly prone to metastasis ( Williamson et al . , 2010; Skapek et al . , 2013 ) . At the genomic level , ERMS is characterized by frequent alterations affecting RAS signaling , although this still represents a minority of the cases ( Stratton et al . , 1989; Langenau et al . , 2007; Martinelli et al . , 2009 ) . The defining oncogenic event in ARMS is a t ( 1;13 ) or t ( 2;13 ) chromosomal translocation in which the PAX7 or PAX3 DNA-binding domain , respectively , is fused to the FOXO1 transactivation domain to create a PAX3/7-FOXO1 chimeric oncogene ( Barr et al . , 1993; Galili et al . , 1993; Shapiro et al . , 1993; Davis et al . , 1994 ) . The PAX3-FOXO1 fusion is the most prevalent fusion in the disease , and functions as an aberrant transcription factor that is expressed in the nucleus and deregulates gene expression signatures ( del Peso et al . , 1999; Fredericks et al . , 1995; Barber et al . , 2002; Khan et al . , 1999 ) . This activity is the predominant cellular insult required for transformation . The PAX3/7-FOXO1 oncogenes remain intractable to therapeutic targeting , impeding the development of effective precision medicine therapies . Fusions are notoriously difficult to model in animals , hence the limited availability of vertebrate animal models of this disease . Furthermore , there is a narrow understanding of the cellular origin of RMS , making it difficult to define the expression pattern required for tumorigenesis ( Hettmer and Wagers , 2010 ) . Zebrafish are a complementary model system that can address these genetic and cellular issues . Advantages of zebrafish systems are two-fold: ( 1 ) they provide insight into the underlying biology of how cancer genes behave in a complex environment and ( 2 ) provide a platform for translational drug discovery efforts . Such strengths are intrinsically important for translational models of pediatric disease . Here , we describe human PAX3-FOXO1-driven zebrafish models . We implemented our zebrafish models to provide the appropriate context to understand the behavior of PAX3-FOXO1 during development and tumorigenesis . The tumor presentation spectrum identified three distinct cellular contexts that are susceptible to transformation , generating insight into basic mechanisms of PAX3-FOXO1 tumorigenesis and human rhabdomyosarcoma . By applying our zebrafish RMS model , we found a novel PAX3-FOXO1 target , her3/HES3 . HES3 is a member of the HES family of basic helix-loop-helix transcription factors , which function as direct or indirect transcriptional activators or repressors , thus regulating gene expression and epigenetic identity ( Kageyama et al . , 2007 ) . HES3 is expressed in the developing brain and inhibits differentiation of neural stem cells ( Hatakeyama et al . , 2004 ) . In cancer , HES3 is expressed in glioblastoma cell culture , and co-localizes with additional markers of stemness in the mouse brain ( Park et al . , 2013; Poser et al . , 2013; Katoh and Katoh , 2007 ) . However , its role as a cooperating gene in PAX3/7-FOXO1 fusion-positive rhabdomyosarcoma has never been described . Taken together , this model represents a novel strategy to identify new targets and biomarkers in the context of human disease and contributes to our understanding of RMS biology by defining the earliest tumor initiation events .
To develop a new vertebrate model of PAX3-FOXO1-dependent tumorigenesis , we performed a survey of promoters driving PAX3-FOXO1 expression . These promoters represent ubiquitous ( beta actin , CMV , ubiquitin ) , hematologic ( fli1 ) , muscle ( unc503 ) , neural crest ( mitfa ) expression , and a gene trap approach . Selected promoters were chosen because of their relevance in the disease as implicated lineages for the cell of origin or for their capacity to drive PAX3-FOXO1 at high levels of expression . Further , all promoters had been previously validated as functional in zebrafish , with data from our group underscoring the beta actin promoter as a successful expression system for EWS-FLI1 transgenic models of Ewing sarcoma ( Leacock et al . , 2012 ) . Human PAX3-FOXO1 was integrated into the zebrafish genome utilizing the Tol2 transposon-based system and microinjection in a stable mosaic manner . Genomic integration and transgene expression were tracked using a GFP or mCherry fluorescent protein linked to the coding sequence of PAX3-FOXO1 with a viral 2A sequence ( Figure 1A–D ) . This results in equimolar expression of both genes on the same mRNA transcript , yet translation as independent proteins . Zebrafish were monitored for up to 19 months . Using this strategy , we identified fusion-oncogene driven tumors first based on gross morphology and then by screening suspected tumors for fluorescent signal . Human PAX3-FOXO1 under the control of different promoters had unique survival and transforming properties suggesting cell-of-origin specificity to oncogenesis . Tested promoters had a negative impact on early survival ( up to 30 days of life ) , but to different extents ( Figure 1—figure supplement 1 ) . Only a subset of the promoters that were tested induced tumor formation , including the beta actin ( 9% of injected zebrafish ) , CMV ( 16% of injected zebrafish ) , and ubiquitin ( 1 . 8% of injected zebrafish ) promoters ( Supplementary file 1 ) . The fli1 , unc503 , mitfa , and gene trap approach driving PAX3-FOXO1 were not transforming ( Supplementary file 2 ) . The three promoters ( beta actin , CMV , ubiquitin ) that induced PAX3-FOXO1 transformation in zebrafish had varied requirements for tp53M214K as a sensitizing mutation . The beta actin promoter driving the expression of PAX3-FOXO1 is tumorigenic in zebrafish , both in a wildtype and a p53-deficient genetic background . In tp53 wild-type zebrafish , BetaActin-PAX3FOXO1 tumors began developing at 3 months of age in 5% of injected zebrafish in the head . Diagnosis of tumors was made by hematoxylin and eosin staining and evaluation by light microscopy . Based on this analysis , the majority of beta-actin-driven PAX3-FOXO1 zebrafish tumors were consistent with primitive neuroectodermal tumors ( Figure 1B ) . However , beta-actin-driven PAX3-FOXO1 injected into the tp53M214K/M214K mutant resulted in one undifferentiated sarcoma after 412 days . The CMV promoter restricting PAX3-FOXO1 expression results in RMS tumors that present in the skeletal muscle of the back . Histological analysis of the zebrafish tumors was consistent with human RMS ( Figure 1C ) . This was true only in the context of the tp53M214K mutation , indicating that the tp53 mutation is sensitizing to and favors RMS . These findings are concordant with the Pax3-Foxo1 RMS mouse model , in which Tp53 deletion , or other secondary cooperating mutations are required for RMS development ( Keller et al . , 2004 ) . Further , patients with Li Fraumeni syndrome develop RMS as part of the spectrum of the human disease ( Li and Fraumeni , 1969 ) . The ubiquitin promoter driving PAX3-FOXO1 generated one undifferentiated sarcoma by 378 days of age in a wild-type genetic background ( Figure 1D ) . We performed RNA-seq on zebrafish tumors derived from all three promoters ( beta actin , CMV , ubiquitin ) , and detected PAX3 and FOXO1 junction spanning reads indicating that human PAX3-FOXO1 is expressed in fluorescent tumors ( Figure 1E , Figure 1—figure supplement 2 ) . These data demonstrate that the human fusion-oncogene is transforming in zebrafish when expressed in a variety of cellular contexts and that some , but not all , cellular identities are susceptible to transformation . The beta actin , CMV , and ubiquitin promoter restricting PAX3-FOXO1 expression have unique latency , penetrance , and spectrum of tumor development ( Figure 1F–G; Supplementary file 1 ) . Tumor incidence curves are shown for the beta actin promoter driving PAX3-FOXO1 in a wild-type background , and for the CMV promoter driving PAX3-FOXO1 in the tp53M214K/M214K mutant background ( Figure 1F–G ) . Rhabdomyosarcoma only develops in the tp53M214K background , predominantly with the CMV promoter as compared to the beta actin promoter . Since human PAX3-FOXO1 is active in zebrafish and produces discrete functional readouts , we next applied our model to identify PAX3-FOXO1 targets that are engaged during development to induce tumorigenesis . The PAX3-FOXO1 fusion-oncogene functions as an aberrant transcription factor due to the PAX3 DNA-binding domain being linked to the FOXO1 transactivation domain ( Linardic , 2008; Cao et al . , 2010 ) . To identify activity unique to PAX3-FOXO1 versus the normal PAX3 transcription factor , we performed a detailed analysis of the functional effects of their expression during zebrafish development . We utilized a mosaic model system , in which plasmid DNAs containing the beta actin promoter driving GFP , or GFP-tagged PAX3 or PAX3-FOXO1 were injected into developing zebrafish using the Tol2 transposon system ( Figure 2A ) . The expression of each construct was tracked using a viral 2A linked fluorescent protein , and mosaic integration of the construct can be observed with fluorescence by 24 hr post-injection ( Figure 2B ) . During the first three days of life , PAX3-FOXO1 injected zebrafish exhibited reduced survival as compared to PAX3 injected zebrafish ( Figure 2C ) . Furthermore , PAX3-FOXO1-induced unique embryonic phenotypes , including cyclopia , which was present in 30% of injected embryos . To address if cyclopia was due to PAX3-FOXO1 expression , we co-injected GFP2A-PAX3FOXO1 with a morpholino that targets the expression construct . In this scenario , the morpholino binds the start codon of GFP and inhibits the ribosome’s association with the mRNA transcript , thus inhibiting the translation of the GFP-viral2A-PAX3FOXO1 mRNA . Morpholino activity was tracked by quantifying the reduced number of GFP positive embryos after injection with titrating doses of morpholino ( Figure 2D ) . Co-injection of PAX3-FOXO1 and a morpholino that knocks-down GFP-PAX3-FOXO1 expression eliminated the cyclopia phenotype ( Figure 2D ) . This is significantly different from the uninjected , GFP injected , and the PAX3 injected groups , suggesting that PAX3-FOXO1 is regulating distinct developmental phenotypes . This discrepancy between PAX3-FOXO1 and PAX3 tolerance was evident at the level of individual cells . To demonstrate this , embryos were injected with beta-actin-driven constructs containing: GFP only , GFP-tagged PAX3 , or GFP-tagged PAX3-FOXO1 . Injected zebrafish embryos were allowed to develop for 24 hr , and then dissociated to single-cell suspensions . Fluorescent activated cell sorting ( FACS ) showed that PAX3-FOXO1 expression significantly reduced the number of GFP + cells at 24 hr post-fertilization as compared to PAX3 ( Figure 2E ) . This reduced viability of PAX3-FOXO1 + zebrafish was consistent in injected embryos that were raised to adulthood and screened at 3 months of age for detectable fluorescence . In BetaActin-mCherry or GFP fluorescent controls , fluorescence was observed in over 70% of adult zebrafish . In PAX3-injected zebrafish , mCherry fluorescence was observed in 32% of adults . We found that only 3% of adult zebrafish had detectable GFP-PAX3FOXO1 expression with resulting tumor development or asymmetric skeletal muscle ( Figure 2F–G ) . This was not true for BetaActin-GFP or BetaActin-GFP-PAX3 , which do not generate tumors or affect normal development or survival . One explanation for this disappearance of PAX3-FOXO1 + cells during the course of development is that PAX3-FOXO1-injected zebrafish have a significant increase in the number of cells undergoing apoptosis . At 24 hr post-fertilization , TUNEL staining was performed on embryos injected with either GFP-PAX3 or GFP-PAX3FOXO1 . These same embryos were counter-stained for GFP to detect the expression levels of their respective transgenes ( Figure 2H ) . PAX3-FOXO1-injected embryos had an increase in the number of TUNEL-positive cells undergoing apoptosis as compared to GFP controls and PAX3 in wild-type zebrafish ( Figure 2I ) . Given these data and the propensity to generate RMS we mechanistically evaluated the contribution of tp53 mutations to early PAX3-FOXO1-induced apoptosis ( Figure 1C and G; Figure 2—figure supplement 1A ) . We found that tp53M214K mutant zebrafish embryos are unable to invoke an appropriate response to PAX3-FOXO1 + cells , resulting in a gross reduction in survival by 3 days of age ( Figure 2—figure supplement 1B ) . Further , injected CMV-PAX3FOXO1 + cells are more abundant in tp53M214K injected mutant zebrafish as compared to their wild-type counterparts by 28 hr post fertilization ( Figure 2—figure supplement 1C–D ) . This significant increase in the number of GFP-PAX3FOXO1 + cells is coupled with a reduction in the ratio of TUNEL-positive cells , indicating that tp53M214K mutant zebrafish provide a susceptible environment for PAX3-FOXO1 + cellular persistence and ultimately tumor development ( Figure 2—figure supplement 1E–G; Figure 2—figure supplement 2 ) . This dichotomy of cellular tolerance and apoptosis is not true for wildtype or tp53M214K mutant zebrafish injected with GFP controls , or GFP-PAX3 ( Figure 2—figure supplement 1B–G ) . Therefore , at both an organismal and individual cell level , ectopic expression of PAX3-FOXO1 was less tolerated than the normal PAX3 gene . Understanding the exact mechanisms for cell tolerance , and defining these discrepancies , could identify the earliest events in RMS transformation or alternatively tumor suppressive responses . Given that PAX3-FOXO1 is uniquely tumorigenic and has different embryonic effects than PAX3 this warranted a more thorough study of the specific developmental pathways and targets that were mediating these early outcomes . To accomplish this , we injected zebrafish embryos with DNA expression constructs in which the beta actin promoter drives either a GFP control , or GFP-tagged PAX3 and PAX3-FOXO1 , and allowed the embryos to develop for 24 hr . The GFP + cell population from zebrafish embryos was then FACS sorted , total RNA isolated , and microarrays were used to determine differential gene expression signatures . After comparing differentially expressed genes versus the GFP injected control for PAX3 and PAX3-FOXO1 , the most highly enriched Gene Ontology ( GO ) terms indicated that PAX3 and PAX3-FOXO1 act as transcription factors ( Figure 3A ) . This was expected given their well-documented roles in the literature and suggested that the mammalian forms are active in zebrafish systems . Furthermore , DAVID analysis of the enriched genes during normal development suggested that these sorted cell populations identify different subsets of zebrafish embryonic tissues . PAX3-positive cells were most indicative of the neural plate , neuroectoderm , or ectoderm , whereas PAX3-FOXO1 was indicative of a somite , segmental plate , or optic vesicle and immature eye ( Figure 3B ) . Even though PAX3 and PAX3-FOXO1 were injected under the control of the same promoter , cells expressing these genes represent either ( 1 ) different cell populations or ( 2 ) identical cell populations that have differential transcriptional responses as early as 24 hr post-fertilization . We then performed a principal component analysis ( PCA ) and found that PAX3 and PAX3-FOXO1 expression changes are distinct and show statistical significance in a developmental context ( Figure 3C ) . This does not contradict the fact that there are a large number of expression changes that are similar between these two transcription factors . Therefore , unique developmental and oncogenic targets of PAX3-FOXO1 could be ascertained by directly comparing its activity to that of PAX3 . To determine what these PAX3-FOXO1 developmental targets might be , we looked at the overlap of induced genes for PAX3 or PAX3-FOXO1 versus the GFP controls . We found that PAX3 and PAX3-FOXO1 up-regulated 118 shared genes , whereas PAX3-FOXO1 uniquely induced 529 genes . These 529 genes were of interest because of their association with the fusion-oncogene and potentially tumorigenesis . Of these 529 genes , those without a human ortholog were eliminated from further analysis due to our interest in human disease . This produced a list of primarily developmental transcription factors and homeobox genes , including her3 , pax3b , dbx1a , and lhx1b . The most highly induced gene unique to PAX3-FOXO1 expression was her3 , with a 21-fold increase as compared to the GFP control ( Figure 3D ) . We validated that her3 expression was induced by the human PAX3-FOXO1 fusion-oncogene in zebrafish by injection of GFP control , PAX3 , and PAX3-FOXO1 . Embryos were allowed to develop and then total RNA was isolated from an embryo pool , cDNA transcribed , and a qRT-PCR performed for her3 expression levels . This analysis indicated that her3 induction was unique to PAX3-FOXO1 ( Figure 3E ) . To verify that the up-regulation of her3 was dependent on PAX3-FOXO1 , we co-injected the BetaActin-PAX3FOXO1 construct with a morpholino that inhibits its early expression by targeting the GFP2A-PAX3FOXO1 transcript . At 24 hr of age , PAX3-FOXO1 mRNA levels were measured by qRT-PCR , and co-injection of the morpholino significantly knocked down PAX3-FOXO1 expression ( Figure 3F ) . A qRT-PCR was performed on these same samples to evaluate her3 expression levels , and the results replicated those of PAX3-FOXO1 ( Figure 3G ) . These data suggest that her3 is a novel target of PAX3-FOXO1 in vivo , that her3 is not induced by PAX3 , and that her3 expression is dependent on the PAX3-FOXO1 oncogene . A hallmark of rhabdomyosarcoma is that tumors express early markers along the skeletal muscle lineage such as MYOD1 , MYOG and Desmin; however , these tumors fail to terminally differentiate indicating a developmental arrest ( Parham and Ellison , 2006; Saab et al . , 2011 ) . To determine if this was recapitulated in our zebrafish model , we performed a mosaic co-injection strategy with the beta actin promoter driving GFP-mCherry controls , human HES3 alone , human PAX3-FOXO1 alone , or a co-injection of PAX3-FOXO1 and HES3 ( Figure 4A ) . Co-injection of transgenes linked to highly expressed promoters generated mosaic integration that overlaid with muscle proteins such as myosin by 24 hr post-fertilization ( Figure 4B ) . Based on this observation , zebrafish were injected and allowed to develop for 24 hr , and then fluorescent embryos harvested and total RNA isolated and reverse transcribed for qRT-PCR analysis of muscle marker genes . This survey of temporally stereotyped muscle markers included early markers myod and myog and terminal muscle differentiation markers myl1 and myhz2 . We found that PAX3-FOXO1 did not affect the mRNA levels of myod or myog; however , by 24hpf there was a significant reduction in the expression of myl1 and myhz2 . Surprisingly , overexpression of HES3 alone had a similar effect of inhibiting terminal muscle differentiation , indicating that her3/HES3 contribute to maintaining a more primitive cellular state . Together , PAX3-FOXO1 and HES3 had no significant additive effect , suggesting they are functioning in a linear pathway with her3/HES3 being downstream of PAX3-FOXO1 ( Figure 4C ) . Hence , PAX3-FOXO1 and HES3 inhibit differentiation in vivo in a vertebrate system via a shared mechanism . The majority of the PAX3-FOXO1 and HES3 co-injected cells overlaid indicating that this mosaic system is a powerful model to study the interactions of two genes within the same cell ( Figure 4A and D ) . Given this capacity , we next determined if HES3 modified the behavior of GFP-PAX3FOXO1 + cells during embryogenesis . Zebrafish embryos were co-injected at the single-cell stage with two independent plasmids that were dually integrated into the genome using the Tol2-transposon-based system . The conditions included the beta actin promoter driving: ( 1 ) mCherry and GFP , ( 2 ) GFP and mCherry-HES3 , ( 3 ) mCherry and GFP-PAX3FOXO1 , and ( 4 ) GFP-PAX3FOXO1 and mCherry-HES3 . Both GFP and mCherry images were taken using the same settings at both 24 hr post-fertilization and 72 hr post-fertilization to evaluate the capacity of the cells to survive ( Figure 4D ) . The number of GFP and mCherry positive pixels were calculated independently and plotted for both timepoints ( Figure 4E–F ) . Co-injection of HES3 allowed PAX3-FOXO1 cells to survive and/or proliferate , with a greater number of GFP-positive pixels and thus PAX3-FOXO1 positive cells persisting by 72hpf ( Figure 4E ) . This phenomenon was unique to the co-injection of HES3 + PAX3FOXO1 , as the number of pixels that were positive between 24hpf and 72hpf for mCherry + GFP-PAX3FOXO1 was not significantly different . Additionally , the number of mCherry positive pixels from 24hpf to 72hpf was insignificantly different for mCherry-HES3 + GFP-PAX3FOXO1 indicating that this observation was not globally applicable , but unique to the persistence of PAX3-FOXO1 ( Figure 4F ) . These results suggest that HES3 promotes a more tolerant cellular environment allowing for survival with expression of the highly toxic PAX3-FOXO1 . One hypothesis is that the co-expression of HES3 allows for PAX3-FOXO1 expressing cells to circumvent apoptosis . By implementing our CMV injected mosaic zebrafish models , we found that HES3 overexpression in zebrafish allows for PAX3-FOXO1+ cells to either inappropriately persist or divide , resulting in an increased number of observable GFP-PAX3FOXO1 + pixels by 24 hr post-fertilization ( Figure 4—figure supplement 1A–C ) . We performed TUNEL assays on these same zebrafish embryos and found that HES3 alone does not induce apoptosis . Further , there is no significant difference in the normalized ratio of TUNEL cells for PAX3-FOXO1-injected embryos as compared to PAX3-FOXO1 + HES3 ( Figure 4—figure supplement 1D–E ) . These data indicate that PAX3-FOXO1 + HES3 cooperation facilitates PAX3-FOXO1 + cells to persist during embryogenesis independently of apoptosis inhibition at this timepoint ( Figure 4—figure supplement 2 ) . Impaired skeletal muscle differentiation is a signature feature of human RMS tumors . To evaluate how HES3 is contributing to oncogenesis in fusion-positive RMS , we determined its capacity to inhibit differentiation in mammalian cell culture ( Figure 5A ) . Stable mouse C2C12 myoblast cell lines were generated by transfection and selection of CMV-HES3 or CMV-Empty control , and the resulting cell lines were evaluated for overexpression of HES3 by qRT-PCR ( Figure 5B ) . Myoblasts were viable with HES3 overexpression and were next tested to determine if there was a temporal or functional difference in their myogenic capacity . Cells were seeded onto porcine gelatin coated plates and collected at one and 6 days post plating after being exposed to low-serum concentrations and supplemented with insulin to promote fusion . The expression levels of selected genes were evaluated to represent the spectrum of myogenic differentiation , including MyoD , MyoG , Myl1 , and Myh1 ( Figure 5A ) . By day 6 of differentiation , MyoD , MyoG , Myl1 , and Myh1 exhibited significant inhibition of expression in C2C12-HES3 overexpressing cells as compared to the C2C12-Empty controls . Moreover , Myl1 and Myh1 , which represent terminal muscle differentiation , were the most affected by HES3 overexpression at fusion day 6 , with a 2 . 9–3 . 2 fold decrease in expression levels , as compared to a 1 . 8–1 . 9 fold decrease observed for MyoD and MyoG ( Figure 5B ) . A decrease in the mRNA expression of markers of muscle identity translated to a reduction in the functional capacity to fuse into multi-nucleated myofibers . C2C12-HES3 overexpressing cells and C2C12-Empty controls were differentiated in low-serum conditions , and immunofluorescence performed to determine protein expression and localization of myosin heavy chain ( Figure 5C ) . Both C2C12-Empty and C2C12-HES3 cells exhibited the capacity to fuse into multi-nucleated myotubes that express myosin heavy chain upon terminal differentiation . However , there was a stark difference in the success of the differentiation . After analyzing the percentage of myogenic nuclei , or the nuclei within a myofiber , there was 2 . 3X more myogenic nuclei in C2C12-Empty as compared to the C2C12-HES3 overexpressing cells ( Figure 5D ) . Furthermore , the fusion index ( myogenic nuclei with > 3 nuclei present per myofiber ) indicated a significant reduction in the fusion capacity of HES3 overexpressing cells ( Figure 5E ) . HES3 overexpression also modified the kinetics of C2C12 differentiation , with myosin protein being undetectable at day 4 of fusion in C2C12-HES3 overexpressing cells and detectable in C2C12 controls ( Figure 5—figure supplement 1A–C ) . These contrasting fusion capacities cannot be fully attributed to a dominant negative effect on MyoD and inhibition of MyoD expression during fusion initiation . In fact , there were comparable levels of MyoD protein at day 3 of C2C12 fusion . By day 9 there is a trend towards decreased MyoD in HES3 overexpressing C2C12 myotubes ( Figure 5—figure supplement 2A–C ) . Hence , HES3 is inhibiting the differentiation process and promoting a more primitive cellular state in mammalian systems . To determine if HES3 overexpression confers properties indicative of tumorigenic capacity in cell culture systems , we analyzed HES3’s impact on C2C12 mouse myoblasts and human rhabdomyosarcoma cells . We focused on the human cell line , Rh30 , which is derived from a bone marrow metastasis of ARMS and contains the PAX3-FOXO1 fusion ( Hinson et al . , 2013 ) . Stable Rh30 cell lines were generated by transfection and selection of either the CMV-Empty or CMV-HES3 expression construct . A qRT-PCR verified HES3 overexpression and that PAX3-FOXO1 was present ( Figure 6A–B ) . This overexpression strategy was pursued because of the low levels of baseline HES3 in Rh30 . We examined a panel of rhabdomyosarcoma cell lines and have observed low HES3 levels by RNAseq , and no detectable endogenous HES3 protein expression with commercially available antibodies ( data not shown ) . To determine if HES3 modified the proliferation kinetics of Rh30 , cells were plated at a low density and timepoints evaluated from days 1 to 6 . We found that in this context HES3 accelerated the cellular accumulation of Rh30 ( Figure 6C ) . Additionally , overexpression of HES3 modified the expression of a panel of genes implicated in metastasis and more aggressive disease . Notably , there was > 300 fold down-regulation of Missing in Metastasis ( MTSS1 ) , and an up-regulation of matrix metallopeptidases 3 and 9 ( MMP3 and MMP9 ) ( Figure 6D ) . Complementary results were obtained when C2C12 cells were transfected with either an empty control vector or CMV-HES3 . Stable C2C12-HES3 cells overexpressed HES3 as compared to the C2C12-Empty control ( Figure 6E ) . HES3 was then evaluated for its ability to modulate cellular accumulation . HES3 significantly accelerated the growth kinetics of C2C12 cells when plated at sub-confluent levels over the course of 6 days ( Figure 6F ) . These same cells were plated in suspension in soft agar and cultured for one month to determine the colony formation capacity . HES3 overexpression significantly increased the number of colonies that formed ( Figure 6G ) . This is complemented by data indicating that HES3 alters the expression of genes that are PAX3-FOXO1 direct targets or whose expression is modified by PAX3-FOXO1; including , MyoG , Igf2 , and Hoxc6 ( Figure 6H ) ( Khan et al . , 1999; Lagha et al . , 2010 ) . Potentially , HES3 is acting as a transcriptional activator and repressor , consistent with its role in normal development . We evaluated the clinical significance of HES3 by determining HES3 expression levels in human RMS tumors . We analyzed RNA-Seq and HuEx array data from three independent RMS tumor cohorts , together representing the largest collection of over 200 sequenced RMS tumors ( Chen et al . , 2013; Shern et al . , 2014 ) . We found that HES3 had higher expression levels in fusion-positive RMS as compared to fusion-negative RMS ( Figure 7A–C ) . Significantly , in RMS patients , the overexpression of HES3 is associated with reduced overall survival in PAX3/7-FOXO1 fusion-positive RMS patients , but not in fusion-negative RMS patients ( Figure 7D ) . RMS tumors from Shern et al . ( 2014 ) were evaluated for HES3 expression levels , and then differentially expressed genes expounded between HES3 high and low tumors to identify molecular targets with potential translational benefit . A subset of clinically relevant kinases and molecular targets was identified as being concordantly overexpressed with HES3; namely , FGFR4 , ALK , and PARP1 ( Figure 7E ) . Taken together , these data are consistent with our results showing that her3/HES3 is a PAX3-FOXO1 target , and that HES3 contributes to a more aggressive phenotype in rhabdomyosarcoma .
Despite the identification of the primary oncogenic driver PAX3/7-FOXO1 over 20 years ago , there have been few changes to the primary treatment and prognosis of the disease ( Barr et al . , 1993; Galili et al . , 1993; Shapiro et al . , 1993; Davis et al . , 1994 ) . Developing complementary ARMS animal models will aid in the understanding of disease biology and identify new therapeutic avenues . Zebrafish ARMS models provide a relevant vertebrate developmental context with experimental advantages such as lineage tracing of the ARMS cell of origin , identifying PAX3-FOXO1 transcriptional targets , and providing a platform for a high-throughput drug screens . Here , we describe a genetic zebrafish model of human disease , in which the human fusion-oncogene , PAX3-FOXO1 , is mosaically integrated into the zebrafish genome . We found that the human fusion is oncogenic in zebrafish under the control of the beta actin , CMV , and ubiquitin promoters , resulting in PNETs , RMS , and histologically undifferentiated sarcoma . Surprisingly , PAX3-FOXO1 transformation capacity is not limited to myogenic cells , reflecting the lineage specific control of PAX3 which is relevant in the developing brain and muscle . Specific lineages that have been postulated as possible RMS tissues of origin , such as endothelial , neural crest and differentiated muscle ( represented here by the fli1 , mitfa , and unc503 promoters ) did not develop tumors , highlighting the requirement for high-expressing ubiquitous promoters in this model . Whereas in patient genomic data , PAX7-FOXO1 undergoes copy number gain to increase its expression levels , this is not the case for PAX3-FOXO1 . Likely , PAX3-FOXO1 must increase its expression level for transformation via transcriptional upregulation ( Barr et al . , 1996 ) . To achieve these higher levels of transcription and transformation , we opted to utilize high-level promoters . This threshold was likely not reached with the fli1 , mitfa , and unc503 endogenous promoters , or , the lineages and/or temporal kinetic of transgene expression are not relevant to the disease . In our zebrafish model , the relevant PAX3-FOXO1 RMS cell type is supportive of CMV-mediated gene expression and requires a tp53 mutation . These observations are substantiated by the mouse Pax3-Foxo1 RMS model , in which deletion of Tp53 or Cdkn2a is required for tumor development ( Keller et al . , 2004 ) . In our zebrafish ARMS model , tp53 mutations allow for PAX3-FOXO1 + cells to survive during embryogenesis by inhibiting apoptosis resulting in later tumor formation ( Figure 2—figure supplements 1–2 ) . This may represent a unique necessity in animal models as both fusion-positive RMS vertebrate models of the disease require that tp53 is mutant or lost . In zebrafish , beta-actin-driven PAX3-FOXO1 is sufficient for tumorigenesis , providing a cleaner background for its functional assessment and greater conservation with the genetics of the human disease . One possible explanation is that zebrafish do not have p14 ARF orthologs , making it an unique model to understand the function of the fusion alone . Therefore , we pursued embryonic analyses of PAX3-FOXO1 function using both the beta actin and CMV promoters due to their tumorigenic effects and unique genetic requirements . We found by performing analogous studies with BetaActin-PAX3FOXO1 and CMV-PAX3FOXO1 that these expression constructs behaved comparably in an embryonic environment ( Figure 2 , Figure 2—figure supplements 1–2 , Figure 4 , Figure 4—figure supplements 1–2 ) . Focusing on robust and conserved early oncogenic mechanisms that are consistent across multiple promoter models limited the possibility of artifactual observations . By studying PAX3-FOXO1 signaling in a complex developmental context , we identified a novel fusion-positive RMS biomarker , her3/HES3 , that predicts more aggressive disease . her3 is a transcription factor that contains a DNA-binding domain , an orange domain , and a tetrapeptide motif at the C terminus ( Hans et al . , 2004 ) . In zebrafish and mouse models , her3 is expressed in the neuroepithelial stem cells of the neural plate and inhibits their premature differentiation ( Hatakeyama et al . , 2004 ) . The human ortholog of her3 , HES3 , is a member of the basic helix-loop-helix ( bHLH ) transcription factor family , which includes MYOD1 and MYOG , that are known to be transcriptionally targeted by PAX3-FOXO1 ( 14 , 33 ) . HES3 acts as both a transcriptional repressor and activator , by indirectly or directly influencing transcription depending on its sub-cellular localization . PAX3-FOXO1 regulation of HES3 is likely indirect . Chip-seq data from rhabdomyosarcoma cell culture indicates that PAX3-FOXO1 does not bind the HES3 promoter; however , these results may differ in a dynamic developmental context ( Cao et al . , 2010 ) . Copy number analysis of fusion-positive patient RMS tumors finds no amplification of HES3 , indicating up-regulation is at the transcriptional level . Our data suggest that PAX3-FOXO1 and HES3 function linearly , with HES3 being downstream of PAX3-FOXO1 . The exact mechanisms of regulation will be addressed in future studies . HES3 contributes to more aggressive disease in patients when overexpressed in the context of the PAX3-FOXO1 fusion , likely by a multi-variate platform of: ( 1 ) inhibiting terminal muscle differentiation , ( 2 ) modifying cellular plasticity to make cells that acquire the PAX3-FOXO1 fusion more tolerant of its expression , and ( 3 ) inducing additional transcriptional changes . PAX3-FOXO1 induction of HES3 expression in a mesodermal lineage is ectopic , and HES3 plays no known role in normal somitogenesis or myogenesis . We have found that mis-expression of HES3 during embryonic development inhibits muscle differentiation . We note that HES3 allows for PAX3-FOXO1 cells to inappropriately persist during embryogenesis independently of inhibiting apoptosis . In human fusion-positive RMS , which typically lack TP53 mutations ( Shern et al . , 2014 ) , HES3 overexpression could represent an alternative mechanism to facilitate PAX3-FOXO1 + cell persistence and tumor initiation events . This role for HES3 promoting PAX3-FOXO1 + embryonic persistence is suggestive of HES3’s capacity in normal development to alter cellular plasticity . PAX3-FOXO1 in combination with HES3 could be challenging the epigenetic identity of mutated cells . Finally , we provide evidence that HES3 may induce a more aggressive and/or metastatic program in PAX3-FOXO1 + patient RMS , including the down-regulation of MTSS1 and up-regulation of MMP3 and MMP9 ( 34 , 35 ) . Previous studies indicate that MTSS1 is a target of the PAX3-FOXO1 fusion ( Ebauer et al . , 2007 ) ; therefore , the modification of MTSS1 expression , and additional genes integral in cell invasion could occur in part through PAX3-FOXO1 induction of HES3 . Most significantly , stratifying patient tumor data between HES3 high and low expressing RMS elucidated translational molecular targets with immediate potential applications in the clinic such as co-expression of FGFR4 , ALK , and PARP1 . Overall , this new PAX3-FOXO1 zebrafish model of rhabdomyosarcoma identifies her3/HES3 as a mediator of tumorigenesis . Further application of this model determined mechanisms of PAX3FOXO1/HES3 cooperation that result in more aggressive disease . Given the strengths of the zebrafish system , we can now apply it for drug screening and repurposing efforts . We envision the utility of our zebrafish PAX3-FOXO1 RMS model being applied for transplant experiments , where generated tumors can be propagated and grafted into adult zebrafish and in developing zebrafish embryos for further study . Zebrafish embryos containing PAX3-FOXO1 + tumor cells can be exposed to single or combinations of small molecules that effectively couple efficacy for tumor elimination with toxicity data in a vertebrate whole animal model . Additionally , zebrafish cancer models are a complementary approach to ongoing sequencing efforts of patient tumors . The functional significance of cooperating genetic mutations can be determined quickly and robustly . In our study , we find that human PAX3-FOXO1 and HES3 are active in zebrafish , and that biomarkers identified in this system are relevant in the human disease . We anticipate that additional application of this new PAX3-FOXO1 driven tumor model will elucidate novel tumorigenic mechanisms and avenues for treatment .
Danio rerio were maintained in an Aquaneering aquatics facility according to industry standards . Vertebrate animal work is accredited by AALAC and overseen by the UT Southwestern IACUC committee . AB , WIK , TL , and AB/TL were the wild-type lines used and were obtained from the Zebrafish International Resource Center ( https://zebrafish . org ) . The p53 mutant line , tp53M214K , was a kind gift from Tom Look ( Berghmans et al . , 2005 ) . Mouse Pax3 coding sequence and human PAX3-FOXO1 coding sequence was a gift from Steve Skapek . The HES3 coding sequence with a MYC-DDK C-terminal tag was obtained from ORIGENE ( RC224630 ) . Pax3 , PAX3-FOXO1 , and HES3 were cloned into the Gateway expression system ( Thermo ) by adding 5’ and 3’ ATT sites ( attb2r/attb3 ) with primers and high-fidelity PCR ( Thermo; Supplementary file 3 ) . Purified PCR products were cloned into a 3’entry clone using the described protocol ( Kendall and Amatruda , 2016 ) . The tol2 kit beta actin promoter , cmv promoter , multiple cloning site , hsp70l promoter , 3’ SV40 late poly A signal construct , and pDestTol2pA2 destination vector were used for construct generation and expression in zebrafish ( Kwan et al . , 2007 ) . The ubi promoter was a kind gift from Len Zon ( Addgene #27320 ) ( Mosimann et al . , 2011 ) , the unc503 promoter from Peter Currie ( Addgene #64020 ) ( Berger et al . , 2013 ) , the fli1 promoter and 3’ entry 2A-mCherry from Nathan Lawson ( Addgene #31160 and #26031 ) ( Villefranc et al . , 2007; Villefranc et al . , 2013 ) , and the mitfa promoter from James Lister ( Addgene #81234 ) . Middle entry beta globin intron and splice acceptor was from Koichi Kawakami ( Kawakami et al . , 2004 ) . The plasmids containing a GFP or mCherry viral 2A sequence were a gift from Steven Leach ( Provost et al . , 2007 ) , and were sub-cloned into a middle entry Gateway expression system . Tol2 mRNA was synthesized from pCS2FA-transposase which was from Koichi Kawakami ( Urasaki et al . , 2006 ) . The constructs generated and utilized in the described studies include: BetaActin-GFP2A-pA , BetaActin-mCherry2A-pA , BetaActin-GFP2A-2AmCherry , BetaActin-mCherry2A-Pax3 , BetaActin-GFP2A-Pax3 , BetaActin-mCherry2A-HES3 , BetaActin-GFP2A-PAX3FOXO1 , CMV-GFP2A-pA , CMV-mCherry2a-pA , CMV-GFP2A-Pax3 , CMV-mCherry2A-HES3 , CMV-GFP2A-PAX3FOXO1 , ubi-GFP2A-PAX3FOXO1 , mitfa-GFP2A-PAX3FOXO1 , fli1-GFP2A-PAX3FOXO1 , unc503-GFP2A-PAX3FOXO1 , SpliceAcceptor-GFP2A-PAX3FOXO1 . Zebrafish were injected at the single-cell stage with equimolar ratios of the described DNA constructs . Injection mixes contained 50 ng/µL Tol2 transposase mRNA , 25 or 50 ng/µL of BetaActin-GFP2A-PAX3FOXO1 or CMV-GFP2A-PAX3FOXO1 , and equivalent molar amounts of comparative plasmid DNA , 0 . 1% phenol red , and 0 . 3X Danieau’s buffer . In knock-down experiments , 300 µM of GFP morpholino ( 5’ ACAGCTCCTCGCCCTTGCTCACCAT 3’; Gene Tools ) , was added to the injection mixes in combination with the plasmid DNA . For survival analysis of embryos , the total number of injected fish was counted , and then the resulting dead or alive embryos subsequently determined at 24 , 48 , and 72 hpf . Survival curves were plotted using GraphPad Prism 7 . 0 c ( La Jolla , CA ) . For cellular persistence analysis during embryogenesis , embryos were injected at the single-cell stage with equimolar amounts of plasmid DNA and Tol2 transposase mRNA . At 24 hr post fertilization zebrafish embryos were dechorionated , and then individually imaged at 24 and 72 hr for GFP and mCherry expression using the exact same settings . In ImageJ , the positive pixel threshold was determined and then applied to quantify the number of GFP or mCherry-positive pixels for each embryo . For FACS sorting of GFP-positive zebrafish cells , zebrafish embryos were injected with purified DNA constructs and transposon mRNA at the single-cell stage and allowed to develop for 24 hr , after which they were deyolked and dissociated to single cells ( Manoli and Driever , 2012 ) . Using a MoFlo Cell Sorter ( Beckman Coulter Life Sciences ) live cells were gated , and then GFP + zebrafish cells were sorted out and collected in 1X Phosphate Buffered Saline ( Thermo ) on ice . Total RNA was immediately isolated using the RNeasy Microkit ( Qiagen ) and its integrity confirmed by Nanodrop on a spectrophotometer and a Bioanalyzer RNA Nanochip ( Agilent ) . Total RNA was utilized for the Zebrafish Affymetrix Gene 1 . 1 Microarray strips which was run in-house . Zebrafish with tumors were humanely euthanized and screened under a Nikon SMZ25 fluorescent stereomicroscope to detect the fluorescent protein indicative of transgene expression . Fresh GFP + tumor tissue was resected and snap frozen in liquid nitrogen . Frozen tissue was subjected to DNA isolation with the DNeasy Kit ( Qiagen ) or total RNA isolation using the RNeasy Microkit ( Qiagen ) . The remaining tumor specimen was placed in histology cassettes and fixed in 4% paraformaldehyde/1XPBS for 48 hr at 4°C ( Fisher ) . They were then de-calcified in 0 . 5M EDTA for 5 days and mounted in paraffin blocks for microtome sectioning . Hematoxylin and eosin staining was performed on de-paraffinized slides . For tumor incidence curves , zebrafish were injected and then only those surviving past 30 days of age were included in the analysis . All zebrafish were screened under the fluorescent microscope to determine if they were GFP positive or negative . Zebrafish without GFP fluorescence were considered to be negative for transgene-dependent tumor formation . Zebrafish that were GFP positive were collected as described earlier , and the presence of malignancies confirmed by hematoxylin and eosin staining and visual review by a pathologist . Injected zebrafish embryos were fixed at 24 hr post fertilization in 4% paraformaldehyde/1XPBS for 24 hr at 4°C or for 2 hr at room temperature in scintillation vials . TUNEL staining was performed using the ApopTag Red In Situ Apoptosis Detection Kit ( Millipore ) . A GFP counter-stain was performed with an anti-GFP antibody at 1:1000 ( MBL International Corporation ) and an Alexa Fluor 488 goat anti-rabbit IgG secondary ( H + L ) at 1:500 ( Thermo ) , or an anti-GFP polyclonal antibody directly conjugated to Alexa Fluor 488 at 1:500 ( Thermo ) . Images for rhodamine/GFP were taken using the same settings across embryos on a Nikon SMZ25 fluorescent stereomicroscope , and number of positive pixels in each embryo determined and analyzed in ImageJ . Gene expression signatures of injected zebrafish embryos were compared for GFP controls , Pax3 , or PAX3-FOXO1 using the Affymetrix Zebrafish Gene 1 . 1 ST Array strip ( Cat #901802 ) . Each sample was run with technical triplicates . Microarray analyses were conducted using the R v3 . 3 . 1 environment ( R core Team 2017 ) . R: A language and environment for statistical computing . R Foundation for Statistical Computing , Vienna , Austria ( https://www . R-project . org ) . Expression profiles were extracted and normalized using Robust Microarray Average ( Irizarry et al . , 2003 ) . The intersection of statistically significant up-regulated genes for Pax3 and PAX3-FOXO1 conditions as compared to an injected GFP control was determined . Significance was assessed using a Welsh Two Sample t-tests . Genes unique to PAX3-FOXO1 expression were eliminated if they did not contain a human ortholog ( based on NBCI HomoloGene build 68 database ) , and genes were then rank ordered and prioritized based on fold change . Genes with statistically significant expression changes were analyzed using DAVID ( https://david . ncifcrf . gov/ ) ( Huang et al . , 2009a; Huang et al . , 2009b ) to identify enriched Gene Ontology ( GO ) terms ( http://www . geneontology . org/ ) . Approximately 2 µg of total RNA was utilized for poly-A RNA enrichment and subsequent library preparation . Sequencing was performed on the Illumina NextSeq 500 Sequencing System with 2 × 75 bp paired end reads . For RNA-seq data , adapter removal and quality-filtering was conducted by Cutadapt ( Martin , 2011 ) . Alignment to the reference zebrafish genome , build GRCz10 , was performed by Bowtie2 ( Langmead and Salzberg , 2012 ) , an ultrafast and memory-efficient tool for aligning sequencing reads to long reference sequences . TopHat2 ( Kim et al . , 2013 ) was used for alignment . De novo assembly of reads into transcripts ( including mRNA , lincRNA , and many other RNA species ) and differential expression analysis were performed by Cufflinks and Cuffdiff ( Trapnell et al . , 2012 ) . To detect the human PAX3-FOXO1 fusion , we realigned all reads to the known junction sequence within PAX3-FOXO1 fusion , which is shown in Figure 1—figure supplement 2 . To reduce false positives , we required that ( 1 ) junction-spanning reads should contain no mismatch with known PAX3-FOXO1 junction sequence and ( 2 ) junction-spanning reads with less than 6 bp matches on either gene was discarded , as suggested by an earlier study ( Li et al . , 2011 ) . In cells , the MF20 primary antibody was utilized for detection of myosin heavy chain at a dilution of 1:40 ( Developmental Studies Hybridoma Bank; DSHB ) in combination with an Alexa Fluor 488 goat anti-mouse IgG ( H + L ) secondary antibody at a dilution of 1:500 ( Thermo ) . DNA was detected using ProLong Gold Antifade mounting media with DAPI ( Thermo ) . MF20 staining in cells was performed as in Kendall et al . , 2012 . In zebrafish embryo whole mounts , the MF20 antibody was used for the detection of myosin at a concentration of 1:100 in combination with an anti-GFP polyclonal antibody directly conjugated to Alexa Fluor 488 at 1:500 ( Thermo ) . The secondary antibody used was Alexa Fluor 594 goat anti-mouse IgG ( H + L ) at 1:500 ( Thermo ) . Images were taken on a Keyence BZ-X700 fluorescent microscope . Total RNA from zebrafish embryos , tumor tissues , and cells was isolated using the RNeasy Mini or Microkit ( QIAGEN ) . cDNA was reverse transcribed from 200 ng-2µg of total RNA with the RT2 HT First Strand Synthesis kit ( QIAGEN ) . qRT-PCR was performed on an ABI 7900HT using the SYBRGreen Master Mix ( BioRad ) and a 10 µL total volume in 384 well plates . See Supplementary file 3 for primer sequences . Delta-delta Ct was calculated , and the calibrator plotted on the far left for every graph . Error bars indicate standard deviation and a student’s t-test was performed on normalized Ct replicates to determine significance . The following BioRad qPCR arrays were utilized , M384 Rhabdomyosarcoma for mouse C2C12 cells and H384 Tumor Metastasis ( SAB Target List ) for Rh30 human RMS cells . Input cDNA was synthesized using the RT2 HT First Strand Synthesis kit ( QIAGEN ) including a synthetic mRNA as an internal quality control ( BioRad ) . qRT-PCR arrays were run on the BioRad CFX384 . Plotted is the fold change above and below the mean gene expression levels from all genes on the array . Cells were harvested and lysed in radioimmunoprecipitation assay buffer ( RIPA ) with cOmplete Mini Protease Inhibitor Cocktail inhibitors ( Sigma ) , and protein levels quantified using Qubit 3 . 0 Fluorometer ( Thermo ) . Twenty micrograms of protein was denatured and loaded on a 4–20% gradient gel ( BioRad ) , and then transferred to PVDF membranes at 4°C . Membranes were blocked in Casein + 0 . 1% Tween-20 ( Thermo ) , and incubated overnight at 4°C with the following antibodies with agitation: MYOD monoclonal antibody 5 . 8A at 1:1000 ( Thermo ) , MYC 71D10 monoclonal antibody at 1:1000 ( Cell Signaling ) , alpha-Tubulin DM1A monoclonal antibody at 1:1000 ( Cell Signaling ) , and MF20 at 1:1000 ( Developmental Studies Hybridoma Bank ) . Goat anti-mouse IgG ( H + L ) -HRP conjugate and goat anti-rabbit IgG ( H + L ) -HRP conjugate secondaries were utilized at 1:20 , 000 ( BioRad ) . Signal was detected using SuperSignal West Pico Chemiluminescent Substrate ( Fisher ) . Membranes were imaged on the BioRad GelDoc XR + and quantification was performed in ImageJ . C2C12 and Rh30 were a kind gift from Steve Skapek . All cell lines were mycoplasma tested with the MycoAlert Mycoplasma Detection Kit ( Lonza ) within 6 months of their use and were negative . C2C12 was maintained in DMEM ( Gibco ) with 10% Fetal Bovine Serum ( FBS; Sigma ) and 1X Antibiotic-Antimycotic ( Gibco ) at 37°C in 5% CO2 . Rh30 were maintained in RPMI-1640 ( ATCC ) with 10% FBS ( Sigma ) and 1X Antibiotic-Antimycotic ( Gibco ) at 37°C in 5% CO2 . Differentiation media for C2C12s included DMEM ( Gibco ) with 2% Horse Serum ( Sigma ) and 10 µg/mL of insulin ( Fisher ) . Cells were transfected with Fugene HD Transfection reagent ( Promega ) of 3 µL:1 µg of DNA . Cells were selected for 30 days as sub-populations in 1 mg/mL of G418 ( Thermo ) prior to experiments and were maintained in 1 mg/mL of G418 . C2C12 cells were differentiated by plating 50 , 000 or 150 , 000 cells per well on 0 . 01% porcine/PBS-coated plates ( 24 wells or six welsl ) in growth media + G418 . If plating density varied from this it is indicated in the figure . For immunofluorescence C2C12 cells were plated on porcine coated glass coverslips . After 24 hours , cells were washed in 1XPBS , and differentiation media + G418 added . Cells were then fused for 6 days in differentiation media , with fresh media being added every other day . For fusion timepoints , cells were collected and pelleted after incubation in TrypLE Express Enzyme ( Fisher ) , and frozen at −80°C for later total RNA isolation . For immunofluorescence , cells were fixed directly in the well with a 1:1 ratio of 4% paraformaldehyde:media at 37°C for 15 min . Subsequent washing and staining steps are detailed in Kendall et al . , 2012 . Cells were seeded at 5 , 000 cells per well for C2C12 and Rh30 in a 24-well plate with four replicates per timepoint . Cells were maintained in GM supplemented with 1 mg/mL of G418 ( Gibco ) during the course of the experiment . At each timepoint , cells were fixed with 4% PFA for 15 min at room temperature , and stained with 0 . 0025% crystal violet ( Alfa Aesar ) in 20% methanol ( Sigma ) for 15 min . They were then rinsed and crystals were solubilized with 10% Acetic Acid ( VWR ) . Absorbance was read on a plate reader at 595 nm . Colony formation assay strategy was adapted from Borowicz et al . ( 2014 ) . A bottom layer of 1 . 2% noble agar ( Difco ) in GM was plated with 1 mg/mL of G418 and allowed to solidify , followed by a top layer of 0 . 6% noble agar in GM with 1 mg/mL of G418 with 5000 cells per well in a six-well plate . Each well was maintained in 100 µL of GM + G418 that was changed three times per week . On day 30 , wells were imaged for analysis , with three images being taken at 10X per well and then averaged . Six technical replicates were performed per sub-clonal population . Three biological replicates representing independent sub-clonal populations were included per experiment . Statistical analysis for survival and tumor incidence curves , and Fisher’s exact test , was performed using GraphPad Prism 7 . 0 c ( La Jolla , CA ) . All other calculations were performed using two-tailed student’s t test in Microsoft Excel Version 15 . 38 . Sample sizes are provided in the figures or figure legends . | One of the most common cancers in children and adolescents is rhabdomyosarcoma , a cancer of soft tissue such as muscle , tendon or cartilage . The fusion of DNA on two different chromosomes causes the most aggressive form of rhabdomyosarcoma . The fused DNA produces an abnormal protein called PAX3-FOXO1 . During normal muscle development , a subset of rapidly growing cells eventually slow down and form mature , working muscle cells . It is still unclear how exactly rhabdomyosarcoma develops , but it is thought that PAX3-FOXO1 stops muscle cells from maturing and the cells that grow out of control result in a tumor . Learning how PAX3-FOXO1 hijacks normal muscle development could lead to new treatments for rhabdomyosarcoma . One treatment approach is to slow the growth of a tumor and force the cells to mature . Then , young patients might avoid chemotherapy or radiation treatments and their side effects . Efforts to improve treatment for this type of cancer face several obstacles . Currently , only one vertebrate animal model of the disease is available to test drugs , and it is still not known how PAX3-FOXO1 causes healthy cells to become cancerous . It is also hard to turn off PAX3-FOXO1 itself , so scientists must find additional proteins that collaborate with it to target with drugs . Now , Kendall et al . show that genetically engineered zebrafish with human PAX3-FOXO1 develop rhabdomyosarcoma-like tumors . Experiments on these zebrafish showed that the protein turns on a gene called her3 . Humans have a similar gene called HES3 . In zebrafish or mouse cells , human HES3 interferes with muscle-cell maturation and allows cells that acquire PAX3-FOXO1 to persist during development instead of dying . It also increases the cell growth and cancerous behavior in human tumor cells . Kendall et al . further looked at HES3 levels in tumors collected from patients with rhabdomyosarcoma and found that having higher levels of HES3 increased the risk of death from the cancer . Human rhabdomyosarcoma tumors with high HES3 levels were also more likely to have certain cell-growth and cell-differentiation related proteins . Drugs that turn off or modify the activity of these proteins already exist . Testing these drugs that target processes such as cell growth in the zebrafish with rhabdomyosarcoma-like tumors may help scientists determine if they reduce tumor growth . If they do , additional trials could determine if they would help patients with rhabdomyosarcoma . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cancer",
"biology"
] | 2018 | PAX3-FOXO1 transgenic zebrafish models identify HES3 as a mediator of rhabdomyosarcoma tumorigenesis |
Analysis of brain ultrastructure using electron microscopy typically relies on chemical fixation . However , this is known to cause significant tissue distortion including a reduction in the extracellular space . Cryo fixation is thought to give a truer representation of biological structures , and here we use rapid , high-pressure freezing on adult mouse neocortex to quantify the extent to which these two fixation methods differ in terms of their preservation of the different cellular compartments , and the arrangement of membranes at the synapse and around blood vessels . As well as preserving a physiological extracellular space , cryo fixation reveals larger numbers of docked synaptic vesicles , a smaller glial volume , and a less intimate glial coverage of synapses and blood vessels compared to chemical fixation . The ultrastructure of mouse neocortex therefore differs significantly comparing cryo and chemical fixation conditions .
The renewed interest in electron microscopy and the emergence of serial imaging approaches to capture volumes of biological tissues at an unprecedented scale ( Briggman et al . , 2011; Helmstaedter et al . , 2013; Bock et al . , 2011; reviewed by Briggman and Bock , 2012 ) have driven the re-examination of commonly used preparation methods ( Mikula et al . , 2012; Tapia et al . , 2012 ) . This has not only been necessary to help increase image contrast and improve imaging speed on a larger scale , but also to help computer vision research produce assisted segmentation approaches for reconstructing different features . However , with this re-invigoration of electron microscope technology , and the need for automation to reconstruct complex structures , it is important to understand how chemical fixation alters brain ultrastructure . The distortion of cell morphology by immersing samples in fixative was apparent in the earliest electron microscopy investigations of the brain , leading to experiments that made detailed comparisons between preparation methods ( Schultz et al . , 1957 ) . These paid careful attention to how different fixatives affected cell volume and preserved the ultrastructure . However , the need to consistently preserve entire organs soon led to cardiac perfusion of aldehydes , and staining with buffered osmium , which became the accepted approach for electron microscopy studies ( Karlsson and Schultz , 1965 , 1966 ) , despite the tissue shrinkage it caused . The degree of shrinkage depends on differences in fixative composition , concentration ( Hillman and Deutsch , 1978 ) , and region ( Kalimo , 1976 ) . Shrinkage has significant implications for the measurement of parameters such as the density of structures; for example , synaptic contacts . Few studies have incorporated a shrinkage factor . Kalimo et al . , adjusted for a 16% linear reduction ( Kalimo , 1976 ) , and Kinney et al . 15% shrinkage along each orthogonal axis ( Kinney et al . , 2013 ) . Any correction is typically calculated on the basis of the changes that occur during the embedding process , with the assumption that no alterations occur during the initial tissue fixation . Using rudimentary indicators such as the position of lesion sites ( Schüz and Palm , 1989 ) , or observations of how the brain filled the skull ( O’Kusky and Colonnier , 1982 ) , gave no precise value for how any brain region changed in volume during the preservation process . In the current study , we made quantitative analyses of fresh and chemically fixed brain tissue using discrete landmarks that allow us to assess the extent of volume changes that occur in the somatosensory cortex . A persistent concern of using chemical fixation has also been the discrepancy between what electron microscopy sees and physiological experiments measure , particularly in terms of extracellular space . Even prior to ultrastructural brain imaging , a measurement of ionic concentrations showed that around a fifth of the tissue volume was extracellular space ( Allen , 1955 ) . This was subsequently verified by various in vivo experiments , using techniques such as resistivity measurements and diffusion analysis . The amount of extracellular space varies between brain regions and the stage of development ( Syková and Nicholson , 2008 ) . In the neocortex of the adult rat , for example , it is 18–22% , and at postnatal day 4–7 , 30% and 43% , depending on the cortical layer ( Lehmenkühler et al . , 1993 ) . Yet standard electron microscopy of this tissue using chemical fixation shows considerably less . This mismatch led Anton Van Harreveld to explore this phenomenon , along with alternative methods of brain fixation . To circumvent any reaction to the chemical fixatives , Van Harreveld introduced a method of rapid freezing of exposed live brain , followed by resin embedding at low temperatures , known as freeze substitution ( Van Harreveld et al . , 1965 ) . This technique revealed the physiological levels of extracellular space . Here , we have revisited this method , using high-pressure freezing to preserve tissue volumes , and comparing how the different cellular compartments are affected with chemical fixation . Ultrastructural analysis using serial section electron microscopy showed that chemically fixed tissue has a reduced volume: a significant loss of extracellular space and a decrease in the volume of neurites . However , the volume fraction of astrocytic elements increased following chemical fixation . In cryo-fixed tissue astrocytes showed a less intimate association with synapses , and their endfeet have a reduced coverage of blood vessels . Differences in membrane arrangements were also apparent at the synapse with a lower density of vesicles along the presynaptic membrane in chemically fixed synapses .
We first analysed how chemical fixation altered the volume of the neocortex by comparing fresh and chemically fixed brain sections cut through mouse primary somatosensory barrel cortex ( Figure 1A , Figure 1—figure supplement 1 ) . Coronal and tangential sections show the distinctive barrel pattern in layer IV , corresponding to the arrangement of whiskers on the mouse's muzzle , enabling us to estimate the total volume change . The chemical fixation protocol was a standard cardiac perfusion with buffered paraformaldehyde and glutaraldehyde , used widely as a preparation method for electron microscopy of brain tissue . Fresh sections were prepared from brains that had been removed rapidly from the skulls of decapitated mice and harvested in the same manner as for electrophysiological experiments . 10 . 7554/eLife . 05793 . 003Figure 1 . Chemical fixation reduces cortical volume and extracellular space . ( A ) Coronal sections of fresh ( left ) and chemically fixed ( right ) adult mouse brains . Double-headed arrows overlaying the somatosensory cortex of each section show the position at which the cortical thickness was measured . ( B ) Measurements of cortical thickness show a 16% reduction after chemical fixation ( p = 0 . 00004 , unpaired Student's t-test , left ) . Measurements across tangential sections show 18% shrinkage in the rostrocaudal axis ( p = 0 . 006 , one way ANOVA ) , but not in the mediolateral axis ( p = 0 . 942 , one way ANOVA , right ) . ( C ) TEM of cryo fixed ( left ) and chemically fixed ( right ) neuropil from the adult mouse cerebral cortex show reduction in the extracellular space ( pseudo-coloured in blue ) after chemical fixation . ( D ) Measurements of the volume fraction of extracellular space from serial section analysis showed a six-fold difference between the two fixation techniques ( p = 0 . 003 , one way ANOVA ) . ( E ) Measurements of volumes occupied by extracellular space , neurites , and glia , from serial section transmission electron microscopy sections showed how the different compartments are altered by chemical fixation . Volume occupied by astrocytic processes was significantly increased after chemical fixation ( p = 0 . 01 , one way ANOVA ) . However , there was no change in the volume occupied by axons and dendrites ( p = 0 . 074 , one way ANOVA ) . As the volume of the cortex is reduced by 31% after chemical fixation , these percentages are shown in the bar chart in which the total volume of chemically fixed neuropil is 69% of the cryo-fixed value ( 100% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 00310 . 7554/eLife . 05793 . 004Figure 1—source data 1 . Data values and statistics underlying Figure 1B , D , E . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 00410 . 7554/eLife . 05793 . 005Figure 1—figure supplement 1 . Schematic scale model ( upper image ) representing the fresh somatosensory cortex ( outer cube ) and chemically fixed cortex ( inner cube ) , showing the extent to which the two fixations change the volume of this brain region . Graph ( lower left ) showing the percentage of neuropil volume occupied by astrocytic ( marked ‘Glia’ ) elements and ECS , in the cryo and chemically fixed neuropil . Graph ( lower right ) shows the percentage occupied by axons and dendrites . Data are represented as mean ± SD . One way ANOVA assessed statistical significance . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 00510 . 7554/eLife . 05793 . 006Figure 1—figure supplement 2 . Comparison between chemical fixation ( left hand images; A and C ) and cryo fixation ( right hand images; B and D ) of acute brain slices shows that both fixation conditions are able to reveal significant amounts of extracellular space . However , this can only be clearly seen within 10 microns of the slice surface ( B ) . Deeper into the cryo-fixed slice , the cellular elements appear disrupted with a fine latticed patterning indicating damage caused by ice crystal formation ( D ) . This is not apparent at the same depth in the chemical fixed slice ( C ) . Scale bar is 1 micron . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 006 Chemically fixed coronal sections showed enlarged ventricles and a clear reduction in total area compared to the fresh sections ( Figure 1A ) . The cortical thickness was reduced by 16% , as measured from the pial surface to the start of the white matter ( fresh 1 . 13 ± 0 . 02 mm , N = 6 mice; chemically fixed 0 . 95 ± 0 . 07 mm , N = 14 mice; p = 0 . 00004 , unpaired two-tailed Student's t-test ) . Tangentially cut sections showed 18% shrinkage along the rostrocaudal axis , measured along the barrel rows: A1–A4 , B1–B4 , C1–C4 and D1–D4 ( fresh 0 . 90 ± 0 . 01 mm , N = 3 mice; chemically fixed 0 . 74 ± 0 . 03 mm , N = 3 mice; p = 0 . 006 , one way ANOVA ) . However , no shrinkage was found along the barrel arcs , on the mediolateral axis: A1–D1 , A2–D2 , A3–D3 , A4–D4 , A1–D4 , A4–D1 ( fresh 1 . 18 ± 0 . 13 mm , N = 3 mice; chemically fixed 1 . 19 ± 0 . 11 mm , N = 3 mice; p = 0 . 942 , one way ANOVA ) . Taken together , these changes indicate that chemical fixation induced total volume shrinkage in the somatosensory neocortex of 30% . Using serial section electron microscopy , we compared the neuropil from the chemically fixed brains with tissue samples that had been rapidly excised and cryo fixed using high-pressure freezing ( McDonald and Auer , 2006 ) and resin embedded by freeze substitution ( Sosinsky et al . , 2008 ) . The chemically and cryo-fixed tissue samples were similarly stained with heavy metals giving a suitable contrast to identify all the membranes and large macromolecular structures ( Figure 1C ) . Cryo-fixed neuropil appeared qualitatively different from the chemically fixed tissue . Neuronal and glial processes were smooth and round , appearing to float in extracellular space . With chemical fixation , the neuropil showed markedly less extracellular space with membranes tightly apposed to each other , often with complex concave and convex shapes . Quantification of serial section electron micrographs revealed that the volume fraction of extracellular space in cryo-fixed neuropil was six times more than chemically fixed samples ( Figure 1D; cryo fixation , 15 . 4 ± 5 . 4% , N = 4 mice; chemical fixation , 2 . 47 ± 1 . 5% , N = 4 mice; p = 0 . 003 , one way ANOVA ) . Further analysis of these volumes measured the contribution of the different cellular compartments . This showed that the astrocytic volume fraction in the cryofixed neuropil was half of the value for chemical fixation ( Figure 1E , Figure 1—figure supplement 1; cryo fixed , 7 . 4 ± 1 . 8% , N = 4 mice; chemical fixation , 14 . 4 ± 3 . 3% , N = 4 mice; p = 0 . 01 , one way ANOVA ) . In contrast , the volume fraction occupied by axons and dendrites was similar between the fixation conditions ( Figure 1—figure supplement 1; cryo fixed 76 . 7 ± 5 . 5% , N = 4 mice , chemically fixed 84 . 1 ± 4 . 1% , N = 4 mice; p = 0 . 074 , one way ANOVA ) . Chemical fixation therefore appears to induce an increase in the astrocytic volume fraction . We next compared the structure of synapses under the two fixation conditions ( Figure 2 ) . Synapses were clearly visible in all material ( Figure 2A ) , and measurements from serial images showed that chemically fixed neuropil had significantly higher synapse density than cryo fixed ( Figure 2B; cryo fixed = 0 . 63 ± 0 . 11 µm−3; chemically fixed = 0 . 87 ± 0 . 15 µm−3 , p = 0 . 042 , one way ANOVA , N = 4 mice each group ) . The increased synapse density after chemical fixation is consistent with the overall volume shrinkage of the neocortex ( Figure 1 ) . Measuring the distance from the edge of spine synapses to the nearest cell membrane , showed a three times larger gap after cryo fixation compared to chemical fixation ( Figure 2B; cryo fixation , 166 ± 18 nm , N = 3 mice; chemical fixation , 53 ± 5 nm , N = 3 mice; p = 2 . 7 × 10−8 , unpaired two-tailed Student's t-test ) . As astrocytes are present at many synapses , where they play a role in glutamate uptake , extracellular space homeostasis , and contribute to the regulation of synaptic transmission ( Ventura and Harris , 1999; Oliet et al . , 2001 ) , we counted the proportion of spine synapses enveloped , or partially enveloped , by their processes ( Figure 2B ) . Astrocytic processes at these types of synapses were significantly fewer in cryo-fixed cortex ( cryo fixation , 34 . 0 ± 11 . 0% , N = 4 mice; chemical fixation , 62 . 4 ± 1 . 9% , N = 3 mice; p = 0 . 008 , one way ANOVA ) . The numbers found in the chemically fixed neuropil are in good agreement with previous measurements ( somatosensory cortex [Genoud et al . , 2006]; hippocampus [Harris and Stevens , 1989] ) . Cryo-fixed tissue , in which there is a greater preservation of the extracellular space , therefore , reveals larger volumes around synaptic clefts suggesting that neurotransmitters can diffuse into large volumes of extracellular fluid before encountering other cell membranes . 10 . 7554/eLife . 05793 . 007Figure 2 . Cryo fixation reveals a larger peri-synaptic space and reduced astrocytic coverage . ( A ) Cryo-fixed neuropil shows synaptic contacts with large amounts of surrounding extracellular space . ( B ) Synaptic density measurements show the chemically fixed neuropil to have 38% more synapses ( left graph , p < 0 . 05 , one way ANOVA ) . Dendritic spine synapses ( presumed glutamatergic ) show greater distances between the edge of the contact zone and the nearest membrane compared with chemical fixation ( middle graph , p < 0 . 001 , unpaired Student's t-test ) . Cryo-fixed synapses show less astrocytic coverage ( right graph , p < 0 . 01; one way ANOVA ) . ( C ) Reconstructions from serial electron microscope images , of axonal boutons ( blue ) synapsing with dendritic spines ( grey ) , show the astrocytic processes in the near vicinity ( red ) . In the cryo-fixed synapse ( left ) , the astrocytic process is not squeezed close to the synaptic contact ( indicated with vesicles in yellow ) . In the chemically fixed example ( right ) , the astrocyte tightly surrounds the synapse , where the vesicle-filled axonal bouton contacts the spine behind it . ( D ) Astrocytic processes reconstructed from serial FIBSEM images using the ilastik software ( www . ilastik . org ) show that chemically fixed astrocytic processes ( right ) have a more elaborate morphology with small processes extending from the flattened lamellae compared with the less complex structure of cryo-fixed astrocytes ( left ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 00710 . 7554/eLife . 05793 . 008Figure 2—source data 1 . Data values and statistics underlying Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 008 The astrocytic elements in chemically fixed neuropil typically show less stained material in their cytoplasm compared to neurons . Their membranes appear to lie against the membranes of the surrounding axons and dendrites , giving them concave shapes , with a space-filling appearance , and large numbers of small processes squeezed between the neurites ( Figure 2C ) . Reconstructions showed that cryo-fixed astrocytic processes stained similarly to axons and dendrites and were more rounded in appearance compared to astrocytes after chemical fixation ( Figure 2D ) . The apparent difference in appearance and arrangement of the astrocytic processes between the two fixation conditions was also investigated at the level of the blood capillaries ( Figure 3 ) , where their close association is suggested to play an important role in the regulation of solutes entering through the blood–brain barrier by almost completely surrounding the endothelial cells that form the vessel lumen ( Mathiisen et al . , 2010 ) . By measuring the proportion of vessels that were surrounded by astrocytic endfeet ( Figure 3A , B ) , we found significantly less coverage in cryo-fixed tissue ( Figure 3C; percentage astrocytic coverage: cryo fixed 62 . 9 ± 15 . 0% , n = 11; chemical fixation 94 . 4 ± 6 . 0% , n = 69; p < 0 . 0001 , unpaired two-tailed Student's t-test ) . Cryo-fixed tissue therefore reveals reduced astrocytic coverage of blood vessels suggesting that the abluminal surface of the endothelial cell has far greater direct access via the extracellular space to the neural elements within the brain than previously thought . 10 . 7554/eLife . 05793 . 009Figure 3 . Cryo-fixed capillaries show less astrocytic coverage . ( A ) Electron micrographs of transversely sectioned capillaries show the astrocytic endfeet pseudo-coloured in orange . Cryo-fixed example shows a darkly stained erythrocyte within the vessel lumen . ( B ) Schematic diagram indicates the coverage measured . ( C ) Chemically fixed tissue contains capillaries with more glial coverage ( p < 0 . 0001 , n = 11 vessels cryo , n = 69 vessels perfused , unpaired Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 00910 . 7554/eLife . 05793 . 010Figure 3—source data 1 . Data values and statistics underlying Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 010 Synapses in chemically fixed tissue can be classified according to their morphology , and in the CNS they are typically categorized as either type 1 or type 2 ( Gray , 1959 ) . Type 2 synapses , later characterized as GABAergic , are distinguishable by their pre- and post-synaptic densities showing equal thickness ( Uchizono , 1965; Colonnier , 1968 ) , and their vesicles appearing flattened with darkened centres ( Figure 4A , Figure 4—figure supplement 1 ) . Measurements of the long and short diameters of synaptic vesicles at type 2 synapses in chemically fixed tissue indicate their ovoid shape ( Figure 4B; long diameter ‘y’ , 41 . 0 ± 5 . 4 nm; short diameter ‘x’ , 28 . 2 ± 3 . 7 nm ) . The glutamatergic type 1 synapses in chemically fixed tissue , in contrast , have an obvious asymmetry with larger postsynaptic densities , and round vesicles with clear centres ( Figure 4—figure supplement 1; diameter y , 40 . 0 ± 3 . 8 nm; diameter x , 39 . 1 ± 3 . 7 nm ) . In cryo-fixed neocortex all synapses , on spines and dendritic shafts showed similar symmetry , and vesicle diameters indicated them all as spherical ( Figure 4A , B; diameter y , 38 . 7 ± 4 . 3 nm; diameter x , 38 . 5 ± 4 . 1 nm ) . This suggests that the chemical fixation causes the structural changes that allow this morphological distinction to be made . To check this , and verify that it was not an effect of the dehydration and embedding process , chemically fixed samples were high-pressure frozen and embedded at low temperature , by freeze substitution ( Sosinsky et al . , 2008 ) . This material contained both asymmetric synapses ( vesicle diameter y , 39 . 2 ± 4 . 0 nm; diameter x , 38 . 9 ± 4 . 8 nm ) and synapses with symmetric pre- and post-synaptic densities , and flattened vesicles ( Figure 4; diameter y , 41 . 6 ± 5 . 3 nm; diameter x , 27 . 7 ± 3 . 4 nm ) . The hallmarks differentiating glutamatergic and GABAergic synapses would therefore appear to be induced by chemical fixation . 10 . 7554/eLife . 05793 . 011Figure 4 . Vesicles of symmetric synapses are distorted by chemical fixation . ( A ) Cryo-fixed synapses on a dendritic shaft ( left image ) and on a dendritic spine ( middle image ) show similar rounded vesicles . A chemically fixed , high-pressure frozen and cryo-substituted ( right hand image ) synapse on a dendritic shaft , however , shows typical features of a symmetric ( presumed GABAergic ) synapse with ovoid vesicles . ( B ) Measurements of the short ( x ) and long ( y ) diameters of synaptic vesicles . Synapses in cryo-fixed tissue cannot be classified according to the symmetry of pre- and post-synaptic densities and all synaptic vesicles were round . Asymmetric synapses in chemically fixed tissue show similarly shaped vesicles , as do the vesicles at asymmetric synapses of chemically fixed tissue that is then high-pressure frozen and freeze substituted in resin . The symmetric synapses , seen in chemically fixed tissue , show vesicles with characteristic ovoid shapes irrespective of how they were resin embedded . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 01110 . 7554/eLife . 05793 . 012Figure 4—source data 1 . Statistics underlying Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 01210 . 7554/eLife . 05793 . 013Figure 4—figure supplement 1 . Examples of glutamatergic synapses ( A , B , C ) , situated on dendritic spines , with round clear vesicles; and presumed GABAergic synapses ( D , E , F ) on dendritic shafts showing flattened , dark vesicles . Scale bar is 200 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 013 We next compared the arrangement of synaptic vesicles in the two fixation conditions , measuring the distance of all vesicles within 150 nm of the presynaptic membrane , for synapses found on dendritic spines , larger than 0 . 2 microns and cut perpendicularly to the synaptic cleft ( Figure 5 ) . The average size of the synapses was the same in each group ( Figure 5—figure supplement 1 ) . The overall density of vesicles , within 150 nm of the presynaptic membrane , was the same in each group ( cryo fixed , 37 . 9 ± 2 . 4 µm−1 , N = 3 mice; chemical fixed , 38 . 0 ± 2 . 6 µm−1; N = 3 mice , p = 0 . 85; one way ANOVA ) . There were , however , clear differences in their spatial distribution ( Figure 5B ) . The synaptic vesicle density within 30 nm of the presynaptic membrane was significantly increased in the cryo-fixed samples ( Figure 5B; cryo fixed , 10 . 46 ± 0 . 88 µm−1; chemical fixed , 2 . 99 ± 0 . 53 µm−1; p < 0 . 001 , unpaired Student’s t-test ) . Between 30 and 60 nm this had decreased in cryo-fixed samples ( Figure 5B; cryo fixed , 4 . 01 ± 0 . 56 µm−1; chemical fixed , 8 . 02 ± 0 . 87 µm−1; p < 0 . 001 , unpaired Student’s t-test ) . Beyond 60 nm there were no differences comparing cryo-fixed and chemically fixed synapses . This suggests that cryo fixation exposes two groups of vesicles; a group lying along the presynaptic membrane , and a second lying further back , away from the site of release . 10 . 7554/eLife . 05793 . 014Figure 5 . Cryo fixation preserves larger numbers of vesicles at the pre-synaptic membrane . ( A ) Electron tomography of a 200-nm thick section shows a cryo-fixed ( upper ) synapse with a large number of vesicles close to the presynaptic membrane in comparison with a similar chemically fixed synapse ( lower ) . In each case three sample images are shown from complete tomographic series . Three-dimensional reconstructions of this region ( right hand images ) show all the vesicles ( red ) in relation to the presynaptic membrane ( blue ) . ( B ) Measurements of the distance of vesicles from the presynaptic membrane show that more vesicles are arranged closer ( 0–30 nm ) to the synapse after cryo fixation ( p < 0 . 0001 , unpaired Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 01410 . 7554/eLife . 05793 . 015Figure 5—source data 1 . Data values and statistics underlying Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 01510 . 7554/eLife . 05793 . 016Figure 5—figure supplement 1 . Synapses found on dendritic spines were the same length . Data are represented as mean ± SD . Unpaired Student's t-test assessed statistical significance; p = 0 . 561 . DOI: http://dx . doi . org/10 . 7554/eLife . 05793 . 016
The analysis of coronal slices at the level of the somatosensory cortex shows that chemical fixation reduced the tissue volume by 30% . This volume decrease would show an increase in synapse density of 43% ( 1/0 . 7 ) . Our synaptic density measurements showed a 38% increase between the cryo-fixed and chemically fixed tissue . The difference between these two values ( 43% vs 38% ) might be accounted for by the fact that synapses were only counted in the neuropil , whereas the cortical volume measurements are from the whole tissue that includes cell bodies and blood vessels . We also cannot rule out changes to the tissue that may occur during the freeze substitution procedure as the acetone replaces the water at low temperature prior to any fixation with the osmium . The fact that the chemical fixation process is relatively slow , involving the diffusion of aldehydes from blood vessels into the surrounding tissue , may also raise questions as to whether the short period of ischemia may cause a rapid assembly or disassembly of synaptic contacts . Sustained occlusion of cerebral vessels eventually leads to cell death and removal of synapses ( Kovalenko et al . , 2006 ) . However , correlative in vivo light and electron microscopy studies would argue against this as they show that any anoxic changes initiated by the chemical fixation procedure do not result in the appearances or disappearances of any synaptic features such as dendritic spines or axonal boutons ( Trachtenberg et al . , 2002; De Paola et al . , 2006; Holtmaat et al . , 2006; Knott et al . , 2006 ) . Additionally , imaging the protein PSD95 in vivo also pinpoints all ultrastructurally identified synaptic connections , imaged with retrospective 3D electron microscopy ( Cane et al . , 2014 ) , suggesting that it is unlikely that any synaptic formation or removal is initiated by the chemical fixation process . To check that the high-pressure freezing itself does not cause a significant alteration of the tissue morphology , we fixed acute slices , prepared in the same manner as for electrophysiological recording , with the cryo fixation or by immersion in the chemical fixative . Both groups showed similar tissue quality and levels of extracellular space , at least where ice crystal formation had not disrupted the ultrastructure ( Figure 1—figure supplement 2 ) . In terms of how the different cellular compartments react to the chemical perfusion fixation , the earliest investigations were aware of the discrepancy between physiological and structural measurements . The ‘watery look of the astrocyte’ led to suggestions that if the physiological measurements of extracellular space were correct then ‘methods of fixation here applied must be erroneous…’ ( Schultz et al . , 1957 ) . Observations like this and many others preceded cryo-fixation studies , first undertaken by Anton Van Harreveld . In Van Harreveld's experiments , exposed brain surfaces were instantaneously fixed , using cooled metal plates , to reveal a physiological level of extracellular space . However , Van Harreveld also found that delaying the process until after the onset of anoxia , preserved a neuropil similar to that seen with chemical fixation ( Van Harreveld et al . , 1965 ) . This was the first illustration of how the extracellular space rapidly disappears once the heart has stopped . More recent live imaging of fluorescent diffusion markers showed a similar reduction ( Thorne and Nicholson , 2006 ) suggesting that the process of cardiac perfusion for chemical fixation would induce a normal tissue response to anoxia resulting from the removal of the blood supply . The reduction of extracellular space cannot account solely for the total tissue volume decrease seen after chemical fixation . Analyses to calculate the volume fractions of different cell components show that the neuronal part remains unchanged . However , considering the total tissue volume has shrunk by 30% the neuronal portion has therefore reduced similarly to maintain the same volume fraction . The doubling of the astrocytic volume fraction after chemical fixation could reflect how this cell type reacts in the aftermath of anoxia . Oxygen depletion elicits an energy reduction resulting in ions moving down their concentration gradients , with an increase in extracellular potassium , and wholesale depolarisation and spreading depression throughout the tissue ( Van Harreveld and Malhotra , 1967; Lutz , 1992 ) . The critical role that astrocytes play in buffering potassium in the extracellular space ( Kofuji and Newman , 2004; Binder et al . , 2006 ) and water movement during brain ischemia ( Manley et al . , 2004 ) point to these cells as playing the principal role in removing the extracellular space . The high concentrations of aquaporin transporters along their surfaces ( Benfenati and Ferroni , 2010 ) suggest that astrocytes may be responsible for the removal of extracellular water ensuring that any changes in outside environment are rapidly neutralised . Cryo fixing brain tissue at different stages after the initiation of spreading depression , or ischemia , has shown similar astrocytic morphologies to those seen after chemical fixation ( Van Harreveld and Malhotra , 1967 ) . The swelling of the astrocytic compartment , and removal of the extracellular space in response to chemical fixation , has important consequences for the interpretation of the ultrastructure . The perisynaptic region in chemically fixed tissue shows smaller spaces within which neurotransmitters can diffuse ( Van Harreveld et al . , 1965; Ohno et al . , 2007 ) ( Figure 1 ) , with smaller distances between the edge of the synapse and surrounding elements . This region has variously been described at dendritic spines with a significant presence of astrocytes . In the neocortex , 60–70% of the bouton/spine interfaces are partially or completely surrounded with an astrocytic process ( Genoud et al . , 2006 ) . A similar proportion was found in the hippocampus ( Harris and Stevens , 1989 ) . The larger perisynaptic space suggested by cryo fixation to exist in vivo would give neurotransmitters a greater opportunity to diffuse more rapidly out of the synaptic cleft , into the enlarged extracellular space where there would be a greater dilution , possibly leading to a slowing of their diffusion . The enlarged perisynaptic space could effectively act as a buffer zone , reducing extrasynaptic neurotransmitter concentrations , which could help isolate synapses , reducing the possibility that their activity would influence other extrasynaptic receptors , and minimising synaptic crosstalk . However , careful computational simulations ( Rusakov and Kullmann , 1998; Hrabe et al . , 2004 ) taking into account spatiotemporal dynamics , tortuosity , and binding/unbinding of neurotransmitters to various proteins ( for example , glutamate transporters largely located on astrocytic membranes ) are essential to gain a more detailed understanding of neurotransmitter diffusion , and how this would be affected by the differences in ultrastructure we found after cryo fixation compared to standard chemically fixed tissue . The reduced synaptic density in the cryo-fixed tissue also places them further from each other , adding to their anatomical isolation . Cryo fixation therefore appears to suggest a reduction in the influence of volume transmission revealing an ultrastructure that might favour wiring transmission ( Kullmann , 2000 ) . A more direct access through an enlarged extracellular space was apparent at the blood capillaries , another site where astrocytic processes have a significant presence . Here , their endfeet completely enclose blood capillaries after chemical fixation ( Mathiisen et al . , 2010 ) but cryo fixation reveals a partial coverage of about two thirds ( Figure 5 ) . The abluminal surface of the endothelial cell , therefore , is less insulated from the neuronal elements suggesting that solutes passing through the capillary wall from the blood have more direct access to the neurons . Conversely , substances released from neurons have greater access to blood vessels , where they might contribute to controlling blood flow ( Attwell et al . , 2010 ) . Other changes in the ultrastructure were seen at the synaptic connections . The classical morphology of inhibitory and excitatory synapses: their differences in symmetry and vesicle shapes that have proved useful for classifying types of connections in electron microscopy studies were not apparent in the cryo-fixed tissue . The flattened appearance of synaptic vesicles in GABAergic boutons therefore appears to be induced by the chemical fixation process . Previous studies have shown that delaying chemical fixation by a few minutes alters synaptic structure . Asymmetric and symmetric contacts showed greater curvature and thicker postsynaptic densities ( Martone et al . , 1999; Kovalenko et al . , 2006; Tao-Cheng et al . , 2007 ) . Pre-synaptically , vesicles were located further from the active zone . Images of the frog neuromuscular junction showed that chemical fixation , unlike freezing , could not prevent the fusion of vesicles to the presynaptic membrane ( Heuser et al . , 1976 ) . More recently , the use of high-pressure freezing , simultaneous to synapse stimulation , has revealed how the vesicles fuse and collapse into the pre-synaptic membrane within 30 ms of stimulation ( Watanabe et al . , 2013 ) . Analyses of vesicle arrangement in cryo-fixed CNS synapses in cultured slices ( Zhao et al . , 2012 ) or synaptosome preparations ( Fernández-Busnadiego et al . , 2010 ) have revealed a high concentration of vesicles clustered at the pre-synaptic membrane with the lowest concentration immediately behind this zone , 50–70 nm further back . However , after stimulation , the lowest concentration is seen at the pre-synaptic membrane steadily increasing further away ( Fernández-Busnadiego et al . , 2010 ) . A similar picture is seen in our comparison between chemical and cryo-fixed cortical synapses ( Figure 3 ) . After cryo fixation , the greatest density of synaptic vesicles is at the pre-synaptic membrane , but after chemical fixation the vesicle density increases away from the pre-synaptic membrane suggesting that chemical fixation is unsuitable for capturing all docked vesicles . Cryo fixation is able to reveal two groups of synaptic vesicles: one close to the synaptic membrane , and the other further back , approximately a vesicle width away . This peak in vesicle density close to the presynaptic release site has also been seen at the calyx of Held synapse ( Han et al . , 2011 ) . Whether this heterogenous distribution represents different functional pools is unproven . However , a functional analysis of calyx synapses showed that disruption to the pre-synaptic protein machinery results in less vesicular release , together with a concomitant reduction in the number of vesicles seen close to the presynaptic membrane ( Han et al . , 2011 ) . This would support the idea that vesicles against the presynaptic membrane are docked and associated with their release machinery ( Harlow et al . , 2001 ) where they can be depleted within a few milliseconds ( Felmy et al . , 2003 ) . This arrangement of vesicles with cryo fixation could be revealing the readily releasable pool; those aligned tightly along the membrane . And the second grouping , further back from those in a docked position may , therefore , be those of the recycling and reserve pool , immobilized and clustered together within the body of the bouton by proteins such as actin or synapsin ( Hilfiker et al . , 1999; Siksou et al . , 2007 ) . Cryo fixation , therefore , can provide a view of tissue ultrastructure that is closer to its natural state . This increases the relevance of morphological analyses by revealing an arrangement of cell membranes that more closely matches functional measurements . However , although these analyses highlight how the exquisite organization of the brain's cellular elements is acutely sensitive to chemical fixation , there are difficulties in cryo preserving large volumes of brain tissue . The high water content of the CNS appears to limit the depth to which vitrification can occur ( less than 10–20 microns in our hands at least ) before ice crystal formation causes significant damage to the structure . Large-scale preservation for ultrastructural analysis will therefore continue to rely on chemical fixation approaches .
Adult mice ( C57 BL/6 , 7–10 weeks old ) were decapitated , and the brain immediately removed . This was then immersed in ice-cold artificial cerebrospinal fluid ( ACSF ) composed of ( in mM ) : 125 NaCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 . 5 KCl , 0 . 1 CaCl2 , 3 MgCl2 , 25 glucose , 3 myo-inositol , 2 Na-pyruvate , 0 . 4 ascorbic acid , pH 7 . 4 which was bubbled with 95% O2 and 5% CO2 . Slices were made with a vibratome ( Leica Microsystems VT1200 ) at 150 microns thickness in the coronal or tangential plane . After cutting , each slice was then transferred immediately to a Petri dish containing the same medium and photographed with a stereo microscope ( Leica Microsystems M205C ) . For cryo fixation , immediately after decapitation , the brain was exposed using blunt surgical tweezers , and a piece of the somatosensory cortex cut from the brain using a razor blade . The tissue was then further sliced , and pieces , approximately 200-µm thick , were placed inside a 6 mm diameter aluminium sample holder with a 200-µm deep cavity . To ensure that no air bubbles became trapped inside the sample holder with the tissue , a small drop of 1-hexadecene was added . This was then high-pressure frozen using a Leica EM HPM100 ( Leica Microsystems ) . This entire procedure was completed as rapidly as possible , in less than 90 s from the moment of decapitation . The frozen samples were then stored in liquid nitrogen until further processing . The animals were deeply anesthetized with pentobarbitone ( 0 . 05 mg g−1 ) and perfused via the heart , using a perfusion pump , at a speed of 7 ml/min with 2 . 5% glutaradehyde , and 2% paraformaldehyde , in phosphate buffer ( 0 . 1 M , pH 7 . 4 , 250–300 ml per animal ) . The tubing used for perfusion was back filled initially with 5 ml of 0 . 1 M PBS ( pH 7 . 4 ) to help remove all blood from the circulatory system before the fixative entered . After perfusion , the animal was left for 1 hr and the brain was removed . 70-µm thick slices were vibratome sectioned in PBS ( 0 . 01 M , pH 7 . 4 ) . These slices were then stained and embedded using previously described methods ( Knott et al . , 2011 ) . Briefly , the slices were washed in cacodylate buffer ( 0 . 1 M , pH 7 . 4 , 3 × 5 min ) , post fixed in 1% osmium tetroxide and 1 . 5% potassium ferrocyanide in cacodylate buffer ( 0 . 1 M , pH 7 . 4 , 40 min ) . They were then stained with 1% osmium tetroxide in cacodylate buffer ( 0 . 1 M , pH 7 . 4 ) for 40 min , and then in 1% uranyl acetate for 40 min before being dehydrated in a graded alcohol series , 3 min each change , and embedded in Durcupan resin . Frozen tissue was stained , dehydrated , and embedded at low temperature using a low temperature-embedding device ( AFS2; Leica Microsystems ) . Frozen samples were transferred to this device in liquid nitrogen and were then initially exposed to 0 . 1% tannic acid in acetone , for 24 hr at −90°C , followed by 12 hr in 2% osmium tetroxide in acetone at −90°C . The temperature was then raised over 4 days to −30°C , and then the liquid replaced with pure acetone and the temperature increased to −10°C over 16 hr . Finally , the tissue samples were mixed with increasing concentrations of resin over 8 hr whilst the temperature rose to 20°C . They were then added to 100% resin for 2 hr and then placed in silicon moulds for 24 hr at 65°C for the resin to harden . Cryo-fixed tissue that showed no artifacts of fixation , such as ice crystal damage , was selected by cutting semi-thin ( 0 . 5 mm thick ) sections ( 2 × 2 mm ) of the resin-embedded material and staining these with toluidine blue . With transmitted light microscopy , an area of well-fixed tissue could be identified and these regions were further trimmed and either used for TEM or FIBSEM . All quantitative analysis of the serial images was carried out using the FIJI software ( http://pacific . mpi-cbg . de/wiki/index . php/Fiji ) . Image alignments were performed using the stackreg plug-in . Synapse density measurements were carried out using serial images through volumes of neuropil , containing no cell bodies or blood vessels . The images were displayed in the TrakEM2 software within FIJI ( Cardona et al . , 2012 ) and counted when they were completely contained within the displayed volume or were touching the left , top , and upper sides of the counting frame . Those touching the right , bottom , and lower sides were excluded . The section thickness was measured by using the mitochondria as cylindrical objects , and counting over how many sections they were present when lying parallel to the imaging plane ( Fiala and Harris , 2001 ) . Measurements of volumes occupied by the different compartments were made by manually segmenting each of the different elements in the serial images using the TrakEM2 software . Area lists were made of the different compartments; neurites ( axons and dendrites ) , astrocytes , and extracellular space , and drawn in each of the serial images . Distance measurements of cortical shrinkage , glial coverage of blood vessels , synapse and vesicle sizes , and vesicle and membrane separations were made using the same software . Models of cellular elements shown in Figure 3 were made using the interactive segmentation tool in the ilastik software ( www . ilastik . org ) from serial FIBSEM images ( Straehle et al . , 2011 ) . These models were then imported into the 3D modelling software Blender ( www . blender . org ) for final composition and rendering . | For many years , scientists have used chemicals to preserve brain tissue to observe its fine structure using high power microscopes . Korogod et al . now show that these chemicals , or fixatives , cause the tissue to shrink , giving the false impression that the cells are tightly packed together . This has led to misinterpretations of how the brain is structured . For example , components such as the synapse , used by neurons to communicate with each other , are bathed in a watery environment , rather than being tightly enclosed by neighbouring cells as previously thought . Electron microscopy is the only imaging method that is able to see the detailed structure of the nervous system , including synaptic connections . The technique fires a beam of electrons through a sample held in a vacuum and creates images at a higher magnification than light microscopes . However , the electron beam and the vacuum damages live cells and tissues . Therefore , samples must be ‘fixed’ to preserve them before they are imaged with these methods . However , the standard method for fixing brain tissue uses chemical ‘fixatives’ , even though these cause shrinkage , and distort the cells . Korogod et al . used an alternative method of fixation—freezing—to better preserve tiny pieces of mouse brain in their natural state . This was achieved with a technique called ‘high pressure freezing’ that combines jets of liquid nitrogen with very high pressures to instantaneously preserve small samples without causing damage through the formation of ice crystals , or any shrinkage and distortion . Once frozen , the samples of mouse brain are encased in resin , and then imaged with the electron microscope . A comparison between the two preservation techniques showed that chemical fixatives remove the watery environment , or extracellular fluid , that surrounds the cells in the brain , squashing them together . The synapses were surrounded by large amounts of extracellular fluid , but cryo fixation also revealed that these sites of communication between neurons also contained many more vesicles—the packets containing the chemicals that pass signals across the synapse . Another type of cell , the glial cell , that supports and helps to maintain neurons , was also strongly distorted by the chemical fixation . These were understood to tightly wrap around synapses , as well as blood vessels , but cryo fixation showed this to be less prominent . This study illustrates that our understanding of how brain's cells are arranged has ignored the effects of the chemicals used to preserve them . Although cryo fixation is only able to preserve tiny samples , it reveals a truer picture of their natural structure , giving scientists a better understanding of how the brain works . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Ultrastructural analysis of adult mouse neocortex comparing aldehyde perfusion with cryo fixation |
Caspase-8 is a key player in extrinsic apoptosis and its activity is often downregulated in cancer . However , human Caspase-8 expression is retained in some tumors , including glioblastoma ( GBM ) , suggesting that it may support cancer growth in these contexts . GBM , the most aggressive of the gliomas , is characterized by extensive angiogenesis and by an inflammatory microenvironment that support its development and resistance to therapies . We have recently shown that Caspase-8 sustains neoplastic transformation in vitro in human GBM cell lines . Here , we demonstrate that Caspase-8 , through activation of NF-kB , enhances the expression and secretion of VEGF , IL-6 , IL-8 , IL-1beta and MCP-1 , leading to neovascularization and increased resistance to Temozolomide . Importantly , the bioinformatics analysis of microarray gene expression data derived from a set of high-grade human gliomas , shows that high Caspase-8 expression levels correlate with a worse prognosis .
The downregulation of apoptotic pathways is a hallmark of cancer ( Hanahan and Weinberg , 2011 ) . Caspase-8 is a central player in the apoptotic cascade triggered by death receptors stimulation ( Juo et al . , 1998 ) ; consistently , its expression ( Pingoud-Meier et al . , 2003 ) or its apoptotic activity ( Cursi et al . , 2006; Safa et al . , 2008 ) are often reduced in cancer . The observation that Caspase-8 is retained in many tumors ( reviewed in [Stupack , 2013] ) suggests a dual role for Caspase-8 in cancer . The identification of several non-canonical functions of Caspase-8 that are independent of its enzymatic activity and of apoptosis , supports this idea . Indeed , Caspase-8 modulates cell adhesion and migration , suggesting that in cancer cells Caspase-8 may be rewired from apoptosis to alternative pathways that sustain tumor growth ( reviewed in [Graf et al . , 2014] ) . We recently demonstrated that Caspase-8 promotes the proliferation and neoplastic transformation of glioblastoma ( GBM ) cell lines ( Fianco et al . , 2016 ) . Interestingly , large-scale gene expression approaches have demonstrated Caspase-8 upregulation in GBM compared to normal tissue; in particular , the mesenchymal subtype of GBMs is characterized by high Caspase-8 expression ( Verhaak et al . , 2010 ) . The fatal nature of GBM is strongly associated with its extensive angiogenesis ( Kargiotis et al . , 2006 ) , and with its capacity to infiltrate throughout the brain tissue and to resist to chemotherapy ( Dunn et al . , 2012 ) . Tumor neoangiogenesis is strongly supported by an inflammatory microenvironment that also promotes the proliferation of tumor cells and the survival of malignant cells and alters responses to chemotherapeutic agents ( Mantovani et al . , 2008 ) . Consistently , in vitro and in vivo studies have identified high levels of IL-8 , IL-6 and IL-1beta in the conditioned media ( CM ) of several GBM cell lines and in microenvironment of clinical samples ( reviewed in Yeung et al . , [2013] ) . This often depends on overactive EGFR signalling , which stimulates NF-kB , AP-1 and cEBP transcription factors , thereby promoting the expression of IL-8 and IL-6 ( Bonavia et al . , 2012; Inda et al . , 2010 ) . The work of several laboratories has identified Caspase-8 as an activator of NF-kB in B cells downstream of antigen receptors ( Su et al . , 2005 ) and Toll-like receptors ( Lemmers et al . , 2007 ) , as well as in T cells ( Bidère et al . , 2006 ) . These observations , along with the pivotal role of NF-kB in modulating cytokine production , in shaping tumor microenvironment and in promoting angiogenesis and GBM progression ( reviewed in Karin et al . [2002] , Dunn et al . [2012] , Yeung et al . [2013] and Nogueira et al . [2011] ) , prompted us to investigate whether high Caspase-8 expression in GBM promotes these functions .
To investigate the possible role of Caspase-8 in GBM angiogenesis , we sub-cutaneously injected mice with matrigel-containing conditioned media ( CM ) from U87MG ( U87 ) cells , in which Caspase-8 expression was genetically silenced ( shC8 ) or not ( shcontrol , named CTR ) ( as shown in Figure 1—figure supplement 1 ) . Matrigel plugs containing CM from U87CTR induced a strong angiogenic response as evidenced by macroscopic analysis and haemoglobin content , similar to that detected in the positive control where VEGF has been added to the media . Importantly , matrigel plugs containing CM from U87 shC8 displayed a significant reduction of both angiogenesis in vivo and the haemoglobin content ( Figure 1A and B ) . 10 . 7554/eLife . 22593 . 003Figure 1 . Caspase-8 expression promotes tumor growth and neoangiogenesis in vitro and in vivo . ( A , B ) Caspase-8 expression promotes the ability of conditioned medium ( CM ) from U87 cells to induce neo-angiogenesis in vivo . Representative images illustrate the macroscopic analysis ( A ) and quantification of Hb content ( B ) of Matrigel plugs containing CM from Sh Control ( CTR , n = 23 ) or Sh Caspase-8 ( ShC8 , n = 21 ) cells . The negative ( Neg , n = 14 ) and positive ( Pos , n = 20 ) controls contained heparin alone or heparin plus VEGF , respectively . The values of biological replicates ( n ) for each condition are shown as single dot , and are expressed as optical density ( OD 540 nm ) /g of the Matrigel plug . The Mann-Whitney test ( independent samples ) was used for statistical analyses . In all experiments , the volume of CM from different samples was normalized on the number of cells for each sample counted when CM was collected . ( C ) Comparison of the tumor size between U87 CTR and U87shC8 mouse xenografts . Quantitative analysis , by Kruskal-Wallis test with Bonferroni correction , of the volume of tumors measured at 3 and 6 weeks after cell injection . Each plot graphically shows the central location and scatter/dispersion of the values of each group: the line series in the rectangular-shaped boxes indicate the median value of the data and the end of the vertical lines indicate the minimum and the maximum data value . The means and their confidence intervals are shown in the diamond-shaped box . P-value was calculated according to the independent samples t-test . Each dot corresponds to the tumor value of one mouse . ***p<0 . 001 . ( D ) The microvessel density , determined immunohistochemically by the means of an anti-CD31 antibody recognizing murine endothelial cells , evidenced the presence of a significantly higher number of vessels in CTR cells ( evaluated as mean ± SD in CTR tumors ) than in shC8 tumors ( ***p<0 . 001 ) . Original magnification 40X , scale bar 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 00310 . 7554/eLife . 22593 . 004Figure 1—source data 1 . Caspase-8 mRNA is efficiently silenced in shC8 and shC8#2 cell lines compared to CTR cells . Statistical analysis of quantitative real time RT-PCR Figure 1—figure supplement 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 00410 . 7554/eLife . 22593 . 005Figure 1—source data 2 . Caspase-8 expression promotes the ability of conditioned medium ( CM ) from U87 cells to induce neo-angiogenesis Figure 1B Statistical analysis of the quantification of Hb content ( Figure 1—figure supplement 2A , C ) . Statistical analysis of the quantification of HUVEC cells proliferation and of tubulogenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 00510 . 7554/eLife . 22593 . 006Figure 1—source data 3 . Caspase-8 expression promotes tumor growth in mouse xenograft experiments . Statistical analysis of tumor growth data from U87 and U87shC8 samples . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 00610 . 7554/eLife . 22593 . 007Figure 1—source data 4 . Caspase-8 expression promotes neovascularization in vivo . Statistical analysis of vessel content analysis from immunohystochemistry experiments ( Figure 1D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 00710 . 7554/eLife . 22593 . 008Figure 1—figure supplement 1 . Caspase-8 mRNA and protein expression is efficiently silenced in shC8 and shC8#2 cell lines compared to CTR cells . ( A ) Quantitative real-time RT-PCR for Caspase-8 mRNA on U87CTR ( CTR ) , U87shC8 ( shC8 ) and U87shC8#2 ( shC8#2 ) cells . Relative qQuantities were calculated by normalizing for TBP . Representative results of a single experiment with n = 3 biological replicates , each one performed in technical duplicate are shown as mean ± SD ( ***p<0 . 001 ) . Three independent experiments were consistent . ( B ) Detection of Caspase-8 protein expression by western blot analysis of whole lysates of CTR , shC8 and shC8#2 cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 00810 . 7554/eLife . 22593 . 009Figure 1—figure supplement 2 . Caspase-8 expression promotes the ability of conditioned medium from U87 cells to induce endothelial cells proliferation and Capillary Tube-Like Network Formation . ( A ) HUVEC proliferation evaluated after exposure for 72 hr to CM from Sh Control ( CTR ) or Sh Caspase-8 ( ShC8 ) cells . Endothelial cells incubated in serum free medium ( neg ) or complete medium ( pos ) were used as negative or positive control , respectively . The results represent the mean ± SD . Error bars represent a SD between two independent experiments , each of them performed in six replicates . Student’s t test was used for statistical analyses . ( B ) Representative images of Capillary Tube-Like Network Formation Assay and quantification of sprouts formation , estimated by measuring the cumulative length of the sprouts ( C ) , on Matrigel . Bar scale: 100 μm . Endothelial cells incubated in serum free medium ( neg ) or complete medium ( pos ) were used as negative or positive control , respectively . The values were expressed as mean of cumulative length of the sprouts ± SD . Error bars represent a SD among biological duplicates , with at least four technical replicates each , of a representative experiment . Student’s t test was used for statistical analyses . In all experiments , the volume of CM from different samples was normalized on the number on cells for each sample counted when CM was collected . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 009 In agreement with this finding , CM from U87shC8 cells was 2 . 5-fold less potent in triggering the proliferation of Human Umbilical Vein Endothelial Cells ( HUVECs ) compared to CM from U87CTR cells ( Figure 1—figure supplement 2 ) . Moreover , when exposed to CM from U87CTR cells , HUVECs formed tube-like structures resembling a capillary plexus; conversely , partially organized and rounded endothelial cells were observed after the addition of CM from U87shC8 cells ( Figure 1—figure supplement 2 ) . To uncover whether Caspase-8 dependent regulation of angiogenesis correlates with the modulation of tumor growth in vivo , we compared the tumorigenic potential of U87 cells in which the expression of Caspase-8 was genetically inhibited or not . As reported in Figure 1C , after three weeks , U87 cells formed tumors of an average volume of ~1 , 254 mm3 in all injected mice , whereas U87shC8 cells exhibited a drastically reduced capacity to form tumors ( mean tumor volume: ~148 mm3 ) , which were detected in only 29% of injected mice . After 6 weeks , the mean tumor volume generated by U87sh8 cells was ~344 mm3 and tumors were identified in 67% of injected mice ( Figure 1C ) . In order to evaluate whether the decreased tumor growth observed in shC8 mice compared to CTR was associated with a lower vascular density , we evaluated neovascularization in CTR and shC8 tumors . Tumor vessels were visualized using anti-CD31 mAb , which permitted an assessment of the vascular density of the tumors . Figure 1D shows that there was a significant difference in vascularization between CTR and shC8 tumors; the mean number of vessels per mm2 field was 20 . 52 ( ±0 . 62 ) for CTR and 1 . 21 ( ±0 . 41 ) for shC8 ( p=0 . 001 ) . Overall , these data suggest that Caspase-8 expression in GBM is required for tumor growth , maximal proliferation of GBM-associated endothelial cells and angiogenesis , probably through the production of secreted factors . To measure whether Caspase-8 expression affects the profile of secreted cytokines and growth factors , we used LUMINEX multiplex bead assay , which allows the simultaneous quantification of several cytokines from the same sample ( Khalifian et al . , 2015 ) . High levels of IL-1β , IL-6 , IL-8 , MCP-1 , VEGF-A and TNFalpha were present in the CM of U87CTR , as is consistent with previous reports ( Albulescu et al . , 2013 ) , and the genetic inhibition of Caspase-8 caused a strong decrease in the levels of these cytokines ( except TNF alpha ) in the CM ( Figure 2A ) . Accordingly , we observed a dramatic reduction of the respective mRNAs upon Caspase-8 downregulation ( Figure 2B ) . We obtained similar results in U87 cells where Caspase-8 expression was inhibited using a different interfering sequence ( Figure 2—figure supplement 1 ) , as well as in U251 , another GBM cell line ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 22593 . 010Figure 2 . Caspase-8 promotes the expression and secretion of cytokines and growth factors . ( A ) Concentrations ( pg/ml ) of IL-1β , IL-6 , IL-8 , MCP-1 ( CCL-2 ) , VEGF-A and TNF-α were measured by Luminex assay , in the supernatants of U87 cells upon stable genetic silencing of Caspase-8 expression ( Sh C8 ) or not ( ShCTR , named CTR ) . Data were plotted as mean ± SD and statistical significance was estimated by Unpaired T-test , ***p<0 . 001 . Error bars represent a SD between three independent experiments , each of them performed in technical duplicate . In all experiments , the volume of CM from different samples was normalized on the number of cells for each sample counted when CM was collected . ( B ) Quantitative real time RT-PCR on U87CTR cells and U87shC8 cells . Relative quantities were calculated normalizing for TBP . Representative results of a single experiment with n = 3 biological replicates , each one performed in technical duplicate , are shown as mean ± SD ( ***p-value<0 . 001 ) . Three independent experiments were consistent . ( C ) Correlation between Caspase-8 and IL-6 , IL-8 , IL1β , MCP-1 and VEGF expression in human glioblastoma . Pearson correlation coefficients computed between gene expression profiles for CASP-8 , IL-6 , IL-8 , IL1β , MCP-1 and VEGF , in 174 glioblastoma RNA-Seq samples retrieved from the Cancer Genome Atlas . The correlation coefficient between expression profiles is proportional to the circle radii in the matrix , and additionally color-coded using the color scale reported to the right of the matrix . The plot was generated using the /corrplot/ R package ( https://cran . r-project . org/web/packages/corrplot ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01010 . 7554/eLife . 22593 . 011Figure 2—source data 1 . Caspase-8 promotes the secretion of cytokines and growth factors . Statistical analysis of Luminex Experiments ( Figure 2A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01110 . 7554/eLife . 22593 . 012Figure 2—source data 2 . Caspase-8 promotes mRNA expression of cytokines and growth factors . Statistical analysis of quantitative real time RT-PCR Figure 2B and Figure 2—figure supplement 1A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01210 . 7554/eLife . 22593 . 013Figure 2—source data 3 . Data collection for the analysis of the correlation between Caspase-8 and IL-6 , IL-8 , IL1β , MCP-1 and VEGF expression in human glioblastoma , Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01310 . 7554/eLife . 22593 . 014Figure 2—source data 4 . Correlation between Caspase-8 and IL-6 , IL-8 , IL1β , MCP-1 and VEGF expression in human glioblastoma . The Pearson correlation coefficients and corresponding p-values between Caspase-8 expression and those of the different cytokine and growth factor genes are displayed . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01410 . 7554/eLife . 22593 . 015Figure 2—figure supplement 1 . Silencing of Caspase-8 triggers the downregulation of IL-6 , IL-8 and VEGF mRNA expression . ( A ) Quantitative real time RT-PCR for IL-6 , IL-8 and VEGF mRNAs in U87 CTR and U87 ShC8#2 . ( B ) Quantitative real time RT-PCR for IL-6 , IL-8 and VEGF mRNAs in U251 Sh control and Sh Caspase-8 ( shC8 ) cell lines . Relatives quantities were calculated to TBP and are relative to U87 Sh CTR cells . n = 3 biological replicates , each performed in technical duplicate ***p-value<0 . 001 , **p-value<0 . 01 , NS = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01510 . 7554/eLife . 22593 . 016Figure 2—figure supplement 2 . Ectopic expression of a dominant negative IKBα triggers the downregulation of IL-6 , IL-8 , IL-1β , MCP-1 and VEGF mRNA expression . Quantitative real time RT-PCR on U87 stably overexpressing a dominant negative IKBalpha construct ( IKB alpha SR ) or not . Relative quantities were calculated by normalizing for TBP . Representative results of a single experiment with n = 3 biological replicates are shown as mean ± SD ( ***p-value<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 016 To clarify the significance of our findings , we interrogated mRNA levels in a RNA-seq data set derived from 174 human GBM patients , which is available in Cancer Genome Atlas ( TCGA ) . For each gene , we built a profile of normalized expression data from the 174 samples , and we then compared these profiles using the Pearson product-moment correlation coefficient . Interestingly , the Caspase-8 expression profile positively correlated with those of IL-6 , IL-8 , IL1β and MCP-1 ( Figure 2C and Figure 2—source data 4 ) , ranging from 0 . 25 when comparing the Caspase-8 and IL-6 GBM expression profiles , to 0 . 53 for Caspase-8 and MCP-1 . All of these correlations were statistically significant ( P-value < 0 . 0008 for each comparison ) , whereas there was no significant relationship between the Caspase-8 and VEGF expression profiles ( Pearson correlation coefficient 0 . 09 , p-value 0 . 22 ) . Overall these results identified a novel role of Caspase-8 in promoting neoangiogenesis . This is in agreement with previous studies on Caspase-8 null mice , which showed defects in the angiogenesis programme during development ( Scharner et al . , 2009 ) . The expression of the aforementioned cytokines is finely tuned by several transcription factors , among which is NF-kB . In gliomas , NF-kB is often constitutively activated ( Cahill et al . , 2016 ) , NF-kB-regulated genes are induced ( Bonavia et al . , 2012 ) , and the expression of these genes correlates inversely with patient prognosis ( Nogueira et al . , 2011 ) . Consistently , the overexpression of a dominant negative mutant of IKBalpha ( IKBaphaS32A/S36A , named IKBapha SR ) that inhibits NF-kB signalling severely decreased the expression of IL-6 , IL-8 , IL1β , MCP-1 and VEGF-A mRNAs in U87 cells ( Figure 2—figure supplement 2 ) . Silencing of Caspase-8 in U87 cells did not reduce the amount of NF-kB but strongly decreased its nuclear localization ( Figure 3A and B ) . Similar results were also obtained in U251 GBM cells ( Figure 3C ) . Preliminary immunohistochemistry analysis in formalin-fixed paraffin-embedded tissues suggests a different distribution of NF-kB in CTR ( nuclear and cytoplasmic ) and in shC8 ( mainly cytoplasmic ) tumors ( Figure 3—figure supplement 1 ) . Consistently , we could detect a significant reduction of VEGF and IL-8 mRNA levels in shC8-derived tumor samples ( Figure 3—figure supplement 1 ) . Our results provide the first evidence that Caspase-8 promotes NF-kB activity in GBM in vitro and in vivo . Interestingly , cancer-associated missense mutations of Caspase-8 resulting in stronger activation of NF-kB have been identified recently in head and neck squamous cell carcinoma ( Ando et al . , 2013 ) . 10 . 7554/eLife . 22593 . 017Figure 3 . Caspase-8 expression promotes NFkB translocation into the nucleus . ( A ) Immunostaining α-p65-NFkB in U87 Sh Caspase-8 and control cell lines . Bar scale: 25 μm . The bar chart represents the cytoplasmic ( cyt ) /nuclear ( nuc ) ratio of cell fluorescence intensity ( FI ) obtained using the Image J program . ***p-value<0 . 001 . ( B , C ) Western blot analysis of fractionated cell lysates of U87 ( B ) and U251 ( C ) cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01710 . 7554/eLife . 22593 . 018Figure 3—source data 1 . Caspase-8 promotes NFkB nuclear localization in U87 GBM cells . Statistical analysis of Image J Quantification of NFkB cellular localization ( Figure 3A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01810 . 7554/eLife . 22593 . 019Figure 3—source data 2 . Caspase-8 promotes IL8 and VEGFA mRNA expression in tumors derived from mouse xenograft experiments . Statistical analysis of quantitative real time RT-PCR ( Figure 3—figure supplement 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 01910 . 7554/eLife . 22593 . 020Figure 3—figure supplement 1 . Caspase-8 expression promotes NFkB activity in vivo . ( A ) NF-kB immunohistochemical expression performed in formalin-fixed paraffin-embedded tissues evidences a different distribution of NFkB in shC8 ( mainly cytoplasmic ) with respect to CTR ( nuclear and cytoplasmic ) tumor samples ( original magnification 40X , scale bar 50 µm ) . ( B ) Quantitative real time RT-PCR for Caspase-8 , VEGF and IL-8 mRNAs in tumor mass caused by subcutaneous injection of U87 CTR control and U87 shC8 cells . Relative quantities were normalized to TBP and calculated relative to U87 Sh CTR . n = 4 ( CTR ) or n = 6 ( shC8 ) biological replicates , each performed in technical duplicate . ***p-value<0 . 001 , **p-value<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 020 Having demonstrated that Caspase-8 expression promotes neoplastic transformation in vitro ( Fianco et al . , 2016 ) and sustains tumor growth and its microenvironment ( Figures 1 , 2 and 3 ) , we asked whether Caspase-8 expression affects GBM cells response to therapy . For several decades , the typical treatment for GBM has been radiotherapy; more recently , Temozolomide ( TMZ ) has been incorporated into the standard treatment as an essential component ( Stupp et al . , 2005 ) . As shown in Figure 4A and in Figure 4—figure supplement 1 , the down-regulation of Caspase-8 expression significantly sensitized U87 cells to the cytotoxic effects of TMZ . Interestingly , the expression of the dominant negative mutant of IKBalpha , IKBaphaSR , had a similar effect ( Figure 4B ) , suggesting that Caspase-8 exerts its function via NF-kB signalling , possibly through secreted cytokines . Indeed , as shown in Figure 4C , CM of control cells that endogenously express Caspase-8 was sufficient to restore resistance to TMZ in cells silenced for Caspase-8 . Importantly , CM from IKBalphaSR-expressing cells , like that from Caspase-8-deficient cells , lost such a protective capability ( Figure 4D ) . These results support the conclusion that Caspase-8 expression in GBM cells triggers resistance to TMZ via an autocrine loop and suggest that the level of Caspase-8 expression may correlate with prognosis . To test this hypothesis , we analyzed the survival of 77 high-glioma patients ( of which 15 were censored ) , whose gene expression was measured using microarrays ( Phillips et al . , 2006 ) . We classified the patients on their Caspase-8 expression , dividing the Caspase-8 expression distribution into three quantiles and considering the first quantile as low expression and the last quantile as high expression . Analysis of the survival curves shows that patients that have higher Caspase-8 expression levels have a lower survival chance than those with lower expression ( Chi-squared 10 . 5 on 1 degree of freedom , p-value 0 . 00117 ) , supporting our hypothesis ( Figure 4E ) . In addition , we investigated the survival rates of the same 77 patients classified into three distinct subtypes: mesenchymal ( 23 cases ) , proneural ( 30 cases ) and proliferative ( 24 cases ) tumors . Patients belonging to each subtype were stratified in terms of high and low Caspase-8 expression , based on the distribution of Caspase-8 probe intensities specific for each subtype . The significantly different Caspase-8 expression levels in the proneural , mesenchymal and proliferative GBMs justified the use of different thresholds for each subtype . For patients classified as proneural , those having high Caspase-8 expression show significantly lower survival probability ( Chi-squared 7 . 2 on 1 degree of freedom , p-value 0 . 00732 ) , while no difference was observed for the mesenchymal and proliferative subtypes ( Figure 4—figure supplement 2 ) . Overall , we identify a novel function of Caspase-8 ( Figure 4F ) , which supports a double agent role of Caspase-8 in cancer . The classical role of Caspase-8 in apoptosis may account for the correlation between loss of Caspase-8 expression and unfavourable prognosis in medulloblastoma ( Pingoud-Meier et al . , 2003 ) . Conversely , tumors that are strongly dependent on NFkB activity and cytokine production , such as GBM , may have a selective advantage in retaining Caspase-8 expression . In these contexts , targeting of Caspase-8 expression or of its tumorigenic functions represents a novel therapeutic approach . 10 . 7554/eLife . 22593 . 021Figure 4 . Downregulation of Caspase-8 increases sensibility to Temozolomide ( TMZ ) . ( A , B ) Viability assay represented in the histogram as mean ± SD . U87CTR control cell lines , U87shCaspase-8 and U87 IKBalphaSR , were incubated in the presence of Temozolomide ( TMZ 0 . 5 mM for 72 hr ) or not . The viability of TMZ-treated cells was assessed with the CellTiter 96 Aqueous One Solution Cell proliferation assay , and was represented as the percentage of inhibition of viability measured in cells without TMZ treatment . Data are represented in the histogram as mean ± SD . Error bars represent a SD between three ( A ) or two ( B ) independent experiments , each of them performed at least in technical triplicate . Student’s t test was used for statistical analyses . **p<0 . 01 ( A , B ) . ( C , D ) U87shCaspase-8 were incubated or not in the presence of Temozolomide ( TMZ 0 . 5 mM for 72 hr ) dissolved in conditioned media derived from U87 ShCtr ( CTR ) , U87 shCaspase-8 ( shC8 ) or U87IKBa SR . Data are represented in the histogram as mean ± SD . Error bars represent a SD between two ( C ) or three ( D ) independent experiments , each of them performed in technical triplicate . Student’s t test was used for statistical analyses . *p-value<0 . 05 ( C ) , NS=not significant . In all experiments shown in panel A–D , the volume of CM from different samples was normalized on the number on cells for each sample counted when the CM was collected . ( E ) Survival curves of high-grade glioma classified based on Caspase-8 expression levels . Glioma patients were classified as low Caspase-8 expression ( green curve ) and high Caspase-8 expression level ( red curve ) , as described in the text . The Kaplan-Meier test supports a significant difference ( p-value 0 . 00117 ) between the survival rates of the two groups , with patients having a low Caspase-8 expression showing a higher survival probability . ( F ) Proposed model depicting the link between Caspase-8 and cytokines in glioblastoma . Caspase-8 promotes NFkB nuclear localization and sustains the production of VEGF , IL-6 , IL-8 , IL-1β and MCP-1 . This pathway promotes neoangiogenesis and triggers resistance to Temozolomide . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 02110 . 7554/eLife . 22593 . 022Figure 4—source data 1 . Caspase-8 downregulation increases sensibility to Temozolomide ( TMZ ) . Statistical analysis of experiments ( Figure 4A , B , C , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 02210 . 7554/eLife . 22593 . 023Figure 4—source data 2 . Survival curves of high-grade glioma classified on the basis of Caspase-8 ( CASP8 ) expression levels . Data collection . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 02310 . 7554/eLife . 22593 . 024Figure 4—source data 3 . Caspase-8 downregulation by two different shC8 constructs increases sensibility to Temozolomide ( TMZ ) . Statistical analysis of experiments ( Figure 4—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 02410 . 7554/eLife . 22593 . 025Figure 4—figure supplement 1 . Caspase-8 downregulation by two independent interference sequences increases sensibility to Temozolomide ( TMZ ) . Viability assay represented in the histogram as mean ± SD . U87CTR control cells ( CTR ) , U87shC8 ( shC8 ) and U87shCaspase-8#2 ( shC8#2 ) cells were incubated or not in the presence of Temozolomide ( TMZ 0 . 5 mM for 72 hr ) . The viability of TMZ-treated cells was assessed with the CellTiter 96 Aqueous One Solution Cell Proliferation assay , and is represented as the percentage of inhibition of viability measured in cells treated without TMZ . Data are represented in the histogram as mean ± SD . Error bars represent SD between two independent experiments each of them performed at least in technical triplicate . Student’s t test was used for statistical analysis . ***p<0 . 001 . In all experiments , the volume of CM from different samples was normalized on the number on cells for each sample counted when the CM was collected . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 02510 . 7554/eLife . 22593 . 026Figure 4—figure supplement 2 . Survival curves and Caspase-8 expression levels in three GBM subtypes . Survival of patients classified as proneural ( A ) , proliferative ( B ) and mesenchymal ( C ) and stratified into low and high Caspase-8 expression groups on the basis of the expression distribution . ( D ) Caspase-8 ( CASP8 ) expression distributions , estimated as microarray probe intensity , in samples belonging to the proneural ( blue curve ) , mesenchymal ( orange curve ) and proliferative ( pink curve ) subtypes . DOI: http://dx . doi . org/10 . 7554/eLife . 22593 . 026
Caspase-8 was stably genetically silenced in U87 ( originally obtained by ATCC ) and U251 cell lines as previously described ( Fianco et al . , 2016 ) . All cell lines were maintained in DMEM supplemented with 10% fetal bovine serum and were routinely tested negative for mycoplasma contamination . Following thawing , cells were used for no longer than one month . Human umbilical endothelial cells ( HUVEC; PromoCell GmbH , Heidelberg , Germany ) were cultured as previously reported ( Gabellini et al . , 2013 ) . The sequences of Caspase-8 used for interference are: shC8 5’-ATCACAGACTTTGGACAAA-3’ , shC8#2 5’-GCCTGGATGTTATTCCAG-3’; the sequence used as control , shcontrol ( CTR ) , is: 5’-GGATATCCCTCTAGATTA-3’ . The pMIGIKBalphaS32A/S36A ( IKBalpha SR ) construct was kindly provided by Y . Ciribilli and A . Inga ( Brockman et al . , 1995; Bisio et al . , 2014 ) . Anti Caspase 8 ( MBL 1:1000 ) RRID:AB_590760; anti Tubulin ( Sigma-Aldrich 1:2000 ) RRID:AB_10013740; anti NF-kB ( p65 ) ( Santa Cruz 1:1000 ) RRID:AB_632037; anti SP1 ( Santa Cruz 1:1000 ) RRID:AB_2171050; anti rat mAb CD31 ( clone SZ31 , Dianova GmbH 1:10 ) RRID:AB_2631039; anti-NF-kB p65 ( clone E379 , abcam , 1:1000 ) RRID:AB_776751; TMZ: temozolomide . All procedures involving animals and their care were authorized and certified by the decree n . 26/2014 of the Italian Minister of Health following the relative guide lines . For in vivo Matrigel assays , 60 μl 10× concentrated CM from the different cell lines , obtained using Centricon-3 concentrators ( Merck Millipore , Billerica , MA ) were mixed with 600 μl of Matrigel ( BD Bioscience , San Jose , CA ) , supplemented with heparin ( 19 . 2 U; Schwarz Pharma SpA , Milan , Italy ) . This medium was injected subcutaneously into the flank of 8-week-old C57BL/6 mice ( provided by the Animal Care Unit of the Regina Elena Cancer Institute , Rome , Italy ) . The negative and positive controls contained heparin alone or heparin plus VEGF ( 60 ng/mice; R&D Systems , Minneapolis , MN ) , respectively . After 5 days , the angiogenic response was evaluated by macroscopic analysis at autopsy , and by measurement of the hemoglobin ( Hb ) content in the pellet of matrigel as previously reported ( Gabellini et al . , 2013 ) . The values were expressed as optical density ( OD at 540 nm ) /100 mg of matrigel . Each group consisted of ten animals . The experiments were repeated three times . In vitro HUVEC cell proliferation was evaluated by a colorimetric assay at the end of treatment as described previously ( Gabellini et al . , 2013 ) . Endothelial capillary tube-like network formation was assessed using Matrigel as described previously ( Gabellini et al . , 2013 ) . Briefly , 24–well microtiter plates were coated with 300 µl/well unpolymerized matrigel ( 10 mg/ml ) and allowed to polymerize at 37°C . Endothelial cells were plated ( 5 × 104 cells/well ) in 1 ml of serum free medium ( negative control ) , complete medium ( positive control ) , or conditioned medium ( CM ) obtained from sh control ( CTR ) or sh Caspase 8 ( shC8 ) cells . After 8 hr , cell growth was observed through a reverted , phase-contrast photomicroscope and photographed . Experiments were repeated at least three times , and each sample was tested in triplicate . Angiogenic activity was quantified by measuring the cumulative length of the sprouts using digital imaging software ( Image J ) to analyze ten fields per experimental group and experiment . Concentrations of IL1-β , IL-6 , IL-8 , MCP-1 and VEGF-A were simultaneously determined in supernatants of shCtr ( CTR ) and ShC8 cells ( n = 3 biological replicates per group and two technical duplicates for each sample ) using a custom-made human magnetic Luminex assay kit ( E-Bioscience ) . The assay procedure was performed according to the manufacturer's instructions and the plate was read on a Luminex-200 instrument ( Luminex Corp . , Austin , TX ) . Data were calculated by generating a calibration curve using the recombinant cytokines specified above , diluted in the cell culture medium used for culturing cells . Concentrations of each analyte were calculated using a standard 5P-logistic weighted curve generated for each target and expressed as picograms per milliliter ( pg/ml ) . Due to out of range readings of undiluted samples , IL-8 concentrations were calculated on 1:10 diluted supernatants . Data were presented as the mean ± SD and the statistical analysis was performed using unpaired T-tests . Cells were plated on coverslips and maintained at 37°C and 5% CO2 for 24 hr before staining . Cells were treated or not with TNFα ( 20 ng/ml ) 20 min before staining . Cells were washed with 1x phosphate buffer salin ( PBS ) three times . They were fixed in 4% paraformaldehyde for 15 min , permeabilized in 0 . 3% triton x100 for 15 min , blocked with 1% BSA for 1 hr at room temperature , and incubated with primary antibody overnight at 4°C . Secondary antibodies were applied for 1 hr at room temperature , stained with Hoechst for 5 min . The primary antibody used was anti NFkB p65 Santa Cruz ( clone C20 ) 1:200 . The secondary antibody was donkey anti rabbit 488 ( Jackson Immune Research ) 1:200 . Images of immunostaining cells were obtained by microscopy using a Olympus BX53 microscope . Quantitative fluorescence data were exported from ImageJ generated histograms into Microsoft Excel software for further analysis and presentation . Cytoplasmic and nuclear staining intensities were compared to give the cytoplasmic/nuclear ratio . One microgram of total RNA isolated by TRIZOL reagent ( Invitrogen , Carlsbad , CA , USA ) was retrotranscribed with MLV-Reverse Transcriptase ( Promega , Madison , WI , USA ) according to standard procedures . Ten nanograms of cDNA were employed to quantify the transcripts by real time RT-PCR using SYBR Select Master Mix ( Applied Biosystem Foster City , CA , USA ) and gene-specific primers , which are listed in supplemental information . Real-time PCR was performed using the 7900HT Fast Real-Time PCR System ( Applied Biosystem ) . Relative quantity ( RQ ) was calculated normalizing for TBP and using a U87CTR sampleas calibrator . Mean values and standard deviations of RQ were generated from three biological replicates . Each experiment was performed for two technical replicates . The following primer sequences were used: Primer sequences listIL-6-FW5’- CAGGAGCCCAGCTATGAACT -3’IL-6-REV5’-GAAGGCAGCAGGCAACAC- 3’IL-8-FW5’-GGTGCAGTTTTGCCAAGGAG-3’IL-8-RV5’-TGGGGTGGAAAGGTTTGGAG-3’VEGF-FW5’-CCTTGCTGCTCTACCTCCAC-3’VEGF-RV5’-CAACTTCGTGATGATTCTGC-3’CCL2-MCP1-FW5’-CTTCATTCCCCAAGGGCTCG-3’CCL2-MCP1-RV5’-GCTTCTTTGGGACACTTGCTG-3’TNFA-FW5’-GGGACCTCTCTCTAATCAGC-3’TNFA-RV5’-TCAGCTTGAGGGTTTGCTAC-3’TBP-FW5’-TGCCCGAAACGCCGAATATAATC-3’TBP RV5’-TGGTTCGTGGCTCTCTTATCCTC-3’CASP8-FW5’- CAGCAGCCTTGAAGGAAGTC -3’CASP8-RV5’-CGAGATTGTCATTACCCCACA-3’ Cell extracts were prepared in IP buffer ( 50 mM Tris–HCl [pH 7 . 5] , 250 mM NaCl , 1% NP-40 , 5 mM EDTA , 5 mM EGTA , 1 mM phenylmethylsulfonyl fluoride , 25 mM NaF , 1 mM sodium orthovanadate , 10 µg/ml TPCK , 5 µg/ml TLCK , 1 µg/ml leupeptin , 10 µg/ml soybean trypsin inhibitor , 1 µg/ml aprotinin ) . For nuclei/cytoplasm cell fractionation , cells were in hypotonic buffer ( 10 mM HEPES [pH 7 . 5] , 10 mM KCl , 0 . 1 mM EDTA , 0 . 1 mM EGTA , 1 mM DTT , and protease and phosphatase inhibitors at concentrations described below ) and incubated for 15 min on ice . NP-40 ( 0 . 6% final concentration ) was added , and nuclei were harvested by centrifugation at 12 , 000 g at 4°C for 30 s . The cytoplasmic fraction was recovered , and nuclear proteins were extracted from the pellet in nucleus buffer ( 20 mM HEPES [pH 7 . 5] , 0 . 4M NaCl , 0 . 1 mM EDTA , 0 . 1 mM EGTA , 1 mM DTT , and protease and phosphatase inhibitors at concentrations described below ) for 1 hr at 4°C on a rotating wheel . For immunoblotting , 50–100 µg of proteins were separated by sodium dodecyl sulfate ( SDS ) polyacrylamide gel electrophoresis ( PAGE ) , blotted onto nitrocellulose membrane , and detected with specific antibodies . U87 cell lines ( U87 CTR and U87 Sh Caspase 8 ) were seeded in 96-well plates ( 1 , 000 cells/well ) and treated for 72 hr with TMZ 0 . 5 mM or DMSO as control . Cell viability was analysed by CellTiter 96 Aqueous One Solution Cell proliferation assay ( Promega ) as previously described ( Stagni et al . , 2015 ) . For in vivo tumorigenicity , female CD-1 nude ( nu/nu ) mice , at 6–8 weeks old and 22–24 g in body weight , were purchased from Charles River Laboratories ( Calco , Italy ) . 4 × 106 U87CTR and U87ShC8 cells were injected subcutaneously into the flank of these mice ( 24 for each group ) . Two different experiments ( the first one with 8 and the second one with 16 animals for each group ) were performed . The mice were observed daily , and their tumor volume ( mm3 ) was calculated as length × width 2 × π/6 . The animals were sacrificed 3 ( CTR ) and 6 ( shC8 ) weeks after cell injection . The results were analysed by pooling together the two experiments to evaluate tumor growth at three weeks ( CTR and shC8 ) and at six weeks ( shC8 ) . The three groups were compared using the Kruskal-Wallis test with Bonferroni correction . Immediately after sacrifice , the tumors were removed: half of each tumor was frozen in Trizol and stored at −80°C and the remaining half was fixed in 4% buffered formalin and paraffin embedded for immunohistochemical analysis . Microvessel density and NFkB expression were evaluated on tumor xenograft paraffin-embedded sections by staining endothelial cells using a CD31 anti rat mAb and the rabbit mAb anti-NF-kB , respectively . Immunoreactions were revealed by ULTRATEK HRP ( Scy Tek Laboratories , UT , USA ) for CD31 and by Bond Polymer Refine Detection in an automated stainer ( Leica Biosystem , Milan , Italy ) for NFkB . A collection of 174 RNA-Seq data samples from patients diagnosed with glioblastoma multiforme was retrieved from the Cancer Genome Atlas ( TCGA ) . The data were produced by the University of North Carolina Cancer Genomic Characterization Center ( CGCC ) using the Illumina HiSeq 2000 platform , and made available in TCGA as Level 3 ( preprocessed ) data . The TCGA data were retrieved from the Genomic Data Commons ( GDC ) using the following search query: Disease Type IS Glioblastoma Multiforme AND Primary Site IS Brain AND Program Name IS TCGA AND Project Id IS TCGA-GBM AND Access IS open AND Data Category IS Gene expression AND Data Format IS TXT AND Data Type IS Gene expression quantification AND Experimental Strategy IS RNA-Seq AND Platform IS Illumina HiSeq Among the files retrieved by this query , we employed the files reporting RSEM normalized gene expression . More details on how to retrieve these data are provided in the Supplementary Materials ( Supplementary file 1 ) . Data processing was carried out using the SeqWare Pipeline project's MapspliceRSEM workflow ( version 0 . 7 ) ( O'Connor et al . , 2010 ) . Gene-level expression data were estimated using RSEM ( Li and Dewey , 2011 ) and normalized to set the upper quartile count at 1 , 000 for gene level . Correlation between gene expression profiles in the 174 samples was computed as the Pearson product-moment correlation coefficient , setting the p-value threshold at 0 . 01 . Microarray gene expression data , obtained using the Affymetrix U133A chip on a set of 77 high-grade gliomas ( Phillips et al . , 2006 ) and for which patient follow-up was available , were retrieved from the Gene Expression Omnibus ( GEO series Id GSE4271 ) in the form of MAS5-normalized intensities . The distribution of Caspase-8 expression was divided in three equal-size quantiles; patients whose Caspase-8 expression was in the first quantile were classified as 'Low Expression' , while those whose Caspase-8 expression was in the third quantile were classified as 'High Expression' . Patient stratification depending on Caspase-8 expression was computed for all patients together , and also independently for the three subtypes in which the patients were classified ( proneural , mesenchymal or proliferative ) . Survival curves and the Kaplan-Meier estimator were computed and plotted using the R package survival ( https://cran . r-project . org/web/packages/survival/index . html ) . | Cancer cells are different to normal cells in various ways . Most cancer cells , for example , delete or switch off the gene for a protein called Caspase-8 . This is because this protein is best known for promoting cell death and stopping tumor cells from growing . However , some cancers keep the gene for Caspase-8 switched on including glioblastoma , the most aggressive type of brain cancer in adults . This begged the question whether this protein may in fact promote the development of tumors under certain circumstances . Glioblastomas are often highly resistant to chemotherapy and can communicate with nearby cells using proteins called cytokines to promote the formation of new blood vessels . The new blood vessel allows the tumor to readily spread into healthy brain tissue , which in turn makes it difficult for surgeons to remove all the cancerous cells . As a result , glioblastomas almost always return after surgery , and so there is strong need for new effective treatments for this type of cancer . Fianco et al . have now investigated whether Caspase-8 helps glioblastomas to grow and form new blood vessels . One common method to study human cancer cells is to inject them into mice and watch how they grow , because these experiments mimic how tumors develop in the human body . When mice were injected with human glioblastoma cells with experimentally reduced levels of Caspase-8 , the cells grew poorly and did not form as many new blood vessels as unaltered glioblastoma cells . Further experiments showed that , when grown in the laboratory , glioblastoma cells with less Caspase-8 were more sensitive to a chemotherapeutic drug called temozolomide . These findings confirm that Caspase-8 does boost the growth and drug resistance of at least one cancer . When Fianco et al . analyzed clinical data from patients affected by glioblastoma , they also observed that those patients with high levels of Caspase-8 often had the worse outcomes . Previous studies conducted in white blood cells showed that Caspase-8 activated a protein complex called NF-kB , which in turn led to the cells releasing cytokines . Fianco et al . have now verified that Caspase-8 promotes NF-kB activity also in glioblastoma cells , and that this causes the cancer cells to release more cytokines . As such , these findings reveal a clear link between Caspase-8 and the formation of new blood vessels by glioblastomas . Future studies are now needed to understand why Caspase-8 promotes cell death in some cancers but the formation of new blood vessels in others . Indeed , Caspase-8 might become a target for new anticancer drugs if it is possible to inhibit its cancer-boosting activity without interfering with its ability to promote cell death . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"cell",
"biology",
"cancer",
"biology"
] | 2017 | Caspase-8 contributes to angiogenesis and chemotherapy resistance in glioblastoma |
Intracellular pH ( pHi ) dynamics is increasingly recognized as an important regulator of a range of normal and pathological cell behaviors . Notably , increased pHi is now acknowledged as a conserved characteristic of cancers and in cell models is confirmed to increase proliferation and migration as well as limit apoptosis . However , the significance of increased pHi for cancer in vivo remains unresolved . Using Drosophila melanogaster , we show that increased pHi is sufficient to induce dysplasia in the absence of other transforming cues and potentiates growth and invasion with oncogenic Ras . Using a genetically encoded biosensor we also confirm increased pHi in situ . Moreover , in Drosophila models and clonal human mammary cells we show that limiting H+ efflux with oncogenic Raf or Ras induces acidosis and synthetic lethality . Further , we show lethality in invasive primary tumor cell lines with inhibiting H+ efflux . Synthetic lethality with reduced H+ efflux and activated oncogene expression could be exploited therapeutically to restrain cancer progression while limiting off-target effects .
Dysregulated pH is a common characteristic of cancer cells , which have a lower extracellular pH ( pHe ) and higher intracellular pH ( pHi ) than normal cells . The lower pHe of tumors , confirmed by pH-sensitive PET radiotracers , MR spectroscopy and MRI ( Zhang et al . , 2010 ) , contributes to local metastatic invasion ( Cardone et al . , 2005; Rofstad et al . , 2006; Stock et al . , 2008; Stock and Schwab , 2009; Busco et al . , 2010; Estrella et al . , 2013 ) . Increased pHi enables a number of cancer cell behaviors , including promoting cell proliferation ( Pouysségur et al . , 1982; Kapus et al . , 1994; Putney and Barber , 2003 ) , glycolytic metabolism ( Reshkin et al . , 2000; Dietl et al . , 2010; Webb et al . , 2011 ) , migration ( Denker and Barber , 2002; Patel and Barber , 2005; Frantz et al . , 2008; Stock and Schwab , 2009 ) , and invasion ( Reshkin et al . , 2000; Hinton et al . , 2009 ) , as well as limiting apoptosis ( Matsuyama et al . , 2000; Lagadic-Gossmann et al . , 2004 ) . The higher pHi of cancer cells is paradoxical because increased production of metabolic acids generated by aerobic glycolysis would be predicted to lower pHi . However , many cancers have elevated expression or activity of proteins that facilitate increased pHi , including carbonic anhydrase 9 ( Swietach et al . , 2007 ) , H+-ATPases ( Martinez-Zaguilan et al . , 1993; Sennoune et al . , 2004; Hinton et al . , 2009 ) , the ubiquitously expressed Na+-H+ exchanger NHE1 ( McLean et al . , 2000; Miraglia et al . , 2005; Chiang et al . , 2008; Yang et al . , 2011 ) and the monocarboxylate transporter family members MCT1 and MCT4 ( Pinheiro et al . , 2010; Halestrap , 2013 ) . Therapeutic targeting of these proteins to reduce H+ efflux and lower pHi has been suggested for limiting cancer progression ( Webb et al . , 2011; Harguindey et al . , 2013 ) , based primarily on findings with xenograft models and isolated cells . Xenograft tumor growth is suppressed by inhibiting NHE1 ( Lagarde et al . , 1988; Yang et al . , 2011 ) or MCT1 ( Sonveaux et al . , 2008; Colen et al . , 2011 ) activity . In cell models , inhibiting NHE1 activity reduces viability of breast cancer ( Reshkin et al . , 2003 ) and leukemic cells ( Rich et al . , 2000; Reshkin et al . , 2003 ) , and silencing MCT4 expression with RNA interference decreases survival of renal carcinoma cells ( Gerlinger et al . , 2012 ) . Although previous studies suggest important roles for pH dynamics in regulating cancer cell behaviors , whether increased pHi is sufficient or necessary for cancer progression in vivo remains unresolved . To better understand how H+ efflux and increased pHi affect cancer cell behaviors we asked two distinct but related questions . First , is increased pHi by NHE1 over-expression in the absence of other transforming signals sufficient to induce dysplasia ? Second , is H+ efflux by NHE1 necessary for oncogene-induced dysplasia ? We found that over-expression of Drosophila melanogaster Dnhe2 , an ortholog of mammalian NHE1 , causes dysplasia , increases proliferation and facilitates oncogene-induced cell invasion in vivo . Additionally , we found that reducing H+ efflux genetically or pharmacologically limits oncogene-induced increases in proliferation and has synthetic lethality with oncogenic Raf in the Drosophila retina as well as in human mammary epithelial cells expressing oncogenic RasV12 .
To address our first question we overexpressed Dnhe2 in Drosophila eye imaginal discs . Three NHE isoforms have been identified in Drosophila ( Dnhe1-3 ) , and sequence analysis of the C-terminal cytoplasmic domains suggests DmNhe2 is most homologous to the mammalian plasma membrane NHE1 , one of nine mammalian isoforms ( Figure 1A ) . Conserved features include a lysine/arginine-rich region ( KR motif ) in the juxtamembrane segment that binds phosphatidylinositol 4 , 5-bisphosphate in the plasma membrane ( Putney and Barber , 2003 ) , a conserved binding motif for the calcineurin homologous protein CHP ( Lin and Barber , 1996; Pang et al . , 2001 ) , a glutamic acid residue in the transmembrane domain that is essential for H+ efflux ( E358 in DmNhe2 and E266 in human NHE1 ) and consensus sites for phosphorylation by Akt and ATM/ATR kinases ( Figure 1A , and Figure 1—figure supplements 1 , 2 ) . The only previous report on DmNhe2 function indicates a role in charge-dependent plasma membrane recruitment of Dishevelled to establish planar cell polarity ( Simons et al . , 2009 ) . 10 . 7554/eLife . 03270 . 003Figure 1 . Dnhe2 over-expression is sufficient to induce dysplasia and hyperproliferation . ( A ) Amino acid sequence analysis of NHE C-terminal cytoplasmic domains reveals that Drosophila melanogaster ( Dm ) Nhe2 has the highest homology with mammalian NHE1 compared with DmNhe1 and 3 . ( B and C ) Scanning electron micrographs of the exterior surface of adult Drosophila eyes of the indicated genotypes . Anterior is to the right in all figures unless otherwise indicated . ( D ) Histological sections reveal an asymmetrical trapezoidal arrangement of rhabdomeres , the light-sensing organelles of the R cells , visible as dark circles; seven of the eight rhabdomeres are visible in each section . ( E ) Traced diagram showing ommatidial orientation , where blue trapezoids indicate wild type , black circles indicate missing R cells , and yellow circles indicate abnormal symmetrical R4/R4 cell specification . ( F ) Single confocal slices from live , pupal eyes expressing GMR > mChpH showing fluorescence of pHluorin ( top ) , mCherry ( middle ) and ratiometric images ( bottom ) . ( G ) pHi was calculated from ratio values using a standard curve and graphed ( black lines show mean ± SEM ) . Calculated pHi values are: wild type ( 7 . 3 ± 0 . 06 , n = 17 ) , GMR > Dnhe2 ( 7 . 7 ± 0 . 04 , n = 15 ) ; GMR > Dnhe2E358I ( 7 . 2 ± 0 . 04 , n = 8 ) . ( ***p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 00310 . 7554/eLife . 03270 . 004Figure 1—figure supplement 1 . DNhe2 is the ortholog of human NHE1 . ( A ) Amino acid sequence alignment of the C-terminal cytoplasmic tails of Homo sapiens NHE1 and Drosophila melanogaster DNhe2 . Conserved motifs and key regulatory residues are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 00410 . 7554/eLife . 03270 . 005Figure 1—figure supplement 2 . Heterologously expressed DNhe2 in NHE1-deficient CCL39 Chinese hamster lung fibroblasts ( termed PS120 cells ) localizes to plasma membrane protrusions , as has been shown for mammalian NHE1 ( Putney and Barber , 2003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 00510 . 7554/eLife . 03270 . 006Figure 1—figure supplement 3 . Dnhe2 over-expression increases pHi in photoreceptor neurons . The pHi in photoreceptor neurons is significantly higher with Dnhe2 over-expression ( 7 . 8 ± 0 . 08 , n = 11 ) compared with wild type ( 7 . 5 ± 0 . 08 , n = 9 ) . ( *p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 006 We used the GMR-GAL4 driver to over-express Dnhe2 ( GMR > Dnhe2 ) in wandering third larval instar ( wL3 ) eye imaginal discs , a developmental stage with proliferating cells that marks the onset of retinal pattern formation . GMR > Dnhe2 expression induced an externally rough adult eye with phenotypes consistent with dysplasia in the underlying epithelium ( Figure 1B ) . In wild type Drosophila , the adult eye is a precisely patterned epithelial structure with hexagonal unit eyes , termed ommatidia , uniformly arrayed across the surface ( Figure 1B , C , left ) . Wild type ommatidia contain eight photoreceptor neurons ( R cells ) , forming a polarized , asymmetrical trapezoid ( Figure 1D , tracing in Figure 1E , left ) . In GMR > Dnhe2 flies , ommatidial organization was disrupted and individual facets were irregular in shape and size ( Figure 1B , C , middle ) . In section , GMR > Dnhe2 retinae had misoriented ommatidia , abnormal cell fates and missing R cells ( Figure 1D , E , middle ) . This phenotype is similar to disrupted tissue organization with expression of a short , unregulated isoform of Dnhe2 in a subset of retinal cells using sevGAL4 ( Simons et al . , 2009 ) . We also used eyelessGAL4 to express Dnhe2 earlier in eye development , which caused a very mild rough eye phenotype . To resolve whether increased H+ efflux by Dnhe2 is necessary for disrupted tissue architecture we generated transgenic flies expressing a mutant Dnhe2E358I , analogous to the NHE1E266I mutation that abrogates H+ efflux ( Denker and Barber , 2002 ) . Externally , GMR > Dnhe2E358I retinae had subtle defects in patterning , including rare bristle placement defects ( Figure 1B , C , right ) . In section , no patterning defects were evident ( Figure 1D , E , right ) , suggesting that dysplasia in GMR > Dnhe2 eyes is dependent on ion transport . To determine whether Dnhe2 over-expression increased pHi we generated transgenic flies expressing a genetically encoded , ratiometric mCherry-pHluorin pH sensor that was previously used to measure pHi in cultured cells ( Koivusalo et al . , 2010; Choi et al . , 2013 ) . After ∼24 hr of transgene expression there was no detectable difference in pHi between wild type and GMR > Dnhe2 in wL3 eye imaginal discs ( data not shown ) . However , in pupal eyes at 42 hr after puparium formation ( 42H apf ) , pHi in GMR > Dnhe2 was significantly higher than wild type in both apical non-neural cells ( Figure 1F , G ) and in photoreceptor neurons ( Figure 1—figure supplement 3 ) . In contrast , pHi in pupal eyes expressing the mutant GMR > Dnhe2E358I was not different than wild type ( Figure 1F , G ) , indicating that this mutant lacks H+ efflux . The delay between onset of transgene expression and increased pHi could be due to accumulation of functionally processed DNhe2 protein . These data suggest that patterning defects in GMR > Dnhe2 retinae reflect constitutively increased pHi , although we cannot rule out possible effects of decreased pHe or altered sodium levels . To determine the underlying cause of the adult rough eye phenotype , we examined GMR > Dnhe2 retinae at earlier time points in development . In wild-type pupal eyes , precise retinal organization is apparent on the apical surface with immunolabeling of the adherens junction protein beta-catenin ( Figure 2A , schematic drawing on right ) . Four central cone cells ( blue ) display stereotyped contacts , and are enwrapped by two semi-circular primary pigment cells ( orange ) to form the ommatidial core . Individual ommatidia are separated by a single layer of secondary pigment cells ( yellow ) , with tertiary pigment cells ( green ) and bristles ( purple ) at alternating vertices to form the hexagonal shape of ommatidia; these cells are collectively referred to as lattice cells . GMR > Dnhe2 pupal eyes had severely disrupted cell shapes and tissue organization , and ommatidia overall were smaller than in wild type . Ommatidia had aberrant numbers of cone and primary pigment cells ( pink ) . Some ommatidia were fused ( red ) , reflecting fused lenses seen in adult eyes ( Figure 1B ) . Filamentous actin organization ( labeled by rhodamine-conjugated phalloidin , purple , Figure 2A ) in wild type eyes showed increased labeling in cone cells and lattice cells compared to the primary pigment cells . In GMR > Dnhe2 pupal retinae , overall phalloidin levels were lower , and the differences between cell types were less pronounced . 10 . 7554/eLife . 03270 . 007Figure 2 . Dnhe2 over-expression disrupts cell shape and tissue organization and increases proliferation . ( A ) Confocal micrographs of pupal retinae show disruption of cell shapes and cell–cell contacts as outlined by beta-catenin labeling ( green ) and rhodamine-phalloidin ( magenta ) in wild type ( top ) and GMR > Dnhe2 pupal retinae ( bottom ) . Schematic diagrams are shown to the right with cell types designated by colors: cone cells ( blue ) , primary pigments cells ( orange ) , secondary pigment cells ( yellow ) , tertiary pigment cells ( green ) , and bristle cells ( purple ) . ( B ) Three-dimensional tissue architecture is shown through the depth of the epithelium . Cells are labeled as follows: DNA label Hoescht ( cyan ) ; rhodamine-phalloidin ( magenta ) ; pan neuronal marker Elav ( yellow ) . In wild type retinae , four cone cell nuclei are seen ( marked with ‘c’ ) and F-actin labeling in the center of ommatidia . 3 µm lower , the stereotyped photoreceptor cell ( R ) organization where R1/6/3/4 are labeled and F-actin staining in the center of each ommatidium is seen . In the third slice , 3 µm lower , R2 , 5 and 8 can be seen . And finally , on the basal side of the epithelium , the F-actin asters that mark the centers of each ommatidium and the single Elav-positive nucleus of the bristle complex found in wild type . ( C ) Immunolabeling of wL3 eye imaginal discs showing proliferating cells ( phosH3 , magenta ) and adherens junctions ( beta-catenin , green ) . ( D ) Quantification of proliferating cells posterior to the SMW ( mean ± SEM: wild type ( 9 . 4 ± 0 . 9 , n = 14 ) , GMR > Dnhe2 ( 16 . 7 ± 0 . 71 , n = 11 ) . ( E ) Confocal slices through whole mount wing discs with engrailed-GAL4 ( enGAL4 ) driver driving expression of GFP ( green ) , or GFP and Dnhe2 . Wing discs are labeled with rhodamine-phalloidin ( magenta ) to show tissue architecture . ( F ) An enlarged view of the wing pouch area , labeled with en > GFP ( green ) and phosH3 ( purple ) . ( G ) Percent of phosH3+ cells in the posterior compartment ( mean ± SEM: en > GFP [56 . 3 ± 1 . 5 , n = 4]; en > GFP , Dnhe2 [68 . 4 ± 1 . 2 , n = 4] ) . ( *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 007 Disorganization was present through the depth of the pupal retinal epithelium . At the level of cone cell nuclei ( Figure 2B , labeled ‘c’ ) , 4 nuclei per ommatidium are present in wild type , and phalloidin labels the center of each ommatidium . The number and arrangement of nuclei were aberrant with GMR > Dnhe2 expression , although phalloidin labeling appeared similar to wild type . Optical sections ∼3 µm deeper in the epithelium reveal R cell nuclei arranged in a characteristic pattern revealed by the neuronal antigen Elav , where four R cells ( R1/3/6/7 ) are visible in wild type . We saw strong phalloidin labeling in the center of each ommatidium , marking the nascent rhabdomeres . In GMR > Dnhe2 pupal eyes , the number and arrangement of R cells in this section were disrupted , and we did not see the phalloidin signal in the center of the ommatidia . Optical sections ∼3 µm deeper in wild type pupal eyes show the stereotyped arrangement of the other four R cells ( R2/5/7/8 ) . In GMR > Dnhe2 retinae , this organization was disrupted , and there are more Elav-negative nuclei present than in wild type . On the basal side of the WT pupal eyes , phalloidin-labeled asters mark the centers of ommatidia , and Elav-negative nuclei of lattice cells were seen . In GMR > Dnhe2 eyes , phalloidin-labeling was decreased and did not show the same pattern as in wild type . Additionally , Elav-positive nuclei persist in this layer , suggesting severe disorganization of R cells . There were also disorganized Elav-negative nuclei , which may be lattice cells . From these data , increasing pHi in the developing eye causes profound defects in tissue architecture that affect all cell types in the developing Drosophila eye , and shows severe disruption of patterning throughout this complex , multilayered epithelium . These phenotypes are consistent with dysplasia , and are similar to effects of expressing activated , oncogenic Raf , as shown below . Consistent with increased pHi being a conserved proliferative signal from yeasts ( Orij et al . , 2012 ) to mammals ( Pouysségur et al . , 1982; Kapus et al . , 1994; Putney and Barber , 2003 ) , cell proliferation was substantially increased with GMR > Dnhe2 expression . A regulated band of proliferation in wL3 eye imaginal discs termed the second mitotic wave ( SMW ) is present posterior to the morphogenetic furrow , a physical indentation in the epithelium that serves as a spatial and temporal benchmark of eye development . Posterior to the SMW , cells divide stochastically . With GMR > Dnhe2 expression , we found a significant increase in the number of proliferating phospho-histone H3 ( phosH3 ) positive cells within this posterior region ( Figure 2C , D ) . We also expressed Dnhe2 in the posterior compartment of the developing wing imaginal disc using an engrailed driver ( enGAL4 ) . In en > GFP , Dnhe2 expressing wing discs , there was overgrowth and dysplasia only in the expressing compartment , while non-expressing tissue resembled wild type ( Figure 2E ) . In wild type wing discs , the number of phosH3 positive cells was relatively similar in the posterior and anterior compartments ( Figure 2F , G ) . With en > Dnhe2 expression , however , this distribution was markedly different with significantly more proliferating cells in the posterior compared with anterior compartment ( Figure 2F , G ) . Together , these data suggest that Dnhe2 acts in a cell-autonomous manner to promote proliferation and dysplasia . NHE1 activity promotes directed cell migration ( Denker and Barber , 2002; Patel and Barber , 2005; Stock and Schwab , 2009 ) and invasion ( Stock and Schwab , 2009; Busco et al . , 2010 ) in cultured cells; however effects on these behaviors in vivo remain unresolved . We tested for synergy between Dnhe2 and oncogenic RasV12 using the ptc-GAL4 driver to express transgenes in a stripe of cells along the anterior/posterior compartment boundary in wL3 wing discs; this assay induces mosaic transgene expression to distinguish autonomous and non-autonomous effects , and can reveal effects on tissue organization , proliferation and invasive/migratory cell behaviors ( Vidal et al . , 2006; Petzoldt et al . , 2013 ) ( Figure 3 ) . Expression of RasV12 induced a wider ptc > GFP stripe compared to control ( ∼1 . 6-fold increase ) , while expression of Dnhe2 alone did not ( Figure 3A–C ) . However , co-expression of RasV12 and Dnhe2 markedly increased the ptc > GFP stripe width greater than twofold ( Figure 3C ) . We also observed a significant increase in the overall size of the wing disc with ptc > RasV12 , Dnhe2 compared with either transgene alone or wild type ( Figure 3A , D ) . To determine invasive phenotypes , we identified single GFP-positive cells in stacks of confocal images . In control and Dnhe2-expressing discs we found only a single instance of a few ( Rofstad et al . , 2006; Stock and Schwab , 2009; Zhang et al . , 2010 ) single GFP-positive cells , and no examples in RasV12-expressing discs ( Figure 3E ) . However , in ptc > RasV12 , Dnhe2 discs , we found examples of invasive cells in all discs examined . Further , we found several different types of invasive behaviors , including basal expansion of the ptc stripe , single invasive cells and streams of cells migrating out of the stripe into neighboring tissue in the posterior compartment ( Figure 3E ) . These data show cell autonomous effects of Dnhe2 expression on growth and migration behaviors , as well as non-autonomous effects on tissue growth . 10 . 7554/eLife . 03270 . 008Figure 3 . Dnhe2 co-expression enhances RasV12-mediated growth and invasion . ( A ) Confocal images of third instar wing imaginal discs expressing ptc > GFP plus indicated transgenes ( green ) and rhodamine-phalloidin ( purple ) fluorescence in XY maximum projection images and ( B ) XZ single slices below ( posterior to the right ) . ( C ) Quantification of ptc > GFP stripe width ( in µm; mean ± SEM: ptc > GFP [26 . 4 ± 1 . 3 , n = 6] , ptc > GFP , Dnhe2 [27 . 4 ± 4 . 0 , n = 6] , ptc > GFP , RasV12 [43 . 8 ± 4 . 5 , n = 6] , ptc > GFP , RasV12 , Dnhe2 [60 . 7 ± 3 . 9 , n = 7] ) . ( D ) Quantification of total area of wing imaginal discs ( in 103 pixels; mean ± SEM , ptc > GFP [137 . 3 ± 9 . 6 , n = 9] , ptc > GFP , Dnhe2 [107 . 8 ± 8 . 9 , n = 9] , ptc > GFP , RasV12 [93 . 5 ± 4 . 1 , n = 12] , ptc > GFP , RasV12 , Dnhe2 [196 . 5 ± 41 . 4 , n = 8] ) . ( E ) Invasive cells were identified as GFP-positive cells that were isolated from other GFP-positive cells in three-dimensional tissue architecture . For each example , shown are XY ( upper left ) , YZ ( upper right ) and XZ ( lower left ) single confocal slices . Invasive phenotypes were rarely seen in ptc > GFP , ptc > GFP , Dnhe2 discs , or ptc > GFP , RasV12 wing discs . In ptc > GFP , RasV12 , Dnhe2 discs , examples of basal stripe expansion ( 4/7 ) , multiple single , invasive cells ( 5/7 discs ) and streams of invading cells ( 2/7 ) were seen . ( *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 008 To address our second question on whether H+ efflux is necessary for oncogene-induced dysplasia , we generated precise deletions of the Dnhe2 using FLP-mediated recombination of FRT-bearing transposon insertions that flank the Dnhe2 coding region ( Figure 4A , B ) . This strategy avoided excision of a neighboring gene CG9257 that caused larval lethality in the existing Dnhe21 allele ( Simons et al . , 2009 ) . Deletions were screened by PCR by using primer pairs to test for the presence of coding sequences as well as for the recombinant P-element generated by recombination ( Figure 4B ) . Five independent excision lines were obtained , and all showed identical phenotypes . Homozygous Dnhe2null flies had ∼20% survival to eclosion ( Figure 4C ) , and surviving adult flies appeared morphologically normal . The lethality of Dnhe2null flies was rescued by expression of Dnhe2 using the muscle-specific mef2-GAL4 driver , but not using the neuronal elav-GAL4 driver ( Figure 4C ) . 10 . 7554/eLife . 03270 . 009Figure 4 . Deletion of Dnhe2 is semi-lethal with a genetic requirement in muscle . ( A ) Schematic diagram of Dnhe2 genomic locus , showing Dnhe2 exons ( black rectangles ) , neighboring genes CG9257 and Dap160 ( grey and white rectangles , respectively ) and P-elements P{XP}Nhe2[d05535] ( white triangle ) and PBac{WH}f02217 ( black triangle ) . Following FLPase-mediated recombination , all coding exons of Dnhe2 were excised , leaving a hybrid P-element ( black and white triangle ) . PCR primer pairs 1 and 2 are within Dnhe2 coding sequence , while primer pair 3 covers the recombinant P-element . ( B ) Sample PCR reactions from genomic DNA preps isolated from homozygous Dnhe2null , heterozygous and wild type adult Drosophila . Primer pairs 1 and 2 yield product only in wild type or heterozygotes . Primer pair 3 is generated in Dnhe2null homozygotes or heterozygotes . ( C ) Dnhe2null flies show 17 . 7% survival to adulthood . Rescue experiments that restore Dnhe2 expression in neurons ( Dnhe2null; elavGAL4/UASDnhe2 ) do not rescue ( 13 . 7% ) , but expression in muscles ( Dnhe2null; mef2GAL4/UASDnhe2 ) rescues to 43 . 6% survival . Controls ( GAL4 driver alone ) do not show any rescue effects with 16 . 0% survival for Dnhe2null; elavGAL4/+ and 8 . 2% survival for Dnhe2null; mef2GAL4/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 009 To test the effects of deleting Dnhe2 with expression of an activated oncogene we initially used GMR-GAL4 to drive activated RasV12; however GMR > RasV12 caused pupal lethality , which precluded phenotypic analysis in adult flies . We therefore induced expression of constitutively active , oncogenic Raf ( Brand and Perrimon , 1994 ) ( GMR > Raf ) . Retinal architecture in homozygous Dnhe2null flies was organized and regular , resembling wild type ( Figure 5A , B ) . The external surface of GMR > Raf eyes was disorganized with indistinct facets and a rough exterior retinal surface ( Figure 5A , B ) . Some ommatidia showed thin or absent lenses . In section , GMR > Raf had ommatidial defects with mis-orientation and occasional missing rhabdomeres ( Figure 5C , tracing in Figure 5D ) . Additionally , GMR > Raf rhabdomeres were morphologically distinct from wild type , appearing more rectangular , especially in R1 , R3 , R5 and R6 . Externally , Dnhe2null; GMR > Raf appeared identical to GMR > Raf ( Figure 5A , B ) , except that the eye color was noticeably lighter and yellowed . Tangential retinal sections revealed a complete loss of tissue architecture and cell structure , with histological features of coagulative necrosis , including loss of rhabdomere morphology , absence of distinct cell membranes and diffusion of pigment granules ( Figure 5C ) . Live , dissected whole-mount adult eyes revealed a significant reduction of the cohesive , red-pigmented retinal epithelial layer , such that the remaining Dnhe2null; GMR > Raf tissue was thin and transparent in appearance ( data not shown ) . These observations support a synthetic lethal interaction between loss of Dnhe2 and expression of oncogenic Raf . 10 . 7554/eLife . 03270 . 010Figure 5 . Synthetic lethality with oncogenic Raf and loss of Dnhe2 . ( A and B ) Scanning electron micrographs of the exterior surface of adult Drosophila eyes . Expression of GMR > Raf causes a rough eye phenotype in the absence and presence of Dnhe2 . ( C ) Histological sections reveal a trapezoidal arrangement of rhabdomeres in wild type and Dnhe2null adult eyes . GMR > Raf eyes show ommatidial rotation defects and abnormal photoreceptor recruitment . Histological sections through Dnhe2null; GMR > Raf eyes show a complete absence of tissue organization consistent with coagulative necrosis . ( D ) Traced diagrams show ommatidial orientation . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 010 Synthetic lethality of Dnhe2null; GMR > Raf was progressive , as determined by examining earlier developmental stages . In wL3 eye discs , Dnhe2null tissue morphology was indistinguishable from wild type ( Figure 6A ) ; however there were fewer proliferating cells than in wild type as indicated by phosH3 staining ( Figure 6A , B ) . Expression of GMR > Raf led to disrupted tissue morphology , with irregular ommatidial spacing and folding of the eye disc , presumably due to increased proliferation ( Figure 6A , B ) . Compared with GMR > Raf the morphology of Dnhe2null; GMR > Raf appeared less disrupted ( Figure 6A ) and hyperproliferation was suppressed ( Figure 6B ) . Pupal eyes were assayed 42H apf for expression of the neuronal antigen Elav , and we found that cells in all genotypes retain their fates as differentiated neurons ( Figure 6C ) . Beta-catenin immunolabeling showed distinct differences in the morphology of cell contacts ( Figure 6D , line drawing in Figure 6E ) . Dnhe2null pupal eyes appeared wild type: individual ommatidia had 4 cone cells and 2 primary pigment cells visible on the apical surface , and a single row of secondary pigment cells separating neighboring ommatidia . However , cell morphologies were slightly abnormal ( Figure 6D , E ) . Secondary pigment cells were thicker than in wild type , and cell shapes and cell–cell contacts were irregular . With expression of GMR > Raf , cell morphologies were profoundly aberrant , with extra and mis-shapen cone cells and ommatidia mis-aligned along the D/V axis , as seen in adults . Additionally , immature cone cell contacts were maintained and secondary pigment cells were absent between some pairs of neighboring ommatidia . These defects were enhanced in Dnhe2null; GMR > Raf pupal eyes , where all cell types appeared smaller and more cells had abnormal morphologies . Additionally , there were more ommatidia with extra cone cells . Immunolabeling pupal eyes for cleaved caspase 3 ( CC3 ) , a marker of apoptotic cells , showed no staining in wild type , Dnhe2null or GMR > Raf eyes . However , in Dnhe2null; GMR > Raf eyes there were a few CC3–positive cells in all retina examined , even though the normal phase of cell death concludes ∼12 hr before this time point ( data not shown ) . CC3 staining was not abundant enough to indicate widespread cell death at 42H apf . Our data suggest that Dnhe2 deletion causes progressive defects in Raf-expressing cells , which culminates in synthetic lethality prior to eclosion . 10 . 7554/eLife . 03270 . 011Figure 6 . Progressive synthetic defects with oncogenic Raf and loss of Dnhe2 . Phenotypic analyses are shown for the indicated genotypes at third larval instar ( wL3 ) and mid-pupal developmental stages . ( A ) wL3 eye discs labeled for phospho-Histone3 ( green ) to indicate dividing cells , and E-cadherin to show adherens junctions ( purple ) . ( B ) The number of phos-H3 cells posterior to the morphogenetic furrow were counted , and shown as % of wild type ( mean ± SEM ) : w1118 ( n = 12 ) , Dnhe2null ( 83 . 6 ± 4 . 2 , n = 8 ) , GMR > Raf ( 121 . 7 ± 3 . 3 , n = 8 ) , Dnhe2null; GMR > Raf ( 68 . 2 ± 3 . 8 , n = 7 ) . ( C ) In pupal eyes , Elav labels photoreceptor neurons in all genotypes . ( D ) In 42H apf pupal eyes , adherens junctions are labeled with βeta-catenin , with schematic drawing shown in ( E ) . Individual ommatidia have four central cone cells that are enwrapped by two primary pigment cells . These ommatidial cores are insulated from each other by a single row of interommatidial cells ( IOCs ) . This organization is disrupted with expression of GMR > Raf , and further enhanced with deletion of Dnhe2 . ( *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 011 Changes in pHi correlated with the progressive morphological defects described above . We measured pHi using the mCherry-pHluorin biosensor in live wL3 eye imaginal discs , approximately 24 hr after the onset of oncogenic Raf expression ( Figure 7A ) . Compared with wild type , the pHi in Dnhe2null eye discs was significantly lower and GMR > Raf was significantly higher ( Figure 7B ) . In contrast , pHi in Dnhe2null; GMR > Raf discs was not different than wild type ( p > 0 . 05 ) . This suggests that the increased pHi with expression of oncogenic Raf requires Dnhe2 activity . We also measured pHi in live 42H apf pupal eyes , and found that pHi in GMR > Raf was still significantly higher than wild type pupal eyes ( Figure 7C , D ) . In contrast to larval eye discs , however , the pHi of Dnhe2null; GMR > Raf was significantly lower than wild type and lower than GMR > Raf alone ( Figure 7C , D ) . These data suggest that pHi progressively decreases through pupal morphogenesis in Dnhe2null; GMR > Raf retinae , and this could contribute to the synthetic lethality observed in adult eyes . 10 . 7554/eLife . 03270 . 012Figure 7 . Dnhe2 deletion progressively decreases pHi in Raf-expressing but not wild type cells . pHi was determined using GMR > mCh-pH . ( A ) Single confocal slices from live , third larval instar eye imaginal discs showing fluorescence of pHluorin ( top row ) , mCherry ( middle row ) , and ratiometric images ( bottom row ) . Ratiometric images were generated by dividing the fluorescence intensity of pHluorin by mCherry . ( B ) pHi was calculated from ratio values using a standard curve and graphed ( mean ± SEM ) . Estimated pHi values in third larval instar retinal tissue are: wild type ( 7 . 1 ± 0 . 08 , n = 10 ) ; Dnhe2null ( 6 . 9 ± 0 . 03 , n = 11 ) ; GMR > Raf ( 7 . 4 ± 0 . 13 , n = 11 ) ; Dnhe2null; GMR > Raf ( 7 . 1 ± 0 . 06 , n = 9 ) . ( C ) Single confocal slices from live , pupal eyes , as described above . ( D ) Estimated pHi values in pupal retinal tissue are: wild type ( 7 . 1 ± 0 . 07 , n = 9 ) , Dnhe2null ( 7 . 3 ± 0 . 08 , n = 7 ) ; GMR > Raf ( 7 . 5 ± 0 . 12 , n = 7 ) ; Dnhe2null; GMR > Raf ( 6 . 8 ± 0 . 05 , n = 7 ) . ( *p < 0 . 05 , ***p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 012 We also found synthetic lethality in clonal cells with loss of H+ efflux and oncogene expression . In clonal human mammary MCF10A cells we found that loss of NHE1 activity lowers pHi and increases cell death selectively in cells expressing oncogenic RasV12 . We first used an estrogen receptor-induced H-RasV12 ( ER-RasV12 ) and suppressed NHE1 expression with shRNA . Tamoxifen-induced Ras expression markedly increased rate of pHi recovery from an acid load ( Figure 8A , Figure 8—figure supplement 1 ) . Tamoxifen increased pHi in ER-Ras cells and with NT shRNA but not in vector control cells ( Figure 8B ) . However , with NHE1 shRNA pHi was significantly lower ( Figure 8B ) , indicating NHE1 is necessary for increased pHi with oncogenic Ras . Tamoxifen treatment increased NHE1 expression ( Figure 8C ) , while NHE1 shRNA treatment significantly reduced NHE1 protein expression ( Figure 8C; two different NHE1 shRNA constructs showed similar results ) . A similar NHE1-dependent increase in pHi was shown using MCF10A cells stably expressing H-RasV12 ( sRasV12 ) and 5 ( N-ethyl-N-isopropyl ) amiloride ( EIPA ) to pharmacologically inhibit NHE1 activity ( Figure 8E , F ) . As a control , we show that EIPA treatment does not effect Ras expression ( Figure 8G ) . Although pHi was higher in MCF10A cells with RasV12 and inhibited NHE1 activity than in Drosophila Dnhe2null; GMR > Raf pupal eyes , measuring pHi of cultured cells requires washes that could remove less adherent , dying cells with a lower pHi . Quantifying dying cells by flow cytometry using fluorescently labeled AnnexinV showed NHE1 shRNA or EIPA significantly increased cell death with RasV12 ( p < 0 . 05 ) ( Figure 8D , H ) but not in control cells . These results combined with our in vivo findings in Drosophila indicate that limiting H+ efflux causes synthetic lethality with oncogene expression . 10 . 7554/eLife . 03270 . 013Figure 8 . Decreased NHE1 expression or activity with RasV12 expression prevents increased pHi in human MCF10A cells and shows synthetic lethality . ( A ) In MCF10A cells expressing ER-RasV12 the rate of pHi recovery from an acid load , an index of H+ efflux , is substantially greater in tamoxifen ( Tmx ) -treated controls ( solid black line ) and with non-targeting ( NT ) shRNA ( solid blue line ) than in the absence of Tmx ( dotted black and blue lines , respectively ) or in vector controls ( solid and dotted green lines ) . In NHE1 shRNA cells pHi recoveries are nearly abolished in the absence ( dotted red line ) or presence ( red solid line ) of Tmx . ( B ) Tamoxifen increases steady-state pHi in control ( mean ± SEM for all condition: control 7 . 41 ± 0 . 02 , +Tmx 7 . 65 ± 0 . 02 ) and NT shRNA control ( 7 . 45 ± 0 . 02 , +Tmx 7 . 65 ± 0 . 02 ) cells but not in NHE1 shRNA cells ( 7 . 36 ± 0 . 03 , +Tmx 7 . 20 ± 0 . 03 ) . ( C ) Expression of NHE1 is increased upon tamoxifen ( Tmx ) treatment of ER-RasV12 cells . Treatment of cells with NHE1 shRNA decreases NHE1 expression . ( D ) Cell death , indicated by Annexin V-positive cells is significantly greater with NHE1 shRNA in ER-RasV12 expressing cells ( 1 . 9× increase over NT shRNA ) , but not in control cells ( 0 . 87× ) . ( E ) The rate of pHi recovery from an acid load is substantially greater in MCF10A cells stably expressing RasV12 ( sRasV12 , red solid line ) compared with control ( WT , solid black line ) or vector expression ( solid blue line ) . The NHE1 inhibitor EIPA abolishes pHi recoveries in control ( dotted black line ) and sRasV12 cells ( dotted red line ) . ( F ) Steady-state pHi is significantly higher in sRasV12 cells compared with control cells . With EIPA treatment , steady-state pHi is significantly lower in sRasV12 ( control 7 . 72 ± 0 . 02 , +EIPA 7 . 27 ± 0 . 02 ) but not in control cells ( control 7 . 53 ± 0 . 03 , +EIPA 7 . 46 ± 0 . 03 ) . ( G ) sRasV12 expressing cells show increased expression of Ras compared to either control or vector cells . Ras expression is unaffected by inhibition of NHE1 . ( H ) Treatment with EIPA induces cell death only in sRasV12 expressing cells ( 2 . 3× increase over control ) , but not in vector ( 0 . 73× ) or control ( 0 . 95× ) cells . Sample were pooled from three independent cell preparations ( for NHE1 shRNA , two different shRNA constructs were tested in two cell preparations and pooled ) . ( *p < 0 . 05 , **p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 01310 . 7554/eLife . 03270 . 014Figure 8—figure supplement 1 . RasV12 expression increases NHE1 activity and pHi . ER-RasV12 expressing cells show increased NHE1 activity in the presence of tamoxifen ( Tmx , solid red line ) , but not without Tmx ( red dotted line ) . This effect is not seen in vector control cells ( black solid and dotted lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 014 Decreasing H+ efflux also increased lethality of invasive tumor cells having a complex mutational landscape compared with only H-RasV12 expression as described above . We used SW620 metastatic colorectal cells , which have an activating mutation in K-Ras ( G12V ) , mutations in the tumor suppressors SMAD4 ( frameshift ) , p53 ( R273H , P309S ) , and adenomatous polyposis coli ( APC ) ( Q1338* ) , as well as MDA-MB-231 invasive breast metastatic cells , which have a different activating mutation in K-ras ( G13D ) , a mutant B-Raf ( G464V ) and mutations in the tumor suppressors cyclin-dependent kinase inhibitor 2A ( deletion ) , and p53 ( R280K ) . In both cell lines , EIPA treatment for 72 hr significantly decreased pHi ( Figure 9A ) and increased cell death compared with controls ( Figure 9B ) . These data indicate that loss of H+ efflux limits survival of cells containing multiple oncogenic lesions and suggest a broad applicability of limiting H+ efflux to increase cell death across cancer types . 10 . 7554/eLife . 03270 . 015Figure 9 . Decreased NHE1 activity in tumor-derived cell lines induces cell death . ( A ) pHi is decreased with EIPA treatment in both SW620 metastatic colorectal tumor cells ( control 7 . 36 ± 0 . 02 , +EIPA 7 . 2 ± 0 . 03 ) , and MDA-MB-231 metastatic breast cells ( control 7 . 58 ± 0 . 03 , +EIPA 7 . 11 ± 0 . 01 ) . ( B ) Cell death as determined by a trypan blue exclusion assay is also increased in both cell lines with EIPA treatment over control: SW620 2 . 3×; MDA-MB-231 2 . 6× . Data are from four or five independent cell preps . ( *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03270 . 015
With the advent of low-cost whole genome sequencing , personalized chemotherapeutic treatments currently dominate discussion of new cancer therapies . However , the balkanization of cancers by causative mutation or tissue of origin ignores the shared disease physiology in most cancers as highlighted by the ‘hallmarks of cancer’ ( Hanahan and Weinberg , 2011 ) . Dysregulated pHi is an emerging hallmark of cancer that is associated with traditional hallmarks , including resisting cell death , sustaining proliferative signaling , activating invasion and metastasis , deregulating cellular energetics and tumor-promoting inflammation . Therefore , targeting pHi-regulatory proteins in cancer cells could address distinct pathological characteristics in parallel . Our data provide an in vivo functional corollary of the established increased pHi and upregulation of H+ efflux mechanisms in cancer cells . Building on previous reports that NHE1 activity and expression is increased in human cancers ( McLean et al . , 2000; Miraglia et al . , 2005; Chiang et al . , 2008; Yang et al . , 2011 ) , we show that engineered over-expression of an NHE1 ortholog in vivo increased pHi by ∼0 . 3 pH units , which is comparable to the increase seen with oncogene transformation . Moreover , over-expression of Dnhe2 was sufficient to induce dysplasia and hyper-proliferation phenotypes , and it enhanced oncogene-mediated invasion . Further , co-expressing RasV12 and Dnhe2 had a strong , synergistic effect on growth and invasion . A similar synergistic effect between expression of RasV12 and the C subunit of the V-ATPase , Vha44 was recently reported ( Petzoldt et al . , 2013 ) , which supports the ability of increased H+ efflux to enhance RasV12-induced phenotypes . However , Vha44 enhances invasion with RasV12 but not activated Src or Abl , suggesting an oncogene-specific synergy . Important to determine in future studies are the relative effects of increased pHi on growth and invasion , and whether decreased pHe contributes to the non-autonomous effects we observed in imaginal discs . Our data also indicate that reducing H+ efflux in cancer cells suppresses multiple pathological characteristics in parallel . Previous studies identified distinct responses to reduced H+ efflux , such as tumor growth in xenograft models ( Lagarde et al . , 1988; Sonveaux et al . , 2008; Colen et al . , 2011; Yang et al . , 2011 ) or viability of isolated cancer cells ( Rich et al . , 2000; Reshkin et al . , 2003 ) . We now show progressive effects of limiting H+ efflux with oncogene expression , including initially suppressing hyper-proliferation and dysplasia , followed by decreasing pHi , and culminating in a synthetic lethal interaction with histological features of coagulative necrosis . Coagulative necrosis is characteristically found within the central region of solid tumors ( Searle et al . , 1975; Lagarde et al . , 1988 ) , and is thought to be induced by extreme physiological cellular stresses . Removing Dnhe2 in an otherwise wild type genetic background decreased pHi but had no effect on larval tissue morphology . Dnhe2null retinal pHi in pupae was slightly higher than in larvae but not significantly different compared with wild type . These data suggest that in the absence of Dnhe2 , pHi homeostasis during metamorphosis can be maintained by alternative ion transport mechanisms , changes in cellular buffering capacity and/or metabolic changes . While our Drosophila studies evaluate synthetic lethality in the ~20% Dnhe2 homozygous mutant ‘escaper’ flies that survive to eclosion , our studies in several transformed cell lines also show a synthetic lethality interaction , suggesting a conserved mechanism . The synthetic lethal interaction we observed suggests H+ efflux is necessary for dissipating oncogene-generated acids , most likely from increased glycolytic metabolism . Our studies further show that these lethal interactions occur in two genetically complex , physiologically adapted invasive tumor cell lines . Our data build on the long-standing observation of increased pHi in cancer cells and suggest several new directions for future investigation . First , our findings indicate that the biosensor pHluorin can be used to resolve pHi dynamics in vivo during tumor formation and metastasis and whether there are spatially distinct pHi dynamics within a tumor that might inform us about tumor properties and plasticity . Second , the rough eye phenotype induced with GMR > Dnhe2 provides a new model for genetic screens to identify pH-sensitive pathways and previously unrecognized pH sensors regulating cell growth and tissue architecture . Dysplasia with increased pHi in the absence of oncogenes is likely in part dependent on cancer-promoting pH sensors ( Schönichen et al . , 2013 ) , including focal adhesion kinase ( FAK ) and the actin regulatory protein cofilin that support tumor growth and metastasis ( Wang et al . , 2007; Cance et al . , 2013 ) and have increased activity at higher pHi ( Frantz et al . , 2008; Choi et al . , 2013 ) . Third , the synthetic lethality we show suggests that therapeutic strategies targeting H+ efflux transporters or pH sensors with selective roles in oncogene-induced responses could limit metastatic progression with minimal off-target effects .
Standard Drosophila genetic techniques were used . Flies were cultured at 25°C unless otherwise noted . Stocks were obtained from the Bloomington Drosophila Stock Center: UAS-RasV12 ( 4847 ) , w1118 ( 5905 ) , mef2-GAL4 ( 27 , 390 ) , eyGAL4 ( 8219 , 8220 ) , UAS-Raf ( 2033 ) ; or the Harvard stock center: P ( XP ) Nhe2d05535 and PBac ( WH ) f02217 . Other stocks used were GMR-GAL4 ( gift from R Cagan ) and enGAL4 and ptc-GAL4 , UAS-GFP ( gifts from D Casso ) . Dnhe2 was originally cloned from a Drosophila S2 cell-derived cDNA library ( GenBank AF235935 ) . We isolated a 3 . 7 kb Dnhe2 cDNA from S2 cells that is essentially identical to the C isoform predicted in Flybase , with the following polymorphisms: A157T , T283I , insL568 , missing exon 17 ( 72 amino acids ) , 12 additional amino acids at the C terminus ( SKGEFQHTGGRY ) . Dnhe2 cDNA was cloned into pUAST , and transgenic flies were generated ( BestGene , Inc . , Chino Hills , CA ) and reported here as UAS-Dnhe2 . To generate UAS-Dnhe2E358I , the cDNA was amplified by PCR , adding a 5ʹ KpnI site and a 3ʹ HA-tag followed by an ApaI restriction site , and subcloned into pCR-TOPOXL . The E358I mutation was generated using Invitrogen QuikChange XL Site Directed Mutagenesis Kit and the following primers: Nhe2 E358I F: gtcgtctttgggatatccttgctgaacgatgccgtcacggttg . Nhe2 E358I R: caaccgtgacggcatcgttcagcaaggatatcccaaagacgac . The mutagenesis also introduced an EcoRV restriction site to facilitate colony screening . pEGFP-mCherry-pHluorin was obtained from S Grinstein ( Koivusalo et al . , 2010 ) . pEGFP-mCherry-pHluorin was digested with EcoRI and NotI , and a ∼1 . 5 kb mCherry-pHluorin containing fragment was gel purified and ligated into pUAST . Transgenic flies were generated ( BestGene , Inc . ) , and are herein referred to as UAS-mCherry-pHluorin . To generate Dnhe2null alleles , FLP-FRT recombination was used to generate w− deficiencies from Exelixis transposon insertions P ( XP ) Nhe2d05535 and PBac ( WH ) f02217 . Excisions were further screened by PCR for the presence of the resulting hybrid residual element ( XP:WH ) by PCR using element specific primers as described ( 51A and 5A1 ) ( Parks et al . , 2004 ) . Lethality was calculated as percentage of expected homozygous null flies to eclose as adults ( number of flies scored: Dnhe2null [1050]; Dnhe2null; mef2GAL4/UASDnhe2 [1231] ) . Adult heads were removed , bisected sagittally and fixed overnight in 2% glutaraldehyde + 4% paraformaldehyde in 0 . 1 M sodium phosphate buffer , pH 7 . 4 . Samples were processed in a critical point dryer , mounted on carbon tape ( Ted Pella , Redding , CA 16 , 084–6 ) or with colloidal silver ( Ted Pella , 16 , 034 ) on an SEM stub ( Ted Pella , 16 , 221 ) and splutter coated with gold-palladium ( Ted Pella , 22–2 ) . Images were collected on a JEOL Neoscope Scanning Electron Microscope , and cropped in Photoshop . Pupal and larval retinal and wing tissue were dissected in PBS , fixed in fresh 4% paraformaldehyde , washed in PBS + 0 . 1% Triton X-100 , immunostained and mounted in N-propyl gallate or Prolong Gold ( Life Technologies , Grand Island , NY , P36930 ) . Antibodies used were: mouse anti-beta-catenin ( cat# N2 7A1 ARMADILLO; 1:10 ) , rat anti-ELAV ( cat# Rat-Elav-7E8A10; 1:20 ) , rat anti-Ecadherin ( cat# DCAD2; 1:10 ) , rat anti-Elav ( cat# Rat-Elav-7E8A10; 1:20 ) , Developmental Studies Hybridoma Bank , Iowa City , IO; rabbit anti-cleaved-caspase 3 ( cat# 9661 , 1:100 ) , mouse anti-phospho-histone 3 ( cat# 9706 , 1:100 ) , Cell Signaling Technologies , Danvers , MA . In Figure 1 , cells were scored as posterior to the SMW if they lay posterior to the 10th ommatidial row behind the morphogenetic furrow . Continuous ptcGAL4-mediated expression of either RasV12 or Dnhe2 induced early larval lethality , therefore flies were raised at 18°C to limit transgene expression until the L2/L3 transition , when they were shifted to 25°C to permit transgene expression for 48 hr prior to dissection . ptc-GFP stripe widths were measured at the dorsal-ventral boundary . To identify invasive cells , confocal Z-stack images were analyzed in three dimensions using NIS Elements software ( Nikon ) . When potential invasive cells were identified , the location was marked with a crosshair , and XZ and YZ projections were examined to determine whether these GFP-positive invasive cells were isolated in three-dimensional space . Adult tissue dissection , fixation and resin embedding were performed as described ( Wolff , 2000 ) . Tissue was cut into 0 . 5 µm sections on a microtome and contrast stained with toluidine blue . Phase contrast images were collected on a Zeiss Axiophot microscope using a 63× oil objective and a Hamamatsu ORCA-ER digital camera . Images were cropped , pseudo-colored and brightness/contrast adjusted in Photoshop . Transgenic Drosophila lines expressing the pH sensor soluble mCherry-pHluorin ( Koivusalo et al . , 2010 ) ( UAS-mCherry-pHluorin ) were crossed to GMR-GAL4 to express the pH biosensor specifically in the developing eye . pH measurement were determined as described ( Grillo-Hill et al . , 2014 ) . Briefly , retinae were dissected from wandering third instar larvae or from 42H apf pupae in HCO3 buffer ( 25 mM NaHCO3 , 115 mM NaCl , 5 mM KCl , 10 mM glucose , 1 mM MgSO4 , 1 mM KHPO4 , 2 mM CaCl2 ) and mounted in MatTek glass-bottom dishes . Fluorescent images were collected on an inverted spinning disk confocal microscope ( Stehbens et al . , 2012 ) . NIS Elements software was used to subtract background and measure fluorescence intensity in pHluorin and mCherry channels . For larval measurements , fluorescence intensity measurements were collected from 3 ROI per eye disc , with each ROI comprising ∼7 ommatidia . For pupal measurements , fluorescence intensity values in single cells were measured for 10 cells per type per eye; cell types were identified by their stereotyped morphology at either the apical surface ( cone cells , primary , secondary and tertiary pigment cells ) or in the middle of the retinal epithelium ( photoreceptor neurons ) . Ratiometric calculations and statistical analyses were performed in Microsoft Excel . To generate a standard curve for each genotype , tissue was incubated for 20–40 min in nigericin buffer ( 25 mM HEPES , 105 mM KCl , 1 mM MgCl2 , 20 µM nigericin [Invitrogen] ) at defined pH points ( ∼6 . 5 , ∼7 . 0 and ∼7 . 5 ) . We did observe a difference in pHi in wild type between our two sets of experiments shown in Figure 1 and Figure 3 , which we attribute to alterations in our imaging systems made between the sets of experiments . Importantly , significant differences were reproducibly seen between genotypes in both series of studies . Estrogen-inducible H-RasV12 ( ER-RasV12 ) cells were a gift from Z Gartner . Control , stable H-RasV12-expressing ( sRasV12 ) and vector control MCF10A cells were a gift from J Debnath . Cell lines were authenticated by DNA sequencing ( DDC Medical , Fairfield , OH ) . Cells were cultured as described ( Debnath et al . , 2003 ) . SW-620 and MDA-MB-231 cells were obtained from ATCC ( Manassas , VA ) and cultured in DMEM-H21 high glucose media + 10% FBS at 37°C with 5% CO2; cells were serum-starved ( 0 . 2% FBS ) for experiments . NHE1 activity was determined as the rate of pHi recovery from an NH4Cl-induced acid load in a nominally HCO3--free buffer as described ( Denker et al . , 2000 ) . Cells were treated with the NHE1 inhibitor 5 ( N-ethyl-N-isopropyl ) amiloride ( EIPA ) at 10 µM for 48 or 72 hr prior to assay . For NHE1 shRNA treatment , lentiviral shRNA constructs ( TRCN0000044650 and TRCN0000044648 ) were purchased from Sigma–Aldrich ( Saint Louis , MO ) . After 2 days of lentiviral infection , cells were selected in 5 µg/ml puromycin for 72 hr and acid load selection was performed as described ( Pouysségur et al . , 1984 ) and plated for experiments within 2 weeks of initial lentiviral infection . pHi measurements were performed as described ( Denker et al . , 2000 ) . FITC-labeled AnnexinV was performed according to manufacturer's directions ( Becton , Dickinson and Company , Franklin Lakes , NJ , cat #560931 ) on a BD Accuri C6 Flow Cytometer . Trypan Blue labeling was performed according to manufacturer's directions ( Life Technologies ) and cells were counted using a hemocytometer . | An individual can develop cancer if cells in their body gain genetic mutations that enable the cells to divide more rapidly and move—or metastasize—to other tissues and organs . These mutations can alter the chemistry of the cell; for example , the inside of a cancer cell is much more alkaline ( has a higher pH ) than the inside of a normal cell . This helps the cancer cells to grow and divide rapidly , and move to other parts of the body , but it is not clear how important this change in pH within the cell is for the development of cancer . Previous studies have shown that many cancer cells increase the amounts or activities of the proteins that remove hydrogen ions from the cell , and so make the inside of a cell more basic . Here , Grillo-Hill et al . increased the amount of one of these proteins , NHE1 , in fruit flies to study its effects on healthy cells . The experiments showed that the larval cells that produced more NHE1 were more alkaline than normal cells , and this caused the cells to become less organized and grow more rapidly . Grillo-Hill et al . also found that NHE1 can work together with the oncogene RasV12 that promotes cancer development to enhance migration within the developing wings of the larvae . Next , Grillo-Hill et al . tested whether the change in pH is necessary for cells to become cancerous . Flies missing the gene that makes NHE1 appeared normal . However , if these cells also have the oncogene Raf—which , like RasV12 , promotes cancer—the inside of the cells gradually became so acidic that they died . Grillo-Hill et al . also found that blocking the activity of NHE1 in human cancer cells could lead to increased death but does not increase the death of normal cells . Grillo-Hill et al . 's findings show that increasing the internal pH of healthy cells leads to characteristics similar to those found in cancer cells . Furthermore , cancer cells can die if they become more acidic . This suggests that targeting pH levels could guide the development of treatments for cancer that selectively kill cancer cells while leaving normal cells unharmed . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2015 | Increased H+ efflux is sufficient to induce dysplasia and necessary for viability with oncogene expression |
Immunity to malaria is often considered slow to develop but this only applies to defense mechanisms that function to eliminate parasites ( resistance ) . In contrast , immunity to severe disease can be acquired quickly and without the need for improved pathogen control ( tolerance ) . Using Plasmodium chabaudi , we show that a single malaria episode is sufficient to induce host adaptations that can minimise inflammation , prevent tissue damage and avert endothelium activation , a hallmark of severe disease . Importantly , monocytes are functionally reprogrammed to prevent their differentiation into inflammatory macrophages and instead promote mechanisms of stress tolerance to protect their niche . This alternative fate is not underpinned by epigenetic reprogramming of bone marrow progenitors but appears to be imprinted within the remodelled spleen . Crucially , all of these adaptations operate independently of pathogen load and limit the damage caused by malaria parasites in subsequent infections . Acquired immunity to malaria therefore prioritises host fitness over pathogen clearance .
Mechanisms of host resistance can eliminate pathogens , but it is disease tolerance that functions to preserve life . Tolerance mechanisms of host defense do not have a direct impact on pathogen load and instead act to minimise tissue damage caused by the pathogen and the immune response targeting it . They also function to protect vital homeostatic processes , such as energy metabolism , under conditions of infection-induced stress ( Martins et al . , 2019 ) . We have taken tremendous steps to understand acquired resistance mechanisms that can eliminate pathogens and provide sterile immunity . On the other hand , it is unclear whether mechanisms of disease tolerance can persist after pathogen clearance to provide an alternative strategy of acquired immunity . And it is these mechanisms that are likely to be at the forefront of host defense when sterile immunity cannot be generated . We propose that immunity to severe life-threatening malaria is underpinned by acquired mechanisms of disease tolerance . The majority of malaria-induced deaths occur in children under the age of 5 infected with Plasmodium falciparum ( Weiss et al . , 2019 ) . A landmark prospective study in Tanzania followed 882 children from birth and showed that the risk of developing severe malaria is highest in the first few infections of life , and very few children ( <1 . 8% ) have more than one severe episode ( Gonçalves et al . , 2014 ) . These data therefore support the longstanding view that immunity against severe malaria is acquired rapidly – often before 12 months of age ( Gupta et al . , 1999; Marsh and Snow , 1999 ) . Crucially , this study further showed that children who survive severe malaria are frequently reinfected and experience episodes of febrile malaria with similar or even higher parasite densities ( Gonçalves et al . , 2014 ) . Immunity to severe forms of malaria is therefore not due to improved parasite elimination ( resistance ) but instead underpinned by the improved ability of the host to limit the pathological consequences of infection ( tolerance ) . There is a growing body of evidence that shows mechanisms of disease tolerance are required to survive a first malaria episode: for example , acute infection causes hypoglycaemia and so the ability to maintain blood glucose levels within dynamic range – in crosstalk with iron metabolism ( Weis et al . , 2017 ) – can determine life or death ( Cumnock et al . , 2018 ) . Furthermore , the induction of heme oxygenase-1 by nitric oxide ( Jeney et al . , 2014 ) leads to the detoxification of free heme , which is released from ruptured red cells ( Seixas et al . , 2009 ) – this protects renal proximal tubule epithelial cells and prevents acute kidney injury ( Ramos et al . , 2019 ) . Nonetheless , whilst these metabolic adaptations are doubtless essential for the survival of naive hosts there is no evidence as yet that they can be recalled and applied more effectively in subsequent infections to provide clinical immunity . Instead , the most effective way to induce persisting mechanisms of disease tolerance may be through host control of inflammation ( Medzhitov et al . , 2012 ) , which can limit collateral tissue damage and avert fatal metabolic perturbations . In malaria , the ability to control systemic inflammation may also minimise the detrimental effects of parasite sequestration by reducing activation of the endothelium and restricting available binding sites in the microvasculature ( Schofield and Grau , 2005 ) . Field data support the idea that dampening the inflammatory response to P . falciparum affords protection; lower levels of circulating pro-inflammatory cytokines are found in children in Malawi who survive severe malaria compared to those who succumb ( Mandala et al . , 2017 ) . And critically , inflammation can be reduced in the absence of improved parasite control; Ghanaian children in a high transmission area have higher parasite densities but less systemic inflammation and fewer febrile episodes than children in a lower transmission setting ( Ademolue et al . , 2017 ) . The blood cycle in malaria unleashes a plethora of parasite-derived as well as host tissue damage-associated signals , all of which can be sensed by innate immune cells . Additionally , pronounced changes in host physiology ( such as hypoxia and acidaemia ) are hallmarks of severe disease ( von Seidlein et al . , 2012 ) . Together these diverse signals trigger monocytes and macrophages to produce pro-inflammatory molecules , many of which have been associated with a poor prognosis including the prototypical myeloid-derived cytokines TNF and IL-6 ( Mandala et al . , 2017 ) . Macrophages can even directly cause severe anaemia through extensive bystander phagocytosis of healthy uninfected red cells in the spleen and impairment of erythropoiesis in the bone marrow ( Jakeman et al . , 1999; Pathak and Ghosh , 2016 ) . Given these key roles in pathogenesis , it may be possible to quickly reduce the risk of severe malaria by modifying the myeloid response to blood-stage infection . Importantly , an emerging body of literature shows that the response of myeloid cells is not hardwired but can be reprogrammed by pathogens and their products . This was first demonstrated by stimulating bone marrow-derived macrophages with LPS in vitro , which reduces the production of pro-inflammatory cytokines and increases the release of antimicrobial effector molecules upon re-stimulation ( Foster et al . , 2007 ) . The ability of monocytes and macrophages to adapt their response to repeated pathogen encounters through cell intrinsic modifications is termed innate memory ( Netea et al . , 2016 ) . And although myeloid cells are usually short-lived and terminally differentiated , memory can nevertheless be imprinted through the epigenetic reprogramming of either long-lived tissue-resident macrophages ( Wendeln et al . , 2018 ) or progenitor cells in the bone marrow ( Kaufmann et al . , 2018 ) . Innate memory has been shown to operate independently of pathogen load ( Dominguez-Andres and Netea , 2019; Seeley and Ghosh , 2017 ) and human monocytes produce less pro-inflammatory cytokines when stimulated after an episode of febrile malaria as compared to before ( Portugal et al . , 2014 ) , suggesting that they are intrinsically modified by infection . We therefore hypothesised that innate memory – leading to the functional specialisation of monocytes and macrophages to limit inflammation and associated pathology – offers the most compelling explanation for how immunity to severe malaria can be acquired so quickly and without the need for enhanced parasite control .
To ask whether malaria can functionally reprogramme myeloid cells , we must first understand their response to acute infection in a naive host; we started by mapping their dynamics in our severe model of disease . We found that the bone marrow quickly prioritises myelopoiesis by increasing the number of granulocyte monocyte progenitors ( GMP ) ( Figure 1D , see Supplementary file 1 for gating strategies ) . Consequently , an increased number of inflammatory monocytes and neutrophils are released into circulation and recruited into their target organ – the spleen ( Figure 1D and Figure 1—figure supplement 2A ) . Furthermore , megakaryocyte erythroid progenitors ( MEP ) appear de novo in the spleen ( Figure 1E ) ; this extramedullary mechanism of erythropoiesis likely represents a division of labour in an attempt to compensate for the loss of erythroid progenitors in the bone marrow ( Pathak and Ghosh , 2016 ) . We also observed major histological changes in tissue structure and integrity with reduced cellularity in the bone marrow contrasting starkly with marked splenomegaly , which was accompanied by a complete loss of organisation between red and white pulp ( Figure 1F and Figure 1—figure supplement 2B–C ) . Remarkably , we found that long-lived prenatally seeded tissue-resident macrophages in bone marrow and spleen ( Hashimoto et al . , 2013 ) rapidly disappear during acute infection ( Figure 1G ) . Since red pulp macrophages are the only cells that can store and recycle iron in the spleen their disappearance thus means that ferric iron , which can be revealed histologically with Prussian Blue staining , is completely absent at the peak of infection ( Figure 1—figure supplement 2D–E ) . We could further demonstrate that patrolling monocytes ( Carlin et al . , 2013 ) , often regarded as the tissue-resident macrophages of the vasculature ( Mildner et al . , 2017 ) , also disappear early in infection ( Figure 1G ) . These findings therefore place inflammatory monocytes at the centre of the acute phase response , since they now provide the only route through which to phagocytose and clear infected red cells . We therefore carefully characterised the fate and function of inflammatory monocytes in the spleen by RNA sequencing ( Figure 2A ) and used clueGO to reveal the complexity of their response to a first encounter with malaria parasites ( Bindea et al . , 2009; Mlecnik et al . , 2014 ) . ClueGO assigns significant gene ontology ( GO ) terms based on differential gene expression and groups them into functional networks by relatedness . When we merged all linked nodes into supergroups ( see Materials and methods ) we found that more than one third of all GO terms were related to host defence ( Figure 2B and C ) . Furthermore , clueGO identified interferon signaling as an upstream regulator of monocyte fate ( Figure 2B ) ; in agreement , interferon-inducible guanylate binding proteins were highly upregulated ( Figure 2—figure supplement 1A ) . We next used core lineage signatures to predict the likely outcome of monocyte differentiation in the spleen , where they can be instructed to become either inflammatory macrophages or monocyte-derived dendritic cells ( Menezes et al . , 2016 ) . This revealed that monocytes initiate a transcriptional programme that is typical of terminally differentiated inflammatory macrophages ( Figure 2D ) . They upregulate their ability to sense parasite- and host-derived danger signals by increasing transcription of diverse pattern recognition receptors ( Figure 2—figure supplement 1B ) , and upregulate expression of the hallmark cytokines and chemokines associated with a type I inflammatory response ( Figure 2E ) . Furthermore , they enhance their capacity to engage T cell immunity by upregulating all major components of the antigen processing and presentation machinery ( for class I and class II MHC ) , and attempt to fine-tune T cell activation by increasing their expression of co-stimulatory molecules and checkpoint inhibitors ( Figure 2F–G ) . Notably , a clear Warburg effect – the metabolic switch from oxidative phosphorylation to glycolysis described when monocytes are stimulated with LPS in vitro ( Cheng et al . , 2014 ) – was not observed in vivo in response to malaria . Instead , the key enzymes involved in both metabolic pathways were transcriptionally induced ( Figure 2—figure supplement 1C–D ) . We next looked at all of these parameters of monocyte and macrophage biology in our mild model of malaria . Surprisingly , despite substantial differences in parasite density and patterns of sequestration ( Figure 1B and Figure 1—figure supplement 1D ) the response of myeloid and progenitor cells in bone marrow , blood , and spleen was remarkably similar between AS and AJ ( Figure 2—figure supplement 2A–C ) . Moreover , the fate and function of spleen monocytes was essentially identical – a direct pairwise comparison identified only a single differentially expressed gene ( Kelch34 ) between the two models . In turn , clueGO analysis revealed that the four largest supergroups in monocytes isolated from AJ-infected mice ( Figure 2B–C ) also dominated the response to AS ( Figure 2—figure supplement 2D–E ) . It therefore appears that parasite genotype , pathogen load , and the sequestration of infected red cells in immune tissues does not fundamentally alter the myeloid response to acute infection . Instead , the haematopoietic switch that promotes myelopoiesis in the bone marrow ( and relocates erythropoiesis to the spleen ) , the disappearance of tissue-resident macrophages and the differentiation of monocytes into inflammatory macrophages may all be part of an emergency response that is unavoidable in a naive host . Malaria parasites can persist for many months ( or even years ) in humans ( Felger et al . , 2012 ) and the fitness costs of maintaining emergency myelopoiesis over these time frames would be exceptionally high . We therefore asked how the host adapts to an ongoing infection that can not be cleared . In the chronic phase of P . chabaudi , the pathogen load can reach up to 1000 parasites per μl blood ( Figure 1B ) and insoluble malaria pigment accumulates throughout the red pulp of the spleen ( Figure 1F ) . Despite this abundance of parasite-derived signals , the spleen stops extramedullary erythropoiesis and creates new structural demarkations between red and white pulp ( Figure 1E–F ) . Furthermore , the bone marrow reduces the production of granulocyte monocyte progenitors , which in turn reduces the number of inflammatory monocytes and neutrophils trafficking into the blood and spleen ( Figure 1D and Figure 1—figure supplement 2B ) . At the same time , resident macrophages begin to repopulate their tissue niches ( Figure 1G ) and ferric iron stores are re-established in the spleen ( Figure 1—figure supplement 2E ) . These data provide compelling evidence that naive hosts quickly adapt to tolerate malaria parasites and return the myeloid compartment towards a healthy uninfected baseline . In agreement , the transcriptome of monocytes during chronic infection is almost indistinguishable from uninfected controls ( Figure 2A ) . Monocytes are no longer programmed to differentiate into inflammatory macrophages upon their arrival in the spleen ( Figure 2D ) and they silence transcriptional signatures of the acute phase response ( Figure 2E–G and Figure 2—figure supplement 1A–B ) . Notably , we find no evidence that they engage mechanisms to suppress inflammation , such as regulatory cytokines ( IL-10 and TGFβ ) or inhibitors of T cell activation ( PD-L1 ) ( Figure 2E and G ) . And furthermore , we find no evidence that they have an alternative fate such as inflammatory hemophagocytes ( Figure 2—figure supplement 1E ) , which have been implicated in chronic anaemia ( Akilesh et al . , 2019 ) . Instead , monocytes simply adopt a state of quiescence during chronic infection despite persisting parasitaemia . Importantly , we can show that quiescence is reversible – when monocytes are removed from the spleen of chronically infected mice and stimulated in vitro with LPS their inflammatory response is comparable to monocytes from uninfected mice ( Figure 2—figure supplement 3A–B ) . In both cases , we clearly observe the Warburg effect and transcription of a plethora of inflammatory cytokines , chemokines and co-stimulatory molecules , together with the induction of Nos2 – the best biomarker of an inflammatory macrophage fate ( Figure 2—figure supplement 3C–D ) . Clearly , monocytes are not in a permanent refractory state during chronic infection; instead , their activation and differentiation in response to parasites and their pyrogenic products must somehow be silenced in the remodelled spleen . This likely represents just one mechanism through which the host minimises inflammation to resolve collateral tissue damage . We next asked whether tolerance could persist in the absence of live replicating parasites to provide long-lived protection . To answer this question , we developed a novel reinfection model that allowed us to exactly match parasite densities between first and second infection . To this end , mice were first infected with the avirulent parasite genotype AS to induce chronic recrudescing parasitaemia and then drug-treated after 40 days of infection to clear circulating and sequestered parasites . One month later , mice were infected for a second time but now with the virulent genotype AJ ( Figure 3A ) . In this model , parasite burden ( Figure 3B ) and the dynamics of red cell loss ( Figure 3C ) are both matched between first and second infection – this eliminates pathogen load as a confounding factor when analysing acquired mechanisms of disease tolerance . In contrast to a first malaria episode , the bone marrow does not prioritise the production of myeloid cells upon reinfection ( Figure 3D ) and preserves its cellularity and structural integrity ( Figure 3—figure supplement 1A ) . Consequently , the number of inflammatory monocytes and neutrophils released into circulation does not increase and nor does their accumulation in the spleen ( Figure 3E and Figure 3—figure supplement 1B ) . Furthermore , the spleen maintains its boundaries between red and white pulp and does not promote extramedullary erythropoiesis ( Figure 3D and Figure 3—figure supplement 1C ) . And whilst first infection obliterates tissue-resident macrophages these cells are resistant to malaria-induced ablation second time around ( Figure 3F ) ; this allows the host to retain ferric iron stores ( Figure 3—figure supplement 1D ) . A single malaria episode is therefore sufficient to disarm emergency myelopoiesis in the bone marrow and protect terminally differentiated macrophages in the spleen . What's more , tissue architecture is preserved and key homeostatic processes are maintained in tolerised hosts . Nevertheless , we reasoned that the myeloid compartment cannot be entirely quiescent during reinfection since mice are able to control replication of the virulent parasite genotype AJ . We therefore isolated spleen monocytes and examined their transcriptional programme in the acute phase of second infection . We identified more than 3000 differentially expressed genes and found that most of these were unique to reinfection ( Figure 4A ) ; remarkably , monocytes did not differentiate into inflammatory macrophages and all functions associated with this fate were silenced ( Figure 4B–C , Figure 4—figure supplement 1A–C ) . Host control of inflammation extended beyond the boundaries of the spleen with circulating levels of CXCL10 and IFNɣ also attenuated at the peak of second infection ( Figure 4D ) . We therefore explored alternative monocyte fates , such as those associated with immune regulation , wound healing , and tissue repair . However , spleen monocytes were not polarised towards alternatively activated macrophages and nor did they induce signature genes associated with myeloid-derived suppressor cells ( Gabrilovich , 2017 ) or reparative Ly6Clo monocytes ( Jung et al . , 2017; Figure 4E ) . Furthermore , they did not upregulate anti-inflammatory cytokines or checkpoint inhibitors ( Figure 4C and Figure 4—figure supplement 1C ) . To make sense of their complex transcriptional profile we therefore turned once more to clueGO . In contrast to first infection , we found minimal enrichment of GO terms linked to host defence; instead , all major supergroups in second infection related to regulation of cell cycle and nuclear division ( Figure 5A–C ) . This localised proliferation of monocytes may support a critical spleen response without having to engage increased production and recruitment from the bone marrow . So what is the function of spleen monocytes in tolerised hosts ? First off , they upregulate the expression of two alarmins ( S100a8 and S100a9 ) ( Figure 5D ) that form the heterodimer calprotectin; this acts as an endogenous TLR4 ligand to silence the inflammatory response to pathogen-derived or damage-associated signals . Notably , increased expression of these alarmins in early life prevents hyperinflammation to protect neonates from sepsis ( Ulas et al . , 2017 ) . Similarly , Chi3l1 ( encoding the prototypical chitinase-like protein ) is also upregulated in second infection and can attenuate inflammasome activation to minimise collateral tissue damage ( Dela Cruz et al . , 2012 ) . In both cases , these mechanisms promote host control of inflammation without impairing pathogen clearance . Monocytes also appear to minimise the effects of hypoxia and oxidative stress on their environment . For example , in first infection monocytes upregulate Sod2 , which encodes a mitochondrial protein required to safeguard inflammatory macrophages from the reactive oxygen species that they produce . But in second infection they instead upregulate Sod3 , which can be secreted to inactivate extracellular reactive oxygen species in the surrounding tissue ( Yao et al . , 2010; Figure 5D ) . In much the same way , monocytes minimise host stress by upregulating Ednra , which can scavenge endothelin-1 to prevent vasoconstriction and hypertension ( Czopek et al . , 2019 ) , and they increase expression of Vegfa to promote angiogenesis . Evidently , the transcriptional profile of spleen monocytes suggests that they take on a tissue protective role in second infection . In support of this argument , monocytes differentially regulate heme-iron metabolism to deal with the release of toxic free heme and reactive iron ( Fe2+ ) from ruptured red cells ( Figure 5E ) . In brief , they increase their transcription of a core set of genes required to sequester extracellular iron ( Ltf encoding lactotransferrin ) , import sequestered iron for detoxification ( Tfr1/Tfr2 encoding the transferrin receptors ) and then export detoxified iron for use in the production of new red cells ( Slc40a1 encoding ferroportin ) . Notably , monocytes do not appear to increase their iron storage capacity , which has been associated with tissue damage in malaria ( Gozzelino et al . , 2012 ) . And nor do they upregulate expression of heme oxygenase-1 ( Hmox1 ) , which is required to detoxify free heme . These data therefore indicate that monocytes specifically enhance iron recycling to promote tolerance to hemolysis – a major source of stress in malaria . Crucially , this only occurs during reinfection , despite an identical red cell loss in naive and tolerised hosts ( Figure 3C ) . Taken together , these data clearly show that spleen monocytes initiate a transcriptional programme designed to promote tolerance to malaria parasites upon reinfection . This is achieved in two ways – first by minimising inflammation to reduce collateral tissue damage and second by engaging pathways that can impart stress tolerance on their environment . Significantly , mice were first infected with the avirulent parasite genotype AS , suggesting that parasite-derived signals may be sufficient to redirect monocyte fate . We moved on to ask whether the fate and function of monocytes could be underpinned by metabolic reprogramming . Cellular metabolism has emerged as a key determinant of monocyte-to-macrophage differentiation ( O'Neill et al . , 2016 ) and so we looked again at transcriptional control of the key enzymes involved in glycolysis and oxidative phosphorylation . We found that both pathways were comparably induced during first and second infection with one notable exception: monocytes switched from upregulating the glucose transporter Slc2a1 in first infection to Slc2a3 in second infection ( Figure 2—figure supplement 1C–D and Figure 4—figure supplement 1D–E ) . Slc2a3 encodes the facilitative GLUT3 transporter , which unlike most other transporters can continue to import glucose under hypoglycaemic conditions ( Simpson et al . , 2008 ) . Switching to GLUT3 may therefore constitute a cell-intrinsic adaptation that allows monocytes to tolerate infection-induced stress . When we looked deeper into the transcriptional control of cell metabolism we found that monocytes also differentially expressed their mitochondrial carrier proteins ( Figure 5F ) . These membrane-embedded proteins provide the cellular wiring that connects metabolic reactions in the cytosol with the mitochondrial matrix by transporting metabolites , nucleotides , and co-enzymes across the inner mitochondrial membrane ( Palmieri , 2013 ) ; in this way , mitochondrial carrier proteins facilitate the complex crosstalk between all major metabolic pathways . In first infection , spleen monocytes primarily upregulate a carrier protein ( encoded by Slc25a1 ) whose major substrate is citrate , a metabolite known to accumulate in inflammatory macrophages . In contrast , monocytes upregulate Slc25a29 during reinfection and this carrier protein shuttles arginine , which can be fluxed through the arginase pathway to promote tolerance and wound healing ( O'Neill et al . , 2016 ) . By re-wiring their mitochondria , monocytes may thus be enabling their specialised tissue protective functions in tolerised hosts . Host control of inflammation may provide a quick and effective way to establish disease tolerance ( Medzhitov et al . , 2012 ) ; so far , we have shown that inflammation is minimised in the bone marrow ( preventing emergency myelopoiesis ) , blood ( decreasing plasma interferon ) , and spleen ( diverting monocyte fate ) . To directly show that this coincides with a reduction in pathology we measured circulating levels of Angiopoietin-2 . This vascular growth factor is the best biomarker of endothelium activation and dysfunction in human malaria and the most accurate prognostic marker of mortality in children ( Yeo et al . , 2008 ) . We found that in first infection Angiopoietin-2 levels increased by more than an order of magnitude but in second infection – with an identical parasite burden – levels did not deviate from a healthy uninfected baseline ( Figure 5G ) . A single malaria episode can therefore induce host adaptations that promote disease tolerance and provide long-lived clinical immunity . So how do tolerised hosts control inflammation independently of pathogen load ? To begin to answer this question , we looked once again at the functional specialisation of monocytes in second infection . Our data are consistent with a model of innate memory , whereby myeloid progenitors in the bone marrow are epigenetically reprogrammed during first infection to intrinsically modify the response of monocytes to reinfection . To test this hypothesis , we asked if malaria induces heritable histone modifications that alter the epigenetic landscape of inflammatory monocytes before their release from the bone marrow . And since tolerance can persist in the absence of parasitaemia , we isolated monocytes from once-infected mice one month after drug cure ( day 70 , see Figure 1A ) . Crucially , this was exactly the same time-point at which we had performed all of our reinfection studies ( Figure 3A ) . In this experiment , we interrogated the distribution of histone modifications genome-wide using ChIPseq and asked whether ( i ) transcription start sites were marked with H3K27ac to activate transcription ( ii ) enhancers or superenhancers were marked with H3K4me1 to promote gene expression or ( iii ) DNA was condensed into heterochromatin by H3K9me3 to silence gene expression . We used the motif discovery software HOMER ( Heinz et al . , 2010 ) to identify peaks and visualised peaks with the genomics exploration tool Integrative Genomics Viewer ( Thorvaldsdóttir et al . , 2013 ) . In the first instance , we looked at the histone modification profiles of genes that define monocyte function in first and second infection; for example , genes associated with inflammation versus proliferation . Subscribing to the notion that ChIPseq reveals qualitative ( not quantitative ) differences ( Ma et al . , 2018; Orlando et al . , 2014 ) , we simply asked whether these genes were marked or not marked . Our prediction was that genes that were upregulated during first infection but silenced during reinfection ( tolerised genes ) would lose marks associated with active transcription or be condensed into inactive heterochromatin . Conversely , specialised genes ( those upregulated exclusively during second infection ) would gain marks to promote transcription . Remarkably however , we found that in almost every case the histone modification profiles of tolerised and specialised genes were identical between monocytes isolated from once-infected mice and uninfected controls ( Figure 6A–B and Figure 6—figure supplement 1 ) . Even when we used HOMER to call differentially modified regions ( DMR ) and quantify differences between once-infected and control mice , we found little evidence to support epigenetic reprogramming of monocytes – of the 2848 tolerised/specialised genes identified by RNAseq 95% had no detectable histone modifications ( Figure 6C ) . And in those rare cases where a DMR was called , HOMER assigned a low confidence peak score ( Supplementary file 2 ) . Innate memory can not therefore easily explain the widespread transcriptional changes that lead to the functional specialisation of monocytes in tolerised hosts . An alternative explanation is that monocyte fate is imprinted within the spleen; after all , tissue printing is a key route to organ-specific identity during monocyte to macrophage differentiation ( Scott et al . , 2016; van de Laar et al . , 2016 ) . We therefore looked for transcriptional evidence of long-lived changes in spleen monocytes that can persist after parasite clearance . We found 111 differentially expressed genes in monocytes isolated from once-infected mice compared to uninfected controls; remarkably , most of these genes were not differentially expressed during acute or chronic infection . Instead , this transcriptional signature was unique to the memory phase and was further enhanced upon reinfection ( Figure 7A ) . This included the transcription factor Maf , which regulates macrophage programming in vivo ( Kang et al . , 2017; Liu et al . , 2020 ) , and Sirpa , which regulates recognition of self ( Bian et al . , 2016 ) . The majority of genes , however , related to cell cycle and nuclear division – specialised functions of spleen monocytes in tolerised hosts . Indeed , we found remarkable overlap in the top GO terms identified in memory and second infection ( Figure 7B ) . A critical part of the transcriptional programme designed to promote tolerance to malaria parasites is therefore already engaged in monocytes prior to reinfection . And this transcriptional signature does not appear to require epigenetic reprogramming in the bone marrow . These data thus provide compelling evidence that malaria may remodel the spleen to imprint tolerance .
In this study , we show that mechanisms of disease tolerance can persist after pathogen clearance to minimise tissue damage , stress and pathology during subsequent infections . These inducible mechanisms of tolerance therefore provide memory and constitute an alternative strategy of acquired immunity . Crucially , these adaptations function to preserve key homeostatic processes and protect life , and are therefore likely to be the primary form of host defense when sterile immunity can not be generated . This is particularly relevant in malaria , where even partial control of parasite densities is not usually demonstrable until adolescence ( Marsh and Kinyanjui , 2006 ) . It seems likely that many complementary strategies of tolerance will need to cooperate to provide protection and we identify three mechanisms in the myeloid compartment alone – emergency myelopoiesis is disarmed; tissue-resident macrophages become resistant to malaria-induced ablation; and spleen monocytes eschew an inflammatory macrophage fate to take on a tissue protective role . These adaptations operate independently of pathogen load and coincide with a reduction in systemic inflammation; what's more , tissue integrity is preserved , ferric iron stores are maintained and endothelium activation/dysfunction is avoided during reinfection . Acquired mechanisms of disease tolerance can therefore avert hallmark features of severe malaria . Host control of inflammation thus appears to provide an effective route to disease tolerance and monocyte activation is clearly not hardwired . However , their functional specialisation in tolerised hosts is not underpinned by epigenetic reprogramming of progenitor cells in the bone marrow but instead seems to be imprinted within the remodelled spleen . Tissue printing of immune cell function is well recognised , and allows monocytes to take on a remarkably diverse range of organ-specific roles . This has been best characterised during the differentiation of monocytes into long-lived tissue resident macrophages , in which local signals imprint specialised functions that are unique to every tissue ( Scott et al . , 2016; van de Laar et al . , 2016 ) . In this way , tissue printing maximises the plasticity of myeloid cells . One of many challenges when resolving the acute phase response to malaria is to refill the red pulp macrophage niche , which was obliterated by acute infection; this is achieved through the recruitment and differentiation of bone marrow-derived monocytes ( Lai et al . , 2018 ) . We find that red pulp macrophages isolated from once-infected mice are transcriptionally identical to those from uninfected controls ( Figure 7—figure supplement 1 ) , which shows that tissue printing can precisely match the transcriptional programme of recruited monocytes with the functional requirements of the tissue – in this case , to re-establish ferric iron stores . It is clear then that the tissue niche imprints organ-specific identity onto monocytes that take residence to maintain homeostatic processes . But similar mechanisms must also operate to control the fate of monocytes that do not take residence and instead are recruited to deal with injury or infection ( Guilliams et al . , 2018 ) . In this scenario , how might the tissue niche be remodelled by malaria to promote tolerance ? A niche is often considered to be a small self-contained tissue scaffold that provides the physical structure required to target specialised signals to resident cells or cells transiting the niche . Nevertheless , a niche can also be a dynamic and expansive space , such as the open circulatory system created by the pulp cords and sinuses of the spleen ( Guilliams et al . , 2020 ) . Malaria causes pronounced splenomegaly and disruption of architecture , including a complete loss of boundaries between red and white pulp . During chronic infection , splenomegaly begins to resolve and these boundaries are re-established but the spleen may not be put back exactly as it was before – it may acquire new and altered features . For example , malaria pigment accumulates throughout the red pulp , and these persistent insoluble deposits may regulate the activation and differentiation of monocytes arriving in the remodelled spleen . In support of this idea , malaria pigment can suppress the oxidative burst ( Schwarzer et al . , 1992 ) and interferon-induced upregulation of class II MHC ( Schwarzer et al . , 1998 ) in human monocytes in vitro . Alternatively , re-assembly of the stromal cell networks that compartmentalise the spleen may alter cell patterning and thereby change the distribution and density of key signals that control monocyte fate ( Bonnardel et al . , 2019; Koliaraki et al . , 2020 ) . Or perhaps the tissue niche required to promote tolerance is already present in the spleen ( even in a naive host ) but is lost when the spleen becomes enlarged and disorganised in first infection . In this example , it is the preservation of spleen architecture that facilitates the functional specialisation of monocytes in tolerised hosts . In every scenario , monocytes arriving in the remodelled spleen receive a new combination of signals that can minimise inflammation and induce their tissue protective functions ( see Figure 7C for a working model ) . Importantly , tissue remodelling and the accumulation of malaria pigment is observed in the human spleen ( Buffet et al . , 2011 ) , and we therefore suggest that this is a conserved strategy through which malaria imprints tolerance . This will almost certainly benefit the parasite as well as the host; after all , if you can minimise host sickness behaviour you will likely maximise the probability of onward transmission . Other changes to the niche include the appearance of malaria-specific memory B and T cells , which provide effective adaptive immunity in our reinfection model ( peak AJ parasitaemia is reduced from 25% in naive hosts to 0 . 5% in tolerised hosts ) . Nonetheless , this relatively low pathogen load is more than sufficient to promote the differentiation of spleen monocytes into inflammatory macrophages , as demonstrated in naive hosts infected with P . chabaudi AS . So how might adaptive mechanisms of immunity influence the myeloid response independently of pathogen load ? One route through which B cells may regulate monocyte fate is through the production of opsonising antibodies , which recognise merozoites and/or infected red cells . Indeed , it has been shown that phagocytosis of immune complexes in the context of Toll-like receptor signaling can induce M2 ( or regulatory ) macrophages ( Guilliams et al . , 2014 ) . However , in these experiments M2 macrophages increased production of the pyrogenic cytokines IL-1β , IL-6 and TNF , and malaria-specific opsonising antibodies have been directly shown to promote inflammation in human macrophages ( Osier et al . , 2014; Zhou et al . , 2012 ) . Instead , the generation of memory T cells that reduce the production of IFNɣ may be a more simple explanation; in agreement , we find that circulating levels of IFNɣ are substantially reduced in second infection . Nevertheless , signaling through the IFNɣ receptor is not a prerequisite for the differentiation of inflammatory macrophages ( Xue et al . , 2014 ) and the early source of IFNɣ in naive hosts is not T cell-derived ( Artavanis-Tsakonas and Riley , 2002 ) . Changes in cytokine production may therefore need to be much more extensive if T cells are to push monocytes away from an inflammatory fate . In this context , it is notable that malaria-specific memory T cells have been shown to adopt new effector functions upon recall ( Zander et al . , 2017 ) ; as such , they may be well placed to modify the availability of diverse signals within the tissue niche . Future studies should therefore aim to uncouple the relative contribution of T cell signals versus structural changes ( inc the appearance of malaria pigment ) in redirecting monocyte fate . This may be achieved by transplanting spleens from tolerised to naive hosts and then infecting with the avirulent parasite genotype AS . In this setting , it will be possible to a ) provide definitive evidence that monocyte fate is imprinted within the spleen and b ) assess the role of memory T cells by injecting donor mice with a depleting anti-CD3 antibody prior to transplantation . Interpretation of these experiments may be complicated however by the fact that recipient mice would still be expected to trigger emergency myelopoiesis and lose their tissue resident macrophages . Systemic inflammation ( and the disruption of homeostatic processes ) may override any local signals in the spleen that favour tolerance – reinfection is characterised by diverse adaptations in distinct tissues after all , and presumably these will need to work in unison to provide clinical immunity . It is therefore important that we consider how other host adaptations might be established . Emergency myelopoiesis can be disarmed in the bone marrow; how is this avoided second time around despite an identical pathogen load ? It could be that changes to the tissue niche promote tolerance , in much the same way as described in the spleen , as first infection leads to a loss of cellularity and tissue integrity in the bone marrow , which inevitably requires repair . But it is also possible that cell-intrinsic changes establish tolerance in this case . Current evidence indicates that systemic inflammation , which can push haematopoietic stem cells ( HSC ) down the myeloid path ( Mitroulis et al . , 2018 ) , is triggered in the bone marrow by plasmacytoid dendritic cells ( pDC ) , which produce type I interferon ( Spaulding et al . , 2016 ) . Can you intrinsically modify pDC to attenuate their function ( e . g . by increasing their activation threshold ) ? Or limit their access to parasites and their pyrogenic products ? Or perhaps tolerance operates further upstream – the expression of pattern recognition receptors on HSC could be restricted to prevent their direct activation by malaria parasites . Whatever the mechanism , we believe that disarming emergency myelopoiesis in the bone marrow is just as important as imprinting monocytes with tissue protective functions in the spleen . We also show that long-lived prenatally seeded tissue-resident macrophages become resistant to malaria-induced ablation . Tissue-resident macrophages were often considered to be immune sentinels that provide a first line of defence against invading pathogens but actually they have been shown time and again to be sensitive to stress , and their disappearance in response to injury or infection is well recognised ( Aegerter et al . , 2020; Blériot et al . , 2015; Machiels et al . , 2017; Scott et al . , 2016 ) . Their functions instead mainly relate to development , tissue repair , and homeostasis . So how does malaria cause the death of tissue-resident macrophages in naive hosts and why are they protected in tolerised hosts ? We suggest that the release and catabolism of free heme from ruptured red cells can explain this phenomenon . Free heme is toxic to tissue-resident macrophages ( Cambos and Scorza , 2011; Ferreira et al . , 2008; Haldar et al . , 2014; Theurl et al . , 2016 ) and the detoxification of free heme is known to confer disease tolerance ( Seixas et al . , 2009 ) . In our model , heme accumulation is likely to be similar between first and second infection as parasite burden and red cell loss are comparable . The preservation of tissue-resident macrophages could therefore be a consequence of more efficiently detoxifying free heme to reduce its bioavailability and pathogenicity . Since we find no evidence that monocytes transcriptionally regulate their capacity to metabolise heme during reinfection ( they exclusively upregulate iron recycling ) other cell types would need to carry out this role , including hepatocytes and renal proximal tubule epithelial cells ( Ramos et al . , 2019; Theurl et al . , 2016 ) . In this scenario , metabolic adaptations that operate in non-lymphoid tissues must also be inducible , and would persist after pathogen clearance to provide memory and clinical immunity . Acquired immunity to malaria is often considered to be slow and ineffective – but this is only true if we consider resistance ( the ability to eliminate parasites ) in isolation . Despite being an equally important part of host defense , the central role of disease tolerance in acquired immunity to malaria has not been fully appreciated . We demonstrate that disease tolerance can be attained after a single malaria episode and can persist in the absence of parasitaemia to allow the host to tolerate subsequent infections . We therefore argue that mechanisms of disease tolerance provide an alternative strategy of acquired immunity that functions not to kill parasites but to limit the damage that they can cause . Adaptations in the myeloid compartment minimise inflammation and promote stress tolerance , and it is likely that these will need to be complemented by adaptations in other immune compartments such as memory T cell reservoirs , which orchestrate innate and adaptive immunity . Furthermore , changes to the immune response will need to be complemented by long-lived metabolic adaptations that maintain homeostasis and preserve organ function . Together , all of these adaptations can work in concert to protect host tissues and minimise pathology . The mechanisms of disease tolerance uncovered in this study are therefore likely just the tip of the iceberg but they can begin to explain how children in endemic areas acquire immunity to severe malaria so quickly and without the need for improved pathogen control ( Gonçalves et al . , 2014; Gupta et al . , 1999; Marsh and Snow , 1999 ) .
All animal experiments were conducted in accordance with UK Home Office regulations ( Animals Scientific Procedures Act 1986; project licence number 70/8546 ) and approved by veterinarian services at the University of Edinburgh . C57Bl/6J mice , originally obtained from the The Jackson Laboratory , were bred and housed in individually ventilated cages under specific pathogen-free conditions . Mice had access to water and rat and mouse no . three breeding diet ( Special Diets Services ) at all times . Experimental procedures were initiated when mice were 8–10 weeks of age , following acclimatisation to a reversed 12 hr dark/light cycle ( lights OFF at 07:00 GMT and lights ON at 19:00 GMT ) . Mice were culled either by cervical dislocation or by pentobarbital overdose followed by exsanguination . We transmitted two serially blood passaged Plasmodium chabaudi chabaudi clones ( P . chabaudi AS ( clone 28AS11 ) and P . chabaudi AJ ( clone 96AJ15 ) ) , which were originally obtained from the University of Edinburgh ( http://www . malariaresearch . eu/ ) . Anopheles stephensi mosquitoes ( strain SD500 ) were reared in-house and infected with P . chabaudi according to our previously published protocol ( Spence et al . , 2012 ) . In brief , donor mice were inoculated with serially blood passaged P . chabaudi by intraperitoneal injection of infected red cells . Gametocytes were quantified on day 14 of infection and mice with >0 . 1% gametocytaemia were anaesthetised and exposed to female mosquitoes at 11:00 GMT . To ensure optimal parasite development , mosquitoes were kept at 26 . 0°C ( ±0 . 5°C ) in an ultrasonic humidity cabinet and provided with 8% Fructose and 0 . 05% 4-Aminobenzoic acid ( Sigma-Aldrich ) feeding solution from this point forward . Successful oocyst development was verified in mosquito midguts 8 days later and sporozoites were isolated from mosquito salivary glands on day 15 post-feed . Salivary glands were dissected under a stereomicroscope and transferred to a glass mortar; to maintain sporozoite viability , salivary glands were kept on ice in RPMI supplemented with 0 . 2% Glucose , 0 . 2% Sodium bicarbonate ( Sigma-Aldrich ) , 2 mM L-Glutamine ( Gibco ) , and 10% fetal bovine serum ( Gibco , FBS Performance Plus , heat inactivated and filtered 0 . 22 μm ) for a maximum of 2 hr . Sporozoites were released from salivary glands by gentle homogenisation and washed three times before enumeration . To initiate infection in experimental mice , 200 P . chabaudi sporozoites were intravenously injected into the tail vein . Following sporozoite injection , P . chabaudi develops in the liver for 52 hr before the release of merozoites to kick-start asexual blood-stage replication – each blood cycle takes approximately 24 hr to complete . Mice were closely monitored for the first 14 days of acute blood-stage infection with parasitaemia quantified daily using Giemsa stained thin blood films ( counting at least 10 , 000 red cells ) . Sickness behaviour and core body temperature ( digital rectal thermometer , TmeElectronics ) were also recorded daily . To assess anaemia , erythrocytes in 2 μl blood ( collected from the tail tip ) were counted using a Z2 Coulter Particle count and size analyser ( Beckman Coulter ) . We determined the healthy range in uninfected C57Bl/6J mice housed in our facility to be 8 . 8–10 . 5 × 109 erythrocytes*ml−1 . Anaemia was classified as severe when red cell loss exceeded 50% . Chronic infection was verified after 40 days of blood-stage parasitaemia by quantitative PCR of parasite 18S ribosomal DNA . DNA was extracted from 20 μl blood ( collected from the tail tip ) using Quick DNA Universal Microprep Kit ( Zymo Research ) , and amplified using TaqMan Universal PCR Mastermix ( ThermoFisher ) with 9 μM of forward primer ( 5-GCGAGAAAGTTAAAAGAATTGA-3 ) , 9 μM of reverse primer ( 5-CTAGTGAGTTTCCCCGTGTT-3 ) , and 2 . 5 μM of probe ( [6FAM] - AAATTAAGCCGCAAGCTCCACG - [TAM] ) on a Roche Lightcycler 480 ( 40 cycles of amplification ) . A standard curve of red cells spiked with known numbers of parasites allowed accurate quantification; mice with >5 parasites*μl -1 ( limit of detection ) were considered chronically infected with P . chabaudi , and all other mice were excluded from the study . Chronic infection was cleared using 100 mg/kg chloroquine diphosphate salt ( Sigma-Aldrich , dissolved in water ) administered by oral gavage daily for 10 days . Memory responses were assessed 30 days after the start of chloroquine treatment , and at this time-point once-infected mice were compared to uninfected age-matched controls that received the same schedule of chloroquine treatment . Mice were first infected with P . chabaudi AS by intravenous ( iv ) injection of sporozoites , and those with qPCR-confirmed chronic parasitaemia were chloroquine treated on day 40 of blood-stage infection . Thirty days after the start of chloroquine treatment mice were infected for a second time but now by iv injection of 5 × 105 mosquito-transmitted P . chabaudi AJ blood-stage parasites . Reinfection was initiated by this route to avoid confounding factors that may arise as a result of liver-stage immunity , which can be observed in C57Bl/6 mice after a single infection with P . chabaudi AS ( Nahrendorf et al . , 2015 ) . Note that uninfected age-matched controls received the same schedule of chloroquine treatment as reinfected mice . Platelet-depleted plasma was prepared from heparinised ( Wockhardt ) blood using two consecutive centrifugation steps ( 1000 xg for 10 min followed by 2000 xg for 15 min ) . Plasma was kept cold throughout and aliquots were stored at −80°C . We used commercially available ELISA kits to quantify plasma IFNɣ ( IFNɣ mouse ELISA kit , extra sensitive , Invitrogen ) , CXCL10 ( IP-10 mouse ELISA kit , Invitrogen ) , and Angiopoietin-2 ( mouse/rat Angiopoietin-2 quantine ELISA kit , R&D Systems ) . Absorbance was measured using a Multiskan Ascent ( MTX Lab systems ) or FluoSTAR Omega ( BMG Labtech ) plate reader . Spleens and femurs from P . chabaudi-infected mice and once-infected mice ( and uninfected age-matched controls ) were fixed in 10% neutral buffered Formalin ( Sigma-Aldrich ) for 24 or 48 hr , respectively . Bones were then decalcified for 48 hr using 10% EDTA ( pH 7 . 2 ) with gentle shaking at 55°C . After these steps , tissues were stored in 70% Ethanol and photographed to visualise macroscopic changes resulting from malaria . The spleen and both femurs from each mouse were paraffin-embedded in a single block and 5 μm sections were prepared ( cross-section for spleen and longitudinal section for bone ) . Sections were stained with Hematoxylin and Eosin ( H&E , ThermoFisher ) or Prussian Blue and Neutral Red ( Scientific Laboratory Supplies and VWR ) at the Shared University Research Facilities , University of Edinburgh . Stained slides were assessed using a Nikon A400a bright-field microscope and images were taken with a Zeiss 503 high-resolution colour camera . Images were cropped , white balance was adjusted and the brightness/contrast standardised between samples using Adobe Photoshop CS6 . To quantify the accumulation of infected red cells in the microvasculature of critical organs and tissues we used the same methodology as described in our sequestration and histopathology study ( Brugat et al . , 2014 ) . In brief , we infected mice by intraperitoneal injection of 105 mosquito-transmitted P . chabaudi AS or AJ blood-stage parasites and measured circulating parasitaemia at the peak of schizogony ( 13:00 GMT ) by counting at least 5000 red cells on Giemsa stained blood films . Immediately thereafter , mice were euthanised and exsanguinated – the spleen and both femurs were prepared for histology as described above ( ‘tissue preparation for histology’ ) . In addition , the left lobe of the liver , the left lung and kidney , the duodenum and the heart were fixed in 10% neutral buffered Formalin ( 24 hr ) and then stored in 70% Ethanol . All organs from each mouse were paraffin-embedded in a single block and 5 μm sections were cut and stained with H&E ( longitudinal section for bone and cross section for all other tissues ) . The percentage of infected red cells contained within the microvasculature of every organ was quantified by counting at least 1000 red cells in at least 20 blood vessels , and is displayed relative to peripheral parasitaemia . For high-resolution images of sequestered parasites in chabaudi malaria please see Brugat et al . , 2014 . Sodium heparin ( Wockhardt ) was used as anticoagulant for whole blood samples; spleens were dissociated in C-tubes using a gentleMACS Octo Dissociator ( Miltenyi Biotec ) ; and bone marrow was flushed from femurs using a 27½G needle/syringe loaded with IMDM . Single cell suspensions were filtered through a 70 μm cell strainer and after red cell lysis leukocytes were counted on a haemocytometer; up to 2 × 106 cells per well were placed into a 96 well V bottom plate for staining . A Zombie Aqua Fixable Viability Dye ( BioLegend ) was used to identify dead cells , after which Fc receptors were blocked using TruStain FcX ( anti-mouse CD16/32 , BioLegend ) . Cell surface staining was performed at room temperature ( antibody panels are detailed in Supplementary file 1 ) and for ChIPseq experiments cells were subsequently fixed in PBS with 1% paraformaldehyde and 10% FBS for 10 min at room temperature ( reaction was quenched with 125 mM Glycine [Sigma-Aldrich] ) . Note that across experiments the viability of leukocytes always exceeded 93 . 8% ( no viability stain for cell sorting ) . To confirm the identity of patrolling monocytes we performed an intracellular stain for the transcription factor Nr4a1 ( clone 12 . 14 , eBioscience ) using the FoxP3/Transcription Factor Buffer Staining Set ( eBioscience ) . Cells were acquired on an LSR Fortessa flow cytometer ( BD Biosciences ) or sorted using an Aria II cell sorter ( BD Biosciences , 85 μm nozzle , sort setting ‘purity’ ) . Both cytometers used BD FACS Diva v8 software and data were subsequently analysed using FlowJo v9 – all gating strategies are shown in Supplementary file 1 . Note that CD115 ( Csf1r ) was replaced with CD11c when sorting monocytes as engagement of the Csf1 receptor has been shown to induce transcriptional changes ( Jung et al . , 2000 ) . Samples with a sort purity <95% were excluded from the study . The absolute number of cells in each tissue was calculated from leukocyte counts of single cell suspensions . For bone marrow , we estimated that one femur contains approximately 11% of total mouse marrow ( Colvin et al . , 2004 ) and extrapolated accordingly . For whole blood , we recorded the volume collected during exsanguination and then extrapolated to total circulating blood volume , according to body weight . This approach allows a direct comparison of cell numbers across tissues . Red pulp macrophages ( Lineageneg F4/80pos B220neg CD11bint CD11cint autofluorescent cells ) were flow-sorted from the spleens of uninfected mice and collected into polypropylene tubes containing IMDM supplemented with 20% FBS and 8 mM L-Glutamine . Sorted cells were then spun ( 1000 xg for 5 min ) onto glass slides using a Shandon Cytospin 3 Cytocentrifuge ( ThermoScientific ) and stained with Prussian Blue and Neutral Red ( Sigma-Aldrich ) . Red pulp macrophages were visualised and photographed using a Leica DM1000 light microscope ( ×100 oil objective ) . 30 , 000 inflammatory monocytes ( Lineageneg Ly6Gneg CD11bpos CD11cneg Ly6Chi ) were flow-sorted from the spleens of chronically infected mice ( P . chabaudi AJ ) or uninfected controls and collected into polypropylene tubes containing IMDM supplemented with 5% FBS and 8 mM L-Glutamine . Following a gentle spin ( 450 xg for 10 min , slow brake ) monocytes were resuspended in 90 μl pre-warmed IMDM containing 10% FBS and 8 mM L-Glutamine , and transferred to an ultra-low attachment 96 well flat bottom cell culture plate ( Corning ) . To stimulate cells 0 . 3 ng LPS ( Lipopolysaccharide from Escherichia coli 0111:B4 , Sigma-Aldrich ) was added and cells were incubated for 4 hr at 37°C and 7% CO2 . RNA from both adherent and non-adherent cells was preserved in 1 ml TRIzol Reagent ( ThermoFisher ) for downstream steps . 10 , 000 inflammatory monocytes ( Lineageneg Ly6Gneg CD11bpos CD11cneg Ly6Chi ) were flow-sorted from the spleens of P . chabaudi infected mice , once-infected mice or uninfected controls and collected into 1 . 5 ml eppendorf tubes containing 1 ml TRIzol Reagent . Samples were inverted ten times , incubated at room temperature for 5 min and snap frozen on dry ice; all samples were stored at −80°C prior to RNA extraction . RNA was extracted using a modified phenol-chloroform protocol ( Chomczynski and Sacchi , 2006 ) with 1-Bromo-3-chloropropane and Isopropanol ( Sigma-Aldrich and VWR , respectively ) . Total RNA was quantified and assessed for quality and integrity by Bioanalyser ( RNA Pico 6000 Chip , Agilent ) – all sequenced samples had a RIN value above 8 . cDNA was generated from 2 ng total RNA using the SMART-Seq v4 Ultra Low Input RNA Kit ( Takara Bio ) and amplified using 11 cycles of PCR . Amplified cDNA was purified using Agencourt AMPure XP beads ( Beckman Coulter ) , quantified on a Qubit 2 . 0 Fluorometer ( dsDNA HS assay , ThermoFisher ) and quality assessed by Bioanalyser ( DNA HS Kit , Agilent ) . Libraries were then constructed from 150 pg of cDNA using the Nextera XT DNA Library Preparation Kit ( Illumina ) according to the manufacturer's instructions . Libraries were quantified by Qubit ( dsDNA HS assay ) and fragment size distribution was assessed by Bioanalyser ( DNA HS Kit ) . Using this information , samples were combined to create equimolar library pools that were sequenced on a NextSeq 550 platform ( Illumina ) to yield 75 bp paired-end ( PE ) reads; the median number of PE reads per sample passing QC across all experiments was 4 . 79 × 107 . FastQ files were downloaded from BaseSpace ( Illumina ) and raw sequence data assessed for quality and content using FastQC ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . We aligned paired-end sequences to the Ensembl release 96 murine transcripts set with bowtie2 v2 . 2 . 7 ( Langmead and Salzberg , 2012 parameters: —very-sensitive -p 30 —no-mixed —no-discordant —no-unal ) to obtain sorted , indexed bam files . Counts for each transcript were obtained using samtools idxstats ( http://www . htslib . org/doc/samtools . html ) and transcript counts were imported into the R/Bioconductor environment using the DESeq2 package ( Love et al . , 2014 ) for pairwise comparisons . Lists of differentially expressed transcripts were filtered to retain only those with an adjusted p value ( padj ) <0 . 01 and a fold change >1 . 5 using R v3 . 6; multiple transcripts annotated to the same gene were consolidated by keeping the transcript with the highest absolute fold change . Heatmaps and stacked circular bar charts were generated using the R ggplot2 package ( Wickham , 2016 ) . All RNAseq data are publicly available ( GEO accession number GSE150047 ) . Lists of differentially expressed genes were imported into clueGO v2 . 5 . 4 ( Bindea et al . , 2009; Mlecnik et al . , 2014 ) – a Cytoscape plug-in ( Shannon et al . , 2003 ) . ClueGO identified the significantly enriched GO terms ( GO Biological Process and GO Molecular Function ) associated with these genes and placed them into a functionally organised non-redundant gene ontology network based on the following parameters: padjcutoff = 0 . 01; correction method used = Bonferroni step down; min . GO level = 5; max . GO level = 11; min . number of genes = 3; min . percentage = 5 . 0; GO fusion = true; sharing group percentage = 40 . 0; merge redundant groups with >40 . 0% overlap; kappa score threshold = 0 . 4; and evidence codes used [All] . Each of the functional groups was assigned a unique colour and a network was then generated using an edge-weighted spring-embedded layout based on kappa score . We found that some GO terms were shared between multiple groups and so we manually merged these functionally connected groups to form supergroups , which we named according to the leading GO term ( lowest padj with min . GO level 5 ) . 50 , 000 fixed inflammatory monocytes ( Lineageneg Ly6Gneg CD135neg cKitneg CD11bpos CD11cneg Ly6Chi ) were flow-sorted from the bone marrow of once-infected mice or uninfected controls and collected into polypropylene tubes containing IMDM supplemented with 5% FBS and 8 mM L-Glutamine . Sorted monocytes were pelleted by centrifugation and washed in HBSS ( Gibco ) that was supplemented with protease inhibitors ( complete ULTRA Tablets Protease Inhibitor Cocktail , Roche ) and 5 mM sodium butyrate ( Alpha Aesar ) ; cell pellets were stored at −80°C . Note that for each biological sample we pooled the femurs and tibias from two mice , and three tubes ( each containing 50 , 000 cells ) were collected for every sample so that we could perform chromatin-immunoprecipitation ( ChIP ) with three different antibodies ( H3K27ac , H3K4me1 , and H3K9me3 ) . We performed ChIP using the True MicroChIP Kit ( Diagenode ) . In brief , chromatin was sheared using a Bioruptor Pico Sonicator ( five cycles: 30 s ON - 30 s OFF , fragments 100–300 bp , Diagenode ) and 10% of sheared chromatin was kept as input control whilst 90% was immunoprecipitated overnight using antibodies against H3K27ac , H3K4me1 or H3K9me3 ( all ChIP-grade from Diagenode ) . Protein-A-coated magnetic beads were then added to samples and after a 6 hr incubation unbound chromatin fragments were removed by thorough washing . ChIP and input DNA were decrosslinked and purified using MicroChIP DiaPure columns ( Diagenode ) . Libraries were prepared using the MicroPlex Library Preparation Kit v2 ( Diagenode ) and amplification was monitored in real-time on a LightCycler 480 ( Roche ) to ensure the optimum number of cycles was used . Amplified libraries were quantified by Qubit ( dsDNA HS assay ) and fragment size distribution assessed by Bioanalyser ( DNA HS Kit ) . After equimolar pooling of samples we purified libraries with AMPure XP beads and sequenced on a HiSeq 4000 ( 75 bp PE reads ) or NovaSeq S1 ( 100 bp PE reads ) ( both Illumina ) . A step-by-step guide to our optimised ChIPseq protocol is available at protocols . io ( dx . doi . org/10 . 17504/protocols . io . bja3kign ) . ChIPseq data quality and content were assessed using FastQC; all samples passed initial QC and were aligned to the mm10 genome using bowtie2 v2 . 2 . 7 ( parameters: —very-sensitive -p 30 —no-mixed —no-discordant —no-unal ) . We then used the motif discovery software HOMER ( v4 . 10 Heinz et al . , 2010 ) to turn indexed bam files ( generated using samtools idxstats ) into tag directories of individual ChIP and input samples . Alignments were converted to bedgraph format using the HOMER script makeUCSCfile . Wig format outputs were converted to tdf files to view data in the Integrative Genomics Viewer ( IGV v . 2 . 7 . 2 Thorvaldsdóttir et al . , 2013 ) using igvtools v2 . 3 . 93 ( parameters: toTDF -z 7 f p98 ) . In HOMER , we identified areas of the genome where ChIP read counts were significantly enriched over fragmented , non-immunoprecipitated input DNA ( which indicates the presence of a histone mark ) by calling peaks in ChIP relative to sample-matched input DNA using predefined parameters ( H3K27ac and H3K9me3 used ‘regions’ and H3K4me1 used ‘typical’ and ‘supertypical’ ) . Default settings were used in every case with the exception of fold change over input , which was set to >3 fold for H3K9me3 . Individual samples were then combined in HOMER to create pooled ChIP tag directories . We selected only samples with a high IP efficiency ( >5% for H3K27ac and H3K9me3 , and >10% for H3K4me1 ) to generate pooled tag directories that comprised: four biological replicates for H3K27ac; 2 ( once-infected ) or 3 ( uninfected ) biological replicates for H3K4me1; and one replicate for H3K9me3 . Bedgraphs of pooled ChIP tag directories were generated and converted to tdf format as above using more than 2 . 2 × 108 tags for H3K27ac , 1 . 1 × 108 tags for H3K4me1 and 7 × 107 tags for H3K9me3 . Peaks were called on pooled ChIP samples relative to pool-matched input DNA using the parameters described above . We created tdf files that indicated the position of peaks across the genome with a fixed height bar and visualised the histone modification profile alongside peak location in IGV . To ask whether genes were marked or not marked we looked for the presence or absence of peaks within 10 kb ( H3K27ac and H3K9me3 ) or 100 kb ( H3K4me1 ) of the transcription start site . To identify differentially modified regions ( DMR ) across the genome we again called peaks on pooled ChIP samples but this time instead of using non-immunoprecipitated input DNA to correct for background we called peaks in once-infected mice relative to uninfected controls ( and vice versa ) . In this way , we identified areas of the genome where read counts were significantly enriched in one or the other experimental group . Low confidence peaks were removed by applying a peak score cut off >3 and DMR were annotated to the nearest gene using the script annotatePeaks in HOMER . We then asked how many of the 2848 tolerised/specialised genes identified by RNAseq were annotated with a differentially modified region ( if a gene was annotated with more than one DMR then the region with the highest peak score was retained ) . All ChIPseq data ( inc individual biological replicates and pooled tag directories ) are publicly available ( GEO accession number GSE150478 ) . 10 , 000 red pulp macrophages ( Lineageneg F4/80pos B220neg CD11bint CD11cint autofluorescent cells ) were flow-sorted from mice 100 days after self-resolving P . chabaudi infection or from uninfected age-matched controls . Sorted cells were collected into 1 . 5 ml eppendorf tubes containing 1 ml TRIzol Reagent ( ThermoFisher ) and samples were stored at −80°C prior to RNA extraction . RNA was extracted using a modified phenol-chloroform protocol ( Chomczynski and Sacchi , 2006 ) and treated with Baseline-ZERO DNase to remove genomic DNA ( Illumina ) . DNase-treated RNA was then purified using the RNA Clean and Concentrator-5 Kit ( Zymo Research ) and total RNA was quantified and assessed for quality and integrity by Bioanalyser ( RNA Pico 6000 Chip , Agilent ) . RNA samples were processed for gene expression analysis using the GeneChip WT Pico Kit and Mouse Gene 1 . 0 ST Array ( Affymetrix ) according to the manufacturer's instructions . Microarray data were processed in R/Bioconductor making use of the oligo , pd . mta . 1 . 0 and mta10sttranscriptcluster packages . Data quality was assessed using the arrayQualityMetrics package ( Kauffmann and Huber , 2010 ) ; all samples passed QC and were normalised using robust multi-array average ( RMA ) , which results in log2 expression intensities . Limma ( linear models for microarray data ) and eBayes packages were used for pairwise comparisons to find differentially expressed genes ( DEG ) between groups ( AS vs uninfected , AJ vs uninfected and AJ vs AS ) but yielded zero DEG in all comparisons ( padj <0 . 05 ) . Log2 expression intensities of signature genes for monocytes , tissue resident macrophages and dendritic cells were plotted using the heatmap . 2 ( ) function in R; these genelists were manually compiled from published studies that set out to identify the gene expression profiles that underpin identity in myeloid cells ( Gautier et al . , 2012; Haldar et al . , 2014; Miller et al . , 2012; Okabe and Medzhitov , 2014 ) . Microarray data are publicly available ( GEO accession number GSE149894 ) . All RNAseq , ChIPseq , and microarray data have been deposited in NCBI's Gene Expression Omnibus ( Edgar et al . , 2002 ) and are accessible through GEO SuperSeries accession number GSE150479 . | Malaria is a parasitic infection spread by mosquitoes that causes hundreds of millions of cases each year . People are most likely to die from malaria the first time they are infected – usually when they are young children . Among those who survive , however , few will develop severe symptoms again , even though they are often reinfected with as many ( or even more ) parasites . This indicates that people do not get better at eliminating the parasite . Instead , protection from severe malaria is a form of tolerance - the body learns to limit the damage the infection causes . But exactly which mechanisms have to be engaged to tolerate malaria is unclear . One way to achieve tolerance may be to switch off damaging inflammation . Nahrendorf et al . explored this possibility by comparing the immune response of mice to their first and second infection with malaria parasites . During the first infection of life , immune cells release harmful inflammatory molecules that activate the lining of blood vessels , causing tissue damage and severe symptoms . During the second infection , these immune cells shut down inflammation and instead actively promote tissue health to reduce damage and improve outcome . This change in the immune response occurs despite the fact that the number of parasites is the same in both infections . Nahrendorf et al . also found that the mouse’s immune cells ’remembered’ to tolerate subsequent infections , even after treatment with a drug that kills all malaria parasites . This was possible because malaria permanently altered the spleen , which reprogrammed the response of the immune cells . A single infection is therefore enough to induce long-lived mechanisms of tolerance that can prevent life-threatening disease . These findings have the potential to change the understanding of immunity to malaria , which currently emphasises the importance of killing parasites . New ways to treat and vaccinate people - and to protect young children from severe malaria - may arise by treating tolerance as an equally important form of host defense . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2021 | Inducible mechanisms of disease tolerance provide an alternative strategy of acquired immunity to malaria |
Numerous challenges have impeded HIV-1 vaccine development . Among these is the lack of a convenient small animal model in which to study antibody elicitation and efficacy . We describe a chimeric Rhabdo-Immunodeficiency virus ( RhIV ) murine model that recapitulates key features of HIV-1 entry , tropism and antibody sensitivity . RhIVs are based on vesicular stomatitis viruses ( VSV ) , but viral entry is mediated by HIV-1 Env proteins from diverse HIV-1 strains . RhIV infection of transgenic mice expressing human CD4 and CCR5 , exclusively on mouse CD4+ cells , at levels mimicking those on human CD4+ T-cells , resulted in acute , resolving viremia and CD4+ T-cell depletion . RhIV infection elicited protective immunity , and antibodies to HIV-1 Env that were primarily non-neutralizing and had modest protective efficacy following passive transfer . The RhIV model enables the convenient in vivo study of HIV-1 Env-receptor interactions , antiviral activity of antibodies and humoral responses against HIV-1 Env , in a genetically manipulatable host .
Numerous challenges have impeded the development of a vaccine that protects against HIV-1 infection . Perhaps the most important of these are intrinsic obstacles to the elicitation of antibodies that recognize the trimeric HIV-1 envelope ( Env ) spike and inhibit viral replication ( Burton and Mascola , 2015; Escolano et al . , 2017 ) . Large portions of the HIV-1 Env trimer are conformationally flexible and shielded by glycosylation , and such properties inhibit recognition by antibodies ( Burton and Mascola , 2015; Escolano et al . , 2017 ) . Additionally , large HIV-1 in vivo population sizes and short generation times , accompanied by error prone replication ( ~10−4/base/cycle ) and recombination , means that vast numbers of sequence variants are generated every day in each infected individual ( Coffin , 1995 ) . Thus , while infected individuals generate strain-specific neutralizing antibodies that impose selective pressure on viral populations and influence viral sequence evolution ( Richman et al . , 2003; Wei et al . , 2003 ) , the discrepant evolutionary rates that characterize HIV-1 Env and antibody-generating B-cells ensures that antibodies present in each individual are generally poorly effective against contemporaneous autologous viruses ( Richman et al . , 2003; Wei et al . , 2003 ) . Moreover , HIV-1 sequence diversification has occurred in millions of individual humans over approximately 100 years , yielding vast global diversity of HIV-1 Env proteins ( Korber and Gnanakaran , 2009 ) . This large and evolving population of HIV-1 Env proteins with intrinsic antibody evasion mechanisms makes the elicitation of broadly effective antibodies by vaccines a formidable task ( Mascola and Haynes , 2013 ) . Nevertheless , rare HIV-1 infected individuals generate potent , broadly neutralizing antibodies ( bNAbs ) that are capable of neutralizing many circulating HIV-1 strains ( Klein et al . , 2013; Wu et al . , 2010 ) . However , they typically arise only after years of infection ( Landais and Moore , 2018 ) . The breadth with which bNAbs neutralize HIV-1 strains is likely a function of their rarity , as any frequently occurring bNAbs would drive frequent resistance ( i . e . loss of activity against circulating strains ) . What distinguishes the rare individuals who generate bNAbs , and whether it is possible to generate bNAbs in a significant fraction of humans through vaccination , are key issues confronting the HIV-1 vaccine research field ( Landais and Moore , 2018 ) . Another significant impediment to HIV-1 vaccine development is the availability of a convenient animal model system in which to study antibody elicitation and efficacy ( Hatziioannou and Evans , 2012 ) . HIV-1 host range is confined to humans and chimpanzees , severely curtailing options for testing vaccines and other prevention strategies . To partly circumvent this problem , chimeric retroviruses based on simian immunodeficiency virus ( SIV ) that express HIV-1 Env proteins , termed ( simian HIVs or SHIVs ) , have been developed ( Li et al . , 1992; Luciw et al . , 1995 ) . Following engineering and a period of adaptation to overcome the sub-optimal use of macaque CD4 receptors , these viruses can often replicate persistently and cause AIDS-like disease in macaques ( Del Prete et al . , 2017; Joag et al . , 1996 ) . Additionally , particular minimally modified HIV-1 strains have been adapted to replicate in pig-tailed macaques ( Hatziioannou et al . , 2014; Schmidt et al . , 2019 ) . Though useful , these models require the significant resources associated with investigations in primates . Small animal models based on immunodeficient mice engrafted with human cells and tissues have provided an experimental system for the in vivo testing of antibodies and other molecules as preventative agents ( Hatziioannou and Evans , 2012; Mosier et al . , 1991; Namikawa et al . , 1988; Watanabe et al . , 2007; Wege et al . , 2008 ) . However , these models are also inconvenient and costly , and human cell engrafted mice generate weak and inconsistent immune responses . Moreover , an important advantage of mouse models , that is the ability to genetically manipulate their immune systems , is lost when the viral target cells and immune system are derived from a human graft . Here , we describe the development of a virus/host animal model that incorporates the critical feature of the HIV-1 viral particle ( the Env spike ) , that is the target of antiviral antibodies , and recapitulates key features of HIV-1 entry and tissue tropism . Specifically , we generated recombinant derivatives of the rhabdovirus , vesicular stomatitis virus ( VSV ) , in which the native envelope glycoprotein ( G ) is replaced by HIV-1 Env from various subtypes , including transmitted founder strains . In these Rhabdo-Immunodeficiency viruses ( RhIV ) , replication is entirely dependent on HIV-1 Env , as well as human CD4 and coreceptors on target cells . In parallel , we constructed transgenic mice that express human CD4 and CCR5 , exclusively in mouse CD4-positive cells , at levels mimicking those on human CD4+ T-cells . Infection of these transgenic mice with RhIVs results in rapid , specific depletion of CD4+ T-cells and an acute viremia that resolves , followed by development of antibodies directed against the HIV-1 envelope . The RhIV model thus enables the convenient in vivo study of HIV-1 Env-receptor interactions , and their inhibition by antibodies in a genetically manipulatable host .
HIV-1 itself cannot replicate in murine cells , even when they are engineered to expressed human versions of HIV-1 receptors and essential Tat cofactors that enable HIV-1 entry and transcription in murine cells ( Bieniasz and Cullen , 2000; Mariani et al . , 2000 ) . However , VSV has a very broad tropism and numerous VSV-based chimeric viruses expressing functional , heterologous envelope proteins , including that of HIV-1 , have been generated ( Boritz et al . , 1999; Johnson et al . , 1997; Rabinovich et al . , 2014; Rose et al . , 2001 ) . We generated a panel of chimeric Rhabdo-Immunodeficiency viruses ( RhIVs ) encoding Env proteins from diverse HIV-1 strains including those representative of strains circulating in human populations . Initially we constructed RhIV strains encoding the subtype B Env proteins of HIV-1NL4-3 , a laboratory adapted X4-tropic strain , HIV-1ADA , a macrophage tropic primary isolate and HIV-1AD17 , an R5 tropic transmitted founder ( T/F ) HIV-1 strain . The ectodomain and transmembrane domains of HIV-1 Env were fused to the cytoplasmic tail of VSV-G ( Figure 1A ) . Retaining the VSV-G cytoplasmic tail has the potential to perturb the tertiary structure of the HIV-1 ectodomain , but this strategy improves HIV-1 Env incorporation into VSV particles ( Johnson et al . , 1997 ) . RhIVNL4-3 , RhIVADA , and RhIVAD17 strains were rescued from recombinant DNA , plaque purified and expanded . Western blot analysis of RhIVADA showed that the HIV-1 envelope protein was readily detectable in lysates from infected cells and pelleted virions ( Figure 1B ) . RhIVNL4-3 and RhIVAD17 strains displayed the appropriate receptor specificity when used to infect GHOSTX4 or GHOSTR5 cells ( Figure 1C ) and infection of these cells with RhIV , but not VSV , led to the appearance of large syncytia , in addition to a pronounced cytopathic effect ( Figure 1C ) . We therefore generated a larger panel of RhIV constructs expressing a variety of physiologically relevant envelope proteins from HIV-1 clades A , B , and C , including T/F viruses . All of these viruses replicated well in vitro , reaching titers of ~106 to 107 plaque forming units ( PFU ) /ml ( Figure 1D ) . To facilitate the monitoring of RhIV infection , we generated a RhIVAD17 based reporter virus that encoded GFP ( Figure 1E ) . RhIVAD17 ( GFP ) replicated robustly in GHOSTR5 cells , generating green fluorescent syncytia , as well as in a human T-cell line MT2/R5 engineered to express the CCR5 coreceptor ( Figure 1F and G ) . Live imaging of RhIVAD17 ( GFP ) replication in 293T/CD4/CCR5 cell monolayers suggested a dominant mode of viral spread in cell monolayers via direct cell-cell transmission , with additional viral transmission to distal cells ( Video 1 ) . We also generated RhIV strains expressing nanoluciferase ( nLuc , Figure 1E ) . Infection of TZMbl cells ( a popular target cell for HIV-1 neutralization assays ) with RhIVAD17 ( nLuc ) generated high levels of nLuc within a few hours of infection ( Figure 1H ) . Analysis of a panel of RhIV ( nLuc ) and corresponding HIV-1 ( nLuc ) viruses revealed that sensitivity to the CCR5-binding antagonist maraviroc was similar for each HIV-1 envelope in the context of either HIV-1 or RhIV infection ( Figure 1I ) . We compared the sensitivity of RhIV ( nLuc ) and HIV-1 ( nLuc ) viruses carrying various HIV-1 Env proteins to neutralization by a panel of well characterized bNAbs . The panel targeted various epitopes on the HIV-1 envelope: PG16 and PG9 recognize a quaternary epitope at the apex of the envelope trimer formed by the V2 loop ( Walker et al . , 2009 ) , 10–1074 recognizes a glycosylation dependent epitope in the V3 loop ( Mouquet et al . , 2012 ) , VRC01 and 3BNC117 target the CD4-binding site ( Scheid et al . , 2011; Wu et al . , 2010 ) and 10E8 targets an epitope in the membrane proximal external region ( MPER ) ( Huang et al . , 2012 ) . In general , matched RhIV ( nLuc ) and HIV-1 ( nLuc ) viruses exhibited similar neutralization properties ( Figure 2A and B ) . While most RhIV and HIV-1 strains were sensitive to the quaternary epitope targeting PG16 and PG9 antibodies , the RhIV1054/HIV-11054 , RhIVSF162/HIV-1SF162 and RhIV V1/HIV-1V1 virus pairs each shared the property of being resistant to PG9 and PG16 ( Figure 2A and B ) . The HIV-1V1 strain was unusual in exhibiting near complete resistance to all bNAbs tested , except 10–1074 and this property was preserved in the corresponding RhIVV1 chimeric virus ( Figure 2B ) . There were , nevertheless , some discrepancies in the potencies with which matched HIV-1 ( nLuc ) and RhIV ( nLuc ) viruses were neutralized by bNAbs . In one example , the MPER-targeting bNAb ( 10E8 ) neutralized RhIVCH505 but did not neutralize HIV-1CH505 ( Figure 2B ) . These occasional discrepancies may be the result of the HIV-1 and VSV-G cytoplasmic tails imposing different conformations on the Env ectodomain . Alternatively there may be differences in Env spike density , heterogeneity and distribution on RhIV virions as compared to HIV-1 virions . To generate small animals that had the potential of being infected by RhIVs , we generated transgenic mice expressing human CD4 ( hCD4 ) along with the CCR5 coreceptor . We engineered a construct that contained the murine Cd4 promoter and intron driving expression of human CD4 and CCR5 cDNAs separated by sequences encoding an FMDV 2A site ( Figure 3A ) ( Seay et al . , 2013 ) , with the goal of ensuring that hCD4 would be present exclusively on murine CD4+ cells , and tight linkage between human CD4 and CCR5 expression . Analysis of several independent transgenic mouse lines revealed variable levels of cell surface hCD4 . We selected three transgenic mouse lines , A1 , C18 and B4 that had high , intermediate and low levels of hCD4 expression respectively ( Figure 3B ) . The A1 line mimicked the levels of hCD4 found on human CD4+ T-cells ( Figure 3C ) and was used in subsequent experiments unless otherwise indicated . Levels of CCR5 ( as indicated by fluorescence intensity ) on the CD4+ cells in the A1 mice were also similar to levels of CCR5 on human CD4+ cells . However , as expected ~100% of hCD4+ cells in the blood of A1 mice were CCR5+ ( Figure 3D ) , while the fraction of CD4+ T-cells that also express CCR5 is known to vary according to tissue location in humans ( see discussion ) . FACS analysis revealed that hCD4 , like mouse CD4 , was expressed exclusively on CD3+ cells , but was absent from the CD8+ cell fraction ( Figure 3E ) . Overall , 100% of mouse CD4+ cells ( but no other cells ) in A1 mice expressed hCD4 and CCR5 at levels mimicking human CD4+ T-cells ( Figure 3E ) . Because VSV is extremely sensitive to type-1 interferon ( Müller et al . , 1994 ) , we crossed A1 , C18 and B4 mice to C57BL/6 mice lacking the type one interferon receptor gene ( Ifnar1 ) , generating A1Ifnar-/- , C18Ifnar-/- and B4Ifnar-/- lines . We first infected A1Ifnar-/- mice with 105 PFU of RhIVCH505 by intraperitoneal injection ( i . p . ) . FACS analysis of peripheral blood mononuclear cells ( PBMC ) 4 days later revealed profound and selective depletion of CD4+ T-cells ( Figure 4A ) . Next , we infected a cohort of A1Ifnar-/- mice with 105 PFU of RhIVCH505 and measured CD4+ T-cell numbers and viral RNA levels in lymphoid tissues . This analysis revealed progressive and profound reductions in CD4+ T-cell numbers in PBMC and spleen , and near complete depletion of CD4+ T-cells from thymus and lymph nodes ( Figure 4B ) . Viral RNA levels peaked at between 103 and 106 copies/μg of cellular RNA between day 1 and day 4 after infection , depending on the tissue , with the highest levels ( >106 copies/μg ) found in thymus ( Figure 4B ) . Conventionally , replication and pathogenesis during immunodeficiency virus infections in primates is monitored longitudinally using blood . We next infected A1Ifnar-/- mice ( i . p . ) with 105 PFU of RhIVBG505 , RhIVDu156 , RhIVSF162 , or RhIVCH505 and monitored plasma viremia and CD4+ T-cells in blood . Plasma viremia peaked at between 106 and 108 RNA copies/ml on day 1 after infection , then declined rapidly during days 1–7 and was cleared before day 14 ( Figure 4C , D ) . CD4+ T cells were nearly completely depleted from blood by day 4 , then gradually recovered ( Figure 4C , D ) . Thus , because analyses of blood enabled long term follow up , and appeared to provide a reasonable surrogate for virus replication and perturbation of cell population in tissues , subsequent analyses were performed using blood . Infection of A1Ifnar-/- mice with reduced doses of RhIVBG505 , RhIVDu156 , RhIVSF162 , or RhIVCH505 revealed that 103 PFU established robust infection with profound CD4+ T-cell depletion , albeit with reduced peak plasma viremia ( 105–107 RNA copies/ml , Figure 4—figure supplement 1A ) . Further reductions in RhIV challenge dose resulted in less consistent infection and reduced peak plasma viremia , with 0/2 , 2/2 , 2/2 and 1/2 mice becoming infected with RhIVBG505 , RhIVDu156 , RhIVSF162 , or RhIVCH505 at a challenge dose of 102 PFU ( Figure 4—figure supplement 1A ) . Even fewer mice became infected with lower peak viremia at a challenge dose of 10 PFU . Infection of immunocompetent ( A1Ifnar+/+ ) mice with 105 PFU of RhIVBG505 , RhIVDu156 , or RhIVSF162 yielded robust infection albeit with ~10 to 100-fold reduced peak viremia as compared to A1Ifnar-/- mice , ( Figure 4E ) . At a lower challenge dose ( 103 PFU ) , A1Ifnar+/+ mice gave far less robust viremia than did A1Ifnar-/- mice ( Figure 4—figure supplement 1B ) for all RhIV strains tested . A1Ifnar+/+ mice also revealed apparent differences in the ability of RhIV strains to cause CD4+ T-cell depletion that did not correlate with differences in plasma viremia . RhIVSF162 appeared to cause more profound CD4+ T-cell depletion than RhIVBG505 and RhIVDu156 , despite reaching similar or lower levels of plasma viremia ( Figure 4E , Figure 4—figure supplement 1B ) . We also challenged mice expressing high ( A1Ifnar-/- ) , intermediate ( C18Ifnar-/- ) or low ( B4Ifnar-/- ) levels of CD4 . The RhIVCH505 strain , bearing a T/F HIV-1 Env protein , exhibited exquisite sensitivity to CD4 expression levels . Infection of C18Ifnar-/- mice resulted in 1000-fold lower peak viremia compared to A1Ifnar-/- mice , while B4Ifnar-/- mice appeared completely resistant to RhIVCH505 infection ( Figure 4D ) . Conversely , the RhIVADA strain that bears a macrophage tropic HIV-1 Env protein was comparatively insensitive to variation in CD4 expression levels . Indeed , infection of A1Ifnar-/- and C18Ifnar-/- mice gave approximately equivalent peak viremia , while infection of B4Ifnar-/- mice gave peak viremia that was reduced only ~10 fold compared to the other mouse lines ( Figure 4F ) . Although infection of A1Ifnar-/- mice , and in some cases A1Ifnar+/+ mice , led to high level viremia and profound CD4+ T-cell depletion , in all cases RhIV infection was apparently cleared . Long term follow-up of a group of RhIVCH505-infected A1Ifnar-/- mice showed that CD4+ T-cells exhibited near complete recovery by approximately 40 to 50 days after initial infection ( Figure 4G ) . To test the utility of the RhIV model system in evaluating the protective efficacy of antibodies , we challenged mice with RhIV following administration of bNAbs . In the first experiment , A1Ifnar+/+ mice were injected subcutaneously ( s . c . ) with 1 mg of the bNAbs PG16 or 3BNC117 and challenged the following day with RhIVBG505 . On the basis of previous experiments , this injection is expected to yield 10–100 μg/ml of antibody in the blood of mice ( Klein et al . , 2012 ) . Both antibodies appeared to provide sterilizing protection , in that no viral RNA was detected in plasma of antibody-injected mice and no perturbations in CD4+ T-cells were observed ( Figure 5A ) . In a second experiment , A1Ifnar-/- mice were injected s . c . with increasing doses ( 50 μg −1 mg ) of 3BNC117 and challenged the following day with RhIVCH505 . As was the case with RhIVBG505 , the highest dose of 3BNC117 gave apparently sterilizing protection against RhIVCH505 infection , with undetectable plasma viremia and no CD4+ cell depletion ( Figure 5B ) . At 0 . 5 mg 3BNC117 , 2/3 mice exhibited apparently sterile protection , while a third had barely detectable viremia and minimal CD4+ T-cell depletion . At lower 3BNC117 doses ( 50 μg and 100 μg ) , partial protection was observed , with low-level viremia ( <103 copies/ml ) and clear CD4+ T-cell depletion , that was not as extensive as control animals ( Figure 5B ) . In one animal ( at the 50 μg dose ) apparently complete protection was observed . The finding that mice cleared RhIV infection provided the opportunity to examine whether protective immune responses might occur following RhIV infection and clearance . Therefore we conducted a series of experiments ( Expt #1 through Expt #5 ) , in which mice were challenged three times , several weeks apart , with a single RhIV strain or different RhIV strains ( see Materials and methods ) . First , we infected A1Ifnar+/+ mice ( Expt #1 , n = 2 , Figure 6—figure supplement 1A ) or A1Ifnar-/- mice ( Expt #2 , n = 2 , Figure 6—figure supplement 1B and Expt #3 n=4 Figure 6A ) with RhIVSF162 . As before , RhIVSF162 infection was cleared and CD4+ T-cells recovered . At 42 days ( Expt #2 ) or 49 days ( Expt#1 and Expt#3 ) after the first infection , mice were rechallenged with RhIVSF162 . Following the second challenge , only low-level plasma viremia ( ~103 RNA copies/ml ) was detected , and only in a subset of mice . The magnitude of CD4+ T-cell depletion following the second infection was reduced compared to the first infection ( in A1Ifnar-/- mice ) or absent ( in A1Ifnar+/+ mice ) . Following a third challenge with RhIVSF162 at 84 days ( Expt#2 ) or 91 days ( Expt#1 and Expt#3 ) after the first infection , plasma viremia was undetectable , and only minor perturbations of CD4+T cell numbers were observed ( Figure 6A , Figure 6—figure supplement 1A , B ) . Next , we did similar experiments ( Expt #4a and Expt #4b ) that employed three sequential challenges with RhIV strains bearing different HIV-1 Env subtypes at each challenge . First , we infected A1Ifnar+/+ mice ( Expt #4a , n = 4 , Figure 6B ) and A1Ifnar-/- mice ( Expt #4b , n = 4 , Figure 6C ) with RhIVDu156 ( encoding a subtype C HIV-1 Env ) . As expected , plasma viremia was cleared within 1 to 2 weeks and CD4+T cells were depleted but then recovered . At 42 days and 91 days after the initial RhIVDu156 infection , mice were challenged with RhIVBG505 ( subtype A Env ) and RhIVSF162 ( subtype B Env ) , respectively . The second challenge with RhIVBG505 resulted in only low-level plasma viremia in a subset of mice and an attenuated degree of CD4+ T-cell depletion . The third challenge with RhIVSF162 gave no detectable plasma viremia and minimal CD4+ T-cell depletion ( Figure 6B , C ) . Similar results were obtained in Expt #5 , where mice were infected on three occasions with various combinations of homologous or heterologous RhIV strains with envelopes of various subtypes ( Figure 6—figure supplement 1C , D , E and F ) . Overall , infection with RhIV gave an apparent ‘vaccine’ effect , that is there was immunity to subsequent challenge with homologous or heterologous RhIV strains , that exhibited breadth with respect to the HIV-1 envelope protein encoded by the challenge strain . To begin to ascertain whether HIV-1 Env-specific antibodies contributed to the apparent vaccine effect of initial RhIV infections on subsequent RhIV challenges , we constructed another VSV-derived chimeric virus , termed VSVMLV-E ( GFP ) . The design of VSVMLV-E ( GFP ) was the same as RhIV , except that it encoded an Env ectodomain and transmembrane sequences from ecotropic murine leukemia virus ( MLV-E ) rather than HIV-1 ( Figure 6—figure supplement 2A ) . VSVMLV-E was also equipped with a GFP reporter gene , and replicated well in NIH3T3 cells ( Figure 6—figure supplement 2B ) , yielding cell-free titers of ~106 PFU/ml . We challenged A1Ifnar+/+ mice with VSVMLV-E or RhIVBG505 which resulted in transient plasma viremia of ~104 RNA copies/ml ( VSVMLV-E ) or 105 to 106 RNA copies/ml ( RhIVBG505 ) ( Figure 6D ) . A subsequent challenge , 28 days later , with RhIVBG505 resulted in undetectable plasma viremia , and little or no CD4+ T-cell depletion , whether mice had been previously infected with VSVMLV-E or RhIVBG505 ( Figure 6D ) . Similarly , challenge of A1Ifnar+/+ mice with VSVMLV-E , 28 days after infection and clearance of RhIVBG505 resulted in no detectable plasma viremia ( Figure 6D ) . Given that MLV-E and HIV-1 Env proteins share no sequence similarity , these experiments suggested that the protection against RhIV infection , afforded by a prior RhIV infection ( Figure 6 and Figure 6—figure supplement 1 ) did not require an immune response to the HIV-1 envelope protein . To further explore whether antibody responses might be responsible for the vaccine effect of RhIV infection , we crossed A1Ifnar+/+ mice to µMT-/-mice that lack functional B-cells ( Kitamura et al . , 1991 ) . Then , A1Ifnar+/+ and A1Ifnar+/+ , µMT-/- mice were challenged with RhIVCH505 . Similar and characteristic trajectories of RhIVCH505 plasma viremia and transient CD4+ T-cell depletion were observed in both B-cell competent and B-cell deficient mouse strains ( Figure 6—figure supplement 2C ) . Following rechallenge with RhIVCH505 48 days later , neither A1Ifnar+/+ nor A1Ifnar+/+ , µMT-/- mice exhibited plasma viremia or CD4+T-cell depletion ( Figure 6—figure supplement 2C ) . Thus , a B-cell mediated immune response was not required for the vaccine effect of a prior RhIV infection on subsequent RhIV challenge . Although the above experiments indicated that antibodies were not essential for protection from a secondary RhIV challenge , they did not determine whether or not protective antibodies might be present . We therefore collected plasma from mice that had been repeatedly challenged with RhIV strains in Expt #1 to Expt #5 ( Figure 6A–C and Figure 6—figure supplement 1A–F ) and tested for the presence of antibodies capable of Env binding , neutralization and protection . Antibody binding tests employed subtype A , B and C SOSIP Env proteins ( Sanders et al . , 2013 ) , captured at their C-termini on ELISA plates ( Figure 6—figure supplements 3–5 ) . A1Ifnar-/- mice infected three times with RhIVSF162 ( subtype B , Expt #3 ) elicited antibodies that bound all four of the SOSIP envelope proteins , whose titers increased after the second infection ( Figure 6—figure supplement 3A ) . ELISA titers were higher with the B41 ( subtype B ) and BG505 ( subtype A ) than with the two subtype C SOSIP proteins ( Figure 6—figure supplement 3A ) , partly reflecting sequence similarity between the infecting RhIV strain and the ELISA antigens . Mice infected sequentially with RhIVDu156 , ( subtype C ) , RhIVBG505 ( subtype A ) then RhIVSF162 ( subtype B ) , in Expt #4 generated antibodies with higher titers on two subtype C SOSIP proteins and the BG505 SOSIP protein than the B41 ( subtype B SOSIP ) protein ( Figure 6—figure supplement 3B ) . However , all four SOSIP proteins were recognized and the second infection ( RhIVBG505 ) boosted and apparently broadened the antibody response initiated by RhIVDu156 infection . A1Ifnar-/- mice generated higher titers of Env binding antibodies than A1Ifnar+/+ mice , thus any potential deficit in antibody generation that might have resulted from the absence of type-I interferon signals was overwhelmed by the larger antigen load following infection ( Figure 6—figure supplement 3B ) . Antibody titers following sequential infection with various combinations of RhIV strains in Exp #5 ( Figure 6—figure supplement 4A and B , Figure 6—figure supplement 5A , B ) followed a general pattern that the antigen most closely resembling the initial challenge virus was best recognized in ELISA assays conducted after the first infection , although the BG505 SOSIP appeared to be generally better recognized than the other SOSIP proteins . The second and sometimes the third RhIV challenges with either homologous or heterologous RhIV strains increased titers and broadened ELISA reactivity ( Figure 6—figure supplement 4A and B , Figure 6—figure supplement 5A , B ) . In most cases , broadly Env reactive binding antibodies were elicited after three RhIV infections , regardless of the RhIV stains used in the three challenges . Further analysis of sera from a subset of mice challenged with a variety of RhIV strains ( in Exp #5 ) using commercial antigen-loaded diagnostic reagents ( INNO-LIA HIV I/II Score ) revealed prominent reactivity with epitope ( s ) on gp41 as well as gp120 ( Figure 6—figure supplement 6A ) . Notably , however , these mouse sera did not contain detectable levels of antibodies that could compete with human bnAbs or ( sCD4 ) for binding to the CD4bs , the apex of the Env trimer formed by the V2 loop or a glycosylation dependent epitope in the V3 loop ( Figure 6—figure supplement 6B ) . We next tested neutralization activity of immunoglobulins purified from pooled convalescent sera taken from mice after the three sequential RhIV challenges . In mice that had been challenged three times with RhIVSF162 , ( Expts #1–3 ) weak neutralization activity was observed against HIV-1SF162 , but not against a heterologous strain ( HIV-1CH505 ) ( Figure 7A ) . In mice that were sequentially infected three times with RhIVDu156 , RhIVBG505 and RhIVSF162 , ( Expt #4 ) no neutralization was detected against HIV-1Du156 , HIV-1BG505 or HIV-1SF162 ( Figure 7B ) . Similarly , in mice that were sequentially infected three times with various RhIV strains ( Expt #5 ) , no neutralization was observed against any of the matched HIV-1 strains ( Figure 7C ) . Even three challenges with RhIVBG505 failed to elicit neutralizing activity against HIV-1BG505 . This finding contrasts with the results obtained with RhIVSF162/HIV-1SF162 . Overall , RhIV infection elicited high titers of HIV-1 envelope binding antibodies . However , these antibodies were primarily non-neutralizing . Although most sera from infected mice lacked HIV-1 neutralization activity , it was possible that non-neutralizing antibodies might contribute to protection ( e . g . via antibody-dependent cellular cytotoxicity , ADCC ) . Therefore , we collected sera from mice after three challenges in Expts #1 to #5 and conducted passive protection experiments . Convalescent serum from each infected mouse was injected s . c . into two recipient mice that were challenged i . p . the following day with RhIVSF162 . First , recipients were given serum from Expt #1 - #3 donors that had been infected three times with RhIVSF162 and had weak neutralizing activity that was specific to HIV-1SF162 ( Figure 7A ) . Then , recipients were challenged with either 105 PFU ( Figure 7D ) or 103 PFU ( Figure 7E ) RhIVSF162 . Reduced peak plasma viremia was observed in mice that had received convalescent serum compared to controls ( Figure 7D , E , Figure 7—figure supplement 1A , B , C ) . However , none of the recipient mice were completely protected , and the reduction in plasma viremia was not statistically significant in one of the recipient mouse cohorts ( Figure 7E ) . Serum from mice that had been infected with RhIVDu156 , RhIVBG505 , and RhIVSF162 , ( Expt #4 , Figure 6B and C ) lacked neutralizing activity but was nevertheless weakly protective . Indeed , upon challenge with 103 PFU RhIVSF162 , Expt #4 convalescent serum recipients had lower peak viremia than controls , and three of fourteen mice were completely protected , with no detectable viremia and no depletion of CD4+ T-cells ( Figure 7F , Figure 7—figure supplement 1D ) . Serum from mice that had been infected with various combinations of 3 RhIV strains ( Expt #5 ) also lacked neutralizing activity and was also weakly protective . Upon challenge with 103 PFU RhIVSF162 , convalescent serum recipients again had lower peak viremia than controls , and two out of fourteen mice were completely protected ( Figure 7F , Figure 7—figure supplement 1E ) . Overall , convalescent serum from RhIV infected mice had abundant and broad Env binding activity , and weak protective activity in passive transfer experiments , that that did not correlate with the presence or absence of neutralizing antibodies .
Herein , we describe the development and use of a small animal , chimeric virus-challenge model that captures several key features of HIV-1 infection . Specifically , RhIV strains exhibit the tropism and , to a large extent , the neutralization properties of HIV-1 . The transgenic mice that are a key component of the RhIV model expressed hCD4 and hCCR5 at the same level as human T-cells , and this expression was restricted to T-cells that normally express mCD4 . RhIV infection results in acute , resolving viremia and CD4+ T-cell depletion . This model can therefore be used to test the activity of monoclonal and polyclonal antibodies in an in vivo setting , where the effector functions of antibodies can contribute to their antiviral activity in a way that is difficult to recapitulate in vitro . Because RhIV employs the intracellular replication machinery of VSV , there are some obvious caveats associated with the RhIV/mouse model . First , the replication cycle of VSV is more rapid than that of HIV-1 , with a single cycle of replication typically requiring 6–8 hr ( Cuevas et al . , 2005 ) . Second , the mode of replication ( ‘stamping machine’ with a DNA provirus for HIV-1 , versus geometric RNA replication for VSV ) might affect the propensity to accumulate escape mutations under antibody driven selective pressure ( Safari and Roossinck , 2014 ) . In practice , however , the early onset of peak viremia and rapid clearance of RhIV infection precluded an assessment of RhIV evolution in the presence and absence of selective pressure . The absence of latency in VSV replication is also a key distinction from HIV-1 infection . The mice generated herein differ from humans in that 100% of their CD4+T-cells also expressed CCR5 , because of the tight linkage of CD4 and CCR5 in the transgene construct . In humans , the proportion of CD4+ T-cells that also express CCR5 varies according to tissue source and inflamation; approximately 5–10% of peripheral blood CD4+ T-cells express CCR5 , while most gut associated lymphoid tissue and rheumatoid arthritis synovial fluid CD4+ T-cells are CCR5+ ( Agace et al . , 2000; Qin et al . , 1998 ) . The elevated frequency with which CCR5 is expressed in blood cells in our transgenic mice may accelerate CD4+ T-cell depletion during RhIV infection . A property of RhIV particles that might be relevant to tropism and neutralization properties is spike density ( the number of envelope trimers per virion ) . HIV-1 has a lower spike density than does VSV ( McSharry et al . , 1971; Zhu et al . , 2006 ) . While we were unable to precisely determine the spike density on RhIV particles , our semi-quantitative estimates indicated that the number of envelope proteins on a RhIV particle was greater than that present on an HIV-1 particle but less than that on a VSV particle . Importantly , however , RhIV strains mimicked the diversity of properties associated with parental HIV-1 strains . A RhIV strain generated using a T/F envelope protein ( RhIVCH505 ) exhibited a requirement , typical of that associated with many T/F HIV-1 strains , for the high hCD4 levels found on human T-cells ( Chikere et al . , 2014 ) . Conversely , a RhIV strain constructed using a macrophage tropic envelope ( RhIVADA ) replicated well in mice whose T-cells expressed lower levels of hCD4 ( Joseph et al . , 2014 ) . Most crucially , the neutralization properties of RhIV strains were similar to those of HIV-1 strains bearing a cognate envelope protein . An interesting feature of RhIV infection was that mice seroconverted to the HIV-1 Env proteins and generated a prominent Env binding antibody response , albeit one that was largely non-neutralizing . The low titer neutralizing antibodies that were elicited by repeated RhIVSF162 infection were autologous , and strain specific neutralization was evident only against an easy-to-neutralize HIV-1 strain , SF162 ( Seaman et al . , 2010 ) . The relatively poor generation of neutralizing antibodies should be expected , given the short period of viremia associated with RhIV infection . In HIV-1 infected humans , autologous neutralizing antibodies typically arise after months of persistent HIV-1 infection , with neutralization breadth only developing ( to varying degrees ) over the ensuing years of chronic infection ( Landais and Moore , 2018 ) . Elaborations of this model that would increase the amount of time that replication occurs in the presence of neutralizing antibodies ( for example substitution of HIV-1 Env into other viruses that are capable of replication in mice , or ablation of CD8-mediated cytotoxic responses ) may allow the co-evolution of Env sequence and antibody to be studied . Additionally , the apparently unfavorable nature of the mouse immunoglobulin repertoire may also contribute to the absence of neutralizing antibodies in the current iteration of the RhIV model , as C57BL/6 mice have previously been reported to be unable to generate autologous neutralizing antibodies following BG505 SOSIP protein immunization ( Hu et al . , 2015 ) . Future iterations of the RhIV model could benefit from the crossbreeding of A1 mice to mouse strains with human immunoglobulin repertoires ( Lee et al . , 2014; Murphy et al . , 2014 ) , although the availability of the such mouse strains to the academic community remains restricted . Despite the paucity of neutralizing antibodies generated by RhIV infected mice , primary infection conferred at least partial protection against a second , and especially a third , RhIV challenge . The bulk of this ‘vaccine’ effect was Env-independent and likely cell mediated . This finding suggests that protective immunity against a cytopathic , CD4+ T-cell tropic virus can , at least in principle , be established without protective antibodies . Nevertheless , convalescent serum from animals that had been challenged three times with RhIV strains exhibited modest protective efficacy in passive transfer experiments with an RhIVSF162 challenge . This protective activity was evident even in the absence of in vitro neutralizing activity . It is therefore likely that protection was mediated by an effector-dependent activity of the antibodies present therein , although we cannot exclude the possibility that sub-detectable neutralizing activity might be responsible . The protection afforded by convalescent sera was limited and manifested as modest reductions in acute RhIVSF162 viremia in most mice . In other animal models , and perhaps in the context of human vaccination ( Rerks-Ngarm et al . , 2009 ) , non-neutralizing antibodies against HIV-1 Env may also be responsible for modest protection ( Haynes et al . , 2012 ) . Conversely , we found that broadly neutralizing antibodies readily conferred apparently sterilizing protection upon challenge with RhIVCH505 and RhIVBG505 that bear Env proteins from more representative HIV-1 strains . The apparent protective effect of non-neutralizing antibodies in this model could , potentially , be enhanced by the absence of accessory genes that mediate CD4 or tetherin downregulation . Entrapment of envelope protein or virions on the surface of infected cells or exposure of CD4-induced epitopes may sensitize infected cells to effector-dependent activities . In conclusion , we have developed a virus-host model system that recapitulates some key features of acute HIV-1 infection . The genetic manipulability of both host and virus in this model could permit a wide range of studies on the factors that influence the elicitation of HIV-1 Env specific antibodies and antiviral efficacy of neutralizing and non-neutralizing antibodies and sera in vivo .
Plasmids encoding the full length VSV genome ( pVSV-FL ) as well as individual VSV genes N , P , L , and G were purchased from Kerafast ( VSV-FL+[2] VSV Plasmid Expression Vector System , EH1002 ) . Plasmids encoding individual HIV-1 env genes were obtained from the NIH AIDS regent repository . Alternatively , env sequences were synthesized ( Genart , Thermofisher ) . Chimeric envelope genes were generated using overlapping PCR products , in which the ectodomain and transmembrane domains of each HIV-1 Env ( equivalent to HIV-1 HXB2 amino acids 1–709 ) was fused to the cytoplasmic tail of VSV-G ( amino acids 486–511 , Figure 1A ) . The chimeric Env cDNAs were inserted into pVSV-FL precisely in place of the existing VSV-G encoding sequences to generate pRhIV plasmids encoding chimeric HIV-1/VSV-G envelopes . VSVMLV-E had a similar design , except that MLV-E Env ectodomain and transmembrane domains ( amino acids 1–634 ) were fused to the cytoplasmic tail of VSV-G ( amino acids 486–511 , see Figure 6—figure supplement 2A ) . RhIV viruses were generated by infecting 293 T cells with T7-expressing vaccinia ( vTF7-3 ) at a MOI of 5 , followed by transfection with pRhIV plasmids and plasmids encoding VSV-N , P , L , and G under the control of a T7 promoter . Supernatants were harvested 48 hr post transfection , filtered ( 0 . 2 μm ) to remove the bulk of the vaccinia virus and plaque purified on GHOST R5 cells . Plaque purified virus was expanded on 293T CD4/R5 cells and cell culture supernatant was harvested , passed through a 0 . 2 μm filter and frozen in aliquots . Virus titers ( PFU/ml ) were determined by plaque formation using GHOST R5 cells . For in vitro spreading replication assays ( Figure 1 ) , GHOST R5 cells were infected with RhIV stocks MOI of 10−4 . Thereafter , aliquots of culture supernatants were harvested at the indicated times 15–40 hr after infection and the extracellular virus yield determined by titration and plaque assay on GHOST R5 cells . RhIV derivatives encoding nano luciferase ( nLuc ) were generated by inserting the nLuc encoding sequences ( from pNL1 . 1 , Promega ) into pRhIV plasmids between the envelope and L genes , along with appropriate VSV regulatory sequences . A pRhIV plasmid encoding GFP was similarly generated by inserting the EGFP encoding sequences between the envelope and L genes , along with appropriate VSV regulatory sequences . HIV-1 proviral plasmids expressing various Env genes were generated by inserting individual Env genes into the HIV-1 molecular clone pNL4-3 . Derivatives of these constructs expressing nanoluciferase ( HIV-1 ( nLuc ) viruses ) were generated by inserting the nLuc encoding sequences in place of Nef . HIV-1 viral stocks were generated by transfecting 293 T cells; supernatant was harvested 48 hr post transfection , filtered ( 0 . 2 μm ) , and titered on TZM-bl cells using a nanoluciferase assay . Cells ( 293T , ATCC CRL-3216 ) were stably transduced with a retroviral vector ( LHCX ) into which was inserted sequences encoding human CD4 and CCR5 genes separated by an FMDV 2A site . Single cell clones were selected and tested for CD4 and CCR5 expression by FACS analysis using AlexaFluor 647 anti-human CD4 ( Biolegend ) and PE anti-human CD195/CCR5 ( BD Pharmingen ) . MT2/R5 cells were generated by transducing MT2 cells ( NIH AIDS regent repository Catalogue number 237 ) with a retroviral vector encoding hCCR5 and selecting a single hCCR5+ cell clone . GHOSTX4 and GHOSTR5 cells , that express hCD4 and CXCR4 or CCR5 , respectively , were obtained from the NIH AIDS reagent repository ( Catalogue numbers 3685 and 3944 ) , and subclones thereof were isolated by limiting dilution . Cells were monitored periodically for retrovirus contamination and were tested for mycoplasma and found to be negative . Identity of cell lines was verified by visual assessment of highly characteristic morphology and virus susceptibility . Serial dilutions of Maraviroc ( NIH AIDS Reagent Program ) , purified IgG from RhIV infected mice , or bNAbs ( VRC01 ( from Xueling Wu ) , 10E8 , PG16 , PG9 ( NIH AIDS Reagent Program ) , 3BNC117 , 10–1074 ( from Michel Nussenzweig ) ) were incubated with virus for 1 hr at 37°C prior to the addition of TZM-bl cells . For neutralization assays against RhIV and all Maraviroc assays , TZM-bl cells were seeded the day before in 96-well Flat bottom plates ( Falcon ) . For neutralization assays against HIV-1 , cells were added in suspension to the virus/antibody mixture after the incubation period . After 4 hr ( RhIV ) or 48 hr ( HIV-1 ) of infection , cells were washed twice with PBS before adding 50 μl of 1X Passive Lysis Buffer ( Promega ) . Cell lysates were mixed with an equal volume of Nano-Glo Luciferase Assay Buffer and Substrate ( Promega ) , incubated for at least 3 min at room temperature , then read using a Modulus II Microplate Multimode Reader ( Promega ) . Sequences encoding human CD4 and CCR5 genes separated by an FMDV 2A site were inserted into a construct containing the regulatory elements for CD4-specific transgene expression ( Killeen et al . , 1993 ) . The linearized transgene construct was injected into C57BL/6J embryos ( Rockefeller University Transgenic Services Laboratory ) to generate transgene lines C57BL/6J-Tg ( Cd4-CD4 , CCR5 ) A1Bsz ( CD4/CCR5HI ) , C57BL/6J-Tg ( Cd4-CD4 , CCR5 ) C18Bsz ( CD4/CCR5INT ) , and C57BL/6J-Tg ( Cd4-CD4 , CCR5 ) B4Bsz ( CD4/CCR5LO ) . Individual transgenic lines were maintained in a hemizygous state ( Tg/0 ) in a C57BL/6J background and genotyped for the presence of the transgene by PCR using the following primers: RL413 GAACCTGGTGGTGATGAGAGCCACTCA and RL425 TGCTTGCTTTAACAGAGAGAAGTTCGT . Selected transgenic lines ( termed #A1 , #C18 , and #B4 ) that were chosen based on high , intermediate and low levels of CD4 expression respectively , were also crossed with C57BL/6J Ifnar1 knockout mouse line ( MMRRC #32045 ) ( Müller et al . , 1994 ) to generate corresponding #A1Ifnar1-/- , #C18Ifnar1-/- and #B4Ifnar1-/- mouse lines . Mice derived from C57BL/6 of both sexes were used , and housed under standard conditions prior to infection . Mice were moved to an ABSL-2 facility and were randomly ascribed to experimental groups prior to infection . Initial infections were done at 8 to 12 weeks . Mice were infected with RhIV stocks ( 10 to 105 PFU in 500 μl DMEM ) by intraperitoneal ( i . p . ) injection . Thereafter , blood was collected in EDTA coated tubes ( Sarstedt ) from the facial vein at the indicated timepoints , typically 1 , 4 , 7 , 14 and 21 days after infection , and weekly thereafter for longer term experiments . Plasma was separated from cells and used for extraction of RNA or in ELISA while cells were processed for FACS analysis of cell populations . For analysis of tissues , mice were euthanized using carbon dioxide . Spleen , thymus , and lymph node tissue was removed and processed for RNA extraction or FACS analysis . In experiments involving serum transfer , previously infected donor animals were bled several times , 3 to 4 days apart , serum isolated from each bleed and pooled with serum from the same individual mouse . Thereafter , 200 μl of heat inactivated serum was injected subcutaneously ( s . c . ) into naïve animals one day prior to an i . p . RhIV challenge . For monoclonal antibody protection experiments , antibodies ( 50 μg to 1 mg ) were diluted in PBS to a final volume of 200 μl and administered s . c . one day prior to an i . p . RhIV challenge . All animal studies were conducted in accordance with The Rockefeller University Institutional Animal Care and Use Committee ( IACUC ) . Viral RNA was extracted from 50 μl aliquots of mouse plasma using Trizol LS Reagent ( Ambion ) . Phase separation steps were performed in MaXtract High Density tubes ( Qiagen ) and GlycoBlue ( Invitrogen ) was used as a coprecipitant . After drying , RNA pellets were resuspended in 50 μl RNase-free Molecular Biology Grade Water ( Corning ) . For RT-qPCR , 8 μl of purified RNA solution was used , and RT-qPCR were carried out in one step using the Power SYBR Green RNA-to-CT 1-Step kit ( Applied Biosystems ) and primers RL509 TGATACAGTACAATTATTTTGGGAC and RL510 GAGACTTTCTGTTACGGGATCTGG , that target the VSV-L gene ( Hole et al . , 2006 ) . Duplicate aliquots of RNA were tested using an Applied Biosystems Step-One Plus Real Time PCR machine . A standard curve , generated using a plasmid DNA template , was used to calculate RNA copies/ml . The limit of detection for this assay was a single copy of cDNA per PCR reaction , equivalent to 125 RNA copies/ml of mouse plasma . Mouse blood , spleen , thymus and lymph node were processed for FACS analysis by making a single cell suspension , removing red blood cells by resuspension in red blood cell lysis buffer ( 150 mM NH4Cl , 10 mM KHCO3 , and 0 . 05 mM EDTA ) , then resuspending the resulting pellet in FACS buffer ( PBS , 0 . 2% bovine serum albumin ) . Cells were incubated with anti-CD16/CD32 ( Fc Block ) prior to staining with the following antibodies: FITC anti-CD3 , PerCP-Cy5 . 5 anti-mCD4 , PE anti-CD19 ( BD Pharmingen ) , APC anti-CD8a , and APC/Cy7 anti-hCD4 ( BioLegend ) . Samples were run on either a LSRII ( Becton Dickinson ) or Attune NxT ( Life Technologies ) flow cytometer and data were analyzed using FlowJo ( Tree Star ) . For measurement of neutralizing activity in mouse serum , purified IgG was used . Serum was separated from whole blood by centrifugation and heat inactivated for 1 hr at 56°C . IgG was purified from serum using the Protein G HP Spin Trap/Antibody Spin Trap kit ( GE Healthcare ) , according to manufacturers instructions then dialyzed overnight in PBS at 4°C ( Slide-A-Lyzer , 20 , 000 MWCO , Thermo Scientific ) . Purified IgG solutions were then filtered through a 0 . 2 μm filter and concentrated ( Spin-X UF Concentrator , Corning ) . GHOST R5 cells infected with RhIV were lysed with RIPA buffer ( 150 mM NaCl , 50 mM Tris , pH 7 . 4 , 0 . 1% SDS , 1 mM EDTA , 1% Igepal , 1% sodium deoxycholate ) . Virions were pelleted through 20% sucrose in PBS . Cell and virion proteins were separated on SDS-PAGE gels and blotted onto nitrocellulose membranes . Blots were probed with anti-gp120 ( American Research Products ) and anti VSV-M ( Kerafast ) with IR800 donkey anti goat and IR680 donkey anti mouse ( LiCor ) secondary antibodies . The env genes of HIV-1 strains were synthesized by GeneART ( Thermofisher ) , in a modified form to generate C-terminally His-tagged soluble SOSIP . 664 Env trimers . The previously characterized SOSIP . 664 trimers were derived from the clade A BG505 strain ( PMID: 24068931 ) , the clade B B41 strain ( PMID: 25589637 ) and two clade C strains DU422 and ZM197M ( PMID: 26372963 ) . The env cDNAs were inserted into pAAV-MCS and the resulting Env expression plasmids were transiently transfected into Expi293 cells using the serum free Expi293 Expression System ( Life Technologies ) . Cell culture supernatants were collected at 5 days post-transfection , sterile filtered , and used as a source of Env proteins . Corning Costar 96-well EIA/RIA Plates were coated with anti-His-Tag Antibody ( pAb , Rabbit , GenScript , US ) at a concentration of 1 µg/ml in coating buffer ( 0 . 05 M Carbonate-Bicarbonate , pH 9 . 6 ) overnight . Unbound antibody was removed with wash buffer ( 50 mM Tris , 0 . 14 M NaCl , 0 . 05% Tween 20 , pH 8 . 0 ) and the plates blocked ( 50 mM Tris , 0 . 14 M NaCl , 1% BSA , pH 8 . 0 ) . SOSIP Env proteins were captured via their C-terminal His-Tags from cell culture supernatants at 37°C for 1 hr . After washing steps , serial 1:2 dilutions of heat-inactivated mouse serum or plasma samples ( beginning a 1:300 dilution ) were added to the plate and incubated at 37°C for 2 hr . The plates were washed and blocking buffer containing an anti-mouse HRP-conjugated antibody ( GAM Ab97040 HRP , 1:20 , 000 , Abcam ) was added and incubated at 37°C for 30 min . Unbound antibodies were removed by additional washing steps and bound HRP detected by TMB One Solution System ( Promega ) . After 20 min of incubation the colorimetric reaction was stopped by adding 0 . 3M phosphoric acid and spectrophotometric readings recorded at 450 nm . Human bNAbs were titrated to give ELISA signals on BG505 SOSIP . 664 coated plates that were in the linear range with respect to bNAb concentration . For competition ELISAs , BG505 SOSIP . 664 coated plates were preincubated for 2 hr with dilutions of IgG that had been purified from mouse sera . Then , the plates were washed and incubated with blocking buffer containing the predetermined concentrations of human bNAbs for 30 min . After washing , plates were incubated with blocking buffer containing an anti-human HRP-conjugated antibody ( Ab97175 HRP , 1:20 , 000 , Abcam ) and bound antibodies detected as above . For detection of antibodies using INNO-LIA HIV I/II Score strips , the manufacturers ( Fujirebio ) procedures were followed , except that Ab97175 goat anti-Hu IgG ( HRP ) 1:20 000 and Ab97040 goat anti-Ms IgG ( HRP ) 1:20 000 were used , as appropriate . Bound antibodies were detected using SuperSignal West Pico PLUS chemiluminescent substrate ( Thermofisher ) . All data is plotted raw , that is individual values for each individual determination and each individual mouse is plotted . The exceptions to this are the qRT-PCR data , in which the mean of technical duplicates is plotted . Animals were allocated randomly to experimental groups . Statistical comparisons between groups in Figure 7D , E , F were done using Graphpad Prism software , and p-values were calculated using a Mann Whitney test . | One of the main obstacles to developing a vaccine against HIV-1 is teaching the immune system to recognize the envelope proteins on the surface of the virus , which are also found on infected cells . Envelope proteins allow HIV-1 to attach to and infect a type of human immune cell known as a T-cell , by interacting with proteins on its membrane called CD4 and CCR5 . Antibodies are proteins produced by the immune system that can stop HIV-1 from spreading . They can recognize and attach to envelope proteins , thus tagging infected cells so the immune system can attack them , and ‘neutralizing’ viral particles to prevent them from infecting more cells . To make a vaccine against HIV-1 , scientists need to teach the immune system how to make neutralizing antibodies . Unfortunately , HIV-1 only replicates in humans and chimpanzees , making it difficult to study how these antibodies are generated . Now , Liberatore et al . have developed a hybrid virus that recreates key features of HIV-1 infection in mice . The interior of these viruses is made up of components from a rhabdovirus , which replicates well in mice , with envelope proteins from HIV-1 incorporated into the viruses’ exterior . Therefore , despite having different replication machinery , these hybrid viruses – nicknamed ‘RhIV’ – are able to infect the cells of mice using the same attachment mechanism as HIV-1 . Next , Liberatore et al . genetically modified mice to produce human CD4 and CCR5 proteins , so RhIV could attach to their T-cells and get inside . The virus rapidly killed the cells it infected , similar to early HIV-1 infection in humans . But , unlike HIV-1 infection in humans , the mice were able to get rid of the virus within a couple of weeks . When the mice were exposed to RhIV a second time , they were partially protected against re-infection . This ‘vaccine effect’ was even stronger if the mice were exposed a third time , making them almost immune to the virus . However , the effect could not be attributed exclusively to antibodies , since mice unable to make antibodies still gained some immune protection after infection with RhIV . The results showed that antibodies produced by the infected mice could recognize HIV-1 envelope proteins , but were unable to neutralize viral particles . Nevertheless , transferring antibodies from infected mice after recovery into healthy mice that had never been exposed to the virus partially protected the healthy mice from infection . This new model system for HIV-1 infection should make it easier to test new types of vaccines in a context where the vaccinated animal can be challenged with RhIV . Additionally , the ability to genetically engineer both the virus and the mouse host – for example by making mice that produce human antibodies – allows further studies into the development of antibodies that recognize the HIV-1 envelope . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Rhabdo-immunodeficiency virus, a murine model of acute HIV-1 infection |
The prefrontal cortex ( PFC ) is thought to orchestrate cognitive dynamics . However , in tests of bistable visual perception , no direct evidence supporting such presumable causal roles of the PFC has been reported except for a recent work . Here , using a novel brain-state-dependent neural stimulation system , we identified causal effects on percept dynamics in three PFC activities—right frontal eye fields , dorsolateral PFC ( DLPFC ) , and inferior frontal cortex ( IFC ) . The causality is behaviourally detectable only when we track brain state dynamics and modulate the PFC activity in brain-state-/state-history-dependent manners . The behavioural effects are underpinned by transient neural changes in the brain state dynamics , and such neural effects are quantitatively explainable by structural transformations of the hypothetical energy landscapes . Moreover , these findings indicate distinct functions of the three PFC areas: in particular , the DLPFC enhances the integration of two PFC-active brain states , whereas IFC promotes the functional segregation between them . This work resolves the controversy over the PFC roles in spontaneous perceptual switching and underlines brain state dynamics in fine investigations of brain-behaviour causality .
Dynamic and flexible changes are among the fundamental key properties of human cognition . Bistable visual perception has been widely used to investigate such cognitive dynamics ( Brascamp et al . , 2018; Sterzer et al . , 2009 ) , and the prefrontal cortex ( PFC ) —in particular , right frontal eye fields ( FEF ) , dorsolateral PFC ( DLPFC ) and inferior frontal cortex ( IFC ) —is thought to be involved in the spontaneous perceptual switching ( Brascamp et al . , 2018; Kleinschmidt et al . , 1998; Lumer et al . , 1998; Lumer and Rees , 1999; Panagiotaropoulos et al . , 2020; Panagiotaropoulos et al . , 2012; Sterzer et al . , 2009; Sterzer et al . , 2002; Wang et al . , 2013; Weilnhammer et al . , 2013 ) . Theoretical work also indicates that top-down signals from the PFC to the visual cortex are essential to perceptual inference ( Hohwy et al . , 2008; Weilnhammer et al . , 2017 ) . These studies imply that inhibitory neural modulation of the PFC should induce behavioural changes in the bistable visual perception . However , no empirical human study has identified such behavioural causality of the PFC ( Brascamp et al . , 2018 ) except for a recent work administering theta-burst transcranial magnetic stimulation ( TMS ) over the right IFC ( Weilnhammer et al . , 2021 ) . Instead , a previous TMS study reported that neural suppression of the right DLPFC did not affect bistable visual perception ( de Graaf et al . , 2011 ) . Other studies claimed that the PFC activity is not essential to the emergence of multistable perception ( Brascamp et al . , 2015; Harrison and Tong , 2009 ) but mere a consequence of it ( Block , 2020; Brascamp et al . , 2015; Frässle et al . , 2014; Knapen et al . , 2011 ) . Why is it so difficult to detect the prefrontal causality in the multistable perception ? Here , given that the whole-brain neural activity during bistable perception is described as large-scale brain state dynamics ( Watanabe et al . , 2014c ) , we hypothesise that causal roles of the PFC should also be dynamically changing during the fluctuation of visual awareness . That is , if the neural activity in the multistable perception can be stated as dwelling in and transitions between a parsimonious number of brain states ( Watanabe et al . , 2014c ) , the detectability of the prefrontal causal effects on the perceptual awareness should be determined by the brain state in which the neural activity pattern is staying when the neural stimulation is administered . If so , such state-dependent behavioural causality should be hardly observed when we intervene in the PFC activity without tracking the brain state dynamics . We directly tested this hypothesis with a brain-state-dependent neural stimulation system ( Bergmann , 2018; Silvanto et al . , 2008; Zrenner et al . , 2016 ) . The system was devised by linking energy landscape analysis ( Ezaki et al . , 2017; Gu et al . , 2018; Kang et al . , 2017; Watanabe et al . , 2014c; Watanabe and Rees , 2017 ) —a data-driven method to identify the neural dynamics during seemingly random behaviours—to an electroencephalogram ( EEG ) -triggered TMS ( Bergmann et al . , 2016; Schaworonkow et al . , 2019; Stefanou et al . , 2018; Zrenner et al . , 2018 ) . To focus on the higher order cortex , we did not adopt a test of binocular rivalry , in which the perceptual fluctuation has often been linked with the neural activity in the lower level brain systems such as the visual cortex ( Haynes et al . , 2005; Lee et al . , 2005; Leopold and Logothetis , 1996; Meng and Tong , 2004; Miller et al . , 2000; Pettigrew and Miller , 1998; Polonsky et al . , 2000 ) . Instead , we used a test of bistable visual perception induced by a structure-from-motion ( SFM ) stimulus ( Figure 1a ) , in which the same visual stimulus is presented to both the eyes of participants , and the higher order cortex is considered to be more involved in the perceptual fluctuation ( Brascamp et al . , 2018; Knapen et al . , 2011; Meng and Tong , 2004 ) .
As a preparation , we examined whether the current EEG system captured qualitatively the same brain state dynamics as those found in our previous fMRI study employing the same SFM stimulus ( Watanabe et al . , 2014c ) . To this end , we recorded gamma-band EEG signals from the seven cortical regions ( Watanabe et al . , 2014c; Figure 1b ) of 65 healthy adults while they were experiencing the SFM-induced bistable visual perception ( left panel of Figure 1c ) and applied an offline energy landscape analysis to the data ( Control Experiment I ) . First , we confirmed that a pairwise maximum entropy model—a basis of the energy landscape analysis—was well fitted to the data in all the participants ( fitting accuracy >84 %; Figure 2a ) . Based on the model , we then found that all the participants had almost the same energy landscape structure consisting of three major brain states: Frontal-area-dominantly-active state ( F state ) , Visual-area-dominantly-active state ( V state ) and Intermediate state ( Int state ) ( e . g . Figure 2b for Participant 1 ) . As in our previous fMRI work , these three states were significantly separated and independent from each other ( i . e . the energy barrier >1; Figure 2c ) and dominated almost all the energy surfaces ( Figure 2d ) . Such model-based estimations were consistent with the actual EEG data . Even in the neural data , almost all the brain activity patterns could be classified into either of the three major brain states ( the other states < 0 . 1 %; Figure 2e ) . The degree to which each of the three brain states occupied the energy landscape ( i . e . the basin size ) was significantly correlated with how frequently the brain state appeared in the EEG data ( r64 >0 . 57; Mean Average Percentage Error , MAPE , < 6 . 8 %; Figure 2f ) . Moreover , the neural dynamics between these three major brain states were qualitatively equivalent to those seen in our fMRI study: there was almost no direct transition between F and V state ( < 0 . 4%; Figure 2g , h ) , and thus the brain activity pattern always used Int state as a stepping stone to travel between F and V state ( Figure 2i ) . Finally , we confirmed that , as shown in our fMRI work , the length of such neural travel between F and V state via Int sate was a good indicator to predict the individual behavioural differences in the percept duration ( r64 = 0 . 67; Figure 2j ) . Furthermore , this brain-behaviour association was preserved at an individual level: even within each participant , the length of the F-Int-V-Int-F travel calculated for each run accurately predicted the percept duration in the run ( e . g . Figure 2k for Participant 1; for all the participants , r19 >0 . 54 , MAPE <7 . 2% , Figure 2l , m ) . Note that these energy landscape analyses used no behavioural information to identify the brain state dynamics; thus , the significant brain-behaviour correlations are not consequences of circular analysis . In sum , these findings show that , at least with an offline analysis , the current EEG system can identify qualitatively the same brain state dynamics underpinning the SFM-induced bistable visual perception as seen in our previous fMRI study ( Watanabe et al . , 2014c ) . These results have sufficient information to categorise each neural activity pattern at each timepoint into either of the three major states . In the Control Experiment II , we implemented such classification information into an online EEG analysis ( Stefanou et al . , 2018; Zrenner et al . , 2018; right panel of Figure 1c ) and examined whether the online analysis could track the bran state dynamics accurately and enable us to administer TMS while the brain activity pattern was dwelling in a specific brain state . First , we confirmed that the brain state identification in the online analysis was significantly similar to that obtained by the offline analysis ( similarity >82 . 3 % for the three major brain states; Figure 3a ) . Then , by linking this online EEG signal processing to monophasic TMS , we succeeded in triggering a burst of inhibitory TMS only when the neural activity pattern was staying in a specific major brain state ( accuracy >84 . 8% , Figure 3b; latency <0 . 8 ms , Figure 3c ) . By applying this state-dependent TMS over the three PFC regions ( i . e . DLPFC , IFC and FEF; Figure 4 ) , we found that the three prefrontal areas had different causal behavioural effects on the spontaneous perceptual switching in a brain-state-dependent manner ( N = 34; F8 , 33=25 . 9 , p < 10–3 for the main effect in a two-way ANOVA; Figure 4b ) . The neural inhibition of DLPFC during F state prolonged the percept duration ( t33 = 9 . 4 , PBonferroni < 0 . 001 in a post-hoc paired t-test , Cohen’s d = 1 . 7 ) , whereas that during V or Int state induced no behavioural change ( t33 <1 . 7 , PBonferroni > 0 . 05 , d < 0 . 3 ) . Regarding IFC , the F-state-dependent TMS destabilised the visual perception ( t33 = 10 . 4 , PBonferroni < 0 . 001 , d = 1 . 8 ) , whilst neither V- nor Int-state-dependent TMS induced any behavioural effect ( t33 <1 . 3 , PBonferroni > 0 . 05 , d < 0 . 29 ) . The F-state-dependent TMS over FEF reduced the percept duration ( t33 = 10 . 5 , PBonferroni < 0 . 001 , d = 1 . 9 ) , whereas no behavioural change was observed in the other FEF TMS conditions ( t33 <0 . 8 , PBonferroni > 0 . 05 , d < 0 . 39 ) . If some behavioural causalities are detectable when the whole-brain neural activity pattern is dwelling in specific brain states , others may emerge when the neural activity pattern has finished travelling a particular brain state trajectory . We then tested this hypothesis and found such state-history-dependent causality ( F5 , 33=85 . 6 , p < 10–3 for the main effect in a two-way ANOVA; Figure 4c ) . Here , we focused on Int state because it is the sole brain state that had two incoming pathways ( i . e . a path from F state and one from V state ) . The TMS over DLPFC during Int state immediately after F state shortened the percept duration ( t33 = 12 . 9 , PBonferroni <0 . 001 in a post-hoc paired t-test , d = 2 . 7 ) , whereas the TMS over DLPFC during Int state right after V state prolonged it ( t33 = 13 . 3 , PBonferroni <0 . 001 , d = 2 . 3 ) . For IFC , the Post-F Int-state-dependent TMS enhanced the perceptual stability ( t33 = 9 . 1 , PBonferroni <0 . 001 , d = 1 . 6 ) , whilst that in Post-V Int state weakened it ( t33 = 12 . 7 , PBonferroni <0 . 001 , d = 2 . 3 ) . No change in the percept duration was observed when we administered TMS over FEF during Post-F or Post-V Int state ( t33 <0 . 58 , PBonferroni > 0 . 05 , d < 0 . 19 ) . Given that no significant behavioural change was induced by longer and stronger TMS ( here , 30 min quadripulse TMS ) ( Hamada et al . , 2009; Hamada et al . , 2007; Watanabe et al . , 2015; Watanabe et al . , 2014a ) over the same PFC regions ( N = 14; F2 , 13=0 . 16 , p = 0 . 85 for the main effect in a two-way ANOVA; Figure 4d ) , these results suggest that the causal behavioural roles of the PFC areas in the bistable visual perception become explicit and measurable only when we intervene in the neural activity in state-/state-history-dependent manners . These brain-state-dependent behavioural responses imply that the underlying neural mechanisms could be accounted for by brain state dynamics . . To reveal such brain mechanisms , we first formulated working hypotheses on the neural effects of TMS by conducting numerical simulation on how local brain inhibition would affect the energy landscape structure and , resultantly , the brain state dynamics . In particular , we calculated changes in the heights of the energy barriers between the three major brain states ( Figure 5a ) , because the barrier heights are associated with the dwelling time in the brain states and inversely correlated with the transition frequency between them ( Watanabe et al . , 2014c ) . As to DLPFC ( Figure 5b ) , the numerical simulation showed that the TMS should increase the energy barrier heights between F and Int state ( F4 , 33=340 . 1 , p < 10–3 for the main effect in a two-way ANOVA; t33 >15 . 2 , PBonferroni <0 . 001 in post-hoc paired t-tests , d > 2 . 6 ) and decrease the barrier height from Int to V state ( t33 = 38 . 0 , PBonferroni <0 . 001 , d = 6 . 7 ) . As a result , the ratio of the Int-to-F barrier to the Int-to-V barrier should significantly increase ( t33 = 27 . 1 , PBonferroni <0 . 001 , d = 4 . 8 ) . Regarding IFC ( Figure 5c ) , the neural inhibition of the region should alleviate the energy barriers between F and Int state ( F4 , 33=89 . 10 , p < 10–3 for the main effect in a two-way ANOVA; t33 >11 . 6 , PBonferroni <0 . 001 , d > 2 . 0 ) and that from Int to V state ( t33 = 48 . 9 , PBonferroni <0 . 001 , d = 8 . 6 ) . The magnitude of the decrease in the Int-to-F barrier height should become larger than that in the Int-to-V barrier , which would result in a significant decrease in the ratio of the Int-to-F barrier to the Int-to-V barrier ( t33 = 61 . 2 , PBonferroni <0 . 001 , d = 10 . 7 ) . For FEF ( Figure 5d ) , its neural suppression should lower the energy barrier from F to Int state ( F4 , 33=32 . 4 , p < 10–3 for the main effect in a two-way ANOVA; t33=25 . 3 , PBonferroni < 0 . 001 , d = 4 . 4 ) but induce no significant change in the other barriers ( t33 <1 . 7 , PBonferroni > 0 . 05 , d < 0 . 27 ) . These structural changes in the energy landscapes indicate how the TMS affected the brain state dynamics ( Figure 6a ) . For example , the higher F-to-Int energy barrier—which is predicted to occur after TMS over DLPFC ( Figure 5b ) —would enhance the segregation between F and Int state , impede the transition from F to Int state and prolong the dwelling in F state . Moreover , such longer F-state dwelling should be observed in F-state-dependent TMS condition the most clearly . By the same logic , the relatively lower Int-to-V barrier , which is also expected to occur after TMS over DLPFC , should increase the Int-to-V transitions . Regarding the TMS over IFC , the lower F-to-Int energy barrier should shorten the F-state dwelling , whereas the relatively lower Int-to-F barrier should increase the Int-to-F transitions . In the TMS-over-FEF condition , the lower F-to-Int barrier should reduce the F-state dwelling . We tested and confirmed these hypotheses by measuring the dwelling time of the three major brain states and transition frequencies between them for all the state-dependent TMS conditions . In the experiments administering TMS over DLPFC , the increase in the F-to-Int energy barrier was correlated with the longer F-state dwelling seen in the F-state-dependent TMS ( t33 = 11 . 6 , PBonferroni <0 . 001 , d = 2 . 0; r33 = 0 . 59 , p < 0 . 001; Figure 6b ) . Also , the higher Int-to-F energy barrier was associated with more frequent Int-to-V transitions that was seen in the Int-state-dependent TMS ( t33 = 3 . 3 , PBonferroni <0 . 05 , d = 1 . 8; r33 = 0 . 52 , p < 0 . 001; Figure 6c ) . In the sessions administering TMS over IFC , the lower F-to-I energy barrier was correlated with the shorter F-state dwelling that was measured in the F-state-dependent neural suppression ( t33 = 6 . 9 , PBonferroni <0 . 001 , d = 1 . 2; r33 = 0 . 69 , p < 0 . 001; Figure 6d ) . The lower Int-to-F barrier accurately predicted more frequent Int-to-F transitions ( t33 = 13 . 0 , PBonferroni <0 . 001 , d = 2 . 8; r33=–0 . 58 , p < 0 . 001; Figure 6e ) that were seen in the Int-state-dependent TMS condition . In the TMS-over-FEF conditions , the lower F-to-Int barrier was associated with the shorter F-state dwelling that was observed in the F-state-dependent TMS session ( t33 = 26 . 1 , PBonferroni <0 . 001 , d = 4 . 8; r33 = 0 . 57 , p < 0 . 001; Figure 6f ) . These results clarify TMS-induced effects on the brain state dynamics during the bistable visual perception and demonstrate that such neural effects are underpinned by the structural changes in the energy landscapes . How long did such neural effects continue after each TMS ? To answer this question , we tracked the magnitudes of the correlations between the empirical neural effects and the numerically-calculated energy barrier changes ( Figure 6b , c , d , e , f ) when we were sliding the time window that was used to empirically quantify such neural effects ( Figure 7a ) . This analysis detected that all the correlations began to weaken approximately 1 . 5 s after each TMS ( Figure 7b ) . This result indicates that the current TMS-induced neural effects started to decay within ~1 . 5 s after the stimulation , and the brain state dynamics began to return to the original forms in such a time scale . Finally , we evaluated associations between these behavioural , numerical and neural responses by mediation analysis and found that the causal behavioural effects are induced by transient changes in the brain state dynamics and attributable to structural changes in the energy landscapes ( P < 0 . 05 for all the indirect effects , α×β; Figure 8 ) . As to the state-dependent behavioural changes , the longer percept duration seen after the F-state-dependent TMS over DLPFC was largely due to the prolonged F-state dwelling , which was induced by the higher F-to-Int energy barrier ( Figure 8a ) . In contrast , the de-stabilisation of visual perception seen after the F-state-dependent TMS over IFC/FEF was caused by the shorter F-state dwelling , which was attributable to the lower F-to-Int energy barrier ( Figure 8b , c ) . The state-history-dependent behavioural causality could also be seen as a consequence of the transient changes in the brain state dynamics , given that the percept duration is closely linked with the length of the F-Int-V-Int-F travel ( Figure 2j , k ) : According to the mediation analysis , the shorter percept duration seen after the Post-F Int-state-dependent TMS over DLPFC was caused by the increase in the Int-to-V transitions , which was triggered by the larger gap between the Int-to-F barrier and Int-to-V barrier ( Figure 8d ) . This statistical result is reasonable because the relatively lower Int-to-V energy barrier would facilitate the Int-to-V transitions transiently , accelerate the completion of the F-Int-V travel and shorten the percept duration . Conversely , the longer percept duration yielded by the Post-V Int-state-dependent TMS over DLPFC is interpretable as a result of the relatively higher Int-to-F energy barrier’s impeding the Int-to-F transitions , accelerating the backward moves to V state and slowing down the completion of the F-Int-V travel ( Figure 8e ) . The mediation analysis showed that the state-history-dependent causal roles of IFC were also accountable by the same logic . The longer percept duration yielded by the TMS over IFC during Post-F Int state could be regarded as a behavioural manifestation of the slowdown of the F-Int-V travel due to the more frequent backward Int-to-F transitions , which was originated from the relatively lower Int-to-F energy barrier ( Figure 8f ) . In contrast , the shorter percept duration induced by the TMS over IFC during Post-V Int state can be seen as results of the acceleration of the F-Int-V travel due to the more frequent forward moves from Int to F state , which was induced by the lower Int-to-F energy barrier ( Figure 8g ) . The main behavioural findings were replicated in a small but independent cohort ( N = 14; t13 >2 . 9 , P < 0 . 01 in one-sample t-tests; Figure 9 ) . Moreover , another independent experiment ( N = 15 ) showed that , as in our previous work ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) , the excitatory TMS induced behavioural effects opposite to those yielded by the inhibitory stimulation ( t14 >2 . 8 , P < 0 . 01 in one-sample t-tests; Figure 9b ) , which added indirect but empirical support for the current observations . To examine the potential of future clinical applications of this brain-state-dependent neural stimulation system , we conducted another two-month longitudinal experiment ( N = 63 ) and found accumulative effects of this closed-loop TMS system . Despite its weak power—approximately 78 % of the power of the similar TMS protocol ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) — , the behavioural effects became larger along with the weekly TMS sessions ( t12 >3 . 2 , PBonferroni < 0 . 05 in paired t-tests; Figure 10a ) . Moreover , the F-state-dependent TMS affected the baseline perceptual stability , which was detectable even one week after the 2-month TMS experiments ( F6 , 12=16 . 4 , P < 10–3 for the main effect in a two-way ANOVA; PBonferroni < 0 . 01 in post-hoc one-sample t-tests; Figure 10b ) .
This study has demonstrated that causal roles of the three PFC regions—right DLPFC , IFC and FEF—in spontaneous perceptual inference were behaviourally detectable when we tracked the brain state dynamics underpinning such perceptual fluctuation . The state-/state-history-dependent behavioural causality was explained by the large-scale brain state dynamics and attributable to the energy landscape structures . Moreover , the current findings suggest distinct functions of the PFC regions in terms of the brain state dynamics: the activation of DLPFC enhances the functional integration between the Frontal and Intermediate state , whereas the IFC activity promotes the functional segregation between the two brain states; the FEF activity stabilises Frontal state . These prefrontal functional differences can be interpreted in more conventional neuropsychological contexts , if we presume that ( i ) Frontal state is involved in the generation of top-down signals , ( ii ) Visual state is related to bottom-up signal generation and ( iii ) Intermediate state is a place for the interactions of the top-down and bottom-up information . Given these assumptions , the DLPFC activity can be seen as a factor to enhance the generation and flow of top-down information , whereas the IFC activity facilitates those of bottom-up information . The activation of FEF can be regard as a support for the top-down signal generation . At first glance , this neuropsychological interpretation is well fitted to widely-known concepts on two attention systems in the brain ( Baldauf and Desimone , 2014; Corbetta et al . , 2008; Corbetta and Shulman , 2002 ) : the dorsal attention system including the FEF is mainly involved in preparing and applying top-down attention , whereas the ventral attention system including the IFC is associated with processing of bottom-up attention signals . However , as argued in a recent TMS study ( Weilnhammer et al . , 2021 ) , it may be difficult to fully describe the PFC roles during bistable perception from the perspective of such attention systems only . For example , if the main function of the IFC in this experiment is to receive bottom-up attention signals from the visual cortex , distribute them to the fronto-parietal areas and direct the attention to a different perceptual inference , the neural suppression of IFC should prolong the percept duration . By contrast , in this study , the inhibition of IFC activity did not always lengthen the percept duration but often shortened it ( Figure 4b , c ) . Given these , it would be more reasonable to infer that the top-down and bottom-up signals , which are supposed to be generated and communicated in the brain state dynamics , contain not only attention-related information but also other types of neural signal such as prediction error in predictive coding paradigm ( Brascamp et al . , 2018; Hohwy et al . , 2008; Weilnhammer et al . , 2021; Weilnhammer et al . , 2017 ) . By the same logic , it may be difficult to regard the current findings as evidence supporting the notion that the DLPFC and IFC are irrelevant to the bistable perception itself but only involved in reporting the perceptual states ( Brascamp et al . , 2015; Frässle et al . , 2014 ) . If these prefrontal regions are related to reporting the perceptual awareness , the neural suppression of the PFC areas should always prolong the percept duration . In reality , the inhibitory TMS over PFC could shorten the percept duration ( Figure 4b , c ) , which indicates the possibility that the DLPFC and IFC are closely involved in the bistable perception itself . In the meantime , we have to be careful to generalise this indication to other types of bistable perception . In fact , it is not SFM but binocular rivalry that the previous study used to show the negligible associations between the PFC activity and bistable perception ( Brascamp et al . , 2015; Frässle et al . , 2014 ) . To resolve this situation , future studies would have to examine brain-state-dependent behavioural causality of the PFC in SFM-induced bistable perception using the non-report paradigm ( Tsuchiya et al . , 2015 ) . In the bistable visual perception paradigm , a line of previous TMS studies reported behavioural causal roles of the parietal cortex in the right hemisphere ( Carmel et al . , 2006; Kanai et al . , 2011; Kanai et al . , 2010; Vernet et al . , 2015; Zaretskaya et al . , 2010 ) , whereas no investigation found such effects in the prefrontal cortex except for a recent one ( Weilnhammer et al . , 2021 ) . The current findings propose a neurobiological account for this difficulty in detecting the prefrontal causality . As shown in Figure 4b , c , the behavioural responses to the TMS depended on the timing of the neural stimulation . Without tracking the brain state dynamics , the behavioural responses to the TMS would be close to the average of such brain-state-dependent effects and likely to be observed as null results ( e . g . Figure 4d ) . If this is the case , why could the recent study successfully detect prolonged percept in the bistable visual perception test after applying conventional TMS over the IFC ? ( Weilnhammer et al . , 2021 ) The TMS protocols are different between the previous study and the current work , and further studies are necessary to answer this question directly; but we can speculate the reason as follows . In the recent study , theta-burst TMS was administered to the IFC during rest . Considering that default-mode network is mainly active during rest and the IFC and DLPFC tend to be inactive , we can speculate that the TMS was applied during brain states similar to Visual state . As shown in Figure 4b , such V-state-dependent TMS over the IFC could induce a moderate prolongation of the percept duration ( Cohen’s d = 0 . 3 ) , which may be detected as a significant behavioural effect in the recent study . Methodologically , the current findings can be seen as another work highlighting the fact that certain brain-behaviour causality is changing so dynamically that we cannot behaviourally detect it in conventional neurostimulation methods that do not consider the temporal changes of the brain states ( Bergmann , 2018; Karabanov et al . , 2016 ) . Previous work has addressed this issue by controlling , inferring or monitoring the brain state: some studies controlled the brain state using external stimuli and applied the TMS when a specific sensory stimulus was presented to the participants ( Cattaneo et al . , 2010a; Cattaneo et al . , 2010b; Ezzyat et al . , 2017 ) ; a clinical research adopted emotional states as indicators of the neural activity and determined the timing of the deep brain stimulation based on such inference ( Scangos et al . , 2021 ) ; neurophysiological studies monitored neural activity and determined the timing of brain stimulation based on the frequency , phase or power of the neural signal ( Mrachacz-Kersting et al . , 2019; Polanía et al . , 2018; Schaworonkow et al . , 2019; Stefanou et al . , 2018; Zrenner et al . , 2018 ) . The current TMS method is categorised into the last group and could be seen as advancement of such direct-neural-monitoring-based brain stimulation . Differently from the previous work monitoring a single neural activity ( Mrachacz-Kersting et al . , 2019; Polanía et al . , 2018; Schaworonkow et al . , 2019; Stefanou et al . , 2018; Zrenner et al . , 2018 ) , the TMS system used here can track the brain state using neural activity patterns recorded from multiple remote brain regions . Considering the multi-regional brain state dynamics underpin complex cognitive activities ( Ezaki et al . , 2018; Watanabe et al . , 2014c; Watanabe and Rees , 2017 ) , such a multivariate monitoring approach could be a more effective manner to investigate more physiological brain-behaviour causality . Moreover , the combination of EEG-triggered TMS and energy landscape analysis may become foundation of a novel tool to control seemingly unstable behaviours . As shown in our longitudinal experiment ( Figure 10 ) , this brain-state-dependent neural stimulation system has accumulative effects despite its relatively weak stimulation . Given that the perceptual stability seen in this bistable perception test was relevant to autistic cognitive rigidity ( Watanabe et al . , 2019a ) , this longitudinal observation may become a basis for new clinical non-invasive neural interventions and accelerate the development of such state-dependent brain stimulation methods . These findings may not be directly applicable to other types of multistable visual perception , such as binocular rivalry , which is linked to lower-level brain architectures such as the visual cortex ( Haynes et al . , 2005; Lee et al . , 2005; Leopold and Logothetis , 1996; Meng and Tong , 2004; Miller et al . , 2000; Pettigrew and Miller , 1998; Polonsky et al . , 2000 ) . In fact , a comprehensive behavioural study reported the relative dissimilarity in perceptual switching rate between the current SFM-induced bistable perception and the binocular rivalry ( Cao et al . , 2018 ) . In contrast , the same study found the similarity in between the SFM-induced bistable perception and other fluctuating perception triggered by spinning dancer ( Liu et al . , 2012 ) and Lissajous-figure ( Weilnhammer et al . , 2014 ) . Given this , the current observations might be more applicable to types of bistable perception that requires construction of a 3D image from 2D motion compared to the other types such as the binocular rivalry . To interpret the current observations in psychological contexts such as predictive coding ( Brascamp et al . , 2018; Hohwy et al . , 2008; Weilnhammer et al . , 2021; Weilnhammer et al . , 2017 ) . more model-based neuroimaging studies , attention-tracking behavioural research and theoretical investigations would be necessary . This study has resolved the long-lasting controversy over prefrontal causal roles in multistable perception ( Brascamp et al . , 2018 ) and revealed distinct functions of the PFC in the brain state dynamics that underpin spontaneous perceptual inference . Furthermore , the combination of the brain-state tracking method and neural-activity-dependent brain stimulation system may re-ignite neurobiological investigation on state-dependent dynamic causality ( Silvanto et al . , 2008 ) in human cognition and become another foundation of more effective neural perturbation tools to intervene in some neuropsychiatric conditions .
This study consisted of seven experiments and one numerical simulation ( Figure 11 ) . First , 65 healthy adult individuals underwent two control experiments , in which we recorded their EEG signals while they were experiencing bistable visual perception induced by a structure-from-motion ( SFM ) stimulus ( Figure 1a ) . In the Control Experiment I , The EEG data were used to ( i ) identify individual brain dynamics during bistable visual perception and ( ii ) verify the locations of EEG electrodes and TMS stimulation site . In the Control Experiment II , we examined the accuracy of the brain-state tracking and state-dependent TMS system . Second , 35 of all the 65 individuals participated in the main experiment . In this experiment , using the information about individual brain state dynamics , we administered inhibitory TMS to the participants in state-/state-history-dependent manners . In parallel , we conducted a numerical simulation to predict how inhibitory TMS changed individual energy landscape structures and , resultantly , brain state dynamics . Thirty four of the 35 participants completed this experiment . This sample size of the main experiment was determined on a power analysis ( effect size = 0 . 5; power = 0 . 8; alpha = 0 . 05 for paired t-tests ) . Other 15 participants underwent a replication experiment that re-tested behavioural causality found in the main experiment , and 14 of them completed the experiment . Afterwards , the 14 participants underwent a conventional 30 min quadripulse TMS ( QPS ) ( Hamada et al . , 2007; Watanabe et al . , 2015; Watanabe et al . , 2014a ) experiment , in which they received the QPS over one of the three PFC regions ( DLPFC , IFC , and FEF ) during rest . The other 15 individuals participated in a validation experiment that examined whether excitatory TMS induced behavioural effects opposite to those caused by inhibitory TMS . The 63 individuals who completed either of these experiments participated in a longitudinal experiment , which evaluated the accumulative effects of the current neural stimulation method . All the 65 participants were right-handed adults ( Edinburgh Handedness Inventory laterality score = 75 ± 16 , mean ± sd ) . None of them had neurological , psychiatric or other medical history and was free from any contraindication to TMS experiments ( Wassermann , 1998 ) . Except for the control experiments and 30 min QPS experiment , all the experiments asked the participants to undergo the tests continually over several weeks; therefore , some of the participants did not complete the study allegedly due to their busyness of life . Consequently , the final sample size was 34 for the main experiment ( age = 23 . 1 ± 1 . 8; female = 12 ) , 14 for the replication experiment ( age = 22 . 2 ± 2 . 0; female = 8 ) , 15 for the validation experiment using excitatory TMS ( age = 22 . 7 ± 1 . 7; female = 7 ) , and 63 for the longitudinal study ( age = 22 . 8 ± 1 . 9; female = 27 ) . Including the participants who dropped out , no participant reported adverse effects throughout this study . This study was approved by Institutional Ethics Committees in RIKEN and The University of Tokyo ( 20-132 ) . The TMS protocols used here complied with the guideline issued by the Japanese Society for Clinical Neurophysiology and that by International Federation of Clinical Neurophysiology ( Rossi et al . , 2009 ) . All the participants provided written informed consents before any experiment and were financially compensated for their participation . The test design of bistable visual perception paradigm in this study is essentially the same as that used in our previous work ( Watanabe et al . , 2014c; Watanabe et al . , 2019a ) . The participants were presented with a structure-from-motion ( SFM ) stimulus ( Figure 1a ) , a sphere consisting of 200 sinusoidally moving white dots in a black background ( angular velocity , 120°/s ) with a fixation cross ( 0 . 1° × 0 . 1° ) at the centre of the 27-inch LCD monitor ( BenQ PD2710 , resolution: 2560 × 1440 ) . In each run , the participants were instructed to see the SFM stimulus for 90 s with their chins put on a chin rest . They were asked to push one of the three buttons according to their visual perception: one for upward rotation , another for downward rotation , and the other for unsure or mixture perception . After sufficient training sessions , the participants repeated this run six times in all the experiments except for the control one . The stimulus presentation and response recording were conducted with PsychToolbox three in MATLAB ( MathWorks , Inc ) . The proportion of the mixture perception was sufficiently small in all the participants , all the experiments ( 1 . 4% ± 0 . 5% of all stimulus presentation times , mean ± sd ) . Thus , we focused on the time during which participants were clearly aware of the direction of the rotation . For each participant , we measured the duration of the clear perception and calculated the median of the duration to evaluate their perceptual stability . The median duration was adopted because the perceptual durations showed long-tailed distributions . Throughout entire this study , we recorded EEG signals from seven regions of interest ( ROIs ) to monitor brain state transitions using a TruScan Research EEG system with 32 TMS compatible Ag/AgCl ring electrodes ( Deymed Diagnostic , Czech Republic ) . As in our previous work , the seven ROIs consisted of the right FEF ( x = 38 , y = 0 , z = 60 in MNI coordinates ) , DLPFC ( x = 44 , y = 50 , z = 10 ) , IFC ( x = 48 , y = 24 , z = 9 ) , anterior superior parietal lobule ( aSPL; x = 36 , y = –45 , z = 44 ) , posterior superior parietal lobule ( pSPL; x = 38 , y = –64 , z = 32 ) , lateral occipital complex ( LOC; x = 46 , y = –78 , z = 2 ) and V5 ( hMT/V5; x = 47 , y = –72 , z = 1 ) ( Figure 1b ) . These coordinates are based on the following previous studies: a study by Sterzer and colleagues ( Sterzer et al . , 2002 ) for the FEF; one by Knapen and colleagues ( Knapen et al . , 2011 ) for the DLPFC; one by Kleinschmidt and colleagues ( Kleinschmidt et al . , 1998 ) for the IFC; three studies by Kanai , Carmel and their colleagues ( Carmel et al . , 2010; Kanai et al . , 2011; Kanai et al . , 2010 ) for the aSPL and pSPL; one by Freeman and colleagues ( Freeman et al . , 2012 ) for the LOC and V5 . Using a stereoscopic neuro-navigation system ( Brainsight Neuronavigation , Rogue Research , UK ) and structural MRI brain images , we located the TMS-compatible EEG electrodes right above on these seven ROIs . Also , for the following calculation of Hjorth signals ( Hjorth , 1970; see Section 3 . 2 . 1 ) , we put three other EEG electrodes around each ROI electrode ( i . e . four electrodes were used for one ROI; Figure 12 ) . The other four electrodes ( i . e . 32 electrodes – four electrodes/ROI ×7 ROIs = 4 electrodes ) were located on Fpz , Oz , A1 , and A2 in accordance with 10–20 international system ( Seeck et al . , 2017 ) . These electrodes were firmly placed on the heads of the participants with elastic caps . After confirming that the impedance was less than 5 kΩ in all the electrodes , we recorded the EEG signals with a 32-channel amplifier ( TruScan EEG LT 32ch Headbox , Deymed Diagnostic; 6 kHz of analogue sampling frequency ) , in which the signals underwent low-pass filtering ( cut-off frequency: 1 . 25 kHz ) and were down-sampled to 3 kHz at almost simultaneously ( latency <5 ms ) . To deliver both inhibitory and excitatory stimulation , we used a quadripulse TMS system that can deliver a train of four monophasic magnetic pulses at once from interconnected four magnetic stimulators ( 70mm-diameter double coil , DuoMag MP , DeyMed Diagnostic , Czech Republic ) . To limit the neural effects of TMS into the target site , our EEG-triggered TMS system did not use conventional quadripulse TMS protocols ( Hamada et al . , 2009; Hamada et al . , 2007; Seeck et al . , 2017; Shirota et al . , 2010; Watanabe et al . , 2014a ) in which the TMS was continually administered throughout 30 min and whose after-effects are observable in connections with remote brain areas ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) and known to last for more than 60 min . Instead , in the main experiment , we conducted a single set of four-pulse 20 Hz monophasic TMS for neural suppression , whereas a set of four-pulse 200 Hz monophasic TMS was administered as an excitatory stimulation in the validation study . In addition , to obtain sufficient length of clean EEG data for the following brain-state tracking , we put ≥9 sec intervals between the stimulations and set the intensity of the TMS stimulation at a lower level ( 70 % of active motor threshold , AMT ) compared to that in the conventional quadripulse TMS protocol ( 90 % of AMT ) ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) . In EEG-triggered TMS , such weak stimulations were considered to induce sufficiently large neural effects ( Schaworonkow et al . , 2019 ) . The AMT was calculated based on motor evoked potentials ( MEPs ) of the right first dorsal interosseous ( FDI ) muscle of each participant in prior single-pulse TMS experiments: the AMT was set as the lowest TMS intensity that evoked a small response ( > 100µV ) when the participants maintained a slight contraction of the right FDI ( approximately 10 % of the maximum voluntary contraction ) in more than the half of 10 consecutive trials ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) . The MEPs of the right FDI were recorded using a pair of 9mm-diameter Ag/AgCl surface cup electrodes that were placed over the muscle belly and the metacarpophalangeal joint of the index finger ( Hanajima et al . , 2001 ) . Before offline analyses , the MEP signals underwent a temporal filter ( 100 Hz–3 kHz ) . The optimal place for the single-pulse TMS for the right FDI muscle was determined as the area over which the simulation induced the largest MEP . The mean AMT across the entire participant cohort in the three types of the current TMS experiment was 35 . 8% ± 7 . 9% of the maximum stimulator output . In one TMS condition , we placed the TMS coil over one of the PFC areas in the same method using the stereoscopic neuro-navigation system ( Brainsight Neuronavigation , Rogue Research , UK ) . We confirmed that the coil did not substantially move throughout the experiment by re-measing the location with the neuro-navigation system at the end of each experiment day . The green circles in Figure 4a show such end-of-the-day locations averaged across the four-day sessions in the main experiment . In this study , we analysed the EEG data in both offline and online manners . We described the details of the two types of EEG analysis in the following sections . The offline EEG analysis was applied to data obtained in the control experiment . We conducted the following conventional preprocessing ( Watanabe et al . , 2019b ) to the EEG data using MATLAB ( MathWorks , US ) and EEGLAB ( Delorme and Makeig , 2004 ) . Note that we used no behavioural response , such as the timing of perceptual switch , in the following preprocessing procedure and energy landscape analysis . First , the EEG data were referenced to the average across all the electrodes , down-sampled to 300 Hz and underwent a temporal filter ( 1–80 Hz ) . Then , we conducted an independent component analysis ( ICA ) to remove cardio-ballistic artefacts and other artefacts induced by eye blinks , eye movements and muscle activity . The ICA algorithm was based on short-time Fourier transforms and complex-valued version of FastICA with a robust measure of non-Gaussianity ( Hyvärinen et al . , 2010 ) . Next , we marked epochs whose mean global field power was too large ( > 5 SD of mean power across entire recording ) and excluded those time periods in all the following main analysis . We then filtered the data to delta ( 1–4 Hz ) , theta ( 4–8 Hz ) , alpha ( 8–13 Hz ) , beta ( 13–30 Hz ) , and gamma ( 30–80 Hz ) bands and estimated a Hilbert envelope amplitude for the gamma-band signal ( Deligianni et al . , 2014; Watanabe et al . , 2019b ) . We used the Hilbert envelope amplitude for the gamma band as a neural signal for each electrode , because the aim of the current EEG recording was to trace the brain state dynamics that was seen in our previous fMRI study ( Watanabe et al . , 2014c ) and the gamma-band signal dynamics were correlated with fMRI signals ( Deligianni et al . , 2014; Watanabe et al . , 2019b ) . Finally , after removing autocorrelation , we calculated a Hjorth signal for each ROI in the following sum-of-difference manner ( Hjorth , 1970 ) : Hjorth signal for ROIi = ( electrode just above ROIi – surrounding electrode 1 ) + ( electrode just above ROIi – surrounding electrode 2 ) + ( electrode just above ROIi – surrounding electrode 1 ) . We then conducted the energy landscape analysis ( Ezaki et al . , 2017; Gu et al . , 2018; Kang et al . , 2017; Watanabe et al . , 2014c; Watanabe and Rees , 2017 ) of the preprocessed datasets of the seven ROIs . For each participant , the EEG signals were concatenated across different runs . Then , as in our previous fMRI work on the brain dynamics ( Ezaki et al . , 2017; Watanabe et al . , 2014c; Watanabe et al . , 2014b; Watanabe and Rees , 2017 ) , we binarised each ROI time-series data using the temporal average of the signals as the thresholds . After this operation , a neural activity pattern of the seven ROIs at time t was described such as Vt=σ1t , σ2t , … , σNt , where σit represents a binary activity of ROIi at time t ( i . e . σit=+1∨-1 ) and N denotes the number of the ROIs ( here , N = 7 ) . The first goal of the energy landscape analysis is to calculate a hypothetical energy value for each neural activity pattern Vk1≤k≤2N , N=7 . The energy value is not related to metabolic consumption but an index that shows the stability of each state: the more stable and more frequent neural activity pattern should have the lower energy value . To this end , we fitted a pairwise maximum entropy model ( MEM ) to the seven binary time-series signals ( Watanabe et al . , 2014c; Watanabe and Rees , 2017 ) . We adopted this model because of its simplicity . It consists of two types of parameters: hi and Jij . The hi represents the basal activity of ROIi and Jij indicates a pairwise interaction between ROIi and ROIj . In terms of neurobiology , this model simply assumes that different brain regions have different intrinsic activity hi and different pairs of brain areas have different coupling strengths Jij . Now , how do we determine the two parameters ? In the energy landscape analysis , we determine hi and Jij so that the model-based ROI average activity σim and model-based average pairwise interactions σiσjm is sufficiently close to the average ROI activity ⟨σi⟩ and average pairwise interaction ⟨σiσj⟩ . We calculated the model-based mean ROI activity σim=Σl=12NσiVlPVl and model-based mean pairwise interaction σiσjm=Σl=12NσiVlσjVlPVl , where σiVk is the binary activity of ROIi in the activity pattern Vk and PVk is the appearance probability of an neural activity pattern Vk . This appearance probability is given as PVk=e-EVk/Σl=12Ne-EVl , where EVk=-Σi=1NhiσiVk-1/2Σi=1NΣj=1NJijσiVkσjVk . This formulation of PVk is determined by the principle of maximum entropy . That is , the information entropy of PVk is maximised by making the PVk obey Boltzmann distribution . In other words , to maximise the information entropy of PVk and minimise any possible constraints , we set the PVk in the form of Boltzmann distribution . Based on this definition , we adjusted hi and Jij until these the ⟨σi⟩m and ⟨σiσj⟩m were approximately equal to the empirically obtained ⟨σi⟩ and ⟨σiσj⟩ using a gradient ascent algorithm . The accuracy of this MEM fitting was examined by estimating a Pearson correlation coefficient between the model-based appearance probability and empirically obtained appearance probability and calculating a proportion of Kullback-Leibler ( KL ) divergence in this second-order model ( D2 ) to that in the first-order model ( D1 ) as follows Watanabe et al . , 2014c; Watanabe et al . , 2013; Watanabe and Rees , 2017: ( D1 – D2 ) /D1 . In the control experiment , the Pearson correlation was larger than 0 . 95 and the KL-divergence-based accuracy was larger than 84 % ( Figure 2a ) . Next , we built an energy landscape and searched for major brain states . The energy landscape was defined as a network of brain activity patterns Vk ( k = 1 , 2 , … , 2 N ) with their energy E ( Vk ) , in which two activity patterns were regarded as adjacent if and only if they took the opposite binary activity at a single ROI . We then searched for local energy minima , whose energy values were smaller than those of all the N adjacent patterns . We then examined hierarchal structures between the local minima by building disconnectivity graphs as follows Watanabe et al . , 2014c; Watanabe and Rees , 2017: ( i ) first , we prepared a so-called hypercube graph , in which each node representing a brain activity pattern was adjacent to the N neighbouring nodes . ( ii ) Next , we set a threshold energy level , Ethreshold , at the largest energy value among the 2 N nodes . ( iii ) We then removed the nodes whose energy values were≥ Ethreshold . ( iv ) We examined whether each pair of local minima was connected by a path in the reduced network . ( v ) We repeated steps ( iii ) and ( iv ) after moving Ethreshold down to the next largest energy value . We ended up with a reduced network in which each local min was isolated . ( vi ) Based on the obtained results , we built a hierarchical tree whose leaves ( i . e . , terminal nodes down in the tree ) represented the local minima and internal nodes indicated the branching points of different local minima . Based on this dysconnectivity graph , we then estimated basin sizes of the local minima as follows . We first chose a node i from the 2 N nodes . If any of its neighbour nodes had a smaller energy value than the node i , we moved to the neighbour node with the smallest energy value . Otherwise , we did not move , indicating that the node was a local min . We repeated this protocol until we reached a local min . The initial node i was then assigned to the basin of the local min that was finally reached . This classification procedure was repeated for all the 2 N nodes . The basin size was defined as the fraction of the number of the nodes belonging to the basin . The energy barrier between local minima l and m was defined based on the procedure of building the disconnectivity graph . When we built the disconnectivity graph by lowering the threshold energy level Ethreshold , we searched for the lowest Ethreshold at which the two local minima were still connected . The height of the energy barrier from local min l to m was then defined as the difference between the Ethreshold and the energy value for the local min l , whereas that from local min m to l was defined as the difference between the Ethreshold and the energy value for the local min m . Then , as in our previous work ( Watanabe et al . , 2014c ) , we summarised the local minima as follows: if the energy barriers from the local min l to m was lower than a threshold ( = 1 , here ) and the energy value of the local min l was larger than that of local min m , we regarded the local min l and its basin as elements of the basin of the local min m . By repeating this procedure , we found that in all the participants their brain activity patterns during bistable visual perception could be classified into any of the three major basins , which corresponded to Frontal-area-dominantly-active , Visual-area-dominantly-active and Intermediate state ( F , V and Int state ) . The energy barrier threshold for this summarisation was set at the same value as in our previous work ( Watanabe et al . , 2014c ) . Through this coarse-graining procedure , we defined the three major brain states and calculated structural indices of the energy landscapes ( i . e . the height of the energy barrier ) . In addition , this summarisation allowed us to classify all the nodes ( i . e . brain activity patterns ) on the energy landscape—except for nodes on the saddles—into any of the three major brain states for each participant . The vectors that were not classified into any of the three major brain states were labelled as ‘other state’ . This classification information would be used in the following EEG-triggered state-dependent TMS experiment . Note that a ‘brain state’ in this study is not a so-called ‘miscrostate’ in conventional EEG research; it indicates an activity pattern of multiple ( here , seven ) brain regions or a group of such activity patterns . Although other analyses , such as hidden Markov model ( HMM ) , can also identify brain states ( Baker et al . , 2014; Ezaki et al . , 2021; Miller and Katz , 2010; Vidaurre et al . , 2018 ) , we adopted the energy landscape analysis in this study because it was previously used to identify the brain states underpinning the bistable visual perception ( Watanabe et al . , 2014c ) . In the final part of the energy landscape analysis , we probed the brain state dynamics by a random-walk simulation on the energy landscape ( Watanabe et al . , 2014c; Watanabe and Rees , 2017 ) . This simulation was performed based on a Markov chain Monte Carlo method with the Metropolis-Hastings algorithm ( Girvan and Newman , 2002; Massen and Doye , 2005 ) . In this simulation , any brain activity pattern Vi could move only to a neighbouring pattern Vj . In other words , this simulation allowed the brain activity pattern to change the activity of only one ROI . Technically , first , one of such neighbouring patterns was randomly chosen . Then , whether actual movement occurred or not was determined at the probability Pij=min1 , eEVi-EVj . That is , if the Vi was more unstable than Vj ( i . e . EVi§amp;gt;EVj ) , the brain activity pattern should always move from Vi to Vj . This rule enhanced the movement to local minima . In the meantime , even if the Vi was more stable than Vj ( i . e . , EVi§amp;lt;EVj ) , there was some room to move to Vj , which prevented the brain activity pattern from being trapped in a local minimum forever . For each individual , we repeated this random walk 105 steps with a randomly chosen initial pattern and obtained a trajectory of the brain activity pattern such as V1 , V2 , … , V105 . After discarding the first 100 steps to minimise the influence of the initial condition , we then classified all the Vt into either of the major brain states ( i . e . Frontal , Visual , and Intermediate state ) and converted V101 , V102 , … , V105 to , for example , [Frontal , Frontal , Intermediate , … . , Visual] . Finally , we counted how long each of the major brain states continued in the brain state trajectory ( dwelling time ) and how often one major brain state transited to another major state ( transition frequency ) . Our previous work demonstrated a strong correlation between the percept duration and the length of return travel between Frontal state and Visual state via Intermediate state ( Watanabe et al . , 2014c ) . Given this , we compared the behaviourally observed percept duration to the length of the F–Int–V–Int–F travel that was calculated in the above random-walk simulation . In parallel with the random-walk simulation , we examined empirical brain state dynamics probing the binary neural vectors with seven elements , Vt . For this purpose , we first categorised all Vt into either of the three major brain states based on the classification information that was obtained in the above ‘Offline EEG analysis: structure of energy landscape’ section . The vectors that were not classified into any of the three major brain states were labelled as ‘Other state’ . To reduce the effects of signal fluctuation , we then applied the following temporal smoothing to the time-series of the brain states . First , in a sliding window manner ( window length = 10 ms ) , we calculated the appearance frequency for each brain state in the time window ( Figure 13a ) , which was assigned to each centre time point in the window as the representative appearance frequency for each brain state ( Figure 13b ) . Next , a Gaussian smoothing filter ( FWHM = 10 ms ) was applied to the representative appearance frequency curves ( Figure 13c ) . Based on the resultant appearance frequency values , we chose the most frequent brain state and assigned it as the brain state at the time point ( Figure 13d and e ) . Note that this temporal smoothing eliminated the ‘other state’ . This empirical brain state dynamics would be used to evaluate the accuracy of the following online EEG analysis . The Online EEG analysis was conducted in all the state-dependent TMS experiments , and its protocol was conceptually the same as that in the previous studies ( Chen et al . , 2013; Schaworonkow et al . , 2019; Stefanou et al . , 2018; Zrenner et al . , 2018 ) . The preprocessed EEG signals ( < 1 . 25 kHz and 3kHz-downsampled ) were input to the real-time target PC machine through a DAQ board ( sampling rate , 2 kHz ) , in which Simulink Real-Time model and xPC Target in Simulink ( MathWorks , US ) were running to analyse the EEG signals and trigger the TMS system via a TTL signal . We set the analysis model in the target PC through an Ethernet-connected host PC prior to the experiments for each participant . First , using the ‘sum-of-difference’ manner , we converted the EEG signals from 28 electrodes ( four electrodes/ROI × 7 ROIs ) into seven Hjorth signals at each timepoint . Then , in a sliding-window manner ( window length = 1000 ms , i . e . 2000 time points ) , we extracted the gamma-band signals ( 30–80 Hz ) by applying a fast Fourier transformation ( FFT ) to EEG data in every time window ( Figure 14a ) . The gamma band was chosen because ( i ) this EEG analysis aimed at reproducing our previous fMRI findings ( Watanabe et al . , 2014c ) and ( ii ) the EEG signals in the frequency window is considered to have temporal properties similar to those of fMRI signals ( Deligianni et al . , 2014; Watanabe et al . , 2019b ) . We then estimated the Hilbert envelope function for the gamma-band signals , whose amplitude would be used as neural signals ( Watanabe et al . , 2019b; Figure 14b ) . To reduce the edge effects ( Stefanou et al . , 2018; Zrenner et al . , 2018 ) , we trimmed 100 ms of the neural time-series data on both the edges of the time window ( Figure 14c ) . Using the remaining 800 ms of the neural signals , we then conducted an autoregressive forward prediction based on Yule-Walker equation ( order = 30 ) ( Chen et al . , 2013; Stefanou et al . , 2018; Zrenner et al . , 2018 ) and estimated 280 ms of neural signals that would come after the edge of the trimmed data ( Figure 14d ) . That is , given the 100 ms edge trimming , this calculation predicted the neural signal in a period between T = –100 ms and T = 180 ms when we set T = 0 at the actual recording timing ( i . e . the original terminal edge of the EEG data ) . The length of the forward prediction was set at 180 ms because ( i ) one inhibitory TMS in this study took 150 ms , ( ii ) the temporal smoothing required 5 ms more data points for its sliding-window-based calculations ( see Section Offline EEG analysis: temporal smoothin ) , and ( iii ) we prepared 25 ms buffer for the following signal processing to realise a nearly simultaneous EEG-triggered TMS system ( Chen et al . , 2013 ) . We obtained such 280 msec time-series signal for each ROI and used them in the following analysis . We then conducted online binarisation of the neural signals during the predicted time period . The binarisation threshold was calculated for each ROI based on the EEG data obtained in a control run that was conducted right before every TMS experiment . Technically , we applied the above-stated "Online EEG analysis: preprocessing" procedure to the EEG data during the control run and calculated the average of the Hilbert envelop amplitude for each ROI in each participant . These preprocessing and binarisation processes yielded a binary neural vector with seven elements ( + 1 or –1 ) at each time point in the 280 ms period ( from T = –100 to T = 180 ) . We then categorised the binary neural vectors into either of the three major brain states based on the classification information that was obtained in the offline energy landscape analysis in the control experiment . The binary neural vectors that were not categorised into any of the three major brain states were labelled as ‘Other state’ . Next , to reduce the effects of signal fluctuation , we applied the same temporal smoothing protocol to the resultant brain state vector as that used in the ‘Offline EEG analysis: temporal smoothing’ . As a result of these analyses and smoothing procedures , we obtained a series of representative brain states in a period between T = –95 ms and T = 175 ms . Note that this temporal smoothing eliminated the ‘Other state’ . The state-dependent TMS was performed based on the brain states that were predicted to appear in the forthcoming period between T = 25 ms and T = 175 ms . The TMS was administered only when the target brain state was dominant in the period ( > 90% of the brain states ) . For example , for the Frontal-state-dependent TMS , only if Frontal state was predicted to dominantly appear in more than 90 % of the time period , the real-time analysis machine sent a TTL trigger signal to the TMS device 21 ms after the EEG signals were input to the analysis machine . This 21 msec buffer for the online EEG analysis was chosen because the EEG data took ~3 ms to reach the analysis machine and the TTL signal from the analysis machine took ~1 msec to trigger the TMS . Given these latencies , the TMS was supposed to start almost 25 ms after the EEG recording ( T = 25 ms; Figure 15 ) . In addition , the 21 ms buffer was sufficiently long for the online EEG analysis . By connecting the TTL signal back to the DAQ board of the real-time analysis machine , we evaluated the signal processing delay using the EEG data collected in the Control experiment I and confirmed that the TTL signal reached back to the analysis machine 21 . 3 ± 0 . 02 ms ( mean ± s . d . , 21 . 1 ms–21 . 5 ms ) after the EEG signals were input into the machine . For the neural-history-specific TMS , we looked into both the forthcoming period ( T = 25 ms to T = 175 ms ) and the preceding period ( T = –95 ms to T = 25 ms ) . The TMS was administered only when both the two periods dominantly showed the target brain states . That is , for example , Post-Frontal Intermediate-specific TMS was administered only when Intermediate state was dominant in the forthcoming period ( >90% of the brain states ) and Frontal state was dominant in the neighbouring time window . Note that , for both the brain-state-/history-specific stimulation , once a TMS was administered , we did not conduct the next stimulation until at least 9 s passed . Such an interval gave the current brain-state-dependent TMS system a sufficient length of clean EEG data before the next TMS . In the above sections , we stated all the essential device setups and analysis procedures . In the following sections , we elaborated on the actual designs of the experiments using such devices and analyses protocols . The control experiment consisted of two parts: EEG part and EEG/TMS part . The Control Experiment I was conducted to ( i ) identify the brain state dynamics in the offline EEG analysis and ( ii ) validate the locations of the EEG electrodes , whereas Control Experiment II was performed to ( iii ) verify the accuracy of the online EEG analysis and brain-state-dependent TMS system . Both the parts employed the same 65 individuals and were performed at least two-day intervals . In the Control Experiment I , we collected EEG signals while the participants were conducting the test of bistable visual perception ( 1 . 5 min/run × 10 runs ) . The participants started this EEG recording sessions after sufficient training of the test . For the aim ( i ) , we applied the offline energy landscape analysis to the EEG data and examined whether the brain dynamics seen in our previous fMRI study ( Watanabe et al . , 2014c ) were qualitatively reproduced in the current EEG experiment . This analysis was conducted for the aim ( ii ) as well: if we successfully confirmed the reproducibility , such observations would provide face validation to the locations of the EEG electrodes . The details of this offline energy landscape were stated above ( see ‘Offline EEG analysis’ sections ) . In the Control Experiment II , the participants underwent test of bistable visual perception ( 1 . 5 min/run × 11 runs ) with the brain-state-dependent TMS system , which was almost the same as stated in the sections ( see ‘Device setup: TMS’ and ‘Online EEG analysis’ ) except for the locations of the TMS coil and an EEG electrode . We placed one of the 32 TMS-compatible Ag/AgCl ring electrodes—which was located on A1 in the original setting—on a wooden table that was set remotely from the participants . The TMS coil was placed over the electrode . Using the signal from this electrode , we measured when each TMS stimulation was conducted without causing significant artefacts on the EEG data . The scalp EEG data in the first run were used to calculate the threshold value for binarisation . Therefore , we did not apply TMS in the run . In the rest ten runs , the TMS was performed in the following five different conditions: Frontal-/Intermediate-/Visual-state-dependent and Post-Frontal-/Post-Visual-Intermediate-state-dependent conditions . Each temporal condition was tested in two runs . In the analysis , we performed both the offline and online analyses using the same EEG data . As stated above , the online EEG analysis required ( i ) the classification information to determine which major brain state would be assigned to each neural activity pattern and ( ii ) the threshold to binarise the neural signals . The requirement ( i ) was met by employing the results of the first half of the control experiment . For the requirement ( ii ) , the data during in the first run were used: we conducted the preprocessing part of the online EEG analysis , extracted a Hilbert envelop curve and calculated the mean value of the envelope amplitude for each ROI in each participant . Using the classification information and binarisation threshold , we performed the online analysis with the EEG data recorded during the rest of the runs ( 10 runs ) . This online EEG analysis was the same as that described above ‘Online EEG analysis’ sections , and we obtained a time-series vector representing brain state dynamics . For the sake of comparison , the offline analysis also used the EEG data recorded in the 2nd-11th run and estimated the brain-state dynamics . We applied all the above-mentioned ‘Offline EEG analysis’ procedures to the data except for the last random-walk simulation part . This offline analysis provided us with information about which major brain state should be assigned to each neural activity pattern . Based on the classification information , we then labelled the preprocessed EEG time-series data , which resulted in a time-series vector of brain state dynamics . In sum , these two types of EEG analysis gave us two time-series vectors representing the brain state dynamics for each participant . We then compared the two vectors and estimated how accurately the online analysis-based time-series vector predicted one that was based on the offline analysis . Technically , we counted timepoints whose brain states in the online analysis-based vector were the same as those in the offline analysis-based vector . We repeated such counting for each of the three major brain states in each participant and evaluated the accuracy . In parallel , we examined the temporal accuracy of the TMS . The brain-state-dependent TMS system was designed to administer a TMS train 25 ms after a specific brain state was detected in the EEG signals . Using the signal from the EEG electrode that was placed on the wooden table and covered by the TMS coil , we measured the actual latency from the brain-state detection to the TMS stimulation and calculated the difference between the empirical latency and the presumed time-lag ( i . e . 25 ms ) . We repeated this estimation for all the five temporal conditions in each participant . In the main experiment , we examined causal behavioural effects of state-/state-history-dependent TMS on the median percept duration in a test of SFM-induced bistable visual perception . The participants underwent TMS over three different brain sites ( DLPFC , IFC and FEF ) in five different neural timings ( F- . V- , Int- , Post-F Int- and Post-V Int-state-dependent TMS ) . In addition , we conducted an experiment using sham stimulation , in which mock TMS was applied over one of the three PFC sites . The choice of the target site in the sham condition was randomised and balanced across the participants . These experiments were conducted on four different days with more than one-week intervals . In each day , the participants received TMS over the same brain sites in the five different brain-state conditions ( Figure 16 ) : they underwent six runs of bistable visual perception tests for each condition , between which they had one-min rest and one control run without TMS . The order of the brain-state conditions was randomised and balanced across the participants . Before and after the TMS conditions , the participants also performed longer control runs . Behaviourally , we calculated the median percept duration for each TMS condition and control session . The behavioural effects of the TMS were evaluated as the ratio of the median percept duration in the TMS conditions to that in the control runs conducted at the beginning of the entire main experiment . Neurobiologically , we examined TMS’s effects on the brain state dynamics . Using the results of the online EEG analysis , we evaluated the dwelling time lengths for the three major brain states and transition frequencies between them after each TMS administration . To reduce the artefacts induced by the TMS , we discarded EEG signals recorded between the beginning of the TMS ( T = 25 ms ) and 100 ms after the end of the TMS ( T = 175 ms + 100 ms = 275 ms ) and analysed the data in the first 4 s time window in the remaining data ( i . e . the period from T = 275 ms to T = 4275 ms ) . To investigate the neurobiological changes in the brain state dynamics , we also numerically examined how the inhibitory TMS affected the structures of the individual energy landscape , such as the height of the energy barrier . This simulation was conducted based on the Hi and Jij that were obtained in the offline energy landscape analysis using the EEG data recorded during the control experiment . If ROIi was the target site of the inhibitory TMS , we removed all the binary neural vectors Vt whose element i , σi , was set at +1 ( i . e . active ) . Then , using the same Hi and Jij , we built the disconnectivity graph ( see ‘Offline EEG analysis: disconnectivity graph in energy landscape analysis’ ) and estimated the structural properties of the energy landscape ( see ‘Offline EEG analysis: structure of energy landscape’ ) . When the local minimum that represented any of the major brain states was removed in this simulation , we alternatively used the brain state that included the neighbouring neural vector as the major brain state . The neighbouring neural vector was defined as a vector that was different from the vector for the local minimum only in one element . In such numerical simulation , all the three major brain states were preserved whichever PFC site was inhibited . We compared the results of the numerical simulation , changes in the brain state dynamics and causal behavioural effects . First , we built hypotheses about the changes in the brain state dynamics based on the numerical simulation and tested them by comparing the results of the numerical simulation with the effects on the brain state dynamics . After confirming the validity of the hypotheses , we examined the relationship between the energy landscape changes , effects on the brain state dynamics and causal behavioural responses using the mediation analysis . We examined the replicability of the findings of the main experiments by repeating the main experiment with focusing on the seven conditions that yielded significant behavioural effects . This experiment employed a small but independent cohort ( originally N = 15; N = 14 after one participant dropped out ) . We also examined the validity of the main findings in experiments using the excitatory TMS ( see details on the TMS protocol in ‘Device setup: TMS’ section ) . As shown in our previous work ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) , the excitatory TMS was expected to induce behavioural and neural effects opposite to those yielded by the inhibitory TMS . To test this , we repeated the main experiment using the excitatory TMS with an independent cohort ( N = 15 ) . Like the replication experiment , we focused on the seven TMS conditions that induced significant behavioural changes in the main experiment . For comparison , we examined the behavioural effects of conventional 30 min inhibitory quadripulse TMS ( QPS ) by employing the 14 individuals who completed the replication experiment at least 1 month before . In this experiment , we adopted the inhibitory QPS protocols that were used in our previous studies ( Watanabe et al . , 2015; Watanabe et al . , 2014a ) . This TMS consisted of 360 consecutive bursts with 5 s intervals ( i . e . 30 min ) , and each burst comprised four monophasic 20 Hz TMS whose intensity was set at 90 % of AMT . We conducted this QPS over either of the three PFC regions using the same TMS device as used in the main experiment . In addition , we performed a sham condition; thus , the participants came to the lab four times with more-than-one-week interval . The order of the TMS conditions was randomised and balanced across the participants . In each day , the participants first completed the six control runs of bistable visual perception tests and then underwent the 30 min TMS session over one of the three PFC sites . In the sham condition , the TMS coil was placed over one of the three PFC areas , which was randomly chosen and balanced across the participants . During the TMS session , they were asked to rest with their eyes open . Ten minutes after the end of the TMS , they took the six runs of the bistable visual perception tests . We investigated accumulative effects of the brain-sate-/history-specific TMS by a longitudinal TMS experiment that employed the 63 participants who completed the main , replication or validation experiment . Like the replication and validation experiments , we focused on the seven TMS conditions that induced significant behavioural changes . In addition , we added three sham conditions , in which the TMS was administered over one of the three PFC regions at random timings . In total , the ten conditions were examined . The 63 participants were randomly assigned to two of the ten conditions: each non-sham condition had 13 individuals , whereas two sham conditions had 12 participants and the other sham condition was examined with 11 individuals . The accumulative effect of each TMS/sham condition was evaluated over nine weeks . In the first 8 weeks , the participants underwent weekly TMS/sham experiments . On each day , the participants underwent a TMS session ( six runs of bistable visual perception tests ) and two control sessions ( three runs ) before and after the TMS session . The details of the TMS protocol were the same as those for the main experiment . The brain site for the sham condition was randomly chosen from the three PFC areas for each participant for each session . The target brain sites were not changed during one period . For each day , we analysed the behavioural effects of the state-/state-history-dependent TMS by calculating proportional changes in the median percept durations within each day . We also tracked the behavioural effects over the 2-month experiment and examined whether the magnitude of the behavioural response at each day was significantly different from the first day ( week 0 ) . In the last week ( week 9 ) , the participants underwent the control sessions only . Such baseline responses were used to evaluate whether the two-month weekly TMS affected the baseline perceptual stability . We examined whether the current EEG system had sufficient specificity to distinguish between the neighbouring prefrontal activities ( i . e . the DLPFC activity and IFC activity ) . First , using the EEG data in Control Experiments ( N = 65 ) , we tested whether the EEG signals recorded from the two prefrontal areas were significantly different from each other . Technically , we performed the offline energy landscape analysis after one of the prefrontal activities was deliberately replaced with the other activity . If this replacement would not disturb the original observations , it would be highly likely that the current EEG system could not detect the difference in the neural activity between the DLPFC and IFC . Inversely , if the original results were distorted , we could infer that the current EEG system has sufficient detectability of differences between the two prefrontal activities . As a result , the significant brain-behaviour correlation seen in the original analysis ( r = 0 . 67; Figure 2j ) deteriorated when we set the DLPFC and IFC activities at the same values . If the IFC activity was replaced with the DLPFC activity , the correlation coefficient was reduced to 0 . 19 ( Figure 17a ) . When we used the IFC activity for the DLPFC activity , the coefficient was decreased to 0 . 23 ( Figure 17b ) . These results support the notion that the current EEG system can detect significant differences in neural activity between the two prefrontal regions . Second , using the EEG data in the main experiment ( N = 34 ) , we compared neural responses to TMS between the DLPFC and IFC ( Figure 17c ) . If our EEG system can have a sufficient SNR , the EEG signals recorded from the IFC would not be affected by TMS over DLPFC . In fact , when we administered TMS over DLPFC , the DLPFC signal showed a significant reduction , but the IFC signal did not ( Figure 17d ) . After TMS over IFC , the IFC signal was significantly decreased , whereas the DLPFC signal was not affected ( Figure 17e ) . In sum , these two analyses indicate sufficient spatial sensitivity and specificity in the current EEG system . In the current study , we adopted a derivation method ( i . e . Hjorth signal calculation ) ( Keren et al . , 2010; Pulvermüller et al . , 1997; Trujillo et al . , 2005; Zion-Golumbic et al . , 2010 ) and independent component analysis ( ICA ) ( Hassler et al . , 2011; Jung et al . , 2000; Lee et al . , 1999 ) to reduce the artefacts of microsaccades on gamma-band EEG signals ( Yuval-Greenberg et al . , 2008 ) . We confirmed the effectiveness of these signal processing procedures in an additional EEG experiment employing 30 healthy individuals ( age = 25 . 1 ± 2 . 1; female = 11 ) . In the experiment , the participants experienced the same bistable visual perception with EEG electrodes placed over slightly different places . Precisely , 28 electrodes were located around the same seven brain regions ( ROIs ) in the same manner as in the original experiment . The other four electrodes were placed around the eyes for electrooculography ( EOG ) in accordance with previous studies ( Croft and Barry , 2000; Elbert et al . , 1985; Shan et al . , 1995 ) : two EOG electrodes at the outer canthi of both the eyes ( HEOGL and HEOGR ) and two at the below and above the right eye ( VEOGI and VEOGS ) . Using an electrode near Pz as a reference , we calculated a radial EOG ( REOG ) signal as follows: REOG = ( HEOGR+ HEOGL + VEOGI+ VEOGS ) /4 – Pz . This REOG signal was used to detect the timings of microsaccades in the same manner as shown in a line of previous studies ( Croft and Barry , 2000; Elbert et al . , 1985; Keren et al . , 2010; Shan et al . , 1995 ) . In parallel with this microsaccade detection , we applied the same preprocessing procedures to the EEG signals recorded from the ROIs . Finally , we compared raw EEG signals with preprocessed signals around the timings of microsaccades and examined whether such preprocessing procedures could remove microsaccade-related artefacts . Technically , we compared the mean power between the microsaccade period ( ± 100 ms around a microsaccade ) and peripheral period consisting of two 100 ms time windows around the microsaccade period ( Figure 18 ) . As a result , we found that both the derivation method and ICA reduced the microsaccade-relevant artefacts . The derivation substantially removed the signal increase around a microsaccade ( e . g . see Figure 18b for effects on DLPFC signal ) , which was confirmed in all the ROI signals quantitatively ( Figure 18c ) . That is , for all the seven ROIs , the mean power in the microsaccade period was not significantly different from that in the peripheral period ( p > 0 . 05 in one-sample t-tests ) . The ICA was effective as well . The procedure reduced the signal increase induced by a microsaccade ( Figure 18d ) in all the ROIs ( p > 0 . 05 in one-sample t-tests; Figure 18e ) . In the case of multiple comparisons , basically , we conducted a two-way ANOVA ( participant× condition ) and post-hoc t-tests with Bonferroni correction . | A cube that seems to shift its spatial arrangement as you keep looking; the elegant silhouette of a pirouetting dancer , which starts to spin in the opposite direction the more you stare at it; an illustration that shows two profiles – or is it a vase ? These optical illusions are examples of bistable visual perception . Beyond their entertaining aspect , they provide a way for scientists to explore the dynamics of human consciousness , and the neural regions involved in this process . Some studies show that bistable visual perception is associated with the activation of the prefrontal cortex , a brain area involved in complex cognitive processes . However , it is unclear whether this region is required for the illusions to emerge . Some research has showed that even if sections of the prefrontal cortex are temporally deactivated , participants can still experience the illusions . Instead , Takamitsu Watanabe proposes that bistable visual perception is a process tied to dynamic brain states – that is , that distinct regions of the prefontal cortex are required for this fluctuating visual awareness , depending on the state of the whole brain . Such causal link cannot be observed if brain activity is not tracked closely . To investigate this , the brain states of 65 participants were recorded as individuals were experiencing the optical illusions; the activity of their various brain regions could therefore be mapped , and then areas of the prefrontal cortex could precisely be inhibited at the right time using transcranial magnetic stimulation . This revealed that , indeed , prefrontal cortex regions were necessary for bistable visual perception , but not in a simple way . Instead , which ones were required and when depended on activity dynamics taking place in the whole brain . Overall , these results indicate that monitoring brain states is necessary to better understand – and ultimately , control – the neural pathways underlying perception and behaviour . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Causal roles of prefrontal cortex during spontaneous perceptual switching are determined by brain state dynamics |
Insufficient protein-folding capacity in the endoplasmic reticulum ( ER ) induces the unfolded protein response ( UPR ) . In the ER lumen , accumulation of unfolded proteins activates the transmembrane ER-stress sensor Ire1 and drives its oligomerization . In the cytosol , Ire1 recruits HAC1 mRNA , mediating its non-conventional splicing . The spliced mRNA is translated into Hac1 , the key transcription activator of UPR target genes that mitigate ER-stress . In this study , we report that oligomeric assembly of the ER-lumenal domain is sufficient to drive Ire1 clustering . Clustering facilitates Ire1's cytosolic oligomeric assembly and HAC1 mRNA docking onto a positively charged motif in Ire1's cytosolic linker domain that tethers the kinase/RNase to the transmembrane domain . By the use of a synthetic bypass , we demonstrate that mRNA docking per se is a pre-requisite for initiating Ire1's RNase activity and , hence , splicing . We posit that such step-wise engagement between Ire1 and its mRNA substrate contributes to selectivity and efficiency in UPR signaling .
Proteins that travel along the secretory pathway first enter the lumen of the endoplasmic reticulum ( ER ) as unfolded polypeptides . Assisted by ER-resident enzymes , they undergo oxidative folding , modification , and assembly reactions . When properly folded , they are packaged into ER exit vesicles and travel to their final destination in the endomembrane system , on the cell surface or , after secretion , outside of the cell . Proteins that do not reach maturity are degraded by the proteasome after retrotranslocation into the cytosol ( via ER-associated degradation ) or by autophagy ( Ellgaard and Helenius , 2003; van Anken and Braakman , 2005; Bernales et al . , 2006 ) . Homeostasis in ER protein folding is achieved by fine-tuning the balance between the protein folding load and the protein folding capacity in the ER lumen ( Mori , 2009; Kimata and Kohno , 2011; Walter and Ron , 2011 ) . In yeast , Ire1 is the only known ER-stress sensor that responds to an accumulation of misfolded proteins in the ER lumen and transduces this information across the ER membrane . On the cytosolic side , Ire1 activation results in the non-conventional splicing of HAC1 mRNA , which is cleaved by Ire1's RNase domain at two splice sites , releasing a single intron ( Sidrauski and Walter , 1997 ) . Upon ligation of the severed exons , the spliced mRNA is translated to produce the Hac1 transcription activator that drives expression of UPR target genes to mitigate ER-stress ( Travers et al . , 2000; Walter and Ron , 2011 ) . Ire1 is activated through higher-order oligomerization ( Kimata et al . , 2007; Aragón et al . , 2009; Korennykh et al . , 2009 ) . Two ER-lumenal domain ( LD ) interfaces , IF1L and IF2L ( ‘L’ for lumenal ) , which were identified in the crystal structure of yeast Ire1 LD and validated by mutagenesis , mediate oligomeric assembly of the LD ( Credle et al . , 2005; Kimata et al . , 2007; Aragón et al . , 2009; Gardner and Walter , 2011 ) . Dimerization via IF1L yields a composite groove extending across the LDs of two Ire1 molecules ( Credle et al . , 2005 ) . Unfolded stretches of polypeptides bind within this groove of Ire1 LD , causing its oligomerization in vitro ( Gardner and Walter , 2011 ) . Proximal activation of the UPR in vivo coincides with the dissociation of Kar2 ( the yeast homolog of the ER-lumenal hsp70 chaperone BiP ) from Ire1 ( Kimata et al . , 2004; Pincus et al . , 2010; Walter and Ron , 2011 ) . Yet , Ire1 mutants with impaired Kar2 binding still respond to ER-stress , although the threshold for activation is lowered ( Kimata et al . , 2004; Pincus et al . , 2010; Walter and Ron , 2011 ) . Thus , misfolded proteins likely are direct ligands that activate Ire1 , while Kar2 fine-tunes the signaling ( Pincus et al . , 2010; Gardner and Walter , 2011 ) . On the cytosolic side of the ER membrane , Ire1 contains both a kinase and an RNase domain , which are tethered to the transmembrane domain via a linker ( Mori , 2009; Kimata and Kohno , 2011; Walter and Ron , 2011 ) . Three cytosolic assembly interfaces , IF1C , IF2C , and IF3C ( ‘C’ for cytosolic ) , were identified from the crystal structures of Ire1 kinase/RNase oligomers . IF1C creates back-to-back dimers of the kinase/RNase domains ( Lee et al . , 2008; Korennykh et al . , 2009 ) that stack onto each other with an axial rotation via IF2C and IF3C to form filaments with a helical arrangement of seven Ire1 dimers per turn ( Korennykh et al . , 2009; Walter and Ron , 2011 ) . The lumenal and cytosolic domain filaments predicted by the crystal structures have a different pitch and thus for steric reasons cannot be collinear . Instead , a two-dimensional arrangement of the two filaments , featuring ∼20–30 Ire1 molecules , provides a model for the higher-order assembly in vivo ( Korennykh et al . , 2009; Figure 1B ) . This model is compatible with the size of Ire1 foci observed by fluorescence microscopy ( Aragón et al . , 2009 ) and is sterically feasible despite the twists of the filaments on either side of the planar membrane , owing to the flexibility and length ( >100 Å ) of the linker domains on either side of the membrane , which can relieve the strain . Alternatively and not mutually exclusive , Ire1 clusters may be dynamic , such that constant rearrangements of the Ire1 molecules in clusters sustain transient intermittent oligomerization events on either side of the membrane . 10 . 7554/eLife . 05031 . 003Figure 1 . Oligomerization of Ire1's cytosolic domain is required for UPR signaling but not for Ire1 cluster formation or HAC1 mRNA recruitment . ( A ) Schematic of S . cerevisiae Ire1 . The ER-lumenal portion of Ire1 is divided in an N-terminal domain ( I , gray ) , a core lumenal—ER-stress-sensing—domain ( cLD , light blue ) , and BiP binding domain ( V , dark green ) , which is tethered via a transmembrane ( TM , orange ) stretch to Ire1's cytosolic portion that is composed of a linker ( L , brown ) , a kinase ( K , ochre ) , and an RNase ( R , purple ) domain ( Walter and Ron , 2011 ) . The activation loop ( light green ) and the αF–αEF ( pink ) loop protrude from the kinase domain ( Lee et al . , 2008; Korennykh et al . , 2009 ) . ( B ) A model architecture of a 24mer Ire1 cluster after oligomerization on either side of the ER membrane . Left: oligomerization via ER-lumenal interfaces IF1L ( tan ) and IF2L ( steel blue ) ( top ) and via cytosolic interfaces IF1C ( indian red ) , IF2C ( sea green ) , and IF3C ( plum ) ( bottom ) . The 24 Ire1 molecules are labeled ( A–H ) A′–H′ , and A′′–H′′ . IF1C-mediated back-to-back dimers are between A & B and C & D , etc . IF2C , is formed between Ire1 molecules A and D , C and F , and so on . The third interface , IF3C , is stabilized by a phosphate in the activation loop resulting from Ire1 trans-autophosphorylation . Dimerization via IF3C is therefore directional from B → D → F and from E → C → A , etc . ( Korennykh et al . , 2009 ) . Right: three-dimensional rendering of the same 24 Ire1 molecules colored as in ( A ) . ( C ) Top: schematic of HAC1 mRNA . The HAC1 open reading frame ( ORF ) is divided into two exons ( black ) . The intron ( gray ) base pairs with the 5′ UTR ( gray ) , causing stalling of ribosomes . Ire1 cleaves the intron at the splice sites indicated by blue arrowheads . The 3′ UTR ( gray ) harbors a stem-loop structure with the 3′ BE ( red ) that facilitates recruitment of the HAC1 mRNA to Ire1 foci ( Aragón et al . , 2009 ) . The 5′ m7G cap ( • ) and polyadenylation ( polyA ) signal are indicated . Middle: the green bar depicts the GFP ORF ( green ) that replaces part of the HAC1 sequence in the splicing reporter ( SpR ) . Translation of GFP only occurs when the intron is spliced from the mRNA , because removal of the intron by Ire1's endonuclease activity lifts a translational block caused by base pairing between the intron and the 5′ UTR ( Pincus et al . , 2010 ) . Bottom: 16 U1A binding sites ( violet ) were inserted into the 3′ UTR of the SpR mRNA , bearing the non-fluorescent GFP-R96A mutant ( GFP* , gray ) , downstream of the 3′ BE containing stem-loop . Binding of GFP-tagged U1A protein allows visualization of the mRNA ( Aragón et al . , 2009 ) . ( D ) Wild-type ( WT ) or ire1Δ cells , having a genomic copy of the SpR , were complemented with centromeric empty vector or bearing ire1 IF mutant alleles ( Aragón et al . , 2009; Korennykh et al . , 2009 ) as indicated . Top: SpR assay of cells . GFP fluorescence for 10 , 000 cells was measured by FACS analysis before or after ER-stress induction with 2 mM DTT for 2 hr , as described ( Pincus et al . , 2010 ) ; mean and s . d . are shown ( n = 2 ) . Bar diagrams for IF mutants are color-coded as in ( B ) left . The signal of DTT treated WT was set at 100% , while the signal reached in DTT treated ire1Δ cells due to auto-fluorescence of 14% was set as background ( light gray bar ) . Statistical significance in a Student's t-test of differences in splicing levels as compared with wild-type is indicated ( *p ≤ 0 . 05; **p ≤ 0 . 01 ) . Bottom: viability assay by 1:5 serial dilutions spotted onto solid media with 0 . 2 µg ml−1 of the ER-stress-inducer tunicamycin . Plates were photographed after 2–3 days at 30°C . ( E ) Localization of Ire1–GFP WT or IFL mutants before ( left panel , control ) and after ( right panels , DTT ) induction of ER-stress . ( F ) Schematic of Ire1–GFP and Ire1–mCherry with the fluorescent modules placed in the juxtamembrane region of the cytosolic linker . ( G ) Localization of Ire1–mCherry WT or IFC mutants , as well as SpRU1A mRNA decorated with U1A–GFP , after induction of ER-stress with DTT . ( E , G ) ER-stress was induced with 10 mM DTT for 45 min; imaging was performed in ire1Δ cells , complemented with Ire1 imaging constructs , as described ( Aragón et al . , 2009 ) . Scale bars represent 5 µm . ( H ) Immunoblot of hemagglutinin ( HA ) -tagged Ire1 protein from lysates from strains as in panel ( D ) and ( G ) . A sample from a strain that overexpressed HA-tagged Ire1 from a 2 µ plasmid served as a positive reference . Ire1 is denoted by an arrowhead . A background band , denoted by an asterisk ( * ) , conveniently serves as a loading reference . ( I ) Bar diagrams depict the percentage of Ire1 signal in foci ( red bars ) and the co-localization index expressed in arbitrary units ( yellow bars ) , as described ( Aragón et al . , 2009 ) , for SpRU1A mRNA recruitment to foci of Ire1 variants shown in ( G ) ; mean and s . e . m . are shown , n = 5–8 . There is no statistical significance in a Student's t-test of differences in foci formation and mRNA recruitments as compared with wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 00310 . 7554/eLife . 05031 . 004Figure 1—Source data 1 . ( A ) Source data for Figure 1D and Figure 1H . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 004 We previously have shown that Ire1 oligomerization allows selective recruitment of unspliced HAC1 mRNA to Ire1 clusters by virtue of a bipartite element in HAC1's 3′ untranslated region ( UTR ) ( Aragón et al . , 2009 ) , which we named the 3′ BE . The 3′ BE is effective in targeting mRNA to Ire1 clusters as long as they are translationally repressed ( Aragón et al . , 2009 ) . Stalling of HAC1 mRNA translation is afforded through base pairing between the intron and the 5′ UTR ( Rüegsegger et al . , 2001 ) . Moreover , the in vitro endonuclease activity of Ire1 kinase/RNase domains is highly cooperative , indicating that oligomerization rather than dimerization leads to RNase activation ( Korennykh et al . , 2009 ) . Intriguingly , the capacity of the kinase/RNase domains to oligomerize in vitro depends on a short stretch of the cytosolic linker that extends N-terminally from the kinase domain ( Korennykh et al . , 2009 ) . In this study , we report that although oligomeric assembly of kinase/RNase domains is essential for activation of Ire1's in vivo mRNA processing capacity , it is not required for driving formation and maintenance of Ire1 clusters or for recruitment of its mRNA substrate . Instead , we discovered that a conserved positively charged element in Ire1's cytosolic linker mediates mRNA docking onto Ire1 clusters . Primary docking of the mRNA to this site is required for subsequent processing of the mRNA by Ire1's RNase domain . The staged way in which HAC1 mRNA is channeled to become subject to Ire1's endonuclease activity enhances efficiency and selectivity in the process and , thus , fidelity in UPR signaling .
Efficiency of HAC1 mRNA splicing ( Figure 1C ) depends on clustering of ER-lumenal domains in vivo ( Kimata et al . , 2007; Aragón et al . , 2009 ) and of cytosolic domains in vitro ( Korennykh et al . , 2009 ) . To assess the contribution of cytosolic oligomerization events to Ire1 function in vivo , we analyzed mutations in the interfaces that govern Ire1 kinase/RNase oligomeric assembly by complementing ire1Δ yeast with centromeric plasmids . Driven by its autologous promoter , expression of ( wild-type or mutant ) Ire1 from these plasmids is at near-endogenous levels ( Aragón et al . , 2009 ) . Disruption of IF1C abolished RNase function ( Figure 1D ) as monitored by the loss of expression of a fluorescent reporter protein ( SpR; Figure 1C ) , whose levels report on Ire1 RNase activity ( Pincus et al . , 2010 ) . Consequently , growth under ER-stress conditions was impaired ( Figure 1D ) , as previously reported ( Lee et al . , 2008 ) . Disruption of IF2C likewise disrupted RNase function and survival under ER-stress ( Figure 1D ) , consistent with in vitro analyses ( Korennykh et al . , 2009 ) . Mutations in IF3C led to a milder phenotype , sustaining intermediate levels of splicing and growth ( Figure 1D ) , similar to the lumenal interface mutants that are shown for comparison ( Aragón et al . , 2009; Figure 1D ) . As expected , mutations in Ire1 did not affect growth under non-ER-stress conditions since growth of ire1Δ yeast is then also unaffected ( Aragón et al . , 2009 ) . Under control conditions , Ire1 distributed diffusely throughout the ER but clustered into discrete foci upon ER-stress as visualized with fluorescently tagged Ire1–GFP ( Aragón et al . , 2009; Figure 1E , F ) . HAC1 mRNA is recruited to Ire1 foci under ER-stress via the 3′ BE targeting signal in the mRNA ( Aragón et al . , 2009; Figure 1C ) , which can be visualized in cells expressing HAC1 splicing reporter mRNA containing U1A binding sites ( SpRU1A , Figure 1C ) and GFP-tagged U1A RNA-binding protein ( Brodsky and Silver , 2000; Aragón et al . , 2009 ) . Disruption of lumenal interfaces interfered with Ire1 clustering ( Kimata et al . , 2007; Aragón et al . , 2009; Figure 1E ) and , consequently , mRNA recruitment ( Aragón et al . , 2009 ) . Disruption of cytosolic interfaces also compromised splicing activity of Ire1 and , hence , growth under ER-stress conditions ( Figure 1D ) , but foci formation was unaffected ( Figure 1G ) . Moreover , expression levels of the IFC mutants were comparable to wild-type ( Figure 1H ) , and the percentage of Ire1 in foci as well as the extent of co-localization of mRNA at those foci , as determined by a customized MatLab script ( Aragón et al . , 2009 ) , was similar for wild-type and IFC mutants ( Figure 1G , I ) . Thus , all three cytosolic oligomeric interfaces are key for Ire1 function in vivo but not for Ire1 stability or its capacity to cluster and recruit HAC1 mRNA . Tampering with the oligomeric assembly of the cytosolic domains of Ire1 gravely affected Ire1's endonuclease activity . To explore directly whether enzymatic activity of Ire1 is necessary for foci formation , we extended our assays using site-specific mutations that selectively disrupt Ire1's kinase and RNase activities ( ‘KD’ and ‘RD’ for kinase- and RNase-deficient , respectively ) . In line with previous results ( Papa et al . , 2003; Korennykh et al . , 2011 ) , splicing and survival under ER-stress were impaired in ire1 ( KD ) and abolished in ire1 ( RD ) mutant cells , while mutant Ire1 expression levels were similar to wild-type ( Figure 2A ) . Yet , Ire1 foci formation and mRNA recruitment were unimpeded in either mutant ( Figure 2B , F ) , indicating that neither of the enzymatic activity is required for Ire1 to recruit its mRNA substrate . 10 . 7554/eLife . 05031 . 005Figure 2 . The kinase and RNase domains of Ire1 are dispensable for foci formation and mRNA recruitment . ( A ) Splicing reporter assay before or after ER-stress induction with 2 mM DTT for 2 hr ( top ) , Western blot of Ire1 ( middle ) , and viability assay under ER-stress conditions ( 0 . 2 µg ml−1 tunicamycin; bottom ) were performed in ire1Δ yeast containing a genomic copy of the SpR , complemented with wild-type ( WT ) , kinase dead ( KD ) , RNase dead ( RD ) , and RNase truncation ( ΔR ) mutant alleles of ire1 . Maximal ( 100% ) and background level ( 14% , light gray bar ) fluorescence are set as in Figure 1A . Mean and s . d . are shown ( n = 2 ) . Statistical significance in a Student's t-test of differences in splicing levels as compared with wild-type is indicated ( **p ≤ 0 . 01 ) . The arrowheads denote ( mutant or truncated ) Ire1 protein and the asterisk a background band on the immunoblot as in Figure 1H . ( B , C ) Top: schematic of the mCherry-tagged versions of the same ire1 mutants as in ( A ) as well as a kinase/RNase truncation ( ΔKR ) mutant , color-coded as in Figure 1A , except defective domains are black . ( D ) Schematic of a chimeric mRNA , SL-PGK1-3′ hac1U1A , which is PGK1U1A , bearing in its 3′ UTR the stem-loop structure with the 3′ BE of the HAC1 mRNA and in its 5′ UTR a small stem-loop ( green ) that confers translational repression ( Aragón et al . , 2009 ) . ( B , C , E ) Localization of Ire1–mCherry and of U1A–GFP decorating either SpRU1A ( B , C ) or SL-PGK1-3′ hac1U1A ( E ) mRNA . ER-stress was induced with 10 mM DTT for 45 min; imaging was performed of ire1Δ cells , complemented with Ire1 imaging constructs , as depicted . Scale bars represent 5 µm . ( F ) Bar diagrams depict the percentage of Ire1 signal in foci ( red bars ) and the co-localization index for mRNA recruitment into foci of Ire1 variants shown in B , C , and E ( mean and s . e . m . , n = 5–10 ) . There is no statistical significance in a Student's t-test of differences in foci formation and mRNA recruitments as compared with wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 00510 . 7554/eLife . 05031 . 006Figure 2—Source data 1 . ( A ) Source data for Figure 2A and Figure 2F . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 006 Strikingly , even removal of the entire RNase domain , either alone ( ‘ΔR’ ) or together with the kinase domain ( ‘ΔKR’ ) , left Ire1 foci formation and mRNA recruitment unaffected ( Figure 2C , F , Figure 2—source data 1 ) . These results show that oligomeric assembly of the kinase–RNase domain is dispensable for the Ire1 clustering . Moreover , these findings indicate that Ire1 must harbor an mRNA docking site within the linker that tethers the kinase/RNase to the transmembrane region because it is the only remaining cytosolic portion in the ire1 ( ΔKR ) mutant . A heterologous mRNA , SL-PGK-3′ hac1U1A—which contains the 3′ BE of HAC1 mRNA , but lacks the HAC1 mRNA intron and splice sites , and a small stem-loop in its 5′ UTR to repress its translation ( Aragón et al . , 2009; and Figure 2D ) —was also efficiently recruited to foci in ire1 ( ΔKR ) mutant cells ( Figure 2E , F , Figure 2—source data 1 ) . These results were surprising , since mRNA recruitment to Ire1 serves to engage the splice sites in the HAC1 mRNA with Ire1's endonuclease domain for cleavage , yet neither the endonuclease domain nor the splice sites ( or their context ) are required for mRNA docking onto Ire1 clusters . Instead , our findings indicate that the core elements sufficient for the recruitment of HAC1 mRNA to and docking of the mRNA onto Ire1 clusters are contained in the 3′ BE of the mRNA and in Ire1's cytosolic linker domain . Ire1 is the only ER-stress sensor that is conserved in all eukaryotes . The kinase/RNase domains and the core lumenal ER-stress sensing domain are conserved , but other domains , including the cytosolic linker domain , show negligible sequence conservation ( Figure 3A ) . The linker greatly varies in length between species but consistently harbors an unusually high number of basic and acidic residues . In fungal species , a short basic sequence stretch ( henceforth referred to as ‘[+]-box’ ) displays recognizable homology . In particular , sequence alignment reveals strict conservation of one lysine and two arginine residues as well as three glycine residues that intersperse the basic residues . The [+]-box is flanked on either one or both sides by acidic sequences . 10 . 7554/eLife . 05031 . 007Figure 3 . The cytosolic linker of Ire1 harbors a positively charged motif that is key for mRNA recruitment and splicing . ( A ) Conservation of Ire1 . Top: mapped onto a schematic of Ire1 domains bordered by residues of which the number is denoted , bar diagrams display relative conservation of the Saccharomyces cerevisiae Ire1 protein sequence to homologs ( lower , left ) from other fungal species Kluyveromyces lactis , Candida glabrata , Aspergillus nidulans , Coccidioides posadasii , Gibberella zeae , Magnaporthe grisea , Neurospora crassa , and Schizosaccharomyces pombe , as well as from the animals Caenorhabditis elegans and Drosophila melanogaster , the two paralogues from the plant Arabidopsis thaliana ( a and b ) and from Homo sapiens ( α and β ) . Domains are color-coded as in Figure 1A , except signal peptides ( SP ) are black; light green represents a loop inserted into the kinase domain of the A . thaliana Ire1s and crimson denotes C-terminal extensions in animal Ire1s . Expanded view ( middle ) of the linker domains that are aligned based on the stretch ( gray box ) for which the sequence alignment is shown on the right . Strictly conserved residues among fungal species except S . pombe are boxed . ( A ) lower right , ( C ) top , Basic ( arginine and lysine ) residues are shown in blue and acidic ( aspartate and glutamate ) residues in red . Glutamines replacing arginines and lysines in the Q-box mutant are black . ( A ) lower right , ( B , C ) Position of the [+]-box is indicated . ( B , D–F ) Localization of Ire1–GFP or Ire1–mCherry and of U1A–GFP decorated SpRU1A mRNA . Imaging was performed in ire1Δ yeast complemented with a genomic copy of C-terminally mCherry-tagged ( ΔKR ) or ( ΔKR/Δ[+]-box ) ire1 mutant alleles , as schematically shown ( B top right ) , or with plasmids encoding IRE1 wild-type ( C ) or ire1 linker mutants ( E , F ) , with the fluorescent modules GFP ( E ) and mCherry ( C , F ) placed in the αF–αEF loop , as schematically shown ( D right ) , before ( D upper panels , E , control ) and after ( B , D lower panels , F , DTT ) induction of ER-stress with 10 mM DTT for 45 min . Scale bars represent 5 µm . Bar diagrams depict the percentage of Ire1 signal in foci ( red bars ) and the co-localization index for mRNA recruitment into foci of Ire1 variants ( mean and s . e . m . , n = 5–10 ) ( B , bottom right , G ) . Statistical significance in a Student's t-test of differences in foci formation and mRNA recruitments as compared with wild-type is indicated ( *p ≤ 0 . 05; **p ≤ 0 . 01 ) . ( C ) Schematic of linker domains with mutations or truncations as in ( A ) ( top ) . Splicing reporter assay before or after ER-stress induction with 2 mM DTT for 2 hr ( left , middle ) , Western blot of Ire1 ( left , bottom ) , and viability assay under ER-stress conditions ( 0 . 2 µg ml−1 tunicamycin; right , bottom ) . Assays were performed in ire1Δ yeast containing a genomic copy of the SpR , complemented with either IRE1 wild-type or ire1 linker mutants with mCherry in the αF–αEF loop . Mean and s . d . are shown ( n = 2 ) . Maximal ( 100% ) and background level ( 14% , light gray bar ) fluorescence are set as in Figure 1A . Statistical significance in a Student's t-test of differences in splicing levels as compared with wild-type is indicated ( **p ≤ 0 . 01 ) . The arrowheads denote ( mutant or truncated ) Ire1 protein and the asterisk a background band on the immunoblot as in Figure 1H . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 00710 . 7554/eLife . 05031 . 008Figure 3—Source data 1 . ( A ) Source data for Figure 3B , Figure 3C and Figure 3G . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 008 All fungal species with a [+]-box containing Ire1 linker have a conserved 3′ BE in their HAC1 mRNAs ( Aragón et al . , 2009 ) . The only fungal species we found with a markedly divergent basic motif is Schizosaccharomyces pombe , which lacks a HAC1 gene altogether ( Kimmig et al . , 2012 ) . In both Ire1 paralogs of Arabidopsis thaliana , the linkers harbor a basic motif that diverges from the fungal [+]-box ( Figure 3A ) , but that motif is conserved among plants ( not shown ) . Conversely , animal species lack any such recognizable motif ( Figure 3A ) . Indicative of an important role for the [+]-box in UPR signaling is that it adorns the short linker extension , which facilitated oligomerization and markedly enhanced endonuclease activity of recombinant kinase/RNase domains in vitro ( Korennykh et al . , 2009 ) . To analyze the role for the [+]-box in Ire1 function in vivo , we first truncated Ire1 further such that the [+]-box was deleted ( Figure 3B ) . Since we could not obtain a centromeric plasmid of this construct , as it was toxic for Escherichia coli , we created ire1Δ strains with a genomic copy of the ire1ΔKR or ire1ΔKR/Δ[+]-box mCherry-tagged transgenes in the LEU2 locus . As expected , Ire1 clustering and mRNA recruitment were still at wild-type levels for ire1ΔKR when genomically integrated ( Figure 3B ) , similar to what we observed when ire1ΔKR was expressed from a centromeric plasmid ( Figure 2C , F ) . Further removal of the [+]-box did not affect Ire1 clustering but markedly reduced mRNA recruitment ( Figure 3B ) . This finding suggests that the [+]-box is key for the docking of mRNA onto Ire1 clusters . Moreover , given that the extent of clustering of the ire1 ( ΔKR/Δ[+]-box ) mutant was similar to wild-type , we conclude that Ire1 clustering is driven by the lumenal domain alone: neither cytosolic oligomeric assembly ( as facilitated by the kinase/RNase domains ) nor mRNA docking ( as facilitated by the [+]-box ) is required for Ire1 foci formation . Next , we set out to explore the role of the [+]-box in the context of the full-length Ire1 using the mutants and truncations depicted in Figure 3C . The experiments presented so far employed Ire1 variants bearing GFP or mCherry modules in the linker ( Aragón et al . , 2009; Pincus et al . , 2010; Rubio et al . , 2011; Figure 1F; Figure 2B , C ) . To avoid interference with mutations in the linker , we relocated the fluorescent modules to the αF–αEF loop in the Ire1 kinase domain ( Figure 3D ) . By contrast to the activation loop ( Figure 1A ) , which becomes hyperphosphorylated due to Ire1's kinase activity , the αF–αEF loop is poorly conserved ( Figure 3A ) and dispensable for Ire1 activity in vitro ( Korennykh et al . , 2009 ) . Accordingly , despite the insertion of a fluorescent protein module into the αF–αEF loop , Ire1 still faithfully formed foci and recruited SpRU1A mRNA under ER-stress conditions ( Figure 3D , G ) and was indistinguishable from untagged wild-type Ire1 in splicing efficiency and growth under ER-stress conditions ( Figure 3C ) . Removal of the [+]-box abolished splicing and growth under ER-stress ( Goffin et al . , 2006; Figure 3C , ‘Δ[+]-box’ ) , as did replacement of lysines and arginines within the [+]-box with glutamines ( Figure 3C , ‘Q-box’ ) . These mutants were expressed at similar levels as the wild-type ( Figure 3C ) . However , extensive truncation of the rest of the linker ( ‘Δ55’ ) —to a length comparable to the shortest among all Ire1 homologs ( 61 amino acids; Figure 3A ) —had no effect on growth ( Figure 3C ) . The [+]-box was shown previously to act as a nuclear import signal when inserted into heterologous proteins ( Goffin et al . , 2006 ) . For Ire1 , however , such a role is unlikely , since tampering with the [+]-box had no effect on the distribution throughout the ER on either of non-clustered Ire1 in the absence ( Figure 3E ) or of Ire1 foci in the presence ( Figure 3F ) of ER-stress . Conversely , removal or mutagenesis of the [+]-box did affect the co-localization of the SpRU1A mRNA to Ire1 foci ( Figure 3F , G ) . Shortening of the linker , while leaving the [+]-box intact , did not impact on co-localization of the mRNA with Ire1 clusters ( Figure 3F , G ) . Thus , the splicing and growth defects ( Figure 3D ) correlated with impairments in mRNA recruitment ( Figure 3F ) . We pose that the [+]-box is a key element for Ire1 function in vivo , because it facilitates docking of the mRNA onto Ire1 clusters . To assess which of the positively charged residues within the [+]-box are important for mRNA docking , we next replaced each individual lysine or arginine with threonine . Splicing and growth under ER-stress were abrogated in mutants of either conserved arginine , R647 or R650 ( Figure 4A ) . A third , less-conserved arginine , R645 , appeared almost equally important for Ire1 function ( Figure 4A ) . Replacement of all non-basic residues ( including the three conserved glycines ) in the [+]-box with arginines or lysines ( Figure 4A , ‘all [+] mutant’ ) disrupted splicing and survival upon ER-stress almost to the level of the mutants of the conserved arginines . This observation suggests that a positively charged cluster alone is not sufficient but that the glycines afford proper display ( spacing or three-dimensional positioning ) of the crucial arginine side chains . 10 . 7554/eLife . 05031 . 009Figure 4 . Three arginines in Ire1's linker are essential for mRNA docking . ( A ) Splicing reporter assay before or after ER-stress induction with 2 mM DTT for 2 hr ( top ) and viability assay under ER-stress conditions ( 0 . 2 µg ml−1 tunicamycin; bottom ) . Assays were performed in ire1Δ yeast containing a genomic copy of the SpR , complemented with ire1 mutants having within the [+]-box a single arginine or lysine replaced with a threonine , as indicated , or all non-positively charged residues replaced with arginines or lysines ( KKKRKRKKRKRKKRRK; all [+] ) . Bar diagrams reporting on mutants of positively charged residues are blue with conserved residues in a darker shade; bar diagrams of the all [+] box mutant in brown . Statistical significance in a Student's t-test of differences in splicing levels as compared with wild-type is indicated ( *p ≤ 0 . 05; **p ≤ 0 . 01 ) . Maximal ( 100% ) and background level ( 14% , light gray bar ) fluorescence are set as in Figure 1D . ( B ) Localization of Ire1–mCherry and U1A–GFP decorated SpRU1A mRNA . Imaging was performed in ire1Δ yeast complemented with ire1 linker mutants having the mCherry module in the αF–αEF loop . Scale bar represents 5 µm . ( A , B ) Schematics of [+]-box variants are color-coded as in Figure 3A with arrow heads denoting the position of point mutations in black . ( C ) Co-localization index for mRNA recruitment into foci of Ire1 variants shown in B ( mean and s . e . m . , n = 5–10 ) . Statistical significance in a Student's t-test of differences in foci formation and mRNA recruitments as compared with wild-type is indicated ( *p ≤ 0 . 05 ) . ( D ) Immunoblot as in Figure 1H of Ire1 of lysates from strains in panel ( B ) ; the arrowhead denotes Ire1 protein and the asterisk a background band as in Figure 1H . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 00910 . 7554/eLife . 05031 . 010Figure 4—Source data 1 . ( A ) Source data for Figure 4A and Figure 4C . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 010 Mutation of no other basic residue , including the single conserved lysine ( K642T ) , markedly disturbed splicing or growth under ER-stress ( Figure 4A ) . Accordingly , mRNA recruitment was unaffected in any of these mutants ( not shown ) . By contrast , mRNA recruitment was impaired for each mutant of the three key arginines , as well as for the ‘all [+]’ mutant ( Figure 4B , C ) , though they were expressed at similar levels to wild-type ( Figure 4D ) . The defect in mRNA recruitment for the R647T and R650T mutants in particular was as strong as the impairment caused by the deletion of the entire [+]-box ( Figure 3F ) . We conclude that three arginine residues , R645 , R647 , and R650 , in a proper spatial arrangement are key for the [+]-box to sustain mRNA docking . Removal of the 3′ BE targeting element from the mRNA resulted in a complete loss of co-localization with Ire1 foci ( Aragón et al . , 2009; Figure 5A ) . In contrast , the loss of a functional [+]-box did not completely eliminate co-localization of the mRNA with Ire1 foci; rather , co-localization levels were reduced two- to threefold compared to the wild-type ( Figure 3B , F; Figure 4B , C ) . Apparently , targeting of the mRNA via the 3′ BE permits residual co-localization with the [+]-box mutants . Such residual co-localization was evident even for the ire1 ( ΔKR/Δ[+]-box ) mutant ( Figure 3B ) , implying that 3′ BE-mediated targeting occurs irrespective of any cytosolic portion of Ire1 . 10 . 7554/eLife . 05031 . 011Figure 5 . Step-wise targeting and docking of mRNA are pre-requisite for activating Ire1 . ( A ) Localization of mCherry-tagged Ire1 and SpR Δ3′ BEU1A mRNA decorated with U1A–GFP . ER-stress was induced with 10 mM DTT for 45 min; imaging was performed in ire1Δ cells , complemented with wild-type IRE1 having the mCherry module in the αF–αEF loop ( left ) . Scale bar represents 5 µm . Co-localization index for mRNA recruitment into foci ( mean and s . e . m . , n = 10 ) ( right ) . Statistical significance of the difference in mRNA recruitment of SpR Δ3′ BEU1A as compared with SpRU1A to Ire1–mCherry clusters ( Figure 3D , G ) was tested using a Student's t-test ( **p ≤ 0 . 01 ) . ( B ) Schematic of mRNA docking ‘bypass’ . The U1A module placed in the αF–αEF loop of Δ[+]-box mutant ire1 facilitates binding of HAC1U1A mRNA via its U1A motifs . ( C ) Splicing was measured by quantitative RT-PCR before or after ER-stress induction with 2 mM DTT for 30 min ( top ) and viability assay under ER-stress conditions ( 0 . 2 µg ml−1 tunicamycin ) of hac1Δ/ire1Δ yeast complemented with centromeric plasmids bearing wild-type IRE1 or Δ[+]-box mutant ire1 either untagged or tagged with the U1A module in the αF–αEF loop , as well as with centromeric plasmids bearing wild-type HAC1 , HAC1U1A , or HAC1 Δ3′ BEU1A . For display of RT-PCR results , the signal for hac1Δ/ire1Δ yeast complemented with wild-type HAC1 and IRE1 under ER-stress conditions was set at 100%; mean and s . d . ( n = 2 ) are shown . Statistical significance in a Student's t-test of differences in splicing levels as compared with wild-type is indicated in black and of differences in splicing levels compared with the ‘bypass’ ( HAC1U1A + ire1 Δ[+]-box-U1A ) is indicated in red ( *p ≤ 0 . 05; **p ≤ 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 01110 . 7554/eLife . 05031 . 012Figure 5—Source data 1 . ( A ) Source data for Figure 5A and Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 05031 . 012 Paradoxically , the residual co-localization of the mRNA with Ire1 clusters that have a defective [+]-box ( Figure 3F , 4B , C ) was not sufficient to sustain splicing ( Figure 3C , 4A ) , despite the fact that Ire1's kinase and RNase domains were intact . Since the [+]-box is unlikely to contribute to the core enzymatic function of Ire1 , as it is not conserved in all eukaryotes , we reasoned that the docking mechanism itself is important to initiate splicing activity . To test this idea , we created a synthetic bypass for [+]-box mediated docking by inserting the U1A RNA binding module into the αF–αEF loop of the ire1 ( Δ[+]-box ) mutant . We reasoned that the insertion of the U1A module would allow direct interaction of the Ire1–U1A fusion protein with HAC1U1A mRNA via the U1A-binding hairpins that we had introduced into the 3′ UTR ( Figure 5B ) , which we thus far employed in the mRNAs for visualization purposes ( Figure 1C ) . Strikingly , the U1A module considerably restored splicing and growth under ER-stress conditions of ire1 ( Δ[+]-box ) mutant complemented with HAC1U1A mRNA ( Figure 5C ) . Thus , docking of the mRNA per se to a site distinct from the kinase/RNase domain is key for the mRNA to be processed by Ire1 . The [+]-box is the module that affords such docking for Ire1 , but it can be substituted by an alternative means that provides an mRNA docking platform . We have shown before that the removal of the 3′ BE from HAC1 leads to a substantial reduction of splicing efficiency and growth under ER-stress conditions ( Aragon et al . , 2009 ) . Commensurate with that loss , splicing and growth were restored to a lesser degree when the ire1Δ[+]-box-U1A mutant was complemented with HAC1 Δ3′ BE U1A mRNA than when the 3′ BE was present ( Figure 5C ) . This finding confirms that 3′ BE-mediated mRNA targeting is an event separate from mRNA docking ( whether mediated by the [+]-box or the U1A ‘bypass’ ) . Taken together , these results indicate that the engagement of HAC1 mRNA with Ire1 is a step-wise process . The 3′ BE directs translationally repressed HAC1 mRNA to sites where Ire1 clusters , which is driven solely by oligomeric assembly of its LD . Docking of HAC1 mRNA onto the [+]-box then is a pre-requisite for productively engaging HAC1 mRNA with Ire1's endoribonuclease , whereupon HAC1 mRNA is cleaved .
Oligomeric assembly on both sides of the ER membrane is required for Ire1 function . Unfolded protein binding to the Ire1-LDs leads to Ire1 oligomerization in foci . As we show here , foci formation does not require participation of Ire1's cytoplasmic domains . It coincides with the loss of BiP binding to the Ire1 lumenal domain ( not shown ) and utilizes interfaces IF1L and IF2L . LD-driven clustering concentrates Ire1's kinase/RNase modules on the cytosolic side of the ER membrane and enables formation of the oligomerization interfaces IF1C , IF2C , and IF3C of the cytoplasmic domains that organize and activate the RNase domains . HAC1 mRNA is targeted to the foci via its 3′ BE , as long as the mRNA is translationally repressed ( Aragón et al . , 2009 ) . The molecular machinery for this process remains unknown . Targeting of mRNAs bearing a 3′ BE can occur to Ire1 foci lacking Ire1's cytoplasmic domains , indicating that a putative mRNA targeting receptor ( s ) may exist ( not shown in Figure 6 ) , which senses that Ire1-LDs are clustered . Concentration of RNAs by recruitment to foci allows docking onto the [+]-boxes in the Ire1 cytosolic linker . As we show here , a synthetic bypass can substitute for the [+]-box mediated docking . Docking may fortify oligomeric assembly of the cytosolic Ire1 domains by tethering the [+]-boxes of several clustered Ire1 monomers . Accordingly , the [+]-box may serve a role akin to the arginine-rich domain of HIV1 Rev , which promotes both RNA binding and protein oligomerization ( Zapp et al . , 1991; Daugherty et al . , 2010 ) . HAC1 mRNA is either repositioned from the [+]-box-docked pool ( as depicted ) or newly targeted HAC1 mRNA molecules are recruited and engaged with the active site of the RNase domain . HAC1 mRNA is cleaved and the severed exons are ligated to produce the spliced mRNA product of the reaction . Our findings identify three arginine residues within the [+]-box that are crucial for mRNA docking and splicing . A key contribution of arginine residues is common to RNA binding proteins ( Lazinski et al . , 1989; Tan and Frankel , 1995 ) , suggesting that mRNA binding to the [+]-box is direct . In particular , the [+]-box encompasses a conserved arginine-glycine-glycine ( RGG ) motif , which in the form of repeats adorns several RNA binding protein families , including snRNPs , hnRNPs , and snoRNPs ( Godin and Varani , 2007 ) . Via combined low-affinity interactions , RGG boxes synergize in the assembly of RNA–protein complexes ( Godin and Varani , 2007 ) . Along these lines it is plausible that when Ire1 is monomeric , the [+]-boxes of individual Ire1 molecules may not sustain RNA binding , which can explain the lack of mRNA recruitment under non-ER-stress conditions ( Aragón et al . , 2009; Figure 3D ) . Clustering of [+]-boxes , resulting from Ire1 LD-driven oligomerization , would generate the required synergistic avidity , such that only clustered [+]-boxes form a docking platform for the mRNA . We found that residual targeting of HAC1 mRNA occurs even in the absence of recognizable features conserved among Ire1 proteins at the cytosolic side of the membrane ( ire1ΔKR/Δ[+]-box ) . Moreover , even when mRNA docking was facilitated by the U1A-mediated ‘bypass’ , the 3′ BE still contributed to HAC1 mRNA splicing efficiency . These data indicate that 3′ BE-mediated HAC1 mRNA targeting operates independently from [+]-box-mediated mRNA docking . In the well-studied example of ASH1 mRNA targeting to the bud tip in yeast , three steps can be distinguished: mRNA particle formation ( which requires the She2 and She4 proteins ) , mRNA transport into the bud ( which requires the She1 , Myo4 , and She3 proteins ) , and finally mRNA anchoring at the bud tip ( which requires the She5 , Bni1 , Bud6 , and Aip3 proteins ) ( Beach and Bloom , 2001 ) . By analogy , Ire1 clusters may contain additional ‘anchoring’ factors for HAC1 mRNA . The putative trans-acting factors postulated to bind the 3′ BE cis-acting element and guide HAC1 mRNA to Ire1 clusters and those to mark the clusters and receive the incoming traffic remain unknown , as is the mechanism by which they would respond to ER-stress . Ire1 clustering and splicing of its mRNA substrate are conserved in metazoan cells ( Yoshida et al . , 2001; Calfon et al . , 2002; Li et al . , 2010 ) , but mRNA recruitment via a 3′ BE is not . Rather , XBP1 mRNA ( the metazoan ortholog of HAC1 mRNA ) is targeted to the ER membrane via a hydrophobic signal encoded in the unspliced mRNA ( Yanagitani et al . , 2009 ) . This mechanism limits diffusion of the mRNA substrate to the two-dimensional plane of the membrane and thus may present an evolutionary bypass for docking via the [+]-box , functionally equivalent to the synthetic U1A bypass we describe here . Alternatively , membrane targeting in metazoans may represent the first targeting step that precedes a more specific docking event . Intriguingly , plants may employ both strategies: the unspliced substrate mRNA encodes a protein containing a hydrophobic membrane anchor ( Deng et al . , 2011 ) , allowing pre-recruitment to the ER-membrane , while the conserved basic linker motif in plant Ire1 homologs may serve as a dedicated mRNA docking site for further selectivity in the recruitment process . The step-wise reaction in which mRNA targeting and docking are staged and intertwined with Ire1 activation may help ascertain that Ire1 recognizes HAC1 mRNA as its privileged substrate , even though at least 50 consensus sequence splice site motifs are present in the yeast transcriptome ( Gonzalez et al . , 1999; Niwa et al . , 2005 ) . Both in Drosophila and human cells , IRE1 processing of mRNA targets other than XBP1 mRNA under prolonged ER-stress has been reported as part of an RNA-degradative pathway presumed to reduce the load of proteins entering the ER , called regulated Ire1-dependent decay ( RIDD ) ( Hollien and Weissman , 2006; Han et al . , 2009; Hollien et al . , 2009 ) . In S . pombe , which lacks HAC1 , ER-stress is mitigated by Ire1 exclusively through RIDD and targets primarily mRNAs that encode proteins destined to enter the ER ( Kimmig et al . , 2012 ) . In view of our data , RIDD may be invoked through modulating the stringency of mRNA delivery to and/or docking onto Ire1 clusters and may be precluded in Saccharomyces cerevisiae due to the stringency of the step-wise process . The organization of the engagement between Ire1 and its mRNA substrate ( s ) into multiple steps , that is targeting , docking , priming , and cleavage , thus emerges as integral to upholding selectivity and efficiency of Ire1-mediated mRNA processing and , hence , of UPR signaling .
Standard cloning and yeast techniques were used for construction , transformation , and integration of plasmids and construction of yeast strains ( Sambrook et al . , 1989; Longtine et al . , 1998; Guthrie and Fink , 2002 ) . All mRNA visualization constructs as well as the ire1Δ strain , containing a genomic U1A–GFP copy or not , have been described ( Aragón et al . , 2009 ) . Ire1 variants in all assays were expressed under the control of the autologous promoter at near-endogenous levels either from centromeric pRS315 ( Sikorski and Hieter , 1989 ) derivatives or from a genomic copy integrated from pRS305 ( Sikorski and Hieter , 1989 ) derivatives . Insertion of monomeric yeast-codon-adapted GFP or mCherry modules in the linker and IFL , IF2C , IF1C , KD , and RD mutants of Ire1 have been described ( Papa et al . , 2003; Lee et al . , 2008; Aragón et al . , 2009; Korennykh et al . , 2009; Korennykh et al . , 2011 ) . The IF3C mutant was created from the R899A mutant that was described before ( Korennykh et al . , 2009 ) to contain an additional substitution , K678A , which , based on the contacts in the crystal ( Korennykh et al . , 2009 ) would further eliminate interactions along that interface . The ΔR mutant was truncated from P982 , the ΔKR from L673 , and the ΔKR/Δ[+]-box from K642 . In the Δ[+]-box mutant residues K642–K658 and in the Δ55 mutant residues I579–E633 were deleted . All positively charged residues ( K & R ) within the K642–K658 stretch were replaced by glutamines in the Q-box mutant , or each K or R individually by threonine , while in the all [+] mutant the non-positively charged residues within the same stretch were replaced with arginines and lysines . The GFP , mCherry , or U1A modules were placed into the αF–αEF loop between H875 and S876 of Ire1 containing a S878T substitution that resulted from the cloning strategy . All Ire1 variants were constructed to contain a C-terminally encoded HA-tag . For the ‘bypass’ experiment ( Figure 5C ) , HAC1 mRNA variants were expressed under the control of the autologous promoter at near-endogenous levels from centromeric pRS316 ( Sikorski and Hieter , 1989 ) derivatives . All yeast strains used for this study were based on the W303a derived CRY1 strain ( Aragón et al . , 2009; Pincus et al . , 2010 ) , including the newly constructed strains ire1Δ::KAN/SpR::HIS ( used for all SpR splicing assays ) , ire1Δ::KAN/ire1ΔKR-mCherry::LEU , ire1Δ::KAN/ire1ΔKR/Δ[+]box-mCherry::LEU ( used in Figure 3B ) , and ire1Δ::KAN/hac1Δ::HIS ( used in Figure 5C ) . The SpR copy was integrated into the genome of the ire1Δ strain from a pRS304 ( Sikorski and Hieter , 1989 ) plasmid derivative of pDEP005 ( Pincus et al . , 2010 ) . The mCherry-tagged IRE1 and ire1 ( ΔKR/Δ[+]-box ) copies were integrated from pRS305 ( Sikorski and Hieter , 1989 ) based constructs . All constructs used in this study are listed in Supplementary file 1 . Cells were grown in standard or 2× concentrated synthetic media containing glucose as carbon source . Stress was induced either with DTT or tunicamycin , using concentrations at which differences between ( samples from ) wild-type and UPR deficient yeast are best appreciated , as we empirically established before: 0 . 2 µg ml−1 tunicamycin for viability assays , 5 mM DTT for imaging of Ire1 clusters and mRNA recruitment ( Aragón et al . , 2009 ) , and 2 mM DTT for the splicing reporter assay ( Pincus et al . , 2010 ) . Two days after transformation of SpR harboring ire1Δ yeast ( ire1Δ::KAN/SpR::TRP ) with plasmids bearing Ire1 variants , fresh colonies were resuspended in 2× synthetic media each in 500 µl in 1-ml deep 96-well plates and incubated for 8 hr at 30°C . Fluorescence of samples was then analyzed either before or after the addition of 2 mM DTT and a further 2 hr incubation at 30°C by flow cytometry using a BD LSR-II , as described ( Pincus et al . , 2010 ) . Total RNA was isolated by the hot phenol method and RT-PCR of spliced and total mRNA of HAC1 and derivatives was performed , as described ( Elizalde et al . , 2014 ) . The primers used to amplify spliced and total HAC1 mRNA were ( forward ) CTTGACAATTGGCGTAATCCAGAA ( for spliced ) and ( forward ) CCACGAAGACGCGTTGACTTGCAG ( for total ) and ( reverse ) GCTATATCGTCGCAGAGTGGGTCTG ( for both spliced and total ) . Protein extraction , electrophoresis , and transfer to nitrocellulose for immunoblot analysis of Ire1 variants with anti-HA antibody ( 12CA ) were performed as before ( Pincus et al . , 2010 ) . Endogenous Ire1 levels are at ∼250 molecules/cell ( Ghaemmaghami et al . , 2003 ) . In this study Ire1 variants were expressed at near-endogenous levels under the control of the autologous promoter from centromeric plasmids , and thus signals for HA-tagged Ire1 variants in immunoblots are weak . All imaging and quantitation of images were performed , as described ( Aragón et al . , 2009 ) . In brief , samples for microscopy were taken from yeast that was kept in early log-phase for at least 16 hr in 2× synthetic media before imaging . Microscopy of laser-excited mCherry or GFP was performed with a Yokogawa CSU-22 spinning disc confocal on a Nikon TE2000 microscope , controlled with µmanager and ImageJ . Images were captured with a 100×/1 . 4 NA Plan Apo objective on a Cascade II EMCCD and selected for analysis to contain significant signal above background but no saturated pixels . For display , images were processed in ImageJ and Adobe Photoshop such that the linear range of signals was comparable . Foci of Ire1 variants were determined and the co-localization index for U1A–GFP decorated SpRU1A , SL-PGK-3′ hac1U1A , or SpR Δ3′ BEU1A mRNA recruited to those foci was scored by using a customized MatLab script , as described ( Aragón et al . , 2009 ) . | Proteins are built based on instructions in template molecules called messenger RNAs ( or mRNAs ) , which are copied from the DNA of genes . As they are made , proteins must fold into a specific three-dimensional shape and some proteins pass into a compartment in the cell , called the endoplasmic reticulum , in which they fold . So-called molecular chaperone proteins assist this folding process . From the endoplasmic reticulum , most proteins travel to other destinations within or outside of the cell . If the molecular chaperones in the endoplasmic reticulum are overwhelmed by their protein folding task , unfolded proteins accumulate; a situation that can be harmful to the cell . In eukaryotic cells including yeast , a sensor protein called Ire1 detects when unfolded proteins build up in the endoplasmic reticulum . As a result , the Ire1 sensor proteins join together to form clusters and an mRNA molecule called HAC1 is specifically recruited to the Ire1 clusters . The portions of the Ire1 protein that extend out from the endoplasmic reticulum into the cell proper then bind to HAC1 mRNA and cut a piece out of it . This edited mRNA encodes the instructions to build a protein that in turn boosts the expression of various components—including the appropriate molecular chaperones—that are needed to alleviate the stress caused by an excess of unfolded proteins . Within clusters , individual Ire1 proteins interact through the portions of the protein found on the inside of the endoplasmic reticulum . Now , van Anken et al . show that these interactions are sufficient for forming and maintaining clusters . The interactions between the portions of the Ire1 proteins outside of the endoplasmic reticulum are needed for editing the HAC1 mRNA but not for forming and maintaining the clusters or for recruiting the HAC1 mRNA molecule to bind to Ire1 . Instead , van Anken et al . discovered an mRNA binding site on the Ire1 clusters , which is separate from the part of the Ire1 protein that cuts the mRNA molecules . The Ire1 protein needs to first bind the HAC1 mRNA molecule at this binding site before it can cut it; van Anken et al . suggest that this two-step process helps ensure accurate and efficient editing of the HAC1 mRNA by Ire1 . This process could also help to minimize the chance of other mRNA molecules being edited by mistake . It will be of interest to investigate if similar safety measures are key for endoplasmic reticulum stress signaling mechanisms in humans , and whether these newly discovered steps can be targeted by drugs to treat disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2014 | Specificity in endoplasmic reticulum-stress signaling in yeast entails a step-wise engagement of HAC1 mRNA to clusters of the stress sensor Ire1 |
Correct wiring is crucial for the proper functioning of the nervous system . Molecular gradients provide critical signals to guide growth cones , which are the motile tips of developing axons , to their targets . However , in vitro , growth cones trace highly stochastic trajectories , and exactly how molecular gradients bias their movement is unclear . Here , we introduce a mathematical model based on persistence , bias , and noise to describe this behaviour , constrained directly by measurements of the detailed statistics of growth cone movements in both attractive and repulsive gradients in a microfluidic device . This model provides a mathematical explanation for why average axon turning angles in gradients in vitro saturate very rapidly with time at relatively small values . This work introduces the most accurate predictive model of growth cone trajectories to date , and deepens our understanding of axon guidance events both in vitro and in vivo .
For the brain to function correctly , it must be wired correctly . Indeed , many neurodevelopmental disorders are likely the result of wiring defects ( Yaron and Zheng , 2007; Geschwind and Levitt , 2007; Lin et al . , 2009; Stoeckli , 2012 ) . Axon guidance , where axons grow and navigate to their targets , occurs primarily via the sensing of molecular cues in the environment . A critical mechanism by which such cues act is believed to be concentration gradients , causing axons to be attracted or repelled in particular directions ( Mortimer et al . , 2008; Lowery and Van Vactor , 2009 ) . However , despite major advances in understanding which molecules are involved in this process ( Tessier-Lavigne and Goodman , 1996; Dickson , 2002; Chilton , 2006; Kolodkin and Tessier-Lavigne , 2011 ) , an accurate quantitative model describing how axon trajectories are influenced by such guidance cues is still lacking . In vivo , axon trajectories may potentially be influenced by many cues . In vitro assays allow individual influences , such as that from the concentration gradient of a single guidance factor , to be isolated and quantified . A substantial mystery posed by in vitro axonal chemotaxis assays is the relatively weak turning produced , even over long periods of time . The naive prediction that axons would promptly turn until they become fully aligned with the gradient turns out not to be true . In an early study of chemotactic responses of chick sensory neurons to a gradient of nerve growth factor in a diffusion chamber , only 60% of nerve tips were preferentially directed toward the gradient direction after 46 hr of growth ( Letourneau , 1978 ) . The growth cone turning assay over 1–2 hr produces average turning angles typically ranging from 10 to 25° , with high variability ( Song et al . , 1997; Höpker et al . , 1999; Xiang et al . , 2002; Ming et al . , 2002; Thompson et al . , 2011 ) . A similarly weak response is observed in the Dunn chamber ( Kent et al . , 2010; Dudanova et al . , 2012; Ruiz de Almodovar et al . , 2011; Dudanova et al . , 2010; Yam et al . , 2009 ) . More recent studies using microfluidic technologies over timescales ranging from hours to days have also elicited average axon turning angles only up to 10–15° ( Wang et al . , 2008; Morel et al . , 2012; Taylor et al . , 2015; Sloan et al . , 2015 ) . Why average turning angles are so small , and what this means for axon guidance in vivo , are unclear . One of the key properties of in vitro axon growth that might explain this mystery is that it is often very straight ( Katz et al . , 1984; Katz , 1985 ) . Axons are under mechanical tension from the pull of the growth cone ( Bray , 1973; Bray , 1979 ) , and this tension stimulates the elongation of the axon by stretching ( Bray , 1984; Zheng et al . , 1991 ) . Traction forces generated in the growth cone arise from the coupling of the continuous retrograde flow of actin to the substrate through adhesion receptors ( Franze and Guck , 2010; Betz et al . , 2011; Athamneh and Suter , 2015 ) . For reasons which are not clear , axons tend not to bend and follow the highly random movements of their growth cones . Rather , they usually form a straight line between their tip and a location where they are firmly attached to the substrate ( i . e . a focal adhesion ( Kaverina et al . , 2002 ) ) . We call such locations anchor points; they can be at the soma , at a branch point , or at some other seemingly sporadic location along the axon . Although it is not clear how this tension leads to elongation , the growth cone advances largely in the stretch direction along the axon , resulting in relatively straight paths . To determine quantitatively what effect this might have on axonal trajectories requires mathematical modelling . Growth cone movements were first analyzed in detail in ( Katz et al . , 1984; Katz , 1985 ) . Subsequently , various phenomenological models have been built that differ as to how they treat stochasticity , and mechanisms for directional preference , namely turning or growth rate modulation . One set attempted to fit the dynamics of growth cone movement to a random walk with drift ( Buettner et al . , 1994; Odde and Buettner , 1995; Maskery et al . , 2004; Pearson et al . , 2011 ) . Li et al . simulated trajectories by assuming the turning angle of the growth cone is in proportion to the angle between the neurite and the resultant filopodial tension ( Li et al . , 1995 ) . In ( Borisyuk et al . , 2008 ) , the axon growth angle depends on the tendency to turn toward the gradient angle and noise . The noise term is small ( 2–5° ) , leading to straight paths that resemble axon growth in the tadpole spinal cord . Another set of models has concentrated on how asymmetric receptor binding across the growth cone might be used as the basis of a turning signal ( Goodhill et al . , 2004; Xu et al . , 2005; Mortimer et al . , 2010 ) , but without considering the consequence for whole trajectories . A third group of models considers the possibility that the velocity of the growth cone is influenced by an attractive gradient from the target cells , and chemoattractants and chemorepellants released from other growth cones and itself ( Hentschel and Ooyen , 1999; Krottje and van Ooyen , 2007 ) . However , none of these models has been closely compared with the details of experimentally measured trajectories in gradients , and parameters such as variability in step sizes , the distribution of instantaneous turning angles , and straightness of real paths , have not been addressed . Thus , the question of whether there is a model that can adequately capture all these characteristics of real trajectories remains open . Without such a model , it is difficult to determine if trajectories observed in vivo are in fact consistent with gradient guidance . Here , we present a new computational model for axonal trajectories based on the combined influence of anchor points , a tendency to turn toward the gradient direction , and random noise . We found experimentally that the gradient had no effect on the step sizes; thus , we only model the turning angles . Critically , the model predicts rapid near saturation of average turning angles with time . To test this model quantitatively , we then introduce a new microfluidics assay for studying axonal response to gradients , and using timelapse imaging characterize the behavior of axons over several hours of growth in both attractive and repulsive gradients . We find that our model fits the behavior observed very closely . We then investigate by simulation the effect of increasing the number of anchor points , and find that this increases the average fidelity of turning but at the cost of higher variability . Together , this work both explains why turning response to gradient saturates so rapidly and reveals the quantitative principles that are required to reproduce accurately in vitro axonal trajectories in response to chemotactic gradients . The model identifies straightness as a limiting factor on how much axons can turn and suggests that the frequency of anchor points plays a key role in the axonal turning response to a gradient .
We modelled three basic influences on the direction of axon growth: a tendency to grow straight , the effect of a chemotactic gradient , if present , and random movement noise . In a fixed coordinate system with arbitrary zero angle direction , we define θ ( t ) as the bearing of the growth cone at time step t , ϕ ( t ) as the angle of the vector connecting the growth cone to its anchor point , and Ψ as the gradient direction ( terminology is summarized in Table 1 ) . We define ‘bearing change’ as Δθ ( t ) , the change in θ ( t ) at time step t , distinct from ‘turning angle’ ψturn , the total change in θ from the initial direction of growth over long periods of time . For simulations we identify each timestep as 5 min of real time . The model ( Figure 1 ) is then10 . 7554/eLife . 12248 . 003Table 1 . Summary of model parameters ( GC: growth cone ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 003SymbolMeaningθGC’s current bearingΦGC’s overall angleΨGradient directionΔθBearing changeψturnTurning angle after 80 minaPersistence strengthbBias strengthξNoise in bearing changeσStandard deviation of ξsStep size every 5 minLDistance from origin to GCSStraightness indexrAnchoring rate10 . 7554/eLife . 12248 . 004Figure 1 . Model set-up and the noiseless case . ( A ) The axon starts growing from the soma ( black segment ) at initiation angle ϕ ( 0 ) . At each time point , the bearing is θ ( t ) , and the bearing change between t and t + 1 is Δθ ( t ) . ϕ ( t ) is the angle of the vector connecting the current position of the growth cone with the anchor point . Ψ is the fixed gradient direction . ( B ) The turning angle ψturn at time t is the angle between the initial direction of growth , and the line joining the initial and current positions of the growth cone . ( C ) Simulation of the growth cone angle using Equation ( 1 ) in the noiseless case ( ξ = 0 ) with the same a = 1 and different values of b . The dashed line is the power law ϕ ( t ) ∝t-ba+b . In the long time limit , this law accurately describes the angle of the growth cone . ( D ) Simulations of the trajectories for different combinations of a and b in the absence of noise . Larger b leads to stronger turning . When a = 0 , the growth cone very rapidly aligns with the gradient . The persistence term ( t > 0 ) leads to incomplete turning . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 00410 . 7554/eLife . 12248 . 005Figure 1—source code 1 . The code to simulate the trajectories based on Equation 1 in the noiseless case . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 005 ( 1 ) ∆θ ( t ) =a∠ ( ϕ ( t ) , θ ( t ) ) +b∠ ( Ψ , θ ( t ) ) +ξ where a scales persistence to move in the same direction as the overall direction of the axon , b scales the bias due to the gradient , and ξ is random noise in the bearing changes . The symbol ∠ ( x , y ) denotes the signed angle between the unit vectors with angles x and y , and constrains the resultant angle to be between −π and π . The step size is the distance moved after one time step and will later be estimated empirically . We consider first the noiseless case ( ξ = 0 ) in long- and finite-time regimes , and then consider the effects of noise . Figure 1C shows the results of setting ξ = 0 , with a fixed step size of s = 3 μm , and simulating the model for long times with the same a = 1 and different values of b ( 0 . 1 , 0 . 2 and 0 . 3 ) . Turning angles rapidly saturate , which can be understood analytically ( see 'Materials and methods' ) : in the t→∞ limit , the growth cone angle follows a power law with respect to time ϕ ( t ) ∝t ( α-1 ) or log ( ϕ ( t ) ) = const + ( α−1 ) log ( t ) ( where α = a/ ( a+b ) ) ( Figure 1C ) . This relationship generally holds for t>exp ( 4 ) ≈ 4 h , meaning that for long times , the rate of change of angle decreases and this rate is determined by the power law exponent b/ ( a + b ) . Since comparison with empirical data ( see later ) shows that the biologically relevant regime is b ≪ a , the exponent is generally small . Thus , while ultimately axons in the model do eventually align with the gradient , this process takes an exceedingly long time . This explains the slow and decreasing change in the turning angle over time in the noiseless case . The finite t regime of this equation is difficult to solve analytically , since ϕ ( t ) depends on the entire history of growth cone movements . Simulations using different combinations of a and b are shown in Figure 1D . For the cases of a≠0 , after 150 time steps ( 12 . 5 hr of real time ) , the resultant turning angle was far from completely aligned with the gradient . Although the bias term bent the trajectory in the direction of the gradient , there was a straightening effect due to the persistence term , constantly pulling the growth cone toward the overall growth direction of the axon . As expected , the pull due to the gradient increased with larger b ( Figure 1D ) . Thus , the persistence term prevented the axon from completely aligning with the gradient . Also apparent is that without noise , the trajectories were all very straight ( with straightness index [see 'Materials and methods'] greater than 0 . 98 ) . Thus , the microscopic constraint imposed by the persistence term leads to the macroscopic phenomenon of incomplete turning . When we introduced Gaussian noise into the bearing changes ( in radians ) ( ξ ~ N ( 0 , σ ) , σ=π/4 ) with the same parameters and initial conditions as above for 1000 axons , the behavior was qualitatively similar: after an initial period of relatively rapid turning , turning angles tended to an almost steady state which was not aligned with the gradient even after a long time . However , the average final turning angle was even less than that of the noiseless case . This is because the noise created more random wandering of the growth cone , further reducing the directional effect of the gradient ( Figure 2A–B ) . After 20–40 min , for b = 0 . 1 , the turning angle distribution of the population was 7 ± 25° ( mean ± std ) . Assuming a normal distribution of turning angles , this means that many of the axons were no longer roughly perpendicular to the gradient , and thus only continued to turn extremely slowly . Therefore , over time the influence of the gradient on the whole population of axons weakened . The persistence term also created a resistance against large turns due to the gradient . Increasing b/a increased the turning angle , but did not alter its rapid saturation with time ( Figure 2C ) . Lastly , we examined the effect on the straightness by varying the standand deviation of the noise σ from 0 to π radians . As the steps became more noisy , the paths became less straight ( Figure 2D ) . 10 . 7554/eLife . 12248 . 006Figure 2 . Model results with noise . ( A ) Long-term behavior of growth cones: Simulation of 9 axons with fixed growth rate and noise in bearing changes ( ξ ∼ N ( 0 , π/4 ) radians ) starting at ϕ ( 0 ) = 90° subject to the gradient direction Ψ = 0 with persistence a = 1 and bias b = 0 . 1 ( blue ) , b = 0 . 2 ( red ) and b = 0 . 3 ( black ) after 150 steps ( 12 . 5 hr of real time ) . ( B ) The trajectories with the same parameters without noise . ( C ) The turning angles over time ( mean ± SEM ) of 1000 axons for different values of b ( 0 . 1 , 0 . 2 , 0 . 3 ) and a = 1 . ( D ) Straightness ( mean ± std ) decreases as the noise variance increases . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 00610 . 7554/eLife . 12248 . 007Figure 2—source code 1 . The code to simulate the trajectories based on Equation 1 in the noisy case . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 007 In summary , the noiseless case generated very straight axons and growth cone angles that followed power laws with respect to time in the long time limit . Similarly , in the noisy case , the rate of change of the average turning angle was initially rapid and then slowed down even more rapidly with time . In both cases , the persistence term was a limiting factor on how much and how fast the axons could turn . Thus , this model captures , at least qualitatively , the behavior that axons turn only slightly in gradients , and even for long times do not generally become completely aligned with the gradient . Having established the basic behaviour of the model , we then asked whether it could reproduce in detail real axon trajectory statistics . We therefore analyzed the trajectories of superior cervical ganglion neurons in a new microfluidics device ( Figure 3A–C ) . This device generated linear gradients , by the mixing of high and low solutions of a chemotropic factor . The gradient was visualized using 40 kDa dextran-tetramethylrhodamine ( Figure 3D ) , and gradients were stable for at least 20 hr ( Figure 3E , F ) . 10 . 7554/eLife . 12248 . 008Figure 3 . The microfluidic assay . ( A ) The design of the chamber: the two solutions were pumped into the inlets and mix in the mixing channels before flowing into the growth chamber where the cells are plated . The mixing channels were of height 50 μm and width 50 μm . Scale bar 1 mm . ( B , C ) Photo of the experimental setup: two glass syringes attached to a Harvard pump injected the solutions into the chamber bonded on a 35 mm plastic plate . ( D ) Two solutions , one of which contained 0 . 1% ( v/v ) dextran fluorescently labelled with tetramethylrhodamine , were used to visualize the gradient . Brighter regions indicate higher concentrations . Scale bar 200 μm . ( E , F ) Line-scan measurements of fluorescence intensity across the device show a linear gradient which persists for at least 20 hr ( t = 0h ( E ) and t = 20h ( F ) ) . The shaded errorbars show standard deviations across 10 chambers . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 00810 . 7554/eLife . 12248 . 009Figure 3—source data 1 . The average brightness intensity and noise in the microfluidic chamber at 0 and 20 hr . The average and noise were estimated from 5 min interval timelapse imaging over an 1-hr period . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 009 We measured the response to nerve growth factor ( NGF ) gradients of axons from dissociated P1-P3 SCG neurons . We chose this model system because almost 100% of these neurons express the NGF receptor TrkA ( Wetmore and Olson , 1995; Verge et al . , 1992 ) . Three conditions were investigated: a control without flow or gradient , an attractive gradient of nerve growth factor ( NGF ) , and a gradient of NGF with added KT5720 , which converts attraction to repulsion by lowering levels of cAMP in the growth cone ( Song et al . , 1997 ) . Cells were injected into the growing chambers and grown for 2 hr before gradient onset . In the control condition , cells were grown over several hours with 0 . 3 nM NGF . In the NGF gradient condition , two solutions of concentrations 0 nM and 10 nM NGF were pumped into the growing chamber through the two inlets . Previous work using Scatchard analysis estimated that Kd = 0 . 9 ± 0 . 3 nM ( Wehrman et al . , 2007 ) and showed that SCG neuronal outgrowth is severely inhibited at the saturating NGF concentration of 40 nM ( Ohta et al . , 1990 ) . Given the healthy growth in our assay , it is clear that the concentration in the gradient condition was below saturation point . We analyzed trajectories for 300 axons per condition . These were obtained from 23 individual chambers in the control case , 27 chambers in the NGF gradient case , and 24 chambers in the NGF gradient plus KT5720 case . In most experiments , 2 chambers were run in parallel , so the total numbers of experiments in each case were 12 , 15 , and 13 , respectively . For the measurement of turning angles , we selected only axons that started growing between 70° and 110° relative to the gradient ( when present ) . An asymmetric concentration field of guidance cue across the growth cone leads to turning ( Song et al . , 1997; Hentschel and Ooyen , 1999; Xiang et al . , 2002; Ming et al . , 2002 ) and axons growing in this range experienced between 94% ( i . e . sin 70° ) to 100% ( i . e . sin 90° ) of the maximum possible concentration difference across the growth cone . Thus , we expected the impact of the gradient would be strongest on these axons ( Figure 4A ) . We tracked the growth cones every 5 min for as long as possible until they collapsed or branched or collided with other cells , axons or the edges of the chamber . The SCG axons were clearly attracted in the NGF gradient ( Figure 4B ) . When the protein kinase A inhibitor KT5720 was added to the high and low solutions at concentration 70 nM , attraction was converted into repulsion as previously described ( Song et al . , 1997; Forbes et al . , 2012 ) ( Figure 4B ) . These results confirm that the gradient in the microfluidic assay elicited a guidance response in SCG axons . From the timelapse imaging data , we then selected the subset of axons that did not branch or retract following growth for several hours in the attractive NGF gradient , and measured the turning angles of the population over time . The average turning angle reached the steady state quickly and did not increase significantly with time , matching the prediction of the model ( Figure 4C ) . 10 . 7554/eLife . 12248 . 010Figure 4 . Turning in microfluidic gradients . ( A ) Images of a representative axon initially almost perpendicular to the gradient at the beginning and end of the measurement after 80 min . Scale bar 20 μm . The red dots are the positions of the growth cone . ( B ) Summary of turning angles in the three conditions ( mean ± SEM ) : control 0 . 2 ± 2 . 1° ( n=120 ) , NGF gradient ( 0–10 nM ) 9 . 3 ± 1 . 9° ( n=143 ) , NGF gradient ( 0–10 nM ) + KT5720−8 . 8 ± 2 . 2° ( n=112 ) . *: p < 0 . 01 t-test in both cases . ( C ) The means ( red ) and standard deviations ( blue ) of turning angles of 143 axons over time for the attractive case . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01010 . 7554/eLife . 12248 . 011Figure 4—source data 1 . The turning angles at different time points from the start of the experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 011 One possible way that the gradient could affect the axons is by causing biased branching [c . f . ( Simpson et al . , 2013 ) ] , or changes in branching rates . To test whether the NGF gradient changed the branch extension and retraction rates , we compared the number of branches per cell after 5 hr of growth and did not detect any difference ( p = 0 . 9 Kolmogorov–Smirnov test , Figure 5A ) . We measured the intervals between successive branching events in the same cell in each condition and did not find any difference in the branching rate ( p = 0 . 7 Kolmogorov–Smirnov test , Figure 5B ) . Similarly , the lifetimes of the branches were unaffected by the gradient ( p = 0 . 2 Kolmogorov–Smirnov test , Figure 5C ) . We counted the number of branches pointing up and down the gradient per cell and did not find any difference ( p = 0 . 8 Kolmogorov–Smirnov test , Figure 5D ) . Thus , the gradient had no effect on axon branching and retraction . 10 . 7554/eLife . 12248 . 012Figure 5 . The gradient did not affect branch extension and retraction rates . ( A ) Histogram of the number of cells with different numbers of branches after 5 hr of growth . The number ( mean ± std ) of branches per neuron in the control condition was 4 . 2 ± 1 . 8 ( n=324 cells ) and in the gradient condition was 4 . 4 ± 1 . 9 ( n=297 cells ) , p = 0 . 9 Kolmogorov–Smirnov test . ( B ) The distribution of interval times between two successive branching events of the same cell . The interval ( mean ± std ) in the control condition was 23 . 1 ± 22 . 8 min ( n=315 intervals ) and in the gradient condition was 24 . 1 ± 23 . 5 min ( n=287 intervals ) , p = 0 . 7 KS test . ( C ) Branch lifetime ( mean ± std ) in the control condition was 87 ± 79 min ( n=245 branches ) and in the gradient condition was 92 ± 81 min ( n=213 branches ) , p = 0 . 2 KS test . ( D ) Histogram of the number of branches pointing up the gradient vs down the gradient ( p = 0 . 8 , KS test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01210 . 7554/eLife . 12248 . 013Figure 5—source data 1 . The number of branches per cell after 5 hr of growth in the control and NGF gradient ( Columns A , B ) . In the gradient , we counted the number of branches pointing up and down the gradient ( Columns C , D ) . We measured the time intervals between successive branching events in the same cell in the control and NGF gradient over 5 hr ( Columns E-F ) . For branches that retracted in the 5 hr imaging time , we measured their lifetimes in the control and NGF gradient ( Columns G , H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 013 To test whether fluid flow in the chamber biased the statistics of the steps , axons growing in the gradient condition with fluid flow were divided into four quadrants with different relative angles to the fluid flow: two quadrants growing perpendicular to the flow , one quadrant growing with the flow , and the other growing against the flow ( Figure 6A ) . Comparing the distribution of bearing changes between the four quandrants , and with axons from the control condition without any flow , showed no influence of the flow ( p = 0 . 7 in Figure 6B and p = 0 . 4 in Figure 6C , Kruskal-Wallis test ) . The means of the bearing changes in quadrants 2 and 4 were non-zero , and the bearing changes accumulated over time to result in a non-zero average turning angle of the population . However , these differences in means were very small ( approximately 1° ) , and there were no significant statistical difference among the distributions . Therefore , the positive turning angles in the NGF gradient ( and negative turning angles in the NGF gradient + KT5720 ) were due to the effect of the gradient , not bias from the flow . 10 . 7554/eLife . 12248 . 014Figure 6 . Flow did not affect step statistics . ( A ) Axons growing in different directions were grouped into four quadrants . ( B ) Growth cones’ step sizes in different quadrants . n values refer to the number of steps in each quadrant . There was no significant difference between the quadrants ( p = 0 . 7 Kruskal-Wallis test ) . ( C ) Grow cones’ bearing changes in different quadrants ( p = 0 . 4 Kruskal-Wallis test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01410 . 7554/eLife . 12248 . 015Figure 6—source data 1 . We divided the axons into four quadrants as explained in Figure 6 and measured the bearing changes and stepsizes in each quadrant . This file contains the step sizes ( Sheet 1 ) and and bearing changes ( Sheet 2 ) in the control condition ( Column A ) and in each quadrant of the NGF gradient condition ( Columns B-E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 015 Axon growth is shown in Videos 1–3 and Figure 7A–C . The growth cone often wandered quite randomly , but nevertheless usually the axon segment remained very straight from the growth cone to the cell body or last axon branch point ( Figure 7A–C ) . This implies that often the entire axon segment was pulled sideways across the substrate ( as can be seen directly in the movies ) . Thus , despite the irregular trajectory of the centre of the growth cone , the tension force on the growth cone from the axon was usually pointing directly back to the last anchor point , consistent with the assumptions of the model . To quantify this further , we measured the angle of this 20 μm segment and the angle to the anchor point ϕ and found them to be almost the same ( Figure 7D , E ) . Thus , we can understand the term ϕ as the tension due to the most distal segment . 10 . 7554/eLife . 12248 . 016Figure 7 . Axons were dragged by growth cones . ( A–C ) Timelapse images of three example growth cones . Red arrows point to the putative anchor points and green arrows point to the growth cones . Time is shown in hours and minutes . ( D ) We measured the angle of the neck of the growth cone ( the last 20 μm , black line ) and the overall growth cone angle ( blue line ) after 1 hr from the start of the experiment . ( E ) The two angles were highly correlated , due to the straightness of the axon . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01610 . 7554/eLife . 12248 . 017Figure 7—source data 1 . We measured the angle of the growth cone from its putative anchor point ( Column A ) and compared with the angle of the most distal 20 μm segment of the axon ( Column B ) 1 hr after the start of the experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01710 . 7554/eLife . 12248 . 018Video 1 . Timelapse images of an example growth cone . One-minute interval phase contrast timelapse imaging of a growth cone in a microfluidic chamber . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01810 . 7554/eLife . 12248 . 019Video 2 . Timelapse images of an example growth cone . One-minute interval phase contrast timelapse imaging of a growth cone in a microfluidic chamber . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 01910 . 7554/eLife . 12248 . 020Video 3 . Timelapse images of an example growth cone . One-minute interval phase contrast timelapse imaging of a growth cone in a microfluidic chamber . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 020 The trajectories ( i . e . the locus of the centre of the growth cone ) in three conditions are plotted in Figures 8–10 . Note that these paths are not the same as the final image of the axon , which generally pointed straight back from the final position of the growth cone to the anchor point . Visually , the paths appear mostly straight with occasional large turns , consistent with a long tail for the bearing change distribution . The mean straightness index for the trajectories was S = 0 . 72 ( Figure 11A ) . 10 . 7554/eLife . 12248 . 021Figure 8 . Trajectories of 300 axons growing over 80 min in the control condition , ordered by the initial angle . The red segments indicate the initial direction of the axon and the blue segments show the traces of the growth cones’ trajectories . Scale bar = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02110 . 7554/eLife . 12248 . 022Figure 9 . Trajectories of 300 axons growing over 80 min in the NGF gradient , ordered by the initial angle . Only axons in the box were selected for turning angle measurements as they were almost perpendicular to the gradient , hence most affected by it . Scale bar = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02210 . 7554/eLife . 12248 . 023Figure 10 . Trajectories of 300 axons growing over 80 min in the NGF gradient with 70 nM KT5720 added , ordered by the initial angle . Only axons in the box were selected for turning angle measurements . Scale bar = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02310 . 7554/eLife . 12248 . 024Figure 11 . Trajectories were straight with step sizes and bearing changes independent of each other . ( A ) Distribution of straightness indices of all paths with mean straightness of 0 . 72 . ( B ) There was no correlation between bearing change and step size ( R2 = 0 . 1 , p = 0 . 7 ) . ( C ) The distribution of bearing changes ( blue ) in radians in the control condition can be fitted to a mixture of two von Mises distributions ( red ) P ( x ) =0 . 5exp ( 3cos ( x ) ) 2π I0 ( 3 ) +0 . 03 . ( D ) Step sizes in the control , attractive and repulsive gradients conditions were similar and well-fitted by the gamma distribution P ( x ) ∝ x2 exp ( −x/24 ) ( red ) . ( E ) Step sizes of individual growth cones ( blue ) can be described by gamma distributions ( red ) ( 9 examples shown ) . ( F ) This distribution of the average step sizes ( blue ) of individual growth cones was well-fitted by a Gaussian distribution N ( 0 . 7 , 0 . 24 ) ( red ) . ( G ) Mean square displacement and standard deviation of 300 growth cones growing over 100 mins in the control condition was super-linear , indicating that growth cone trajectories were straighter than predicted by a simple random walk . ( H ) Autocorrelation of bearing changes ( mean ± std ) showed that successive bearing changes were anti-correlated . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02410 . 7554/eLife . 12248 . 025Figure 11—source data 1 . From our 5-min interval tracings , we measured the bearing changes and step sizes in the control condition ( Columns A , B ) , the mean step sizes of all the growth cones ( Column C ) and the step sizes in the NGF gradient and NGF gradient + KT5720 conditions ( Columns D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 025 There was little correlation between the bearing change magnitude and step size ( Figure 11B ) . The distribution of bearing changes in radians was well fitted by a mixture of a von Mises and a uniform distribution ( −π < x < π ) ( Figure 11C ) . That is there was a great deal of randomness in bearing changes , but with a peak in probability near the forward direction . Thus , growth cones tended to move in a straight line instead of turning uniformly randomly . This is inconsistent with the assumptions of several previous models ( Buettner et al . , 1994; Odde and Buettner , 1995; Maskery et al . , 2004 ) . Accumulating across all the growth cones , the distributions of step sizes over 5 min were statistically indistinguishable across the three conditions ( Kruskal-Wallis test p = 0 . 35 ) , and were well fitted by a gamma distribution ( Figure 11D ) . That is , the most likely step size was around 0 . 5 μm/min , but the distribution had a long tail , so that longer step sizes were also seen . The distribution of step sizes for each individual growth cone were also well fit by gamma distributions ( Figure 11E ) . However , individual growth cones had idiosyncratic mean values . The distribution of these mean values could be well fitted by a Gaussian distribution ( Figure 11F ) . Nevertheless , the mean square displacement was clearly not linear , implying that a simple random walk is not suitable to describe the movement ( Figure 11G ) . Successive steps were anti-correlated ( Figure 11H ) , which was not accounted for in a previous model ( Pearson et al . , 2011 ) . This helps the paths remain relatively straight: if successive steps were positively correlated , the paths would become more bent over time . Due to large noise in the bearing changes , bearing changes more than one step apart were uncorrelated . Having established the key statistics of steps from the data , we now asked if the simple model in Equation ( 1 ) could replicate the observed trajectories and explain the phenomenon of saturated turning . We sampled the mean speed vmean of each growth cone from a truncated Gaussian distribution of mean 0 . 7 μm/min and standard deviation 0 . 24 μm/min . At each time point ( 5-min interval ) , the growth cone sampled a step size from the gamma distribution Γ ( 4∕u , vmean*u∕4 ) where u was a uniform random number . The bearing changes evolved according to Equation ( 1 ) . We found that the random noise ξ in bearing changes ( in radians ) could be well fit by the mixture von Mises distribution P ( ξ ) = cexp ( d cos ( ξ ) ) 2π I0 ( d ) + ( 1-c ) where c and d are parameters to be fit . This distribution is not necessarily the same as that of the bearing changes in Figure 11F . As the bearing change is the sum of three random terms , its distribution is broader than the distribution of the noise term . To estimate the four free parameters a , b , c , d , we input the initiation angles ϕ ( 0 ) and used the model to generate the distribution of turning angles ψturn . We then estimated the likelihood function that the turning angle data was generated from the model with the given parameters L ( ψturn|a , b , c , d , ϕ ( 0 ) ) . We found the values of a , b , c , d that maximized the likelihood ( the parameters mostly likely to have generated our empirical turning angle data ) were a = 0 . 7 , b = 0 . 09 , c = 0 . 75 , d = 6 . The statistics over 5000 simulated trajectories using these parameters are shown in Figure 12 , and some example trajectories are shown in Figure 13 . Similarly , we fitted the model to the control data and repulsive gradient and found b = 0 . 002 and b=−0 . 08 , respectively , with other parameters remaining at very similar values as before . Notably , the simulated turning angles changed little over time ( Figure 12A ) , consistent with our preliminary prediction and the data ( Figures 1E , 4C ) . Thus , with realistic step sizes and bearing change noise distributions , the model was able to capture the phenomenon of saturated turning with quantitatively accurate means and variances over time . The distribution of turning angles and step sizes also closely matched the real data ( Figure 12B , C ) . The simulated straightness distribution was also very smilar to the real distribution ( p = 0 . 2 , t-test ) ( Figure 12D ) . The model also captured the distribution of bearing changes , the mean square displacement and the anticorrelation between successive bearing change , which was a consequence of the persistence term straightening the paths ( Figure 12D–F ) . If successive steps were positively correlated , the paths would become more bent over time . This correlation was rapidly lost beyond one time lag because of the large noise . 10 . 7554/eLife . 12248 . 026Figure 12 . Model captured key statistics of trajectories . ( A ) The evolution of simulated turning angles ( mean ± std ) of n=5000 growth cones over time in the attractive gradient condition . ( B ) Simulated turning angles after 16 steps ( 80 min ) had mean 9 . 8° and standard deviation of 24 . 2° , similar to the empirical data in red in Figure 4B . ( C ) Distribution of simulated step lengths ( blue ) , fitted with the empirical distribution ( red ) . ( D ) Straightness of simulated trajectories ( mean 0 . 75 , blue ) , compared with empirical distribution ( red ) ( p = 0 . 2 , t-test ) . ( E ) Simulated bearing changes ( blue ) fitted with the mixture of von Mises distributions given in Figure 11D ( red ) . ( F ) Mean square displacement of simulations ( blue ) and data ( red ) . ( G ) Autocorrelation of simulated bearing changes . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02610 . 7554/eLife . 12248 . 027Figure 12—source code 1 . The code to simulate the trajectories based on Equation 1 with the step sizes and bearing changes described in Section Turning angles over time were well predicted by the model . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02710 . 7554/eLife . 12248 . 028Figure 13 . Simulated trajectories from three conditions . ( A ) control , ( B ) NGF gradient , ( C ) NGF gradient + KT5720 . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 028 Unlike previous models , we did not assume constant steps or a uniform distribution of bearing changes but rather derived these from empirical data . The model was then able to predict the evolution of the average turning angle over time , the straightness profile and the anticorrelation in bearing changes . Most importantly , it could explain the phenomenon of slow and saturated turning , due to a weak bias term relative to the persistence term . A microscopic factor in each step led to a macroscopic phenomenon of limited , variable turning and straight paths . This often overlooked feature of axon growth turned out to be critical in our model in limiting the overall turning . We also found little difference between the attractive and repulsive case , indicating that attractive and repulsive gradients employed similar mechanisms and could not reduce the variability of axon trajectories . The in vitro data we have presented here was well-fitted by assuming the only anchor point is where the axon emerges from the soma or the branch point . However , the in vivo environment is much more complex , and axons may establish anchor points with the substrate at multiple positions as they extend . We therefore investigated in the model what effect this would have on turning angles . We assumed that at each timestep , the probability of that point becoming a new anchor point was fixed , while leaving the evolution at each step as before . The average number of anchor points per timestep ( i . e . 5 min ) is denoted by r . We analyzed two cases: anchoring probabilistically at each time step ( Figure 14A–C ) , and anchoring at regular intervals ( Figure 14D–F ) . We simulated the trajectories for T = 150 timesteps with the same parameters as Figure 2A ( a = 1 , b = 0 . 1 , ξ = 0 or ξ ∼ N ( 0 , π/4 ) radians ) . In both cases , more anchor points led to sharp turns in the trajectories and larger mean turning angle ( Figure 14G ) , since the growth cone now was more free from its initial position . However , it increased the variability in the turning ( Figure 14G ) . Given the same rate of anchoring , whether the growth cone put down new anchor points probabilistically or regularly made little difference to the mean turning . We compared the mean square final angle ⟨ϕ ( T ) 2⟩ which is the sum of the bias ⟨ϕ ( T ) ⟩2 and the variance ( ϕ ( T ) ) . Ideally , the growth cone should completely align with the gradient , that is ϕ ( T ) =0 . Figure 14H shows the bias/variance trade off . Although more anchor points introduced larger variance in the final angle , they achieved greater turning , that is smaller ϕ ( T ) . Figure 14I shows that in the case without movement noise ( ξ = 0 ) and regular anchoring , the growth cone angle ϕ ( t ) also followed a power law in the large t limit , similarly to the case without anchoring ( Figure 1C ) . However , the exponent of this power law was larger , demonstrated by a steeper slope between log ( t ) and log ( ϕ ( t ) ) , meaning that the growth cone aligned with the gradient faster with more anchor points . Thus , increasing the rate of anchoring leads to stronger turning , but increases the variance of the responses . 10 . 7554/eLife . 12248 . 029Figure 14 . More variability with more anchor points . ( A–C ) Trajectories of growth cones with probability of putting down a new anchor r= 0 . 01 , 0 . 05 , 0 . 1 at each timestep and the same parameters as Figure 2A ( a = 1 , b = 0 . 1 , T = 150 timesteps ) . The black plots are without noise in the bearing changes , the blue plots are with noise ξ~N ( 0 , π/4 ) radians in the bearing changes and the red dots are the anchor points . More anchor points lead to higher variability but also stronger turning . The means and standard deviations of turning angles and the values for the noiseless versus the noisy case in brackets for r = 0 . 01 , 0 . 05 , 0 . 1 are 32 ± 9° ( 30 ± 36° ) , 55 ± 8° ( 49 ± 57° ) and 67 ± 5° ( 60 ± 56° ) , respectively . ( D–F ) Trajectories of growth cones with the same rate of putting down new anchor points as A-C but at regular intervals . The means and standard deviations of turning angles and the values for the noiseless versus the noisy case in brackets are 27° ( 24 ± 17° ) , 57° ( 54 ± 51° ) , 69° ( 66 ± 51° ) . ( G ) The means and standard deviations of turning angles after 150 timesteps as a function of the anchoring rate at regular intervals in the noiseless and the noisy case . ( H ) The mean square of the final growth cone angle ( in degrees ) ⟨ϕ ( T ) 2⟩ for different anchoring rates r after 150 steps . ⟨ϕ ( T ) 2⟩ is the sum of the bias term ⟨ϕ ( T ) ⟩2 and the variance term var ( ϕ ( T ) ) . Although more anchor points add more variance to the final angle ( red curve ) , they achieve stronger turning ϕ ( T ) ≈0 ( black curve ) . ( I ) The evolution of ϕ ( t ) over time , for the case of anchoring at regular intervals and no noise in the movement ( ξ = 0 ) . With more anchor points , ϕ ( t ) also follows the power law but with steeper slope , meaning that ϕ ( t ) →0 at a faster rate than the case without anchor points . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 02910 . 7554/eLife . 12248 . 030Figure 14—source code 1 . The code to simulate the trajectories based on Equation 1 with normally distributed noise in bearing changes described in Section Multiple anchor points achieved sharp turns but also increased variability . In the regular anchoring case , the growth cone position after every 1/r steps becomes a new anchor point . In the probabilistic anchoring case , each growth cone position has a probability of r to become a new anchor point . growth-cone-tracker-5min . Growth cone tracking code . The code tracks the position of the growth cone centre every 5 mins from timelapse AVI files . extract-GC-positions . Growth cone position extraction code . The code to extract the position of the growth cone from the tracings . DOI: http://dx . doi . org/10 . 7554/eLife . 12248 . 030
Here , we presented a model of axon trajectories in gradients and helped resolve the mystery of why axon turning angles in gradients saturate over time in vitro , revealing an important factor limiting axon turning . We found that the movement of the growth cone was strongly influenced by the axon’s tendency to maintain a straight trajectory forward , limiting the directional effect of the gradient and preventing the axon from aligning with the gradient even after a long time . Our model predicted that , averaged over a large population of axons , the initial rate of turning drops rapidly over a short period of time ( 20–40 min ) . The model shows that adding more anchor points can give the growth cone more flexibility and produce larger average turning , but also increases the variability . Thus , we predict that different substrates , producing different densities of anchor points , could result in different trajectories for the same the gradient conditions . The application of forces to axons can induce rapid elongation without axonal thinning , and thus stretch can stimulate growth ( Suter and Miller , 2011 ) . Furthermore , stretch can also regulate the mode of growth . When axons are tightly bound to a sticky substrate , stretching only happens at the tip and axons elongate by tip growth . In contrast , if axons grow relatively unattached to the substrate , they will lengthen by stretching due to the pull of the growth cone ( Chang et al . , 1998; O'Toole et al . , 2008 ) , which appears to be the case in our experimental condition . The tension along the axon will cause stimulation of growth in the existing direction producing straighter trajectories . The stiffness of axons is also important ( Rajagopalan et al . , 2010 ) , and stiffer axons will likely have higher persistence due to their more limited ability to bend . This tension results from cytoskeletal coupling with adhesive interactions to the substrate and is critical to growth cone migration ( Heidemann et al . , 1997; Spire , 2009 ) . Although anchor points are an abstraction in our model , their biological implementation may be focal adhesions . Only at these points is the axon firmly attached to the substrate . There is a number of ways in which anchor points could be investigated experimentally in future work . Axons could be stained for proteins such as integrins ( Kaverina et al . , 2002 ) to test whether their distribution is strongly localized to particular points along the axon . We also predict that applying force orthogonal to the direction of axon growth , for instance by using a pipette to puff liquid at different locations along the axon , would cause a deflection of the axon of a size related to the distance from the nearest anchor point . A similar experiment was performed using a glass needle to tow axons ( O'Toole et al . , 2008 ) . It was observed that the distal region of the axon was free of the substrate , while the proximal region was firmly attached . In addition , it could be possible to determine the internal stress field of an axon , as has been done for growth cones ( Betz et al . , 2011 ) : we would expect the stress to in general be different on the two sides of an anchor point . The density of anchor points will depend on the components of the extracellular matrix ( ECM ) . In our experiments , on laminin , they appeared to be rare . This might be because adhesion points are expensive to produce and the axon can grow faster when it is not attached to the substrate ( Chang et al . , 1998 ) . However , the biological factors governing when new anchor points are generated are unknown . Tension is also dependent on cell type and two main properties of the substrate: stiffness and ECM components . Our data comes from peripheral nervous system ( PNS ) neurons growing on a laminin substrate that is hard rather than gel-like , and other cell types of different substrates might have different behaviours . Central nervous system ( CNS ) and PNS neurons have different sensitivities to substrate stiffness due to adaptation to their natural environments ( Koch et al . , 2012 ) , and traction force in vitro increased on stiffer substrates ( Koch et al . , 2012 ) . Substrates with different ECM components differentially promote growth cone motility and point contact formation . For example , growth cones are more highly motile and neurites extend more rapidly on laminin than fibronectin because point contacts have higher turnover rate ( Robles and Gomez , 2006 ) . Overall , our work suggests that without many anchor points , cues additional to gradients may be necessary for axons to reliably find their targets in vivo ( unless the motility noise is for some reason much lower in vivo than in vitro ) . These could include mechanical cues and axon-axon interactions . To understand such interactions , it is important to generate assays with realistic substrates suitable for different cell types . Recent 3D culture models , in which cells are grown with a protein scaffold , can capture some aspects of the tissue environments instead of hard surfaces ( Cullen et al . , 2007 ) . It will be interesting to see how different ECM properties lead to changes in trajectories and whether they can facilitate more reliable turning . In conclusion , we have presented a simple mathematical model which gives accurate quantitative predictions of the properties of axonal trajectories in a microfluidics-based in vitro gradient assay . The model identifies the key importance of anchor points in controlling turning and provides an explanation for why axonal turning in gradients in vitro tend to saturate rapidly at small turning angles . This model provides a predictive framework which can be used to test whether axonal trajectories observed in vivo can be explained purely in terms of gradient guidance , or whether additional guidance mechanisms are also required .
The device was designed in AutoCAD 2013 based on the pattern in ( Dertinger et al . , 2001 ) . The microfluidic master molds were prepared by standard soft photolithography techniques . The silicon wafer ( M . M . R . C . Pty Ltd ) was coated with a 50 μm thick layer of SU-8 2100 relief ( MicroChem , Westborough , MA ) with a spincoater ( EVG ) . The master was printed onto high-precision photoplates ( Konica-Minolta ) with a photoplotter ( EVG ) . Before replica moulding , the silicon master was silanized with vapour phase silane , under vacuum for 1 hr at room temperature to prevent polydimethylsiloxane ( PDMS ) adhering to the master . PDMS base elastomer ( Sylgard 184 , Dow Corning , Midland , MI ) and silicon elastomer curing agent in a 10:1 ( m/m ) ratio were thoroughly mixed and poured on the master to a depth of about 4 mm . The mold with PDMS was degassed in a vacuum chamber for 2 hr and baked at 80°C for 2 hr . The PDMS replica was then peeled from the silicon wafer and cut into individual chambers . Holes were cored in the PDMS chambers using a 0 . 75 mm corer ( Harris Uni-Core ) . To bond the PDMS chamber to a plastic tissue-culture petri dish , the dish and the chamber were plasma treated at 100W at a pressure of 380 mTorr for 40 s . The plate was then covered with ( 3-Aminopropyl ) triethoxysilane ( APTES , Sigma-Aldrich ) solution ( 5% APTES in 70% ethanol ) . The plastic dish was washed thoroughly with water , air dried , and the chamber was pressed onto the dish . To avoid air bubbles forming , the dish was filled with distilled water and degassed in the vacuum chamber for 10 min before use . SCG neurons were harvested from postnatal day 1–3 Wistar rat pups . The SCGs were then cut into thirds , incubated in 0 . 25% trypsin ( GIBCO ) at 37°C for 30 min and then triturated through a flamed-polished Pasteur pipette for 5 min to dissociate individual cells . The growth solution was Opti-MEM solution ( GIBCO ) containing 1X penicillin/streptomycin , 10 μg/mL mouse laminin , 4% ( v/v ) fetal calf serum , 2% B-27 supplement ( Life Technologies ) , and 0 . 3 nM NGF . The cells were suspended in the growth solution and injected in the microfluidic chamber using a 100-μL glass syringe ( SGE Analytical Science ) . Two syringes were filled with either the high ( 10 nM ) or low ( 0 nM ) NGF solution . The high solution contained 0 . 1% ( v/v ) 40 kDa-dextran fluorescently labelled with tetramethylrhodamine to visualize the gradient . After the cells were seeded , the syringes containing the high and low solutions were connected to the chamber using polyethylene tubings . The chamber was moved to an incubated inverted microscope ( Zeiss AxioObserver ) . The syringes were attached to a Harvard PHD pump and the flow rate was set at 10 μL/hr . After the flow had been started , fluorescent images of the chamber were taken in Zen software . Background intensity outside the chamber was subtracted from the images . The average brightness intensity and variations over time were then calculated across the chamber . A gradient of fluorescence confirmed that the NGF gradient had formed in the growth chamber . To generate a repulsive gradient , KT5720 ( Alexis Biochemicals ) , a specific inhibitor of protein kinase A ( PKA ) , was added into both the high and low solutions at a concentration of 70 nM . After the onset of the gradient , the axons were imaged every 5 min for 6 hr using Zeiss Zen software . After data acquisition , axons of 30 μm length , growing in all directions , that did not branch or retract in at least 80 min , were chosen for measurements . All axons were tracked manually using customized MATLAB software ( The MathWorks ) for as long as possible until they branched or retracted . A 5-min time interval was chosen because , for smaller intervals , variability in identifying the centre of the growth cone was larger than the net movement between frames . The point where the axon attaches to the cell body or the main branch was considered the anchor point . The straightness index S is the inverse of tortuosity , and compares the overall net displacement G of a path with the total path length T ( Codling et al . , 2008 ) . Consider a walk that starts at location ( x0 , y0 ) , and after n steps of lengths lj ( j = 1 . . . n ) finishes at ( xn , yn ) . The straightness index is given by: S = GT = ( xn - x0 ) 2+ ( yn - y0 ) 2∑j=1 nlj This index is between 0 and 1 , where 1 corresponds to movement in a straight line and 0 corresponds to a walk that returns to the origin . The closer this index is to 1 , the straighter the trajectory is . Obviously , S depends on the time interval used for tracing but can be used to compare conditions , which all have the same time interval . All parameters of the model are summarized in Table 1 . We consider a model which is a discretized random walk in which we separate the length and directions of the steps ( Figure 1A ) . We discretized the axons at a timestep of 5 min , and , based on hypotheses we test later , only explicitly modelled the turning angles of the steps or ’bearing changes’ . Δθ ( t ) , the ‘bearing change’ at time t depends on the current bearing of the growth cone θ ( t ) , the angle ϕ ( t ) of the vector connecting the growth cone to its anchor point , the gradient direction Ψ and the noise ξ according to Equation ( 1 ) : Δθ ( t ) =a∠ ( ϕ ( t ) , θ ( t ) ) +b∠ ( Ψ , θ ( t ) ) +ξ , where two parameters a and b scale the contributions of the first term representing persistence and the second term representing the bias due to the gradient . The symbol ∠ ( x , y ) denotes the angle difference x-y constrained to take values from −π to π . It is positive for an anticlockwise turn to get from y to x . As the bearing is biased by the gradient direction , the overall growth cone angle ϕ ( t ) will also be biased by the gradient , coupled through the above equation . We first assume there is only one fixed anchor point where the axon initially grew out of the cell body or the main branch . We will later relax this assumption and allow the growth cone to put down new anchor points along its path . We denote the rate of anchor point deposition as r , which is the inverse of the average number of steps per new anchor point . We first assume an initial direction of ϕ ( 0 ) =θ ( 0 ) =π∕2 , a gradient direction of Ψ = 0 , and a fixed step size s every 5 min . In the idealized noiseless case ( ξ = 0 ) as t→∞ , the equation reaches a steady state when Δθ=0 , that is: ∆θ ( t ) =a ( ϕ ( t ) -θ ( t ) ) +b ( 0-θ ( t ) ) = 0 . This gives: θ ( t ) = aa+bϕ ( t ) =αϕ ( t ) with α=aa+b . Defining L to be the distance of the growth cone from its original position , and using the geometry in Figure 1A , we have: tan ( ϕ ( t+1 ) ) =L sin ϕ ( t ) +s sin ( αϕ ( t ) ) L cos ϕ ( t ) +s cos ( αϕ ( t ) ) ≈ tan ( ϕ ( t ) ) + s sin ( αϕ ( t ) ) L cosϕ ( t ) - L sin ϕ ( t ) s cos ( αϕ ( t ) ) L2cos2ϕ ( t ) The approximation above is due to s ≪ L and ϕ ( t ) → Ψ = 0 as t → ∞ . Using the Taylor expansion f ( x0 + δx ) ≈ f ( x0 ) + δxf’ ( x0 ) and d tan−1 ( x ) /dx = 1/ ( x2 + 1 ) , we invert both sides of the above equation to obtain: ϕ ( t+1 ) ≈ tan-1tan ϕ ( t ) + s sin ( αϕ ( t ) ) L cos ϕ ( t ) - L sin ϕ ( t ) s cos ( αϕ ( t ) ) L2 cos2 ϕ ( t ) ≈ ϕ ( t ) + s sin ( αϕ ( t ) ) L cosϕ ( t ) - L sin ϕ ( t ) s cos ( αϕ ( t ) ) L2 cos2 ϕ ( t ) cos2 ϕ ( t ) ≈ ϕ ( t ) + s/L ( sin ( αϕ ( t ) ) cos ϕ ( t ) - cos ( αϕ ( t ) ) sin ϕ ( t ) ) ≈ϕ ( t ) + s sin ( ( α-1 ) ϕ ( t ) ) /L At t→∞ , Δθ ( t ) →0 , meaning the growth direction aligns with the gradient , thus ϕ ( t ) →0 and L≈st due to geometry ( even for the a = 0 case ) , so the above equation can be simplified as dϕ ( t ) dt≈ ( α-1 ) ϕ ( t ) t dϕ ( t ) ϕ ( t ) ≈ ( α-1 ) dtt ln ϕ ( t ) = ( α-1 ) ln t+const Therefore , the long-term turning behaviour of axons in the model is given by the power law ϕ ( t ) ∝t ( α-1 ) . | For your brain to work , millions of nerve cells have to be connected together precisely . To achieve this , growing nerve fibres navigate through the developing brain by following chemical cues . One important such cue is how the concentration of a chemical varies with distance across the brain . This variation is known as a chemical gradient . However we still don't fully understand exactly how nerve fibres use such gradients to steer themselves . Nguyen et al . have built a mathematical model that accounts for the way that nerve fibres respond to chemical gradients , and showed that the model closely matched new experimental data on the growth of nerve fibres from the rat brain . The model implies that , under some conditions , nerve fibres turn surprisingly little in response to a chemical gradient . Nguyen et al . ’s model can now be used to predict nerve fibre responses in many other situations , and could help researchers to understand more about how the brain becomes wired up during development . The model could also reveal more about the conditions that are needed to cause nerve fibres to turn sharply in response to chemical gradients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | A mathematical model explains saturating axon guidance responses to molecular gradients |
The calcium-activated chloride channel TMEM16A is a member of a conserved protein family that comprises ion channels and lipid scramblases . Although the structure of the scramblase nhTMEM16 has defined the architecture of the family , it was unknown how a channel has adapted to cope with its distinct functional properties . Here we have addressed this question by the structure determination of mouse TMEM16A by cryo-electron microscopy and a complementary functional characterization . The protein shows a similar organization to nhTMEM16 , except for changes at the site of catalysis . There , the conformation of transmembrane helices constituting a membrane-spanning furrow that provides a path for lipids in scramblases has changed to form an enclosed aqueous pore that is largely shielded from the membrane . Our study thus reveals the structural basis of anion conduction in a TMEM16 channel and it defines the foundation for the diverse functional behavior in the TMEM16 family .
Calcium-activated chloride channels ( CaCCs ) are important constituents of diverse physiological processes , ranging from epithelial chloride secretion to the control of electrical excitability in smooth muscles and neurons ( Hartzell et al . , 2005; Huang et al . , 2012; Oh and Jung , 2016; Pedemonte and Galietta , 2014 ) . These ligand-gated ion channels are activated upon an increase of the intracellular Ca2+ concentration as a consequence of cellular signaling events . Although CaCC function can be accomplished by unrelated protein architectures ( Kane Dickson et al . , 2014 , Kunzelmann et al . , 2009 ) , the so far best-characterized processes are mediated by the protein TMEM16A ( Caputo et al . , 2008; Schroeder et al . , 2008; Yang et al . , 2008 ) . TMEM16A is a member of the large TMEM16 family of membrane proteins , also known as anoctamins ( Yang et al . , 2008 ) . The family is exclusively found in eukaryotes and contains 10 paralogs in mammals that all share considerable sequence homology ( Milenkovic et al . , 2010 ) ( Figure 1—figure supplement 1 ) . Although it was initially anticipated that all TMEM16 proteins would function as anion channels ( Hartzell et al . , 2009; Tian et al . , 2012; Yang et al . , 2008 ) , it is now generally accepted that only two family members ( the closely related TMEM16A and B ) are ion channels ( Pifferi et al . , 2009; Scudieri et al . , 2012 ) , whereas most others work as lipid scramblases , which catalyze the passive and bidirectional diffusion of lipids between the two leaflets of a phospholipid bilayer ( Brunner et al . , 2016; Malvezzi et al . , 2013; Suzuki et al . , 2013 , 2010; Whitlock and Hartzell , 2017 , 2016 ) . The TMEM16 family shows a new protein fold , as revealed by the structure of the fungal homologue nhTMEM16 , which functions as lipid scramblase ( Brunner et al . , 2014 ) . nhTMEM16 consists of structured cytoplasmic N- and C-terminal components and a transmembrane domain ( TMD ) containing 10 transmembrane helices . As general for the TMEM16 family , the protein is a homo-dimer ( Fallah et al . , 2011; Sheridan et al . , 2011 ) with each subunit containing its own lipid translocation path located at the two opposite corners of a rhombus-shaped protein distant from the dimer interface ( Brunner et al . , 2014 ) . This lipid path is formed by the ‘subunit cavity’ , a membrane-spanning furrow of appropriate size to harbor a lipid headgroup . Since the subunit cavity is exposed to the membrane , it was proposed that its polar surface provides a favorable environment for lipid headgroups on their way across the membrane , whereas the fatty-acid chains remain embedded in the hydrophobic core of the bilayer ( Brunner et al . , 2014 ) . In the vicinity of each subunit cavity , within the membrane-embedded domain , a conserved regulatory calcium-binding site controls the activity of the protein ( Brunner et al . , 2014 ) . In light of the nhTMEM16 structure and the strong sequence conservation within the family , a central open question concerns how the TMEM16A architecture has adapted to account for its altered functional properties . Previous results suggested that the same region constituting the scrambling path also forms the ion conduction pore ( Yang et al . , 2012; Yu et al . , 2012 ) . However , in what way the distinct structural features of a scramblase , which allows the diffusion of a large and amphiphilic substrate , are altered in a channel that facilitates the transmembrane movement of a comparably small and charged anion , remained a matter of controversy . Here we have resolved this controversy by the structure determination of mouse TMEM16A ( mTMEM16A ) by cryo-electron microscopy ( cryo-EM ) at 6 . 6 Å resolution and a complementary electrophysiological characterization of pore mutants . Our data define the general architecture of a calcium-activated chloride channel of the TMEM16 family and reveal its relationship to the majority of family members working as lipid scramblases . The protein shows a similar overall fold and dimeric organization as the lipid scramblase nhTMEM16 . However , conformational rearrangements of helices lining the lipid scrambling path have sealed the subunit cavity , resulting in the formation of a protein-enclosed ion conduction pore that is for most parts shielded from the membrane but that might be partly accessible to lipids on its intracellular side .
We were interested in the structural properties that distinguish ion channels from lipid scramblases in the TMEM16 family and thus decided to investigate the structural properties of the chloride channel TMEM16A by single particle cryo-EM . For that purpose , we generated a stable HEK293 cell-line , which constitutively expresses the ( ac ) isoform of mTMEM16A , and purified the protein at a saturating calcium concentration in the detergent digitonin ( Figure 1—figure supplement 2A , B ) . Images of flash-frozen samples were recorded on a FEI TITAN Krios electron microscope equipped with an energy filter and a K2-summit camera ( Figure 1—figure supplement 2C ) . The three-dimensional structure of the mammalian ion channel at a nominal resolution of 6 . 6 Å was reconstructed from total of 213 , 243 particles picked from 4178 micrographs ( Figure 1—figure supplement 2C , D; Figure 1—figure supplement 3A; and Table 1 ) . Since the resolution did not significantly improve after addition of further images , it is likely limited by the sample . In the resulting electron density map , the main features of the protein are well defined ( Figure 1A , Figure 1—figure supplement 4 and Video 1 ) . Similarities with nhTMEM16 allowed the construction of a poly-alanine model encompassing the secondary structure elements of the TMD and most of the cytoplasmic N- and C-terminal domains ( Figure 1—figure supplement 4A ) . 10 . 7554/eLife . 26232 . 003Figure 1 . mTMEM16A structure . ( A ) Ribbon representation of the mTMEM16A dimer with the EM density ( contoured at 11σ ) superimposed . ( B ) Superposition of mTMEM16A ( blue and red ) and nhTMEM16 ( beige and grey ) . A and B , The view is from within the membrane with the extracellular side at the top . The membrane boundary is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 00310 . 7554/eLife . 26232 . 004Figure 1—figure supplement 1 . Sequence alignment . Sequences of mTMEM16A ( ac ) ( UniProt Q8BHY3 . 2 ) and nhTMEM16 ( NCBI Reference Sequence: XM_003045982 . 1 ) were aligned with Clustal Omega ( Sievers et al . , 2011 ) and edited manually based on the nhTMEM16 structure ( PDBID 4WIS ) . Identical residues are highlighted in green , homologous residues in yellow and residues of the Ca2+ binding site in red . Secondary structure elements of nhTMEM16 are indicated below . The numbering corresponds to mTMEM16A . ( ▼ ) Indicates mutated positions of the narrow neck and ( ▲ ) of the intracellular vestibule . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 00410 . 7554/eLife . 26232 . 005Figure 1—figure supplement 2 . Protein preparation and cryo-EM image processing . ( A ) Gel-filtration profile of purified and de-glycosylated mTMEM16A run on a Superdex200 column equilibrated with the detergent digitonin . ( B ) SDS-PAGE gel of the concentrated peak fractions used for cryo-EM sample preparation . mTMEM16A is labeled ( * ) . The molecular weight of marker proteins ( kDa ) is indicated . ( C ) Representative micrograph of purified mTMEM16A in vitreous ice . ( D ) Representative images of 2D class averages from two-dimensional classification in RELION . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 00510 . 7554/eLife . 26232 . 006Figure 1—figure supplement 3 . Three-dimensional reconstruction of mTMEM16A . ( A ) Angular distribution of particles included in the final 3D reconstruction . The number of particles with respective orientations are represented by length and color of the cylinders . ( B ) Fourier shell correlations ( FSC ) calculated between independently refined half-maps before ( red ) and after ( blue ) masking indicating a final resolution of 6 . 6 Å on the basis of the FSC = 0 . 143 criterion . ( C ) 3D density map of mTMEM16A colored according to the local resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 00610 . 7554/eLife . 26232 . 007Figure 1—figure supplement 4 . EM density of the mTMEM16A channel . ( A ) Stereo view of the mTMEM16A dimer . The protein is displayed as ribbon , selected α-helices are labeled . ( B ) Stereo view of a superposition of the mTMEM16A and the nhTMEM16 dimers . The EM-density ( contoured at 11σ ) is shown superimposed . ( C ) Stereo view of the unmasked EM density at low contour superimposed on the mTMEM16A model . The micelle is highlighted by coloring the density below a resolution threshold of 5 . 5 Å in dark grey . Lines indicate the membrane boundary and arrows the micelle distortion . Compared to A , the view is rotated by −25° around the y axis . ( D ) Close-up of the density in b at low contour near the intracellular region between α-helices 4 and 5 showing the distortion of the detergent micelle . ( E ) EM-density ( contoured at 7σ ) surrounding the two short amphiphilic helices α0a and α0b at the start of the TMD . EM density in B and E was sharpened with a b-factor of −700 Å2 . A and B , view and color coding is as in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 00710 . 7554/eLife . 26232 . 008Table 1 . Statistics of cryo-EM data collection , 3D reconstruction and model building . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 008Data collectionMicroscopeFEI Titan KriosVoltage ( kV ) 300CameraGatan K2-summitCamera modesuper-resolutionEnergy filterpost-column Gatan GIF quantum energy filter ( 20 eV slit ) Defocus range ( µm ) −0 . 5 to −3 . 8Pixel size ( Å ) 0 . 675 ( in super-resolution ) 1 . 35 ( for reconstruction ) Objective aperture ( µm ) 100Exposure time ( s ) 15Number of frames50 or 100Dose rate on specimen level ( e-/Å2 ) 0 . 8 or 1 . 5 per frame ~80 in totalReconstructionSoftwareRELION1 . 4 and RELION2 . 0SymmetryC2Final number of refined particles213 , 243Resolution of polished unmasked map ( Å ) 7 . 85 ÅResolution of polished masked map ( Å ) 6 . 65 ÅMap sharpening B-factor ( Å2 ) −351 ( −700 for model building ) Model StatisticsNumber of residues modeled434SoftwareChimera , Coot , PhenixMap CC ( whole unit cell ) 0 . 552Map CC ( around assigned model ) 0 . 85710 . 7554/eLife . 26232 . 009Video 1 . The mTMEM16A structure . Ribbon representation of the mTMEM16A model with the EM density and the nhTMEM16 structure superimposed . The structures are seen from within the membrane . Ribbons are colored as in Figure 1 and the positions of bound Ca2+ in nhTMEM16 are indicated by green spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 009 The EM-density of mTMEM16A superimposed on the model of the protein is shown in Figure 1A . Due to the presence of Ca2+ , it likely shows the channel in a Ca2+-bound conformation . In light of the irreversible rundown of TMEM16A-mediated currents observed in patch-clamp experiments at high Ca2+ concentrations , it is at this point ambiguous whether this conformation corresponds to a conducting or a non-conducting state of the channel . Within the membrane , the overall dimensions of mTMEM16A are very similar to nhTMEM16 ( Figure 1B , Figure 1—figure supplement 4B and Video 1 ) . All transmembrane helices are well resolved and thus , could be unambiguously allocated . On the extracellular side , the mTMEM16A map contains a substantial amount of unassigned density that can be attributed to extended loops connecting transmembrane α-helices 1–2 ( α1α2 loop ) and transmembrane α-helices 9–10 ( α9α10 loop ) , which are respectively 50 and 65 residues longer compared to nhTMEM16 ( Figure 1A and Figure 1—figure supplement 1 ) . Both loops appear to be structured , folding into a compact extracellular domain ( Figure 2A ) . Notably , this domain harbors six cysteines that have been shown to be indispensable for channel activity ( Yu et al . , 2012 ) and that are thus potentially involved in disulfide bridges . On the cytoplasmic side , the N-terminal domain of mTMEM16A exhibits a similar fold and location with respect to the TMD as its counterpart in nhTMEM16 ( Figures 1B and 2B , C and Figure 1—figure supplement 4B ) . Consequently , there is no interaction between the N-terminal domains of adjacent subunits , which was previously proposed based on biochemical experiments ( Tien et al . , 2013 ) . A 92 residue long extension in mTMEM16A that precedes the folded N-terminal domain ( Figure 1—figure supplement 1 ) appears to be unstructured , but there is unaccounted electron density that cannot be interpreted at the current resolution of the data ( Figure 1—figure supplement 4B–D ) . At the C-terminus , which is 38 residues shorter than its equivalent part in nhTMEM16 , the first α-helix ( Cα1 ) is folded but it has moved away from the dimer axis and thus no longer contacts its symmetry mate ( Figure 2D ) . The remainder of the C-terminus is likely unstructured and , unlike in nhTMEM16 , does not interact with the adjacent subunit . Hence , the interaction of the subunits within the mTMEM16A dimer differs significantly from nhTMEM16 since the cytosolic domains do not contribute to the dimer interface . Instead , interactions are established mainly at the extracellular part of transmembrane α-helix 10 , which is in a similar location as in nhTMEM16 but extends further towards the outside ( Figures 1 and 2D ) . In the TMD , all membrane-spanning segments are well defined including two short amphiphilic α-helices at its N-terminal part that interact with the polar headgroups at the inner leaflet of the lipid bilayer ( Figure 1—figure supplement 4E ) . In general , the transmembrane helices are in comparable locations to their counterparts in nhTMEM16 ( Figure 1B and Figure 1—figure supplement 4B ) and thus account for the overall similarity between both structures . 10 . 7554/eLife . 26232 . 010Figure 2 . Features of the mTMEM16A structure . ( A ) Unassigned EM density ( contoured at 11σ ) of the extracellular α1α2 and α9α10 loops . Connected transmembrane α-helices are shown as ribbon and labeled . ( B ) Structure of the N-terminal domain . Secondary structure elements are shown as ribbon , α-helices are labeled . ( C ) Ribbon representation of the mTMEM16A dimer . The transmembrane domains ( TMD ) of individual subunits are colored in blue and red , respectively , N-terminal domains ( NTD ) in green and the C-terminal domains ( CTD ) in violet . The view is as in Figure 1A . ( D ) Helices α10 of the TMD and Cα1 of the CTD of both subunits of the superimposed dimeric mTMEM16A and nhTMEM16 structures are shown . The view is from within the membrane towards the dimer interface . B and D , Sections of the EM density ( contoured at 7σ ) are superimposed on selected parts of the model . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 010 The pore region of mTMEM16A , also containing the regulatory calcium-binding site , is formed by transmembrane α-helices 3–8 . This region is well defined , except for the loops connecting α-helices 5 and 6 and 6 and 6’ ( Figure 3 , Figure 3—figure supplement 1 and Video 2 ) . Although , at the current resolution , neither the helix-pitch nor side-chains are resolved , there are several structural features that constrain the location of residues and thus allow for their approximate assignment . The placement is facilitated by conserved loops connecting α-helices 4–5 , 7–8 and 8–9 , which are well defined in the cryo-EM map and thus determine the register of the transmembrane segments ( Figure 3—figure supplement 1B ) . We could further constrain the position of the conserved calcium-binding site , as density between α-helices 6 , 7 and 8 coincides with the position of the two bound calcium ions of nhTMEM16 ( Figure 3—figure supplement 1C ) . The ion conduction pore is lined by residues located on α-helices 3–7 ( Figure 3 ) . In contrast to the transmembrane segments close to the dimer interface ( i . e . α-helices 1 , 2 , 9 , 10 ) , several of the pore-forming α-helices have changed their position relative to nhTMEM16 ( Figures 1B and 4 , Figure 4—figure supplement 1A and Videos 3 and 4 ) . These changes are most pronounced for α-helices 3 , 4 and 6 . As a consequence of conformational rearrangements , α4 and α6 , which line the opposite borders of the membrane-accessible subunit cavity of nhTMEM16 , have come into contact at the extracellular part of the membrane to form a protein-enclosed conduit that is shielded from lipids ( Figures 3 and 4 , Figure 3—figure supplement 1D , Figure 4—figure supplement 1A and Videos 3 and 4 ) . Together with α-helices 3 , 5 and 7 , they constitute the narrow neck of an aqueous pore that spans the extracellular two thirds of the membrane ( Figures 3 and 4 , Figure 4—figure supplement 1B and Video 3 ) . Towards the intracellular side , the detachment of α4 and α6 results in the a dilation of the pore to a wide intracellular vestibule that is exposed to both the cytoplasm and the lipid bilayer ( Figure 4—figure supplement 1B , C ) . The resulting gap between both α-helices may cause a local destabilization of the membrane that is also manifested in a distortion of the detergent micelle observed in the density at lower contour ( Figure 1—figure supplement 4C , D ) . 10 . 7554/eLife . 26232 . 011Figure 3 . Pore region of mTMEM16A . Transmembrane α-helices 3–7 constituting the ion conduction pore of a single mTMEM16A subunit are shown as ribbon and labeled . Sections of the EM density ( contoured at 7σ ) are superimposed on the model . Green spheres correspond to the positions of bound Ca2+ in nhTMEM16 . The view in the left panel is as in Figure 1A , the relationship of other panels is indicated . The location of the ion conduction pore is marked by a black line ( left panel ) or an asterisk ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01110 . 7554/eLife . 26232 . 012Figure 3—figure supplement 1 . Pore region and Ca2+ binding site . ( A ) Stereo view of the pore region of mTMEM16A . The view is from within the membrane . The pore region including the Ca2+ binding site encompassing transmembrane α-helices 3–8 is shown as ribbon . EM density ( contoured at 11σ ) of the entire molecule is shown superimposed . ( B ) EM density ( contoured at 7σ ) corresponding to conserved α7α8 , α4α5 and α8α9 loops superimposed on the model . ( C ) EM density ( contoured at 7σ ) corresponding to the Ca2+-binding region superimposed on the model . The Cα positions of conserved residues constituting the binding site are shown as violet spheres and labeled . ( D ) EM density around α-helices 4 and 6 . A and C , Green spheres indicate bound Ca2+ ions identified in nhTMEM16 . C and D , The relationship between different views is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01210 . 7554/eLife . 26232 . 013Figure 4 . Structural relationships between TMEM16 channels and scramblases . Superposition of pore lining helices of mTMEM16A ( blue ) and nhTMEM16 ( beige ) . Ca2+ ions bound to nhTMEM16 are displayed as spheres ( green ) . Views are as in Figure 3 . The location of the ion conduction pore is marked by a black line ( left panel ) or an asterisk ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01310 . 7554/eLife . 26232 . 014Figure 4—figure supplement 1 . Pore geometry . ( A ) Stereo view of a superposition of the pore regions of the ion channel mTMEM16A ( blue ) and the lipid scramblase nhTMEM16 ( beige ) . The perspective is as in Figure 4 ( center ) . ( B ) Stereo view of the pore region of mTMEM16A . ( C ) Stereo view defining positions of basic residues within the pore . Cα positions of selected residues mutated in this study are indicated as spheres and labeled . The perspective is as in Figure 4 ( left ) . B and C , The molecular surface of the pore is shown as grey mesh . Black and grey lines indicate the boundaries of the hydrophobic and polar parts of the membrane , respectively . A and B , Green spheres indicate bound Ca2+ ions identified in nhTMEM16 . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01410 . 7554/eLife . 26232 . 015Video 2 . Pore region of mTMEM16A . Transmembrane α-helices 3–8 constituting the ion conduction pore and the Ca2+ binding site of one mTMEM16A subunit ( blue ) . EM density is superimposed . Green spheres correspond to the positions of bound Ca2+ in nhTMEM16 . The views are as in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01510 . 7554/eLife . 26232 . 016Video 3 . Comparison of pore regions . Superposition of transmembrane α-helices 3–7 of one subunit of mTMEM16A ( blue ) and nhTMEM16 ( beige ) . Helices line the ion conduction pore in the channel and the lipid pathway in the scramblase , respectively . Green spheres correspond to the positions of bound Ca2+ in nhTMEM16 . The views are as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01610 . 7554/eLife . 26232 . 017Video 4 . Helix arrangements in the TMEM16 family . Morph ( cyan , middle panel ) between transmembrane α-helices 3–7 of one subunit of mTMEM16A ( blue , left panel ) and nhTMEM16 ( beige , right panel ) . The morph between both structures emphasizes the different arrangement of helices in a lipid scramblase and an ion channel of the TMEM16 family and does not reflect conformational changes in TMEM16A . Helices line the ion conduction pore in the channel and the lipid pathway in the scramblase , respectively . Green spheres correspond to the positions of bound Ca2+ in nhTMEM16 . The view is similar as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 017 A model of the pore is shown in Figure 5a . Since the current resolution of the data does not permit a quantitative analysis of its geometry , we restrict our description of the pore to its general geometric features . The wide , intracellular entrance narrows above the region constituting the regulatory Ca2+-binding site ( Figure 4—figure supplement 1B ) . Under the assumption that the structure is close to a conducting state , the narrow upper part most likely requires permeating ions to shed their hydration shell . This is consistent with the observation that the anion selectivity of TMEM16A follows a type 1 Eisenman sequence ( Qu and Hartzell , 2000; Schroeder et al . , 2008; Yang et al . , 2008 ) , which favors larger anions with a lower solvation energy . The pore is amphiphilic and contains charged , polar and apolar residues . The low effective affinity of Cl- conduction suggests weak interactions with permeating ions ( Figure 5—figure supplement 1A ) . Due to the absence of a detailed structural representation of the ion conduction path , we focused on the role of long-range coulombic interactions on anion conduction . We have thus mutated basic residues in the pore to alanine ( Figure 4—figure supplement 1C ) and recorded currents in inside-out patches ( Figure 5—figure supplement 1B ) . In these recordings , we can expect deviations from the linear current-voltage relationships of WT in cases where a mutation alters the rate-limiting barriers at either entrance to the narrow part of the pore ( Figure 5—figure supplement 2A ) ( Läuger , 1973 ) . Such behavior has been previously observed for mutations of Lys 588 , where the removal of the positive charge has resulted in a strong outward rectification of the current ( Jeng et al . , 2016; Lim et al . , 2016 ) . In the model of mTMEM16A , this residue is located at the end of the funnel-shaped vestibule close to the neck of the ion conduction path ( Figure 5A and Figure 4—figure supplement 1C ) . In our data , the mutation K588A has resulted in a similar rectification , indicating that the truncation of the positively charged side-chain has perturbed the electrostatic interaction with permeating anions , ( Figure 5B and Figure 5—figure supplement 2B , C ) effectively increasing the energy barrier of negatively charged ions to enter the pore from its intracellular side ( Figure 5—figure supplement 2A , B ) . A similar effect was observed for the nearby mutant K645A , which removes a positive charge from α-helix six at a position that is located slightly further towards the extracellular side ( Figure 5A , C and Figure 5—figure supplement 2B , C ) . In contrast , several mutations of positively charged residues located in the wide intracellular vestibule did not alter the linear current-voltage relationship of WT ( Figure 5—figure supplement 2C–E ) . At the opposite end of the pore , the mutation R535A has resulted in an inward-rectification , indicating that the mutation hampers the entrance of the anion from the outside ( Figure 5A , D and Figure 5—figure supplement 2A–C ) . In between Lys 645 and Arg 535 , the mutation R515A has caused a deviation from the linear current-voltage relationship in both directions ( Figure 5A , E ) . Thus , this positive charge most likely lowers a rate-limiting energy barrier for anion permeation halfway through the narrow part of the mTMEM16A pore ( Figure 5—figure supplement 2A , B ) . This is consistent with the six-fold lower currents measured for this mutant , despite its robust expression at the surface of HEK cells ( Figure 5—figure supplement 1B , C ) . In no case have we seen any change in the reversal potential measured in asymmetric chloride concentrations , which indicates that no single positive charge dominates the strong anion selectivity of the channel ( Figure 5—figure supplement 3 ) . Together with our structural investigations , the electrophysiology data support the notion of a narrow pore in TMEM16A that widens towards the intracellular side . 10 . 7554/eLife . 26232 . 018Figure 5 . Functional properties of mutants of pore lining residues . ( A ) Structure of the pore region of mTMEM16A viewed from within the membrane as shown in the left panel of Figure 3 . The molecular surface of the ion conduction pore is shown as grey mesh , transmembrane α-helices of the pore region as ribbon , the Cα positions of mutated residues as spheres . Black and grey lines indicate the boundaries of the hydrophobic and polar regions of the bilayer , respectively . I-V relationships of pore mutants ( B ) K588A , ( C ) K645A , ( D ) R535A and ( E ) R515A . Currents were recorded from inside-out patches at 1 mM Ca2+ and symmetric Cl- concentrations . Rundown-corrected data were normalized to the response at 120 mV and show mean and s . e . m . of 8–15 independent recordings . Solid lines show fits to a barrier model . The I-V relationship of WT is shown as dashed line for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01810 . 7554/eLife . 26232 . 019Figure 5—figure supplement 1 . Electrophysiology . ( A ) Concentration-conductance relationship of mTMEM16A currents . Slope conductance was measured from inside-out patches excised from HEK293T cells expressing mTMEM16A at −100 mV , 1 mM Ca2+ and different intracellular Cl- concentrations . Data show mean of 8 independent experiments , errors are s . e . m . ( B ) Averaged traces of the rundown-corrected and normalized independent datasets of WT and pore mutants used for the characterization of I-V relationships . Data were recorded from inside-out patches excised from HEK293T cells expressing mTMEM16A at 1 mM Ca2+ and symmetric Cl- concentrations . Currents were corrected for the irreversible rundown of the channel . Top right panel shows the voltage protocol . Scale bars represent the mean amplitude of averaged datasets . ( C ) Fluorescence of HEK293T cells transfected with a YFP fusion construct of the mutant R515A . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 01910 . 7554/eLife . 26232 . 020Figure 5—figure supplement 2 . Permeation model and properties of pore mutants . ( A ) Rectification in a barrier model of ion conduction . Panels show idealized energy profiles on the permeation path containing up to three energy barriers ( left ) and the consequence on I-V relationships ( right ) . Top , left , two barriers of same height results in a linear I-V relationship . Top , right , the increase of the intracellular barrier results in outward rectification of the current . Bottom , left , the increase of the extracellular barrier results in inward rectification of the current . Bottom , right , the increase of a central barrier results in rectification in both directions . ( B ) Energy profile for ion permeation across mTMEM16A derived from the fits of the I-V relationships shown in Figure 5B–E . ( C ) Table of the rectification index ( RI ) of WT and pore mutants calculated as the ratio of the currents measured at 100 and −100 mV . ( D ) Close-up of the intracellular vestibule of mTMEM16A . The view is as shown in Extended Data Figure 6C . The locations of residues mutated in this study are indicated by spheres . ( E ) I-V relationships of mutants in the intracellular vestibule . Currents were recorded as in Figure 5 from inside-out patches at 1 mM Ca2+ and symmetric Cl- concentrations . Data were normalized to the response at 120 mV and show mean and s . e . m . of 8 independent recordings . For WT , solid line shows a fit of the data to a barrier model . For mutants , I-V relationship of WT is shown as dashed line for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 02010 . 7554/eLife . 26232 . 021Figure 5—figure supplement 3 . Ion selectivity of pore mutants . Na+ vs . Cl- selectivity of WT mTMEM16A and pore mutants . For each construct , left panels show I-V plots of the instantaneous current in response to the indicated voltage steps at 150 mM extracellular and the indicated intracellular NaCl concentrations at 1 mM Ca2+ . Right panels show the relation between the intracellular NaCl concentration and the reversal potential ( Erev ) . The line indicates the Nernst potential of Cl− . A , B , Data are mean values of normalized I-V plots from 5 to 12 individual patches , errors are s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 021
The present study has addressed structural relationships within the TMEM16 family . Since the majority of TMEM16 proteins work as lipid scramblases , which catalyze the diffusion of lipids between the two leaflets of a bilayer , it was postulated that the few family members functioning as ion channels may have evolved from an ancestral scramblase ( Whitlock and Hartzell , 2016 ) . However , the way in which TMEM16 channels have adapted to fulfill their distinct functional task has remained unknown . The structure of mTMEM16A reported here has now resolved this question . As anticipated from the strong sequence conservation , the general architecture of each subunit is shared between both branches of the family ( Figure 1B ) . A previous structure-based hypothesis suggested a possible subunit rearrangement in dimeric TMEM16 channels , where both subunit cavities come together to form a single enclosed pore ( Brunner et al . , 2014 ) . Although this hypothesis was already refuted by recent functional investigations , which demonstrated that the protein contains two ion conduction pores that are independently activated by Ca2+ ( Jeng et al . , 2016; Lim et al . , 2016 ) , the ultimate proof for a double barreled channel is now provided by the mTMEM16A structure , which reveals the location of two pores , each contained within a single subunit of the dimeric protein . A different proposition , referred to as the proteolipidic pore hypothesis , postulated that the ion conduction pathway in TMEM16 channels is partly composed of lipids ( Whitlock and Hartzell , 2016 ) . The authors suggested that immobilized lipid headgroups lining the membrane-exposed ion conduction pore may lower the dielectric barrier for permeating ions on their way across the lipid bilayer ( Whitlock and Hartzell , 2016 ) . Our study has also provided strong evidence against this hypothesis . Instead , the model of mTMEM16A shows that α-helical rearrangements have resulted in occlusion of the lipid pathway , while opening up a conductive pore which is largely shielded from the membrane ( Figure 6 and Videos 3 and 4 ) . The only potential access of lipids is provided on the intracellular side where the detachment of transmembrane α-helices 4 and 6 form a funnel-shaped vestibule that is exposed to the cytoplasm and the lipid bilayer ( Figures 5A and 6B ) . The gap between both α-helices may be a relic of an ancestral scramblase , and as suggested by the observed distortion of the detergent micelle in mTMEM16A , possibly destabilizes the bilayer ( Figure 1—figure supplement 4B , D ) . Notably , this gap is also present in nhTMEM16 , where a similar effect of membrane-bending has been proposed to facilitate scramblase activity , as suggested by molecular dynamics simulations ( Bethel and Grabe , 2016 ) . In this respect , it is noteworthy that the intracellular region connecting transmembrane α-helices 4 and 5 has recently been identified to play an important role in lipid scrambling in TMEM16F and was thus assigned the term ‘scramblase domain’ ( Yu et al . , 2015 ) . Whereas TMEM16A itself does not facilitate lipid transport , scrambling activity was conferred to a chimeric TMEM16A protein carrying the ‘scramblase domain’ of TMEM16F ( Yu et al . , 2015 ) or the equivalent region of TMEM16E ( Gyobu et al . , 2015 ) . Although these results emphasize the general role of the intracellular funnel region for lipid interactions , the altered structure of the ‘subunit cavity’ , in particular the absence of a membrane-exposed polar crevice in TMEM16A , leave the mechanism of lipid scrambling in these chimeras ambiguous . 10 . 7554/eLife . 26232 . 022Figure 6 . Mechanistic relationships within TMEM16 family . ( A ) Depiction of the mTMEM16A pore . The molecular surface of the pore region is shown as grey mesh . The boundaries of hydrophobic ( black ) and polar regions ( grey ) of the membrane are indicated by rectangular planes . The positions of positively charged residues affecting ion conduction are depicted as blue and bound Ca2+ ions as green spheres . Hypothetical Cl− ions ( radius 1 . 8 Å ) placed along the pore are displayed as red spheres . ( B ) Schematic depiction of features distinguishing lipid scramblases ( left ) from ion channels ( right ) in the TMEM16 family . The view is from within the membrane ( top panels ) and from the outside ( bottom panels ) . The helices constituting the membrane accessible polar cavity in scramblases have changed their location in channels to form a protein-enclosed conduit . A and B , Permeating ions and lipid headgroups are indicated in red . DOI: http://dx . doi . org/10 . 7554/eLife . 26232 . 022 The structural view of the ion conduction path in mTMEM16A consisting of a funnel-shaped intracellular vestibule that narrows to a tight pore at the extracellular part of the membrane ( Figure 6A ) is supported by our electrophysiology experiments . Analysis of mutants shows minimal influence of basic residues in the wide intracellular vestibule , but pronounced rectification upon similar replacements near the narrow neck of the pore . Remarkably , equivalent mutations of two of these residues ( Arg 515 and Lys 645 ) have previously been described to alter the selectivity between different anions ( Peters et al . , 2015 ) . Assuming that the imaged protein conformation resembles a conducting state , its pore structure suggests that permeating anions have to shed their hydration shell and interact with pore-lining residues ( Figure 6A ) . The low effective affinity of Cl- conduction indicates that there might not be a single strong site for ion coordination , but that the ions might instead weakly interact with the extended pore region ( Qu and Hartzell , 2000 ) ( Figure 5—figure supplement 1A ) . This is consistent with the fact that no single mutation was identified so far that weakened the strong selectivity for anions over cations ( Figure 5—figure supplement 3 ) . Although ion conduction was previously also reported for TMEM16 family members which function as lipid scramblases ( Lee et al . , 2016; Malvezzi et al . , 2013; Yang et al . , 2012; Yu et al . , 2015 ) , it was proposed that these processes are leaks accompanying the movement of lipids ( Yu et al . , 2015 ) , which differs significantly from the selective anion permeation described here for TMEM16 channels . In summary , our work has unraveled how TMEM16 proteins use a similar architecture to exert substantially different functions . Both structures , namely the scramblase nhTMEM16 and the ion channel mTMEM16A , define the structural relationships within the family , whereby a hydrophilic membrane-exposed cavity in TMEM16 scramblases has changed to an aqueous membrane-shielded pore in TMEM16 channels ( Figure 6B and Video 4 ) . Despite the unusual functional breadth of the family , this ligand-gated ion channel turns out to share its mechanism for ion conduction with other , structurally unrelated , channel proteins .
A HEK293 cell-line stably expressing the mouse TMEM16A ( ac ) isoform ( mTMEM16A , UniProt Q8BHY3 . 2 ) containing a 3C cleavage site , a myc- and an SBP-tag at its C-terminus was generated using the Flp-In System ( Flp-In-293 Cell Line , R75007 , Invitrogen ) . Adherent HEK cells constitutively expressing mTMEM16A were grown on 10 cm dishes ( Corning ) at 37°C and 5% CO2 in Dulbecco’s modified Eagles’s Medium ( Sigma ) containing either 10% fetal bovine serum ( FBS , Sigma ) for cell propagation or 5% FBS during protein production . After reaching >80% confluency , cells were harvested by centrifugation at 500 g , washed with PBS buffer ( 137 mM NaCl , 2 . 7 mM KCl , 12 mM phosphate pH 7 . 4 ) and stored at −20°C until further use . For purification , frozen cell pellets from 7 l of adhesion culture were thawed and resuspended in 140 ml buffer A ( 20 mM HEPES pH 7 . 5 , 150 mM NaCl and 0 . 5 mM CaCl2 ) containing protease inhibitors ( cOmplete , Roche ) . All further steps were carried out at 4°C . Protein was extracted in buffer A containing about 1% digitonin ( AppliChem ) for 2 hr under gentle agitation . Insoluble material was removed by centrifugation at 22 , 000 g for 30 min . The supernatant was filtered through a 5 μm filter ( Minisart , Sartorius ) and incubated with 3 ml of Streptavidin UltraLink resin ( Pierce , ThermoScientific ) in batch for 1 . 5 hr . The beads were washed with 60 column volumes of buffer A containing 0 . 12% digitonin ( Calbiochem; buffer B ) and eluted with three column volumes of buffer B containing 4 mM of biotin . Protein was deglycosylated for 2 hr by addition of PNGaseF , and subsequently concentrated ( Amicon Ultra , 100 k ) . The concentrated sample was applied to a Superdex 200 size-exclusion chromatography column equilibrated in buffer B . The following day fractions containing target protein were concentrated to obtain 15 µl of pure protein at a final concentration of 3 mg ml−1 and subsequently used for EM sample preparation . 2 . 5 µl of purified mTMEM16A at a concentration of 3 mg ml−1 were pipetted onto glow-discharged 200 mesh gold Quantifoil R1 . 2/1 . 3 holey carbon grids ( Quantifoil ) . Grids were blotted for 2–5 s with a blotting force of 1 at 20°C and 100% humidity , and flash-frozen in liquid-ethane using an FEI Vitrobot Mark IV ( FEI ) . Cryo-EM data were collected on a 300 kV FEI Titan Krios electron microscope using a post-column quantum energy filter ( Gatan ) with a 20 eV slit and a 100 µm objective aperture . Data were collected in an automated fashion on a K2 Summit detector ( Gatan ) set to super-resolution mode with a pixel size of 0 . 675 Å and a defocus range of −0 . 5 to −3 . 8 µm using SerialEM ( Mastronarde , 2005 ) . Images were recorded for 15 s with an initial sub-frame exposure time of 300 ms ( 50 frames total ) with a dose of 1 . 5 e−/Å2/frame , and later with a sub-frame exposure time of 150 ms ( 100 frames total ) with a dose of 0 . 8 e−/Å2/frame , resulting in a total accumulated dose on the specimen level of approximately 80 e−/Å2 . A total of 5503 dose-fractionated super-resolution images were 2 × 2 down-sampled by Fourier cropping ( final pixel size 1 . 35 Å ) and subjected to motion correction and dose-weighting of frames by MotionCor2 ( Zheng et al . , 2016 ) . The contrast transfer function ( CTF ) parameters were estimated on the movie frames by ctffind4 . 1 ( Rohou and Grigorieff , 2015 ) . Images showing a strong drift , higher defocus than −3 . 8 µm or a bad CTF estimation were discarded , resulting in 4178 images used for further analysis . Image processing was performed using the software package RELION1 . 4 ( Scheres , 2012 ) and at a later stage RELION2 . 0 ( Kimanius et al . , 2016 ) . Approximately 4000 particles were manually picked to generate templates for automated particle selection . Following automated picking in RELION , false positives were eliminated manually or through a first round of 2D classification resulting in 755 , 348 particles . These were subjected to several rounds of 2D classification to remove particles belonging to low-abundance classes . The remaining 522 , 701 particles were sorted during 3D Classification with C2 symmetry imposed . A model was generated from the nhTMEM16 X-ray structure ( Brunner et al . , 2014 ) ( PDBID 4WIS ) , low-pass filtered to 60 Å and used for the first round of classification . In an iterative mode , the best output map was used for subsequent classification or refinement rounds . Similar classes , comprising 377 , 371 particles , were combined and subjected to auto-refinement in RELION . The resulting map was masked and had a resolution of 7 . 35 Å . To further improve the quality of the density map , per-particle alignment of the frames was performed using the polishing algorithms in RELION . The best results were obtained upon inclusion of all dose-weighted frames and application of a running average window of 9 , a standard deviation of 2 pixels on the translations during movie refinement and 200 pixels on particle distance during particle polishing ( Scheres , 2014 ) . Polished particles were subjected to another round of 2D and 3D classification , resulting in a selection of 213 , 243 particles . The final polished , auto-refined and masked map had a resolution of 6 . 6 Å . The final map was sharpened using an isotropic b-factor ranging between −351 Å2 and −700 Å2 and used for model building . Local resolution estimates were calculated within RELION . All resolutions were estimated using the 0 . 143 cut-off criterion ( Rosenthal and Henderson , 2003 ) with gold-standard Fourier shell correlation ( FSC ) between two independently refined half maps ( Scheres and Chen , 2012 ) ( Figure 1—figure supplement 3B ) . During post-processing , the approach of high-resolution noise substitution was used to correct for convolution effects of real-space masking on the FSC curve ( Chen et al . , 2013 ) . A poly-alanine model encompassing the secondary structure elements of mTMEM16A was constructed based on the nhTMEM16 X-ray structure ( Brunner et al . , 2014 ) ( PDBID 4WIS ) . For that purpose the nhTMEM16 structure was initially docked into the EM density using UCSF Chimera ( Pettersen et al . , 2004 ) . The fit of certain fragments as rigid bodies was subsequently improved in Coot ( Emsley and Cowtan , 2004 ) . Long and poorly conserved loop regions and side-chains were removed from the model and residues of mTMEM16A were assigned based on a sequence alignment ( Figure 1—figure supplement 1 ) . Density for conserved short loops and bound Ca2+ ions assisted the assignment of the register for residues of the pore region . The structure was improved locally by real space refinement in Coot ( Emsley and Cowtan , 2004 ) followed by global real space refinement in Phenix ( Adams et al . , 2002; Afonine et al . , 2013 ) maintaining strong secondary structure and symmetry constraints between the two subunits of the dimeric protein ( Table 1 ) . The final model consists of 434 residues and includes the β-strands and α-helices of the N-terminal domain , two peripheral and 10 transmembrane spanning α-helices of the TMD , including short and conserved loop regions , and the first α-helix of the C-terminal domain . It contains residues 123–127 , 167–214 , 242–254 , 278–282 , 295–305 , 315–355 , 409–438 , 486–520 , 535–602 , 633–666 , 681–781 , 855–885 and 892–904 . The molecular surface of the pore was calculated with MSMS ( Sanner et al . , 1996 ) from coordinates where side-chain positions of residues constituting the ion conduction pore were modeled in Coot ( Emsley and Cowtan , 2004 ) . Model building was performed on the final cryo-EM map sharpened with a B-factor of −700 Å2 , as shown in all figures except for Figure 1—figure supplement 4C , D where a B-factor of −351 Å2 was applied . All structure calculations and model building were performed using software compiled by SBGrid ( Morin et al . , 2013 ) . Structure figures ad movies were prepared with DINO ( http://www/dino3d . org ) or UCSF Chimera ( Pettersen et al . , 2004 ) . For electrophysiology , the mTMEM16A ( ac ) cDNA was cloned into a pcDNA3 . 1 plasmid modified for the FX-system ( Geertsma and Dutzler , 2011 ) with a C-terminal YFP/SBP/myc tag . Mutations were introduced by a modified QuikChange method ( Zheng et al . , 2004 ) and confirmed by sequencing . HEK293T cells ( ATCC CRL-1573 ) were transfected with 3 µg of DNA per 6 cm dish using the calcium phosphate precipitation method . Transfected cells were used within 24 to 96 hr after transfection . Inside-out patches were excised from HEK293T cells expressing WT or mutant constructs after the formation of a gigaohm seal . The seal resistance was typically 4–8 GΩ or higher . Recording pipettes were pulled from borosilicate glass capillaries ( O . D . 1 . 5 mm , I . D . 0 . 86 mm ( Sutter ) ) and fire-polished with a microforge ( Narishige ) before use . Pipette resistance was 3–8 MΩ when filled with recording solution . Voltage-clamp recordings were performed using the Axopatch 200B amplifier controlled by the Clampex 10 . 6 software through Digidata 1550 ( Molecular Devices ) . Raw signals were analogue-filtered at 5 kHz through the in-built 4-pole Bessel filter and digitized at 20 kHz . Liquid junction potential was not corrected . Solution exchange was performed using a theta glass pipette mounted on a high-speed piezo switcher ( Siskiyou ) . Experiments were performed at 1 mM Ca2+ on the intracellular side to maximize channel activation . This also minimizes interference by time-dependent relaxation of the current during a voltage step when information on the instantaneous current response is required . The pipette solution contained 150 mM NaCl , 5 . 99 mM Ca ( OH ) 2 , 5 mM EGTA and 10 mM HEPES at pH 7 . 4 ( NaCl buffer ) . Rectification experiments were carried out under symmetrical ionic conditions with a bath solution having the same composition as the pipette solution . For permeability experiments , the NaCl concentration was adjusted by mixing the NaCl buffer with a ( NMDG ) 2SO4 solution containing 100 mM ( NMDG ) 2SO4 , 5 . 99 mM Ca ( OH ) 2 , 5 mM EGTA and 10 mM HEPES at pH 7 . 4 at the required ratio . For high ionic strength , KCl buffer , containing 150 mM KCl , 5 . 99 mM Ca ( OH ) 2 , 5 mM EGTA and 10 mM HEPES at pH 7 . 4 , was used for both bath and pipette solutions to minimize the junction potential . For concentrations above 150 mM Cl− , KCl was dissolved in this solution at the required amounts . Data were background-subtracted before analysis . Background current was obtained by recording in the corresponding solution in the absence of intracellular Ca2+ . I-V data were obtained by measuring the instantaneous current after each voltage jump in a step protocol ( Figure 5—figure supplement 1B ) . To correct for current rundown , the measured instantaneous currents were divided by the fraction of current remaining during the pre-pulse at +80 mV and were expressed as normalized current ( I/I120mV ) . This is important as uncorrected current rundown can give rise to artificial rectification . Potential voltage offset was detected by recording in symmetrical solutions . Only patches with an offset <2 mV were accepted for analysis . The voltage offset was subtracted from the reversal potentials obtained from asymmetric ionic conditions for the same patch whenever possible . This was not possible for a minority of constructs that displayed low current and/or fast rundown . For these constructs , the averaged offset was subtracted from the averaged reversal potentials obtained in asymmetric ionic conditions . Data are presented as mean ± s . e . m . . To analyze the position-dependent effect of mutations on the rectification of the current , we have employed a barrier model akin to that described by Läuger ( 1973 ) . We are aware of the general limitations of barrier models for quantitative interpretations ( Eisenberg , 1999 ) and thus only aim for a phenomenological description . The model assumes the presence of multiple hypothetical energy barriers on the ion conduction path that are not necessarily identical ( Appendix Scheme 1 ) . The equation used to fit the experimental I-V data and to determine the descriptive energy profile of the constructs is shown below . I=zFAezFV2nRTci−coe−zFVRTe−zFVn−1nRT+ ( 1σh ) 1−e−zFVn−2nRT ezFVnRT−1+1σβ The model contains three free parameters ( n , σβ and σh ) that govern the shape of the I-V relation , which , with reasonable constraints , can be reliably determined from our data ( Figure 5B–E and Figure 5—figure supplement 2A , B ) . A is a proportionality factor , n is the number of barriers and σβ and σh are relative rates for outward flux across the innermost and internal barriers compared to the external barrier . For our fit , we used three barriers to describe the observed behavior and determined σβ and σh for the mutant constructs . The relative increase of the barrier height is obtained byΔEa ( in−out ) =−RTlnσβΔEa ( mid−out ) =−RTlnσh where Ea is the activation energy corresponding to the respective rate constant . These parameters were used to construct descriptive energy profiles to illustrate the effect of the mutations and are shown in Figure 5—figure supplement 2A , B . For more details , see Appendix 1 . The electron density map has been deposited in the Electron Microscopy Data Bank under the accession code EMD-3658 and the coordinates of the model in the Protein Data Bank under the accession code 5NL2 . | Cell membranes are made up of two layers of oily molecules , called lipids , embedded with a variety of proteins . Each type of membrane protein carries out a particular activity for the cell , and many are involved in transporting other molecules from one side of the membrane to the other . The TMEM16 proteins are a large family of membrane proteins . Most are known as lipid scramblases and move lipids between the two layers of the membrane . However , some TMEM16 proteins transport ions in or out of the cell , and are instead called ion channels . TMEM16 proteins are found in animals , plants and fungi but not bacteria , and play key roles in many biological activities that keep these organisms alive . For example , in humans , ion channels belonging to the TMEM16 family help keep the lining of the lung moist , and allow muscles in the gut to contract . The structure of a scramblase shows that two protein units interact , with each unit containing a furrow that spans the membrane , through which lipids can move from one layer to the other . However , to date , the shape of a TMEM16 ion channel has not been determined . It was therefore not clear how a protein with features that let it transport large , oily molecules like lipids had evolved to transport small , charged particles instead . TMEM16A is a member of the TMEM16 family that transports negatively charged chloride ions . Using a technique called cryo-electron microscopy , Paulino et al . have determined the three-dimensional shape of the version of TMEM16A from a mouse . Overall , TMEM16A is organized similarly to the lipid scramblase . However , some parts of the TMEM16A protein have undergone rearrangements such that the membrane-exposed furrow that provides a path for lipids in scramblases is now partially sealed in TMEM16A . This results in an enclosed pore that is largely shielded from the oily membrane and through which ions can pass . Additionally , biochemical analysis suggests that TMEM16A forms a narrow pore that may widen towards the side facing the inside of the cell , though further work is needed to understand if this is relevant to the protein’s activity . The three-dimensional structure of TMEM16A reveals how the protein’s architecture differs from other family members working as lipid scramblases . It also gives insight into how TMEM16 proteins might work as ion channels . These findings can now form a strong basis for future studies into the activity of TMEM16 proteins . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Structural basis for anion conduction in the calcium-activated chloride channel TMEM16A |
Dystroglycan is a cell membrane receptor that organizes the basement membrane by binding ligands in the extracellular matrix . Proper glycosylation of the α-dystroglycan ( α-DG ) subunit is essential for these activities , and lack thereof results in neuromuscular disease . Currently , neither the glycan synthesis pathway nor the roles of many known or putative glycosyltransferases that are essential for this process are well understood . Here we show that FKRP , FKTN , TMEM5 and B4GAT1 ( formerly known as B3GNT1 ) localize to the Golgi and contribute to the O-mannosyl post-phosphorylation modification of α-DG . Moreover , we assigned B4GAT1 a function as a xylose β1 , 4-glucuronyltransferase . Nuclear magnetic resonance studies confirmed that a glucuronic acid β1 , 4-xylose disaccharide synthesized by B4GAT1 acts as an acceptor primer that can be elongated by LARGE with the ligand-binding heteropolysaccharide . Our findings greatly broaden the understanding of α-DG glycosylation and provide mechanistic insight into why mutations in B4GAT1 disrupt dystroglycan function and cause disease .
Dystroglycan ( DG ) is a highly glycosylated basement membrane receptor involved in a variety of physiological processes including maintenance of the skeletal muscle-cell membrane integrity and establishment of the structure and function of the central nervous system ( Barresi and Campbell , 2006 ) . DG is composed of a cell-surface α-subunit and a transmembrane β-subunit . α-DG acts as a receptor for laminin-G domain-containing extracellular matrix ( ECM ) proteins such as laminin , agrin , perlecan and neurexin ( Barresi and Campbell , 2006 ) . In addition , it serves as a cellular receptor and entry site for most Old World arenaviruses , including the highly pathogenic Lassa virus ( LASV ) and Clade C New Word arenaviruses ( Cao et al . , 1998 ) . LASV is the causative agent of severe hemorrhagic fever in humans , a disease that has a mortality rate of ∼15% resulting in several thousand deaths each year . α-DG effectiveness as a receptor is dependent on complex post-translational modifications . Besides numerous modifications with N-glycans and mucin-type O-glycans , a highly complicated series of additions to a phosphorylated O-mannosyl glycan moiety in the N-terminal region of the mucin domain are essential for ligand binding ( Kanagawa et al . , 2004; Hara et al . , 2011 ) . Defects in the proper post-translational processing of α-DG result in loss of receptor function , and in a broad spectrum of congenital muscular dystrophies ( CMDs ) that are accompanied by a variety of brain and eye malformations . Collectively , these dystrophies are classified as dystroglycanopathies ( Barresi and Campbell , 2006 ) . To date , over 17 genes have been reported to be directly or indirectly involved in this ‘functional glycosylation’ of α-DG , and have been linked to human disease when mutated ( Mercuri and Muntoni , 2012; Bonnemann et al . , 2014 ) . Recent gene discovery efforts revealed several novel dystroglycanopathy genes with unknown function ( Vuillaumier-Barrot et al . , 2012; Buysse et al . , 2013; Jae et al . , 2013 ) . In our previous work we were able to assign functions to the POMGNT2 ( Protein O-linked mannose N-acetylglucosaminyltransferase 2 ) ( GTDC2 ) , B3GALNT2 and POMK ( Protein O-mannose kinase ) ( SGK196 ) gene products , which contribute to the synthesis of a phosphorylated , O-mannosyl-linked trisaccharide on α-DG ( Yoshida-Moriguchi et al . , 2013 ) ( Figure 1—figure supplement 1 ) . This so-called Core M3 structure ( GalNAc-β3-GlcNAc-β4-Man-α Ser/Thr ) is synthesized in the endoplasmic reticulum ( ER ) , and it is thought to be a platform for further functional modification of α-DG as it passes through the secretory pathway . The glycosyltransferase LARGE was shown to synthesize and transfer repeating units of [–3-xylose–α1 , 3-glucuronic acid-β1–] to α-DG ( Inamori et al . , 2012 ) . This heteropolymer is postulated to be the terminal glycan moiety anchored by the Core M3 structure . It resembles the ligand-binding glycan and its length correlates with the affinity of α-DG for its ligands ( Goddeeris et al . , 2013 ) . However , how the laminin-binding glycan synthesized by LARGE is attached to the Core M3 structure , and which glycans or other molecules contribute to and form part of this linker structure , remains unknown . We set out to elucidate the structure and monosaccharide composition of the α-DG post-phosphoryl modification , applying a strategic and multifaceted experimental approach starting from the terminal end synthesized by LARGE . We used glycosylation-deficient cells , in vitro enzyme assays , deglycosylation strategies , and NMR ( Nuclear Magnetic Resonance ) -based structure analysis as experimental tools . This strategy revealed a β1 , 4 glucuronyltransferase activity for B4GAT1 . We present experimental evidence that this enzyme B4GAT1 , which was previously described in the literature as B3GNT1 ( Sasaki et al . , 1997 ) in fact encodes for a β1 , 4 glucuronyltransferase and not a β1 , 3 N-acetylglucosaminyltransferase as previously thought . This activity contributes to production of the post-phosphoryl glycan linker by transferring a glucuronic acid ( GlcA ) residue onto a xylose ( Xyl ) acceptor . It thereby forms a glucuronyl-β1 , 4-xylosyl disaccharide , the direct acceptor required by the glycosyltransferase LARGE to initiate formation of the terminal heteropolysaccharide that is involved in ligand binding . B4GAT1 enzymatic activity , of both a recombinant form and the endogenous protein in mouse embryonic fibroblasts ( MEFs ) , was further characterized using a newly developed HPLC ( High-Performance Liquid Chromatography ) -based assay for B4GAT1 activity . Our findings contribute to the current understanding of α-DG posttranslational processing , providing mechanistic insights regarding the pathomechanism underlying α-DG glycosylation-deficient CMDs and revealing new therapeutic avenues for blocking entry of pathogenic LASV viruses .
Our previous work showed that the ER-resident enzymes POMGNT2 ( GTDC2 ) , B3GALNT2 and POMK ( SGK196 ) contribute to synthesis of the phosphorylated Core M3 trisaccharide on α-DG , a moiety that is required as platform for further modification with the LARGE mediated laminin-binding glycan ( Yoshida-Moriguchi et al . , 2013 ) . However , a number of additional genes , namely FKTN ( Fukutin ) ( Kobayashi et al . , 1998; de Bernabe et al . , 2003 ) , FKRP ( Fukutin-related protein ) ( Brockington et al . , 2001; Beltran-Valero de Bernabe et al . , 2004 ) TMEM5 ( Vuillaumier-Barrot et al . , 2012 ) and B4GAT1 ( B3GNT1 ) ( Wright et al . , 2012; Buysse et al . , 2013; Shaheen et al . , 2013 ) are known to be crucial for proper α-DG glycosylation , yet how they contribute has not yet been determined ( Figure 1—figure supplement 1 ) . To investigate if these unassigned genes are involved in the pre- or post-phosphorylation process of Core M3 , we expressed Fc-tagged recombinant α-DG ( DGFc340 ) in [32P] orthophosphate-labeled control and glycosylation-deficient cells . DGFc340 is a secreted α-DG deletion construct that contains only the minimal region of the α-DG mucin-like domain ( aa 316–340 ) , that is required for its functional glycosylation followed by a C-terminal fusion tag encoding the heavy-chain constant ( Fc ) moiety of human IgG1 ( to enable purification of the secreted recombinant protein ) ( Hara et al . , 2011 ) . Although only a small subpopulation of the expressed DGFc protein enters the pathway for functional maturation it was demonstrated that this truncated α-DG fusion protein is a valuable tool to study α-DG functional glycosylation ( Hara et al . , 2011 ) . The goal was to test if DGFc340 can be [32P] phosphorylated in fibroblasts with defects in various dystroglycanopathy genes ( Table 1 ) . In our experiment , fibroblasts with defects in FKTN , FKRP , TMEM5 , B4GAT1 ( B3GNT1 ) and LARGE , but not in the phosphate kinase POMK , were able to produce radioactively labeled DGFc340 ( Figure 1A ) , indicating that FKTN , FKRP , TMEM5 , B4GAT1 and LARGE are involved downstream of POMK in the Core M3 post-phosphorylation process . 10 . 7554/eLife . 03941 . 003Table 1 . Summary of features of control and glycosylation-deficient cell linesDOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 003Mutant geneClinical phenotypeCell typeNucleotide variantAmino acidreferenceControl ( human ) noneHuman skin fibroblastCRL-2127 ( ATCC ) POMKWWS/MEBHuman skin fibroblast14bp homozygous deletion ( c . 720_733delGCTGGTGAGTGCG ) ] , homozygousp . Leu241Profs*26 ( Yoshida-Moriguchi et al . , 2013 ) FKTNWWSHuman skin fibroblastc . 385delAp . I129fsX1GM16192c . 1176C > A , heterozygousp . Y392X ( Coriell Cell Repository ) FKRPWWSHuman skin fibroblastc . 1A > G , homozygousp . M1V ( Van Reeuwijk et al . , 2010 ) TMEM5WWSHuman skin fibroblastc . 1101 G > A , homozygousp . G333RunpublishedB4gat1 ( B3gnt1 ) CMDMEFc . 464T > C , compound het with LacZ null allele , B4gat1LacZ/M155Tp . M155T ( Wright et al . , 2012 ) LargemydCMDMEFdeletion of exons 5−7 , homozygous ( Grewal et al . , 2001 ) Control ( mouse ) noneMEF10 . 7554/eLife . 03941 . 004Figure 1 . Postulated α-DG modifying enzymes are involved in post-phosphorylation processes in the Golgi prior to LARGE . ( A ) Phosphorylation of Fc-tagged DGFc340 in the context of α-DG glycosylation defects . Fc-tagged DGFc340 was produced in [32P] orthophosphate-labeled fibroblasts from control and glycosylation-deficient patients and mice ( Table 1 ) . The DGFc340 was isolated from the culture medium by using protein A-agarose and the samples were separated by SDS PAGE . Gels were stained with Coomassie brilliant blue ( CBB ) and analyzed by phosphorimaging ( 32P ) . ( B ) Subcellular localization of α-DG modifying putative glycosyltransferases , as assessed by immunofluorescence . HEK293T cells stably expressing c-Myc-tagged proteins were stained with anti-Myc ( red ) , anti-Giantin ( Golgi marker , green ) and 4 ́ , 6-diamidino-2-phenylindole ( DAPI , nuclei , blue ) . Individual stainings for c-Myc and Giantin are shown in greyscale and a merged image is shown in color . Scale bars indicate 10 µm . ( C ) Quantitative On-Cell protein blot analysis of LARGE-induced α-DG glycosylation hyperglycosylation in glycosylation-deficient cells . α-DG glycosylation status was tested with and without forced LARGE overexpression by adenovirus mediated gene transfer . The On-Cell Western blots were probed with an antibody against the glycosylated form of α-DG ( IIH6 ) . IIH6 On-Cell quantitative data were normalized with DRAQ5 cell DNA dye ( n = 3 ) . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 00410 . 7554/eLife . 03941 . 005Figure 1—figure supplement 1 . α-DG functional glycosylation and known proteins contributing to its synthesis . α-DG core M3 functional glycosylation can be divided in 2 major processing steps . O-mannosyl pre-phosphoryl modification which is carried out by enzymes in the endoplasmic reticulum ( ER ) ( highlighted in the blue box ) and O-mannosyl post-phosphoryl modification by known or putative glycosyltransferases in the Golgi ( highlighted in the red box ) . Both gene products with known function ( black ) and gene products with currently unidentified function ( red ) are indicated . The putative glycosyltransferases B4GAT1 ( B3GNT1 ) , FKTN , FKRP and TMEM5 are proposed to act prior to LARGE which adds a GlcA-Xyl heteropolymer that is responsible for ligand binding . However based on current knowledge it cannot be completely ruled out that they are involved in the modification of the LARGE glycan repeat itself to modulate ligand binding . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 005 Immunofluorescence examination of HEK293T ( Human Embryonic Kidney ) cells stably transfected with Myc-tagged constructs of this set of proteins , revealed that they co-localize with the Golgi-resident marker protein Giantin ( Linstedt and Hauri , 1993 ) ( Figure 1B ) . Previously , Golgi localization was also demonstrated for FKRP ( Esapa et al . , 2002 ) , FKTN ( Esapa et al . , 2002; Xiong et al . , 2006 ) , B4GAT1 ( B3GNT1 ) ( Buysse et al . , 2013 ) and LARGE ( Brockington et al . , 2005 ) . These results indicate that most if not all of the α-DG O-mannosyl post-phosphoryl processing is carried out by Golgi-resident enzymes . The laminin-binding glycan repeat generated by LARGE is hypothesized to be the terminal glycan structure of the α-DG O-mannosyl post-phosphoryl modification ( Figure 1—figure supplement 1 ) . This would suggest that FKTN , FKRP , TMEM5 and B4GAT1 contribute to a post-phosphoryl linker structure , that can serve as an acceptor for the modification with LARGE . Previous work by Kuga et al . , ( Kuga et al . , 2012 ) also had indicated that FKTN and FKRP are part of the α-DG O-mannosyl post-phosphoryl modification pathway . To test our hypothesis , we infected a panel of glycosylation-deficient cells with a LARGE expressing adenovirus construct and analyzed the glycosylation status of α-DG and the degree of hyperglycosylation by On-Cell immunoblotting with monoclonal antibody IIH6 , which recognizes the α-DG laminin-binding glycan transferred by LARGE ( Inamori et al . , 2012; Goddeeris et al . , 2013 ) . As expected , overexpression of LARGE did not produce the IIH6-positive glycan or significantly bypass the glycosylation defect in either FKTN- , FKRP- , TMEM5- or B4GAT1-deficient cells ( Figure 1C ) , supporting the notion that these encoded proteins work prior to LARGE in the O-mannosyl post-phosphoryl modification process . In summary , our data suggest that Golgi-localized putative glycosyltransferases FKTN , FKRP , TMEM5 and B4GAT1 are essential for the synthesis of a linker structure that is connecting the α-DG O-mannosyl phosphate platform with the terminal laminin binding glycan added by LARGE . To further elucidate the structure of the α-DG post-phosphoryl modification , we examined how synthesis of the terminal LARGE glycan was initiated . Although LARGE is known to be a dual glycosyltransferase that synthesizes repeating units of [–3-xylose–α1 , 3-glucuronic acid-β1–] on α-DG , the identities of both the initiating sugar and the acceptor sugar for this laminin-binding polymer remained unknown . To determine which sugar is initially transferred by LARGE , we developed an in vitro glycosylation assay using a recombinant soluble ( transmembrane domain deleted ) form of LARGE ( LARGEdTM ) and the acceptor protein DGFc340 . DGFc340 isolated from the culture medium of Largemyd ( Large-deficient ) MEFs lacks the LARGE modification on phosphorylated O-mannosyl glycans , and is hypothesized to terminate in a glycan acceptor structure that can be recognized by LARGE ( Inamori et al . , 2012 ) ( Figure 2A ) . To determine which of the sugars in the polymer is initially transferred by LARGE we incubated DGFc340 isolated from the Largemyd MEF culture medium with LARGEdTM and [14C]-labeled UDP-xylose ( Xyl ) and/or UDP-glucuronic acid ( GlcA ) radionucleotide sugar donors . The glycosyl-transfer reaction was measured as the transfer of radioactivity onto the DGFc340 acceptor glycoprotein . As negative control we used a DGFc340 mutant construct ( T317A/T319A ) , which lacks the O-mannosylation sites that are the crucial acceptor platform for subsequent synthesis of the laminin-binding glycan ( Hara et al . , 2011 ) . When the radionucleotide sugars were tested individually in the LARGEdTM in vitro assay , the addition of [14C] UDP-Xyl , but not that of [14C] UDP-GlcA radionucleotides , resulted in radioactive labeling of the DGFc340 acceptor ( Figure 2B ) . However , in the presence of both UDP-Xyl and UDP-GlcA , the transfer of radioactivity was significantly increased consistent with the fact that the LARGE glycan is a heteropolysaccharide ( Figure 2A/B ) . These results indicate that xylose is the initial sugar transferred by LARGE , and that this is followed by the transfer of GlcA to form the repeating [–3-xylose–α1 , 3-glucuronic acid-β1–] heteropolymer . 10 . 7554/eLife . 03941 . 006Figure 2 . β-GlcA serves as an acceptor sugar for LARGE modification starting with xylose . ( A ) Schematic diagram showing the α-DG post-phosphoryl modification in the context of control and glycosylation defects . LARGE adds the ligand-binding glycan to α-DG via a proposed glucuronic acid ( GlcA ) acceptor . LARGEdTM catalytic domains Xyl-T ( orange ) and GlcA-T ( blue ) are highlighted in color . Depicted are also the hypothesized terminal sugar structures of glycosylation-deficient cell lines Largemyd ( Large-deficient ) and pgsI-208 ( UDP-xylose deficient ) . Cleavage of terminal β-GlcA by exoglycosidase β-glucuronidase ( β-GUS ) in Largemyd is indicated ( scissor symbol ) . ( B ) Transfer of [14C] radiolabeled Xyl and GlcA to DGFc340 by LARGEdTM . Fc-tagged DGFc340 was produced in Largemyd ( Large-deficient ) MEF cells and isolated from the culture medium using protein A-agarose . The protein A-bound DGFc340 was used as acceptor in a LARGEdTM reaction with radiolabeled [14C] UDP-Xyl and/or [14C] UDP-GlcA sugar donors . The figure represents the transfer of radiolabeled saccharides onto the donor DGFc340 ( n = 3 ) . Error bars represent SD ( C ) β-Glucuronidase pre-treatment of DGFc340 from Largemyd deficient cells impairs LARGEdTM modification . Protein A-bound DGFc340 ( acceptor ) isolated from transfected Largemyd MEFs was digested with β-glucuronidase ( β-GUS ) prior to the LARGEdTM ( enzyme ) reaction , which included UDP-Xyl and UDP-GlcA as sugar ( donors ) . After incubation with LARGEdTM DGFc340 ( acceptor protein ) was subjected to protein blotting with antibodies against the glycosylated form of α-DG ( IIH6 ) , against Fc and against LARGE ( Rb331 ) . ( D ) The ability of LARGEdTM to modify DGFc340 is impaired in the context of sugar donor-deficient CHO mutant cell lines . Fc-tagged DGFc340 was produced in various glycosylation-deficient Lec CHO cells and isolated from the culture medium using protein A-agarose . As in ( C ) protein A-bound DGFc340 acceptor was used in a LARGEdTM reaction and analyzed by protein blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 006 Next we wanted to identify the acceptor glycan used by LARGE to initiate formation of the laminin-binding glycan . Since the Xyl-T ( xylosyltransferase ) activity of LARGE has acceptor specificity for β-linked GlcA during heteropolymer formation , we hypothesized that β-linked GlcA might be the initial acceptor for the glycan added by LARGE . To test this , we pre-treated DGFc340 from Largemyd MEF cells with β-glucuronidase ( β-GUS ) ( Figure 2A/C ) , and assessed its modification by LARGEdTM in an in vitro assay . Subsequent immunoblotting with the LARGE glycan-specific antibody ( IIH6 ) revealed that the pretreatment of Largemyd DGFc340 with β-glucuronidase resulted in a strong reduction of the IIH6 signal ( Figure 2C ) . These data indicate that LARGE uses a β-linked GlcA residue as an acceptor sugar to initiate synthesis of the polymeric glycan . To determine which monosacharides contribute to synthesis of the O-mannosyl post-phosphoryl acceptor for the LARGE glycan , we performed a LARGEdTM assay with DGFc340 acceptor isolated from a panel of sugar nucleotide-deficient CHO ( Chinese hamster ovary ) cells ( Stanley , 1985; Kingsley et al . , 1986 ) . LARGEdTM was able to efficiently modify DGFc340 from Pro5 ( wild-type ) , Lec2 ( CMP-sialic acid-deficient ) , Lec8 ( UDP-galactose-deficient ) and Lec13 ( GDP-fucose-deficient ) cells , suggesting that sialic acid , galactose and fucose do not contribute to functional glycosylation of α-DG ( Figure 2D ) . ldlD cells deficient for UDP-galactose ( UDP-Gal ) and UDP-N-acetylgalactosamine ( UDP-GalNAc ) demonstrated reduced acceptor activity for LARGEdTM , which can be explained by the fact that B3GALNT2 requires UDP-GalNAc for synthesis of the initial Core M3 structure . Similarly , DGFc340 from Lec15 cells , which are deficient for Dol-P-Man synthesis , did not serve as LARGEdTM acceptor because POMT1 ( Protein O-mannosyltransferase ) and POMT2 require the Dol-P-Man sugar donor to initiate the O-mannosyl Core M3 structure . Most interestingly , in the LARGEdTM in vitro assay DGFc340 from UDP-Xyl-deficient pgsI-208 CHO cells was not modified ( Figure 2D ) . It had also been shown that ectopic LARGE expression in pgsI-208 CHO cells did not induce α-DG hyperglycosylation ( Inamori et al . , 2012; Ashikov et al . , 2013 ) , consistent with the fact that UDP-Xyl is essential for synthesis of the LARGE glycan in vivo . However , in our LARGEdTM in vitro assay , pgsI-208 DGFc340 was also not modified by LARGEdTM , despite the presence of both of the required sugar nucleotides , UDP-GlcA and UDP-Xyl . This clearly demonstrated that one or more xylose residues are required on α-DG before it can be functionally glycosylated by LARGE ( Figure 2A ) . Having identified a β-linked glucuronic acid as the terminal acceptor saccharide for LARGE and xylose as component of the α-DG post-phosphoryl glycan modification , we next sought to determine which enzyme is responsible for the hypothesized glucuronyltransferase activity . Among the group of unassigned genes ( FKTN , FKRP , TMEM5 , B4GAT1 ) only the B4GAT1 gene product showed homology to glucuronyl-transferases . In particular it shares 44% similarity with the LARGE GlcA-T ( Glucuronyltransferase ) domain ( CAZy: GT49; Figure 3A ) . This designated B4GAT1 as a promising candidate for the GlcA-T transferase upstream of LARGE . To test this hypothesis , we generated a 6xHis-tagged soluble construct of B4GAT1 ( transmembrane domain deleted , B4GAT1dTM ) , expressed it in HEK293T cells and purified the recombinant enzyme from the culture medium ( Figure 3—figure supplement 1 ) . We then conducted a transfer assay with B4GAT1dTM as the enzyme source , UDP-GlcA as the sugar donor and fluorescently labeled β-xyloside ( 4-methylumbelliferyl-β-D-xyloside , Xyl-β-MU ) as the acceptor . The reaction products were separated by high-performance liquid chromatography ( HPLC ) . A unique product peak was detected only when UDP-GlcA was used as donor ( Figure 3B , Figure 3—figure supplement 2A ) . We also tested the acceptor specificity , which revealed that B4GAT1dTM GlcA-T activity has low preference for α-linked Xyl , but showed >10- fold higher preference and specificity towards β-linked Xyl acceptors ( Figure 3C , Figure 3—figure supplement 2A ) . The fact that the LARGE glycan disaccharide Xyl-α1 , 3-GlcA-MU was a very weak acceptor for B4GAT1dTM GlcA-transfer suggests that B4GAT1 overexpression does not interfere with LARGE mediated synthesis of the laminin-binding glycan . A characterization of the B4GAT1 GlcA-T activity revealed a metal dependence for manganese ( Mn2+ ) divalent cations ( Figure 3—figure supplement 2B ) and a pH-optimum near pH 7 . 0 ( Figure 3—figure supplement 2C ) . The product peak obtained from the enzymatic reaction of B4GAT1dTM with β-Xyl-MU acceptor was isolated , and its analysis by NMR revealed that the GlcA residue was β-linked to the four position of the xylose β-MU ( Figure 3—figure supplement 3 , Figure 3—source data 1 ) . Thus , B4GAT1 possesses xylose β1 , 4-glucuronyltransferase ( GlcA-T ) activity and it is specific for the substrate β-linked Xyl . 10 . 7554/eLife . 03941 . 007Figure 3 . B4GAT1 has xylose β1 , 4 glucuronyltransferase activity . ( A ) Schematic representation of LARGE and B4GAT1 functional domains . GlcA-T ( blue ) , Xyl-T ( orange ) and transmembrane domain ( black ) are indicated . ( B ) Representative HPLC profiles of the reaction product generated in the absence ( top ) and presence ( bottom ) of a UDP-GlcA sugar ( donor ) in a reaction mix containing Xyl-β-MU ( acceptor ) and B4GAT1dTM ( enzyme ) . Samples were separated on an LC-18 column . P , product . S , unreacted substrate . Dotted line , %B buffer . ( C ) Comparison of B4GAT1dTM GlcA-T activity with respect to various xylose-MU acceptor sugars . Relative activity ( % ) with respect to Xyl-β-MU acceptor ( specific activity: 0 . 2 µmol/h/mg ) is shown ( n = 3 ) . Error bars represent SD . ( D ) Comparison of LARGEdTM Xyl-T activity with respect to various monosaccharide and disaccharide GlcA-MU acceptor sugars . Relative activity ( % ) with respect to intrinsic LARGE polymer specific activity with GlcA-β1 , 3-Xyl-α-MU disaccharide acceptor ( 0 . 08 µmol/h/mg ) ( n = 3 ) . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 00710 . 7554/eLife . 03941 . 008Figure 3—source data 1 . Chemical shifts ( ppm ) of the signals in the 1H and 13C NMR spectra of the enzymatic reaction product of GlcA-β1 , 4-Xyl-β-MU of the glycosyltransferase B4GAT1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 00810 . 7554/eLife . 03941 . 009Figure 3—figure supplement 1 . Purification of B4GAT1dTM . ( A ) Schematic representation of B4GAT1 and the B4GAT1dTM construct used in the enzymatic activity assay . The transmembrane ( TM ) sequence was replaced with a 3xFLAG-TEV tag sequence and the C-terminus was modified with a Myc-6xHis-tag . ( B ) Purification of recombinant B4GAT1dTM from bioreactor culture medium . The recombinant protein was expressed in HEK293T cells and purified from the culture medium using Talon metal-affinity resin . The bioreactor medium samples before ( start ) and after the purification ( void ) as well as the eluted purified protein ( elution ) were analyzed by immunoblotting with anti-Myc ( 4A6 ) antibody . CBB , stained with Coomassie Brilliant blue . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 00910 . 7554/eLife . 03941 . 010Figure 3—figure supplement 2 . Basic characterization of the xylose β1 , 4-glucuronyltransferase activity of B4GAT1 . ( A ) Donor sugar specificity of B4GAT1dTM . Representative data from two independent assays , demonstrating relative activity ( % ) of B4GAT1dTM ( enzyme ) GlcA-T toward Xyl-α-MU and Xyl-β-MU ( acceptor ) when tested with various sugar nucleotides ( donor ) . The specific activity set as 100% for acceptor Xyl-β-MU was 0 . 2 µmol/hr/mg . No sugar other than GlcA was transferred to the acceptors to a significant extent . ( B ) Metal dependence of the B4GAT1dTM GlcA-T activity . Activity assay was carried out in the presence or absence of each metal ion or EDTA ( 10 mM ) , and results are shown as relative activity ( % ) . The GlcA-T activity in the presence of Mn2+ ( specific activity: 0 . 26 µmol/h/mg ) was arbitrarily set at 100% ( n = 3 ) . n . d . , not detected . Error bars represent SD ( C ) pH optimum of B4GAT1dTM GlcA-T activity . Data from three independent experiments are shown as relative activity ( % ) . The highest activity ( specific activity: 0 . 22 µmol/h/mg ) in the dataset was arbitrarily set at 100% . The GlcA-T assays were carried out using Xyl-β-MU as acceptor . The buffers used were: acetate for pH 4 . 5––5 . 5 ( open circle ) , MES for pH 5 . 5––6 . 5 ( closed circle ) , MOPS for pH 6 . 5––7 . 5 ( open square ) and Tris–HCl for pH 7 . 5––8 . 5 ( closed square ) . The details of the conditions are presented in the Materials and methods section . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 01010 . 7554/eLife . 03941 . 011Figure 3—figure supplement 3 . NMR analysis reveals that B4GAT1 is a β1 , 4 glucuronyltransferase . ( A ) HMQC spectrum ( top ) and overlay of HMQC ( black ) and HMBC ( green ) spectra ( bottom ) for the B4GAT1 enzymatic reaction product . The cross-peaks are labeled with a first letter representing the subunit designated in C and the rest of the label representing the position on that subunit . The observed interglycosidic cross-peak BH1/AC4 in the HMBC spectrum clearly demonstrates the presence of a 1→4 interglycosidic linkage between the residues B and A . The cross-peak marked with a star represents an impurity . ( B ) TOCSY ( top ) and ROESY ( bottom ) spectra of the B4GAT1 enzymatic reaction product collected with a mixing time of 77 and 300 ms , respectively . The cross-peaks are labeled as in ( A ) . The observed interglycosidic ROEs are indicated in green circles . The ROE data indicate that both residues exist in β-configurations . ( C ) Schematic depiction of the disaccharide structure produced by B4GAT1 , with the sugar units labeled A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 01110 . 7554/eLife . 03941 . 012Figure 3—figure supplement 4 . Test B4GAT1 for GlcNAc transferase activity with iGnT substrate Gal-β1 , 4-GlcNAc-β-MU . ( A ) Using B4GALT1 we synthesized the hypothesized iGnT substrate Gal-β1 , 4-GlcNAc-β-MU by transferring a β1 , 4 Galactose to the acceptor GlcNAc-β-MU . The purified Gal-β1 , 4-GlcNAc-β-MU disaccharide was further analyzed by NMR . ( B ) HMQC spectrum . The folded peak is shown in blue . The cross peaks are labeled with a first letter representing the subunit as designated in the structure shown above the spectra and the rest of the label representing the position on that subunit . Overlay of HMQC ( black and blue ) and HMBC ( green ) spectra . The strong cross peak labeled as BH1/AC4 was detected in the HMBC spectrum , demonstrating the presence of a 1→4 interglycosidic linkage between Gal and GlcNAc . ( C ) TOCSY spectrum collected with a mixing time of 77 ms . ROESY spectrum collected with a mixing time of 300 ms . The observed strong NOEs from BH1 to BH3 and BH5 ( cross peaks labeled as BH1/BH3 and BH1/BH5 ) demonstrate that the Gal has a β-configuration . Similarly , the observed strong NOEs from AH1 to AH3 and AH5 ( cross peaks labeled as AH1/AH3 and AH1/AH5 ) demonstrate that the GlcNAc has a β-configuration . A strong interglycosidic NOE was observed between BH1 and AH4 ( green circle ) , which is consistent with the 1 to 4 linkage as determined from the HMBC spectrum . ( D ) Representative HPLC profiles of the reaction product generated in the absence ( blue ) and presence ( red ) of a UDP-GlcNAc sugar ( donor ) in a reaction mix containing Gal-β1 , 4-GlcNAc-β-MU ( acceptor ) and B4GAT1dTM ( enzyme ) . Samples were separated on an LC-18 column . S , unreacted substrate . Dotted line , %B buffer . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 012 It had previously been shown that , during synthesis of the LARGE heteropolysaccharide , β1 , 3-linked GlcA serves as the acceptor for LARGE Xyl-T ( Inamori et al . , 2012 ) . In the current study we found that a β1 , 4-linked GlcA transferred by B4GAT1 serves as the acceptor glycan for initiation of synthesis of the LARGE glycan , via the addition of a xylose . To assess if LARGE can use one or the other glycosidic linkage β-linked GlcA acceptor with higher efficiency , we tested LARGEdTM Xyl-T activity on two disaccharides GlcA-β1 , 4-Xyl-β-MU and GlcA-β1 , 3-Xyl-α-MU along with the monosaccharide GlcA-β-MU ( 4-methylumbelliferyl-β-D-glucuronide ) . As shown in Figure 3D , LARGE did not distinguish between GlcA-β1 , 3-Xyl and GlcA-β1 , 4-Xyl , as similar activities were measured in the presence of both disaccharide acceptors . However , the length of the acceptor appears to be important , since the disaccharide acceptor showed 25-fold higher activity than the monosaccharide acceptor ( Figure 3D ) . In summary , it is currently unknown why the β1 , 4-linked GlcA acceptor in the initial LARGE acceptor primer and the β1 , 3-linked GlcA in the terminal LARGE glycan have different linkages , and how each linkage contributes spatially to the overall structure , while LARGE shows similar activity towards both acceptors . To elucidate further the role of B4GAT1 in vivo , we isolated MEFs from control , Largemyd ( Large-deficient ) and B4gat1-deficient mice ( Wright et al . , 2012 ) ( Table 1 ) and analyzed the glycosylation status of α-DG . Immunoblotting revealed that whereas control MEFs were positive for functional glycosylation of , and laminin-binding by , α-DG from Largemyd MEFs completely lacked both features ( Figure 4A ) . Also , B4gat1-deficient MEFs demonstrated strongly reduced but detectable residual functional glycosylation and laminin binding , and normal levels of hypoglycosylated α-DG core protein ( Figure 4A ) . Adenovirus-mediated ectopic expression of B4GAT1 did not affect the glycosylation status of α-DG in control and Largemyd MEFs but , as expected , was able to rescue the α-DG glycosylation defect in B4gat1-deficient MEFs ( Figure 4A ) . As demonstrated previously ( Barresi et al . , 2004; Inamori et al . , 2012; Willer et al . , 2012 ) , forced adenovirus-mediated ectopic expression of LARGE in control and Largemyd MEFs induces α-DG hyperglycosylation . In contrast , B4gat1-deficient cells showed only a low level of α-DG hyperglycosylation after LARGE overexpression ( Figure 4A ) , suggesting that in the context of mutant B4GAT1 , only few acceptor sites for LARGE modification are available . Finally , ectopic co-expression of B4GAT1 and LARGE resulted in α-DG hyperglycosylation in all three tested MEF lines ( Figure 4A ) . This result is consistent with the hypothesis that B4GAT1 acts prior to the glycosyltransferase LARGE . 10 . 7554/eLife . 03941 . 013Figure 4 . B4gat1-deficient MEFs have impaired α-DG functional glycosylation and endogenous B4GAT1 activity . ( A ) Functional glycosylation and complementation analysis of α-DG in wild-type , Large- and B4gat1-deficient MEFs . Immunoblots and laminin overlay assay of WGA-enriched cell lysates extracted from WT , Largemyd ( Large-deficient ) and B4gat1-deficient MEFs . As indicated MEFs were uninfected ( mock ) or infected with adenovirus constructs expressing B4GAT1 , LARGE or both ( B + L ) . Antibodies used were: glyco α-DG ( IIH6 ) , core α-DG , core β-DG ( AP83 ) , anti-V5 and anti-LARGE ( Rb331 ) . ( B ) Comparison of endogenous B4GAT1 GlcA-T activity in control , Large- and B4gat1-deficient MEFs . Additionally , B4gat1-deficient MEFs ( B4gat1LacZ/M155T ) complemented with control B4GAT1 expressing adenovirus ( Ad5 ) were tested . Cell lysates were used as enzyme source to measure endogenous B4GAT1 activity . Relative activity ( % ) with respect to control MEFs specific activity ( 91 . 6 pmol/h/mg ) is shown ( n = 3 ) . Error bars represent SD . ( C ) Comparison of endogenous LARGE GlcA-T activity in control , Large- and B4gat1-deficient MEFs . Additionally , Large-deficient MEFs complemented with control LARGE expressing adenovirus ( Ad5 ) were tested . WGA enriched glycoprotein samples were used as enzyme source to measure endogenous LARGE activity . Relative activity ( % ) with respect to control MEFs specific activity ( 0 . 52 pmol/h/mg ) is shown ( n = 3 ) . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 01310 . 7554/eLife . 03941 . 014Figure 4—figure supplement 1 . B4gat1-deficient MEFs have impaired endogenous B4GAT1 activity . Representative HPLC profiles of the reaction product are shown . ( A/B ) Endogenous B4GAT1 enzyme activity of cell lysates from wild-type ( A ) and B4gat1-deficient ( B ) MEFs . B4gat1-deficient MEFs ( B4gat1LacZ/M155T ) show some residual activity <3% ( asterisk ) . ( C/D ) Endogenous LARGE enzyme activity of WGA-enriched cell lysates from control ( C ) and Largemyd ( D ) ( Large-deficient ) MEFs . Largemyd MEFs lack LARGE activity and do not show any residual activity ( asterisk ) . The details of the conditions are provided in the Materials and methods section . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 014 Next , to determine if endogenous B4GAT1 activity was detectable in MEF cells , we subjected samples from control , B4gat1 and Largemyd MEFs to the B4GAT1 enzyme activity assay , using Xyl-β-MU as the acceptor . Whereas control and Largemyd samples showed comparable B4GAT1 transferase activity , only low residual activity ( <3% ) was detectable in B4gat1-deficient cells , and this loss could be restored by ectopic expression of B4GAT1 ( Figure 4B , Figure 4—figure supplement 1A/B ) . Similarly , when we tested LARGE GlcA-T activity in control and glycosylation-deficient MEFs , only Largemyd MEFs lacked LARGE GlcA-T activity; the control and B4gat1-deficient cells were normal ( Figure 4C , Figure 4—figure supplement 1C/D ) . These results suggest that the enzymatic activities of B4GAT1 and LARGE are independent and that each is unaffected by mutations in the gene product of the other . To date , several disease-causing B4GAT1 ( formerly termed B3GNT1 ) mutations have been reported in human patients ( Buysse et al . , 2013; Shaheen et al . , 2013 ) , in an N-ethyl-N-nitrosourea ( ENU ) -induced mutant mouse model ( Wright et al . , 2012 ) and in a genetic screen for modifiers of LASV entry ( Jae et al . , 2013 ) . To test how reported B4GAT1 missense mutations affect the intracellular localization of B4GAT1 and its enzymatic activity , we cloned three mutant B4GAT1-Myc expression constructs ( Figure 5A ) : Mut1 ( N390D ) represents a mutation identified in a patient with Walker-Warburg syndrome ( Buysse et al . , 2013 ) ; Mut2 ( D227N/D229N ) is a mutation in the glycosyltransferase signature DXD motif ( Wiggins and Munro , 1998; Bao et al . , 2009 ) ; and Mut3 ( M155T ) mimics a mutant allele identified in a B4gat1-deficient mouse model with axon guidance defects ( Wright et al . , 2012 ) . Immunoblot analysis confirmed that expression levels were similar for the Myc-tagged B4GAT1 control and all three mutant constructs Mut1-Mut3 in stably expressing HEK293T cells ( Figure 5B ) . Previously , it had been shown that B4GAT1 localizes to the trans-Golgi near the TGN ( trans-Golgi network ) ( Bao et al . , 2009; Lee et al . , 2009; Buysse et al . , 2013 ) . In our immunofluorescence analysis , we found both the control and mutant construct Mut1 to exhibit normal Golgi localization and to co-localize with the Golgi marker Giantin ( Figure 5C ) . B4GAT1 mutations in constructs Mut2 and Mut3 , however , resulted in a high degree of mislocalization to the ER , as judged by overlap of the signal with the ER marker ERp72 ( Figure 5D ) , indicating that the B4GAT1 mutant proteins are misfolded and retained in the ER . Analysis of B4GAT1 enzyme activity in lysates from cells stably overexpressing these constructs revealed strongly reduced activity in the cases of all three mutants , of less than 5% compared to activity levels in wild-type control ( Figure 5E ) . Similarly , none of the B4GAT1 mutant constructs was able to complement and rescue the α-DG glycosylation defect in B4gat1-deficient MEFs ( Figure 5F ) . These findings confirm that the identified B4GAT1 mutations are pathological and have a direct negative impact on B4GAT1 activity regardless of their subcellular localization . Additionally , the finding that the B4GAT1 DXD motif is essential further supports a role for B4GAT1 as a glycosyltransferase , since the DXD motif is thought to be involved in binding carbohydrate sugar-nucleoside diphosphates in manganese-dependent glycosyltransferases ( Wiggins and Munro , 1998 ) . 10 . 7554/eLife . 03941 . 015Figure 5 . Expression analysis and GlcA-T enzyme activity of B4GAT1 mutant constructs . ( A ) Schematic presentation shows B4GAT1 enzyme product with functional domains and B4GAT1 mutations Mut1-Mut3 are indicated . ( B ) Expression analysis of B4GAT1-Myc control and mutant constructs in stable HEK293T cells . Immunoblotting of cell lysates from HEK293T cells stably overexpressing wild-type B4GAT1-Myc and mutant constructs ( Mut1 , Mut2 and Mut3 ) with anti-Myc antibody and β-Actin ( loading control ) . ( C/D ) Subcellular localization of B4GAT1-Myc control and mutant constructs in stable HEK293T cells ( see B ) . B4GAT1-Myc constructs were stained with anti-Myc ( red ) , ( C ) anti-Giantin ( Golgi marker , green ) , ( D ) anti-ERp72 ( ER marker , green ) and 4 ́ , 6-diamidino-2-phenylindole ( DAPI , nuclei , blue ) . Individual stainings for c-Myc Giantin and ERp72 are shown in greyscale , and merged images are shown in color . Scale bars indicate 10 µm . ( E ) B4GAT1 enzyme activity in cell lysates from stable HEK293T cells overexpressing B4GAT1-Myc wild-type and B4GAT1-Myc mutant constructs ( Mut1-Mut3 ) . Relative activity ( % ) with respect to B4GAT1wild-type specific activity ( 19 . 8 nmol/hr/mg ) is shown ( n = 3 ) . Error bars represent SD , Statistical analyses were performed by two-tail Student's t test . **p < 0 . 001 . ( F ) Complementation of B4gat1-deficient ( B4gat1LacZ/M155T ) MEF cells with B4GAT1-Myc control and mutant constructs . B4gat1-deficient MEFs were nucleofected with a wild-type or mutant B4GAT1 expression construct . α-DG functional glycosylation was analyzed by On-Cell-Western analysis . α-DG functional glycosylation was detected with α-DG glyco ( IIH6 ) antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 015 To further characterize the endogenous acceptor for B4GAT1 , we first tested if B4GAT1dTM was able to use DGFc340 from control , Largemyd and B4gat1-deficient MEFs as an acceptor . Similar to the LARGE acceptor experiment in Figure 2B , we used radiolabeled [14C] UDP-GlcA sugar donor and measured transfer of [14C] to the protein A-bound DGFc340 acceptor . As expected , DGFc340 isolated from B4gat1-deficient cells was the only acceptor that incorporated substantial levels of the radioactive label ( Figure 6A ) . This confirmed that only the DGFc340 acceptor from B4gat1-deficient cells resembled the terminal acceptor glycan suitable for B4GAT1dTM to add GlcA . Our B4GAT1dTM in vitro enzyme assay demonstrated acceptor specificity for β-linked xylose ( Figure 3C ) . To corroborate the hypothesis that β-linked xylose also serves as the endogenous B4GAT1 α-DG acceptor we pre-treated DGFc340 from B4gat1-deficient cells with β-xylosidase and measured the transfer of [14C] GlcA by B4GAT1dTM . After β-xylosidase treatment , the ability of B4gat1-deficient DGFc340 to act as an acceptor was strongly reduced; this constitutes indirect evidence that a β-linked xylose is indeed the postulated endogenous acceptor for B4GAT1 ( Figure 6B ) , and that a yet unidentified xylosyltransferase acts upstream of B4GAT1 . 10 . 7554/eLife . 03941 . 016Figure 6 . β-xylose is the endogenous acceptor for B4GAT1 . ( A ) B4GAT1dTM enzymatic transfer of [14C] radiolabeled GlcA to DGFc340 . Fc-tagged DGFc340 ( acceptor ) was produced in control , Largemyd ( Large-deficient ) and B4gat1-deficient MEFs and isolated from the culture medium using protein A-agarose . The protein A-bound Fc340 was used as acceptor in a B4GAT1dTM ( enzyme ) reaction with radiolabeled [14C] UDP-GlcA sugar ( donor ) . The figure represents the transfer of radiolabeled GlcA onto the donor DGFc340 ( n = 3 ) . Error bars represent SD . Statistical analyses were performed by two-tail Student's t test . **p < 0 . 001 . ( B ) β-Xylosidase pre-treatment impairs B4GAT1dTM transfer of [14C] radiolabeled GlcA . DGFc340 ( acceptor ) from B4gat1-deficient MEFs was digested with β-xylosidase prior to the B4GAT1dTM ( enzyme ) transfer reaction with [14C] UDP-GlcA sugar ( donor ) . The figure represents the transfer of radiolabeled GlcA onto the donor DGFc340 ( n = 3 ) . Error bars represent SD Statistical analyses were performed by two-tail Student's t test . **p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 016 To further corroborate our finding that the glycosyltransferase LARGE utilizes a glucuronic acid-β1 , 4-xylose-β disaccharide acceptor as a primer to elongate it with its dual glycosyltransferase and polymerizing activity , we performed NMR structural studies . In our approach towards confirming each individual glycosidic linkage , we first synthesized the tetrasacharide GlcA-Xyl-GlcA-Xyl-MU , starting with the monosaccharide acceptor Xyl-β-MU and extending it in a stepwise manner using recombinant B4GAT1dTM and LARGEdTM as enzymes sources ( Figure 7A ) . 10 . 7554/eLife . 03941 . 017Figure 7 . NMR analyses of the tetrasacharide generated from B4GAT1dTM and LARGEdTM enzymatic reactions . ( A ) Schematic depiction of the tetrasaccharide structure produced by the sequential reactions of B4GAT1dTM followed by LARGEdTM with the sugar units labeled A-D to indicate the order of their addition . ( B ) Overlay of the HMQC ( black ) and HMBC ( green ) spectra of the tetrasaccharide . All cross-peaks in the HMQC spectrum are labeled . Three interglycosidic cross-peaks detected in the HMBC spectrum are also labeled and indicated with red circles . The peaks are labeled with a first letter representing the subunit designated in A , and the rest of the label representing the position on that subunit . ( C ) TOCSY spectrum ( top ) and ROESY spectrum ( bottom ) of the tetrasaccharide . The TOCSY and ROESY spectra were collected with mixing time of 77 and 300 ms , respectively . The cross-peaks are labeled as in B . The observed interglycosidic ROEs are indicated with green circles . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 01710 . 7554/eLife . 03941 . 018Figure 7—source data 1 . Chemical shifts ( ppm ) of the signals in the 1H and 13C NMR spectra of the tetrasaccharide of GlcA-β1 , 3-Xyl-α1 , 3-GlcA-β1 , 4-Xyl-β-MU produced by B4GAT1 and LARGE . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 018 The 1H and 13C resonances of the isolated tetrasaccharide product were assigned by using heteronuclear multiple quantum coherence ( HMQC ) , heteronuclear 2-bond correlation ( H2BC ) , and heteronuclear multiple bond coherence ( HMBC ) spectra ( Figure 7B , Figure 7—source data 1 ) . The detection of the interglycosidic cross-peaks of BH1/AC4 , CC1/BH3 , and DH1/CC3 in the HMBC spectrum ( Figure 7B ) clearly indicates the presence of a 1→4 interglycosidic linkage between sugar residues B and A , a 1→3 interglycosidic linkage between residues C and B , and a 1→3 interglycosidic linkage between residues D and C , respectively . A strong rotating-frame Overhauser effect ( ROE ) was observed from the H1 to H3 and H5 protons of residues A , B , and D in the ROE spectroscopy ( ROESY ) spectrum ( Figure 7C ) , demonstrating that they have a β-configuration . The observed strong ROE from the residue C H1 proton to its own H2 , but not to H3 and/or H5 demonstrates that the residue C has an α-configuration . The inter-residue ROEs observed in the ROESY spectrum are also consistent with the interglycosidic linkage assignments determined from the HMBC spectrum . Therefore , the tetrasaccharide has the glycosidic linkage structure GlcA-β1 , 3-Xyl-α1 , 3-GlcA-β1 , 4-Xyl-β-MU ( Figure 7A ) . These studies show that B4GAT1 possesses β1 , 4 glucuronyltransferase activity , and that LARGE can elongate this primer structure by adding repeating units [-3-Xyl-α1 , 3-GlcA-β1-] to produce a heteropolysaccharide . To further illustrate the complexity of assembling the functional glycan of α-DG , we summarize the current knowledge about the α-DG sugar structures and the contributing genes/enzymes in Figure 8 . 10 . 7554/eLife . 03941 . 019Figure 8 . Model of proposed α-DG O-mannosyl laminin-binding glycan structure and the enzymes that contribute to its synthesis . Post-phosphoryl modification of α-DG requires B4GAT1 ( β1 , 4 glucuronyltransferase ) ; this enzyme generates the acceptor glycan , which serves as a primer for the glycosyltransferase LARGE to initiate synthesis of the laminin-binding glycan . Both gene products with known function ( black ) and gene products with currently unidentified function ( red ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 01910 . 7554/eLife . 03941 . 020Figure 8—figure supplement 1 . B4GAT1 and LARGE expression in human tissues . qPCR revealing ubiquitous B4GAT1 and LARGE expression in all tissues analyzed , with highest expression of LARGE in brain and heart . cDNA was synthesized using random primers and oligo ( dT ) on commercially available human tissues RNAs . For each tissue , B4GAT1 and LARGE were specifically amplified , in triplicate , in the presence of SYBRgreen , and their expressions was normalized to that of the 28S RNA ( normalization control ) . The expression in each tissue is referenced with respect to that in brain . Analyzed tissues: Br ( brain ) , Ey ( eye ) , He ( heart ) , Ki ( kidney ) , Li ( liver ) , Lu ( lung ) , Pa ( pancreas ) , MG ( mammary gland ) , Ov ( ovary ) , Pl ( placenta ) , Pr ( prostate ) , SM ( skeletal muscle ) , Sp ( spleen ) , Te ( testis ) , Th ( thymus ) . Error bars represent . DOI: http://dx . doi . org/10 . 7554/eLife . 03941 . 020
In this study we used a multidisciplinary approach to investigate how the assembly of the α-DG LARGE glycan is initiated , and found that it requires B4GAT1-dependent synthesis of a novel glucuronyl-xylosyl acceptor primer . We show that B4GAT1 is a xylose β1 , 4-glucuronyltransferase , and that it is involved in synthesizing the glycan primer that subsequently can be elongated by LARGE with the ligand-binding glycan . B4GAT1 was initially cloned and described by Sasaki et al . , ( Sasaki et al . , 1997 ) as β1 , 3-N-acetylglucosaminyltransferase ( iGnT , β3GNT1 or B3GNT1 ) , which is essential for the synthesis of poly-N-acetyllactosamine . Furthermore , the B3GNT1 enzyme was proposed to contribute to the i antigen synthesis pathway by transferring N-acetylglucosamine onto a β-galactose acceptor with β1 , 3 linkage ( Sasaki et al . , 1997 ) . In contrast our data reveal a β1 , 4-glucuronyltransferase activity , which we have designated B4GAT1 . We tested B4GAT1dTM with UDP-GlcNAc and the proposed Gal-β1 , 4-GlcNAc-β-MU acceptor , but we were not able to validate any N-acetylglucosaminyltransferase activity ( Figure 3—figure supplement 4 ) . Therefore , we propose to rename the enzyme B4GAT1 , as a new member of the glucuronyltransferase family of proteins . To date , only 2 other enzymes are known to have β1 , 4-glucuronyltransferase activity . These are EXT1 and EXT2 , and both are involved in the synthesis of heparan sulphate proteoglycans ( Lidholt and Lindahl , 1992 ) . Our findings regarding assembly of the LARGE glycan reveal striking similarities to the unique mechanism underlying the synthesis of proteoglycans . Both glycan polymers consist of repeating disaccharides that are synthesized by glycosyltransferases with dual glycosyltransferase activities ( Esko et al . , 2009; Inamori et al . , 2012 ) . Furthermore , in both cases assembly of the terminal heteropolymer glycan is initiated by a disaccharide primer , which is part of a larger glycan linker that anchors the polysaccharide to a protein backbone . Future studies are needed to elucidate the full α-DG glycan structure and determine the roles of the putative glycosyltransferases FKTN , FKRP and TMEM5 in anchoring the α-DG ligand-binding glycan moiety to the phosphorylated Core M3 structure . At this point it cannot be ruled out that other , currently unidentified , genes also contribute to synthesis of the functional α-DG glycan . Similar to their counterparts in other dystroglycanopathy genes , B4GAT1 ( B3GNT1 ) loss-of-function mutations in human patients result in Walker-Warburg Syndrome ( WWS ) ( Buysse et al . , 2013; Shaheen et al . , 2013 ) , the most severe condition in a range of clinically defined CMDs that are accompanied by brain and eye malformations . Milder B4GAT1 mutations with residual enzyme activity are expected to cause a milder Limb Girdle Muscular Dystrophy ( LGMD ) phenotype , but patients with such mutations have not yet been described . B4gat1 ( B3gnt1 ) -null mutations in mice result in early embryonic lethality , at ∼ E9 . 5 ( Wright et al . , 2012 ) , as is the case for reported null mutations in Dag1 ( Williamson et al . , 1997 ) , Pomt1 ( Willer et al . , 2004 ) , and Fukutin ( Kurahashi et al . , 2005 ) . Proper α-DG glycosylation is essential for early embryonic development in the mouse , including formation of the basement membrane , as defects in the Reichert's membrane are the suspected cause of death in α-DG glycosylation-deficient mice ( Williamson et al . , 1997; Willer et al . , 2004; Kurahashi et al . , 2005 ) . However , an ENU-based genetic screen for abnormal CNS axonal tracks identified a viable B4gat1 ( B3gnt1 ) dystroglycanopathy mouse model carrying a p . M155T B4gat1 mutation ( Wright et al . , 2012 ) . The majority of compound heterozygous mice with both a LacZ ( B4gat1LacZ ) null allele and a hypomorphic p . M155T ( B4gat1M155T ) allele die perinatally , but a few survive and develop a characteristic CMD phenotype . In this study we used MEFs isolated from the B4gat1LacZ/M155T mice and measured endogenous B4GAT1 activity . As expected the B4gat1-deficient MEFs were hypomorphic , producing low-level residual B4GAT1 activity ( <3% relative to levels in wild-type control ) ( Figure 4B ) . This corroborates that our B4GAT1 assay can be a valuable diagnostic tool for measuring endogenous activity in patient cells and tissues . The residual B4GAT1 enzyme activity in the B4gat1-deficient MEFs was also reflected when the α-DG glycosylation status was analyzed biochemically , by immunoblotting . Although B4GAT1 endogenous activity was very low , it was sufficient to synthesize low amounts of functionally active α-DG that was capable of binding the ligand laminin ( Figure 4A ) . This finding accounts for the difference between the early embryonic lethal phenotype in B4gat1 null ( B4gat1LacZ/LacZ ) mice and the slightly milder phenotype in B4gat1 hypomorphic ( B4gat1LacZ/M155T ) mice ( Wright et al . , 2012 ) . It is worth noting that α-DG glycosylation is highly tissue specific as well as highly dependent on the developmental stage of the cells/tissue ( Barresi and Campbell , 2006 ) . To date , it is not fully understood what causes the tissue-specific differences in α-DG processing , which are reflected as differences in its molecular weight and its ability to bind laminin ( Goddeeris et al . , 2013 ) . LARGE , the key contributor to assembly of the terminal laminin-binding glycan , and B4GAT1 as the upstream priming enzyme are broadly expressed at the RNA level ( Figure 8—figure supplement 1 ) . Although both genes are similarly expressed in most tissues they are strikingly different in heart with LARGE expression being high and B4GAT1 being low . Based on published case reports it does not appear that LARGE patients ( Longman et al . , 2003; Clarke et al . , 2011; Meilleur et al . , 2014 ) are more prone to cardiac defects than other dystroglycanopathy patients . Also lower B4GAT1 expression in the heart does not present a significant bottleneck for α-DG functional glycosylation as heart α-DG has full ligand binding ability ( Goddeeris et al . , 2013 ) . Therefore , the functional consequences of such uncoordinated expression of B4GAT1 and LARGE are currently unknown . It is more likely that other gene products involved in α-DG functional glycosylation can become limiting factors and that the integration of all involved players account for the tissue-specific differences of this complex and highly controlled synthesis pathway . Furthermore , whether α-DG that is not properly glycosylated possesses an as yet unidentified ligand binding activity remains unclear . LARGE has been shown to be highly tunable in the context of cancer , T-cell development and muscle regeneration ( de Bernabe et al . , 2009; Liou et al . , 2010; Goddeeris et al . , 2013 ) . Repression of LARGE expression is responsible for the defects in DG-mediated cell adhesion that are observed in epithelium-derived cancer cells , and point to a defect of its glycosylation as a factor in cancer progression ( de Bernabe et al . , 2009 ) . Similarly , it was demonstrated that expression of B4GAT1 ( B3GNT1 ) is absent in a IIH6-negative subpopulation ( PC3-L ) of an otherwise IIH6-positive human prostate cancer cell line ( PC3 ) . The loss of B4GAT1 expression and laminin-binding by α-DG in these cells was inversely correlated with the observed malignancy and tumor progression of the prostate cancer when these cells were transplanted into SCID mice ( Bao et al . , 2009 ) . In general , these results emphasize that proper α-DG glycosylation plays a critical role in tumor suppression . Previous data suggested that B4GAT1 ( B3GNT1 ) may be an integral component of various enzyme complexes , working with various glycosyltransferases that are functionally associated and involved in the same biosynthetic pathway . For example , it might work with B4GALT1 ( Lee et al . , 2009 ) in the synthesis of poly-N-acetyllactosamine and with LARGE ( Bao et al . , 2009 ) in synthesis of the α-DG laminin-binding glycan . It was also hypothesized that B4GAT1 may regulate LARGE , as B4GAT1 overexpression promoted formation of the LARGE-generated laminin-binding glycan ( Bao et al . , 2009 ) . However , in light of data presented in this study , in particular the finding that endogenous LARGE activity is not affected in B4gat1-deficient cells and vice versa , it seems more likely that B4GAT1 and LARGE have independent enzyme activities ( Figure 4B/C ) . In an effort to provide additional direct evidence and further corroboration of our conclusion that a xylose present in the α-DG O-mannosyl post-phosphoryl glycan linker serves as endogenous acceptor for B4GAT1 , we performed radioactive metabolic cell labeling with [3H]-xylose . The goal was to show radioactive labeling of DGFc340 expressed in B4gat1-deficient MEFs with [3H]-xylose , which could subsequently be released by β-xylosidase treatment . However , this type of metabolic cell labeling proved to be technically challenging since only an insignificant amount ( ∼0 . 01% ) of the total [3H]-xylose radioactivity was incorporated into the secreted DGFc340 fusion protein even after 4 day long-term labeling ( data not shown ) . It is known that xylose uptake from the media into cells is poor ( Snider et al . , 2002 ) , which in our case becomes the limiting factor and made this experimental approach not feasible . Nevertheless , we feel confident that the sum of our indirect data including B4GAT1 Xyl-β-MU acceptor specificity ( Figure 3C ) , pgsI-208 DGFc340 LARGE acceptor test ( Figure 2D ) , β-xylosidase B4GAT1 acceptor treatment ( Figure 6B ) and finally the in vitro synthesis of a GlcA-β1 , 3-Xyl-α1 , 3-GlcA-β1 , 4-Xyl-β-MU tetrasaccharide by the sequential action of B4GAT1 and LARGE ( Figure 7 ) provide strong and convincing evidence that β-xylose is indeed the endogenous acceptor for B4GAT1 . In conclusion , our study has identified B4GAT1 as a xylose β1 , 4-glucuronyltransferase , and revealed that it contributes to the O-mannosyl post-phosphoryl glycan linker of α-DG by synthesizing a glucuronyl-xylosyl disaccharide . This is the crucial acceptor primer that is targeted by the glycosyltransferase LARGE to initiate formation of a heteropolysaccharide on α-DG that is involved in its binding to ligands . As B4GAT1-deficiency was linked to laminin-binding defects of α-DG in a variety of contexts , our new findings will shed light on the mechanism underlying α-DG glycosylation-deficient CMDs ( Buysse et al . , 2013; Shaheen et al . , 2013 ) and tumors ( Bao et al . , 2009 ) , and is expected to also open new therapeutic avenues for blocking the entry of pathogenic arenaviruses , including the hemorrhagic LASV into human cells ( Jae et al . , 2013 ) .
All tissues and patient cells were obtained and tested according to the guidelines set out by the Human Subjects Institutional Review Board of the University of Iowa; informed consent was obtained from all subjects or their legal guardians ( See Table 1 ) . Cells were maintained at 37°C and 5% CO2 in Dulbecco's modified Eagle's medium ( DMEM ) plus fetal bovine serum ( FBS: 10% in the case of HEK293T cells , 20% in the case of fibroblasts from patient skin ) and 2 mM glutamine , 0 . 5% penicillin-streptomycin ( Invitrogen , Carlsbad , CA ) . Pro5 ( wild-type ) and the glycosylation-deficient CHO ( Lec cells ) mutant cell lines termed Lec2 and Lec8 were purchased from ATCC ( Patnaik and Stanley , 2006 ) . The Lec15 . 2 ( Maeda et al . , 1998 ) and ldlD ( Kingsley et al . , 1986 ) cell lines were kindly provided by Monty Krieger , the Lec13 ( Ohyama et al . , 1998 ) cells by Pamela Stanley and the pgsI-208 ( Bakker et al . , 2009 ) cells by Jeff Esko . These CHO cells were grown and maintained in F12 nutrition mix medium with 10% fetal bovine serum ( Invitrogen ) at 37°C and 5% CO2 . MEFs were generated from E13 . 5 embryos ( Table 1 ) as previously described ( Xu et al . , 2005 ) and were maintained in DMEM supplemented with 10% FBS , 2 mM glutamine , and 0 . 5% penicillin-streptomycin at 37°C in 5% CO2 . Phosphorylation of α-DG in glycosylation-deficient fibroblasts was determined based on the incorporation of [32P] into a secreted Fc-tagged α-DG recombinant protein , as described elsewhere ( Yoshida-Moriguchi et al . , 2010 ) . E1-deficient recombinant adenoviruses ( Ad5CMV-DGFc340 , Ad5CMV-DGFc340mut ( T317A/T319A ) and Ad5CMV-LARGE/RSVeGFP ) were generated by the University of Iowa Gene Transfer Vector Core and have been described previously ( Barresi et al . , 2004 ) . The constructs used to generate the E1-deficient recombinant adenoviruses Ad5CMV-DGFc340 and Ad5CMV-DGFc340mut ( T317A/T319A ) were made from pcDNA3-DGFc340 and DGFc340mut ( T317A/T319A ) ( Hara et al . , 2011 ) . pcDNA3-DGFc340 vectors were digested with KpnI/XbaI , and the resulting fragments were ligated into a KpnI/XbaI-digested pacAd5-CMV-KNpA vector . Similarly , Ad5CMV-B4GAT1-V5/RSVeGFP was generated by PCR amplifying a 1 . 3 kb C-terminal V5-tagged open reading frame fragment corresponding to mouse B4gat1 ( B3gnt1 , NM_175383 ) and cloning it into the XhoI/NotI polylinker region of pAd5CMVK-NpA . The following primers were used to amplify , by PCR , B4gat1-V5: 6823 forward ( 5′-agactcgagaccATGcaaatgtcctacgccat-3′ , XhoI adapter is bolded , start ATG is shown in capital letters ) and 6822 reverse ( 5′-tatgcggccgcCTACGTAGAATCGAGACCGAGGAGAGGGTTAGGGATAGGCTTACCgcatcggtggggagagttgg-3′; the NotI adapter is bolded and the V5-tag is shown in capital letters ) . Cultured cells were infected with viral vector for 12 hr , at an MOI of 400 . We examined cultures 3–5 days after treatment . We used nucleofection as nonviral method for transferring genes into MEF cells . The Human Dermal Fibroblast Nucleofector kit was used according to an optimized protocol provided by the manufacturer ( Amaxa Biosystems , Germany ) . WGA-enriched glycoproteins from frozen samples and cultured cells were processed as previously described ( Michele et al . , 2002 ) . Immunoblotting was carried out on polyvinylidene difluoride ( PVDF ) membranes as previously described ( Michele et al . , 2002 ) . Blots were developed with IR-conjugated secondary antibodies ( Pierce Biotechnology , Rockford , IL ) and scanned with an Odyssey infrared imaging system ( LI-COR Bioscience , Lincoln , NE ) . Laminin overlay assays were performed as previously described ( Michele et al . , 2002 ) . The monoclonal antibodies to the fully glycosylated form of α-DG ( IIH6 ) ( Ervasti and Campbell , 1991 ) , and also the polyclonal antibodies rabbit β-dystroglycan ( AP83 ) ( Duclos et al . , 1998 ) and anti-LARGE ( Rb331 ) ( Kanagawa et al . , 2004 ) were characterized previously . G6317 ( core-αDG ) from rabbit antiserum was raised against a keyhole limpet hemocyanin ( KLH ) -conjugated synthetic peptide of human dystroglycan ( Willer et al . , 2012 ) . Mouse monoclonal anti-Myc ( clone 4A6 ) antibodies were purchased from Millipore ( Billerica , MA ) , mouse monoclonal anti-β-Actin ( Clone AC-74 ) antibodies were purchased from Sigma ( St . Louis , MO ) and mouse monoclonal anti-V5 antibodies were purchased from Invitrogen . HEK293T cells expressing Myc-tagged glycosyltranferases were fixed with 4% paraformaldehyde in PBS , and then permeabilized with 0 . 2% Triton X-100 in PBS for 10 min on ice . After blocking with 3% BSA in PBS , the slides were incubated with anti-c-Myc antibody ( 4A6 , Millipore ) anti-Giantin ( abcam , United Kingdom ) or anti-ERp72 antibody ( Calbiochem , San Diego , CA ) for 18 hr at 4°C . The cells were incubated with an appropriate secondary antibody conjugated to Alexa488 or Alexa555 fluorophore after washing with PBS . 4′ , 6′-Diamidino-2-phenylindole dihydrochloride ( DAPI , Sigma ) was used for nuclear staining . Images were observed using a Zeiss Axioimager M1 fluorescence microscope ( Carl Zeiss , Thornwood , NY ) . The On-Cell complementation assay was performed as described previously ( Willer et al . , 2012 ) . In brief , 2 × 105 cells were seeded into a 48-well dish . The next day the cells were infected with 200 MOI of Ad5CMV-LARGE1/eGFP in growth medium . Three days later , the cells were washed in TBS and fixed with 4% paraformaldehyde in TBS for 10 min . After blocking with 3% dry milk in TBS +0 . 1% Tween ( TBS-T ) , the cells were incubated with primary antibody ( glyco α-DG , IIH6 ) in blocking buffer overnight . To develop the On-Cell Western blots we conjugated goat anti-mouse IgM ( Millipore ) with IR800CW dye ( LI-COR ) , subjected the sample to gel filtration , and isolated the labeled antibody fraction . After staining with IR800CW secondary antibody in blocking buffer , we washed the cells in TBS and scanned the 48-well plate using an Odyssey infrared imaging system ( LI-COR ) . For cell normalization , DRAQ5 cell DNA dye ( Biostatus Limited , United Kingdom ) was added to the secondary antibody solution . Open reading frames ( ORF ) were PCR amplified using the following primer sequences: mB4gat1 ( 1 . 3 kb ) , pTW324: forward 5′-aagGGATCCaccatgcaaatgtcctacgccatccg-3′ ( BamHI restriction site is shown in capital letters and start ATG is bolded ) and reverse 5′-agagcggccgcCTACAAGTCTTCTTCAGAAATAAGTTTTTGTTCGCTAGCcccgcatcggtggggagagttgggg-3′ ( NotI restriction site is bolded , Myc-tag sequence is underlined and NheI restriction site is shown in capital bold letters ) . hFKTN ( 1 . 4 kb ) , pTW322: forward 5′-taaAGATCTaccatgagtagaatcaataagaacgtggttttg-3′ ( BglII restriction site is shown in capital letters and start ATG is bolded ) and reverse 5′-ttcGCTAGCcccatataactggataacctcatcccactc-3′ ( NheI restriction site is shown in capital letters ) . mFkrp ( 1 . 5 kb ) , pTW323: forward 5′-taaGGATCCaccatgcggctcacccgctgctg-3′ ( BamHI restriction site is shown in capital letters and start ATG is bolded ) and reverse 5′-ttcGCTAGCcccaccgcctgtcaagcttaagagtgc-3′ ( NheI restriction site is shown in capital letters ) . mTmem5 ( 1 . 3 kb ) pTW330: forward 5′-taaGGATCCaccatgcggctgacgcggacacg-3′ ( BamHI restriction site is shown in capital letters and start ATG is bolded ) and reverse 5′-ttcGCTAGCcccaactttattattaataaaaaatgaactttc -3′ ( NheI restriction site is shown in capital letters ) . mLarge ( 2 . 3 kb ) , pTW355: forward 5′-taaAGATCTaccatgctgggaatctgcagagggag-3′ ( BglII restriction site is shown in capital letters and start ATG is bolded ) and reverse 5′-ttcGCTAGCcccgctgttgttctcagctgtgagatatttc-3′ ( NheI restriction site is shown in capital letters ) . First a BamHI/NotI digested PCR fragment from mB4gat1 was cloned into the BamHI/NotI multiple cloning site ( MCS ) of a pIRES-puro3-derived vector , in which the NheI site in the MCS was deleted . Subsequently all other genes were digested with either BamHI and NheI or BglII and NheI , and subcloned into a BamHI and NheI-digested mB4gat1-myc pIRES-puro ( pTW324 ) construct . To generate the mouse B4GAT1-Myc Mut1-Mut3 mutant expression constructs , we used the same forward primer A ( 5′-aagGGATCCaccatgcaaatgtcctacgccatccg-3′ ) and a reverse primer D ( 5′- agagcggccgcCTACAAGTCTTCTTCAGAAATAAGTTTTTGTTCGCTAGCcccgcatcggtggggagagttgggg-3′ ) that were used to clone mB4GAT1-Myc ( see mB4gat1-Myc pTW324 cloning ) . Primers A and D bind at the 5′-end and 3′-end of the mB4gat1 coding region . For each mutation we designed overlapping forward ( B1-3 ) and reverse ( C1-3 ) primers that included the respective mutation ( shown in bold capital letters ) : mB4GAT1-Mut1 ( c . 1168A > G , p . N390D ) : B1 5′-ccaaaaggaggctgaaGaccagcgcaataagatc-3′ and C1: 5′-gatcttattgcgctggtCttcagcctccttttgg-3′ . mB4GAT1-Mut2 ( c . 679/685 G > A , p . D227N/D229N ) : B2 5′- ggccaactacgccctggtgattAatgtgAacatggtgcccagcgaagggc-3′ and C2 5′- gcccttcgctgggcaccatgtTcacatTaatcaccagggcgtagttggcc-3′ . mB4GAT1-Mut3 ( c . 464T > C , M155T ) : B3 5′- gcgctagggtcgccaCgcacctcgtgtgcccctc-3′ and C3 5′- gaggggcacacgaggtgcGtggcgaccctagcgc . Using the mB4gat1-Myc ( pTW324 ) expression construct as template , we PCR amplified 5′-fragments , using primer pairs A/B1-3 and 3′-fragments using C1-3/D , respectively . The PCR products were isolated and used as the template DNAs in the second round of amplification with primer pair A-D . The 1 . 3 kb final PCR product was purified and digested with BamHI/NheI and then ligated into pTW324 digested with the same enzymes . The sequence of the insert DNA was confirmed by Sanger sequencing . The construct expressing B4GAT1 without its transmembrane region was generated by amplifying a 1 . 1 kb cDNA fragment of mouse B4gat1 ( B3gnt1 , acc . # NM_175383 ) from mB4GAT1-Myc expression vector pTW324 , using primer pair #8629 ( 5′-ggtGAATTCcacggccaggaggagcagg-3′ ) and #8630 ( 5′- atgACCGGTatgcatattcaagtcttcttcagaaataagtttttgttcgc-3′ ) . EcoRI and AgeI restriction sites included in the primers are indicated in capital letters . The PCR fragment was digested with EcoRI and AgeI and subcloned to generate construct pCMV3xFLAG-TEV-B4GAT1dTM-Myc6xHIS ( pTW351 ) , which expresses a mouse B4GAT1dTM fusion protein ( amino acids 37–415 ) tagged with a N-terminal 3xFLAG and C-terminal Myc6xHis . HEK293T cells were transfected with constructs pTW351 ( B4GAT1dTM ) using FuGENE 6 ( Roche Applied Science , Indianapolis , IN ) . The construct contains an IRES-puromycin resistance cassette and stable cell lines were selected in medium containing Puromycin ( 1 µg/ml , InvivoGen , San Diego , CA ) . Expression and secretion of B4GAT1dTM protein into the culture medium was confirmed by immunoblotting with anti-Myc antibody 4A6 ( Millipore ) . The stable cell lines obtained in this way were adapted to serum-free medium 293SFMII ( Invitrogen ) and cultivated in CELLine bioreactors ( CL1000 , Argos Technologies , Elgin , IL ) . B4GAT1dTM and LARGEdTM secreted into the culture medium by HEK293T cells were purified using the Talon metal-affinity resin ( Clontech , Mountain View , CA ) according to the manufacturer's instructions . The purity of the protein was confirmed by SDS-PAGE and Coomassie Brilliant blue ( CBB ) staining ( Figure 3—figure supplement 1B ) . For the enzyme assay , the eluate was desalted and concentrated using an Amicon Ultra centrifugal filter unit ( Millipore ) . To generate the DGFc340 and DGFc340-mut acceptor proteins we infected control and glycosylation-deficient MEFs and CHO-derived cell lines with Ad5-CMV DGFc340 adenoviral vectors at an MOI of 400 . At 4 days post-infection the secreted proteins were isolated from the culture medium using Protein A-agarose beads ( Santa Cruz , Dallas , TX ) . DGFc340 bound Protein A-agarose beads were washed three times with TBS and Protein A slurry prebound with ∼25 µg DGFc340 was added to the in vitro LARGEdTM assay . Enzyme reactions ( 50 µl ) were carried out at 37°C , with 5 mM UDP-GlcA and 5 mM UDP-Xyl , in 0 . 1 M MES ( 2- ( N-morpholino ) ethanesulfonic acid ) buffer ( pH 6 . 0 ) supplemented with 10 mM MnCl2 , 10 mM MgCl2 , 0 . 2% Triton X-100 and 5 µg purified LARGEdTM protein . The reaction was terminated by adding 5× LSB and boiling for 5 min , The samples were subsequently analyzed by immunoblotting . DGFc340 ( ∼25 µg ) and DGFc340-mut ( ∼25 µg ) bound Protein A-agarose beads were washed with TBS and used in the in vitro LARGEdTM assay . 30 µl enzyme reactions were carried out at 37°C for 20 hr , with 0 . 05 µCi UDP-GlcA [GlcA-14C] ( final conc . 5 . 5 µM ) and 0 . 05 µCi UDP-Xyl [Xyl-14C] ( final conc . 6 . 6 µM ) , in 0 . 1 M MES buffer ( pH 6 . 0 ) supplemented with 10 mM MnCl2 , 10 mM MgCl2 , 0 . 2% Triton X-100 and 5 µg purified LARGEdTM protein . The reaction was terminated by adding 25 µl of 0 . 1 M EDTA . After three washes with TBS the Protein A-agarose-bound DGFc340 samples were analyzed by scintillation counting . The reactions for B4GAT1dTM activity were carried out similarly . Again , 30 µl enzyme reactions were carried out at 37°C for 20 hr and with 0 . 05 µCi UDP-GlcA [GlcA-14C] ( final conc . 5 . 5 µM ) , but in this case 0 . 1 M MOPS ( 3- ( N-morpholino ) propanesulfonic acid ) buffer ( pH 7 . 0 ) supplemented with 10 mM MnCl2 , 10 mM MgCl2 , 0 . 2% Triton X-100 was used , with 0 . 25 µg purified B4GAT1dTM protein . [14C] labeled sugar nucleotides were purchased from ARC ( American Radiolabeled Chemicals , St . Louis , MO ) . Recombinant β-glucuronidase from E . coli was purchased from Sigma ( G8295 ) . Each digest was performed in a 100 µl volume at 37°C for 12 hr in 50 mM NaPO4 , pH 7 . 0 , 5 mM DTT , 1 mM EDTA , 0 . 1% Triton X-100 in the presence of 10 µg ( 100 units ) β-glucuronidase . Recombinant β-xylosidase from E . coli was purchased from Sigma ( X3504 ) . Each digest was performed in a 100 µl volume at 70°C for 60 min in 50 mM sodium acetate at pH 5 . 8 in the presence of 20 µg β-xylosidase . The HPLC-based enzyme assays for B4GAT1-Myc ( 100 µg cell lysates ) and B4GAT1dTM ( 0 . 25 µg purified protein ) were performed using Xyl-β-MU ( 0 . 1 mM ) ( Sigma ) as the acceptor . The samples were incubated for 2 hr for analytical purposes and 24 hr for preparative purposes . 50 µl enzyme reactions were carried out at 37°C , with 5 mM UDP-GlcA , in 0 . 1 M MOPS buffer ( pH 7 . 0 ) supplemented with 10 mM MnCl2 , 10 mM MgCl2 , and 0 . 2% Triton X-100 . The reaction was terminated by adding 25 µl of 0 . 1 M EDTA and boiling for 5 min . The supernatant was analyzed using a LC18 column ( 4 . 6 × 250 mm Supelcosil LC-18 column ( Supelco , Bellefonte , PA ) ) with Buffer A ( 50 mM ammonium formate pH 4 . 0 ) and Buffer B ( 80% acetonitrile in buffer A ) , using a 12% B isocratic run at 1 ml/min using Beckman Gold system ( Beckman Coulter , Inc . , Brea , CA ) . The elution of MU derivatives was monitored by fluorescence detection ( 325 nm for excitation , and 380 nm for emission ) . For the assessment of metal dependence , all ions were used at a concentration of 10 mM in 0 . 1 M MOPS pH 7 . 0 . To test pH-dependent activity testing buffers ranging from pH 4 . 5–8 . 5 were used: 0 . 1 M sodium acetate ( pH 4 . 5–5 . 5 ) , 0 . 1 M MES ( pH 5 . 5–6 . 5 ) , 0 . 1 M MOPS ( pH 6 . 5–7 . 5 ) and 0 . 1 M Tris ( pH 7 . 5–8 . 5 ) . To assess endogenous B4GAT1 GlcA-T activity in MEFs , we solubilized the cells in TBS 1% TX-100 . 100 µg total protein from crude lysates were added to each assay . 50 µl enzyme reactions were carried out for 18hr at 37°C , with 5 mM UDP-GlcA , in 0 . 1 M MOPS buffer ( pH 7 . 0 ) supplemented with 10 mM MnCl2 , 10 mM MgCl2 , and 0 . 2% Triton X-100 . For analysis of substrate specificity Xyl-α-MU ( Sigma ) , Xyl-β-MU ( Sigma ) and Xyl-α1 , 3-GlcA-β-MU were added to the standard enzyme reaction at a concentration of 0 . 1 mM . The HPLC-based enzymatic assay for LARGEdTM ( 5 µg purified protein ) and endogenous LARGE was performed using GlcA-β-MU , GlcA-β1 , 3−Xyl-α-MU and GlcA-β1 , 4−Xyl-β−MU as the acceptor for Xyl-T activity and Xyl-α1 , 3-GlcA-β-MU for GlcA-T activity as described previously ( Inamori et al . , 2012 , 2013 , 2014 ) . For the assessment of endogenous LARGE GlcA-T activity in MEF cells , we solubilized the cells in TBS 1% TX-100 and enriched glycoproteins from crude lysates ( 2 mg total protein ) using WGA-agarose . N-Acetylglucosamine-eluted glycoproteins from WGA-bound glycoproteins were incubated in a volume of 50 μl for 18 hr at 37°C , with 0 . 1 mM MU-acceptor , 5 mM UDP-GlcA in 0 . 1 M MES buffer pH 6 . 0 , 10 mM MnCl2 , 10 mM MgCl2 and 0 . 2% Triton X-100 . The reaction was terminated by adding 25 μl of 0 . 1 M EDTA and boiling for 5 min , and the supernatant was analyzed with an LC-18 column using a 12% B isocratic run . The test B4GAT1dTM for GlcNAc transferase activity Gal-β1 , 4-GlcNAc-β-MU ( 0 . 1 mM ) was used as acceptor . The 50 µl enzyme reactions were carried out as described previously ( Sasaki et al . , 1997 ) at 37°C , with 5 mM UDP-GlcNAc in 0 . 1 M cocodylate buffer ( pH 7 . 0 ) supplemented with 20 mM MnCl2 , 5 mM ATP and 0 . 25 µg B4GAT1dTM enzyme . The reaction was terminated by adding 25 µl of 0 . 1 M EDTA and boiling for 5 min . The supernatant was analyzed using a LC18 column ( 4 . 6 × 250 mm Supelcosil LC-18 column ( Supelco ) ) with Buffer A ( 50 mM ammonium formate pH 4 . 0 ) and Buffer B ( 80% acetonitrile in buffer A ) , using a 16% B isocratic run at 1 ml/min using Beckman Gold system . A large scale reaction was carried out using B4GAT1dTM purified using a metal-affinity resin as described previously for LARGEdTM ( Inamori et al . , 2012 ) . B4GAT1dTM was added to 10 mM of UDP-GlcA and Xylose-β−MU in 50 mM MOPS buffer pH 7 . 0 , 10 mM MgCl2 , 10 mM MnCl2 and 0 . 5% TX-100 and incubated for 48 hr at 37°C with rotation . The sample was then run over a C18 column ( 4 . 6 × 250 mm Supelcosil LC-18 column ( Supelco ) ) with Buffer A ( 50 mM ammonium formate pH 4 . 0 ) and Buffer B ( 80% acetonitrile in buffer A ) using a 16% B isocratic run at 1 ml/min on a Beckman Gold system . The elution of MU derivatives was monitored by fluorescence detection ( 325 nm for excitation , and 380 nm for emission ) . The product in the peak fractions was collected and lyophilized . The dried sample was then brought up in Milli-Q water ( 500 µl ) and lyophilized and this procedure was repeated three times , after which the sample was brought up in Milli-Q water . The product was quantitated based on the standard curve of GlcA-β-MU . This sample was used for NMR studies . The GlcA-β1 , 4-xylose-β-MU disaccharide ( B4GAT1 product ) was added to 10 mM of UDP-Xyl in 50 mM sodium acetate buffer at pH 5 . 5 and with 10 mM MgCl2 , 10 mM MnCl2 , 0 . 5% TX-100 and LARGEdTM attached to metal-affinity resin and incubated for 48 hr at 37°C with rotation . The sample was then run over a LC18 column ( 4 . 6 × 250 mm Supelcosil LC-18 column ( Supelco ) ) with Buffer A ( 50 mM ammonium formate pH 4 . 0 ) and Buffer B ( 80% acetonitrile in buffer A ) using a 16% B isocratic run at 1 ml/min on a Beckman Gold system . The elution of MU derivatives was monitored by fluorescence detection ( 325 nm for excitation , and 380 nm for emission ) . The trisaccharide peak was collected and lyophilized . The lyophilized sample was then brought up in 10 mM UDP-GlcA in 50 mM MOPS buffer pH 6 . 0 , 10 mM MgCl2 , 10 mM MnCl2 and 0 . 5% TX-100 and incubated for 48 hr at 37°C with rotation . It was again run on a C18 column with 16% B isocratic run . The product peak fraction was then collected and lyophilized . The dried sample was brought up in Milli-Q water ( 500 µl ) and lyophilized . This procedure was repeated a total of three times . The last time the sample was brought up in Milli-Q water and the product was quantitated using a standard curve of GlcA-β-MU . This sample was used for NMR studies . A large scale reaction was carried out using recombinant human B4GALT1 ( purchased from R&D Systems cat# 3609-GT , Minneapolis , MN ) . B4GALT1 ( 1 . 5 µg ) was added to 5 mM of UDP-Gal and 3 mM GlcNAc-β−MU in 50 mM Tris buffer pH 7 . 5 , 10 mM MgCl2 and 150 mM NaCl and incubated for 48 hr at 37°C with rotation . The sample was then run over a C18 column ( 4 . 6 × 250 mm Supelcosil LC-18 column ( Supelco ) ) with Buffer A ( 50 mM ammonium formate pH 4 . 0 ) and Buffer B ( 80% acetonitrile in buffer A ) using a 16% B isocratic run at 1 ml/min on a Beckman Gold system . The elution of MU derivatives was monitored by fluorescence detection ( 325 nm for excitation , and 380 nm for emission ) . Over time in the above reaction a peak was seen that ran about 1 . 5 min after the GlcNAc-β−MU peak at 21 . 5 min . This product peak was collected was and lyophilized . The dried sample was then brought up in Milli-Q water ( 500 µl ) and lyophilized and this procedure was repeated three times , after which the sample was brought up in Milli-Q water . The product was quantitated based on the standard curve of GlcA-β-MU . This sample was used for NMR studies . Samples were prepared for NMR by fractionation ( using gel filtration and/or LC-18 chromatography ) as described above , followed by the exchange of hydroxyl hydrogens by lyophilization and dissolution in 10 mM sodium phosphate buffer pH 6 . 5 , in 100% D2O . 1H homonuclear two-dimensional DQF-COSY ( Rance et al . , 1983 ) , TOCSY ( Braunschweiler and Ernst , 1983 ) , and ROESY ( Davis and Bax , 1985 ) experiments , and 1H/13C heteronuclear two-dimensional HMQC , HMBC , and H2BC experiments ( Nyberg et al . , 2005 ) were collected using a Bruker Avance II 800 MHz NMR spectrometer equipped with a sensitive cryoprobe . All NMR spectra were recorded at 25°C . The 1H chemical shifts were referenced to 2 , 2-dimethyl- 2-silapentane-5-sulfonate ( DSS ) . NMR spectra were processed using the NMRPipe software package ( Delaglio et al . , 1995 ) and analyzed using NMRView software ( Johnson and Blevins , 1994 ) . | Dystroglycan is a protein that is critical for the proper function of many tissues , especially muscles and brain . Dystroglycan helps to connect the structural network inside the cell with the matrix outside of the cell . The extracellular matrix fills the space between the cells to serve as a scaffold and hold cells together within a tissue . It is well established that the interaction of cells with their extracellular environments is important for structuring tissues , as well as for helping cells to specialize and migrate . These interactions also play a role in the progression of cancer . As is the case for many proteins , dystroglycan must be modified with particular sugar molecules in order to work correctly . Enzymes called glycosyltransferases are responsible for sequentially assembling a complex array of sugar molecules on dystroglycan . This modification is essential for making dystroglycan ‘sticky’ , so it can bind to the components of the extracellular matrix . If sugar molecules are added incorrectly , dystroglycan loses its ability to bind to these components . This causes congenital muscular dystrophies , a group of diseases that are characterized by a progressive loss of muscle function . Willer et al . use a wide range of experimental techniques to investigate the types of sugar molecules added to dystroglycan , the overall structure of the resulting ‘sticky’ complex and the mechanism whereby it is built . This reveals that a glycosyltransferase known as B3GNT1 is one of the enzymes responsible for adding a sugar molecule to the complex . This enzyme was first described in the literature over a decade ago , and the name B3GNT1 was assigned , according to a code , to reflect the sugar molecule it was thought to transfer to proteins . However , Willer et al . ( and independently , Praissman et al . ) find that this enzyme actually attaches a different sugar modification to dystroglycan , and so should therefore be called B4GAT1 instead . Willer et al . find that the sugar molecule added by the B4GAT1 enzyme acts as a platform for the assembly of a much larger sugar polymer that cells use to anchor themselves within a tissue . Some viruses–including Lassa virus , which causes severe fever and bleeding–also use the ‘sticky’ sugar modification of dystroglycan to bind to and invade cells , causing disease in humans . Understanding the structure of this complex , and how these sugar modifications are added to dystroglycan , could therefore help to develop treatments for a wide range of diseases like progressive muscle weakening and viral infections . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2014 | The glucuronyltransferase B4GAT1 is required for initiation of LARGE-mediated α-dystroglycan functional glycosylation |
Human performance at categorizing natural visual images surpasses automatic algorithms , but how and when this function arises and develops remain unanswered . We recorded scalp electrical brain activity in 4–6 months infants viewing images of objects in their natural background at a rapid rate of 6 images/second ( 6 Hz ) . Widely variable face images appearing every 5 stimuli generate an electrophysiological response over the right hemisphere exactly at 1 . 2 Hz ( 6 Hz/5 ) . This face-selective response is absent for phase-scrambled images and therefore not due to low-level information . These findings indicate that right lateralized face-selective processes emerge well before reading acquisition in the infant brain , which can perform figure-ground segregation and generalize face-selective responses across changes in size , viewpoint , illumination as well as expression , age and gender . These observations made with a highly sensitive and objective approach open an avenue for clarifying the developmental course of natural image categorization in the human brain .
A fundamental function of the human brain is to organize sensory events into distinct classes , that is , perceptual categorization ( Rosch , 2007 ) . This function is well illustrated in vision , the dominant sensory modality in humans: visual categorization in natural scenes occurs extremely rapidly ( Thorpe et al . , 1996 ) and in the near absence of attention ( Li et al . , 2002 ) . Yet , visual categorization is extremely challenging . For instance , categorizing a visual stimulus as a face—arguably the most significant visual stimulus for human social ecology—requires to isolate the face from its natural background scene ( ‘figure-ground segregation’ , Appelbaum et al . , 2006; Peterson , 2014 ) and distinguish the face from the wide range of non-face stimuli in the environment which share visual properties with faces . Moreover , a common response ( i . e . , generalization ) should be given to faces appearing under various viewing conditions ( i . e . , changes of head orientation , size , illumination , etc ) and varying greatly in terms of gender , age , expression , ethnic origin , so on . Despite this challenge , human performance at face categorization is impressive ( Crouzet et al . , 2010 ) , surpassing even the most sophisticated automatic systems ( Scheirer et al . , 2014 ) . Up to now , the ontogeny of face categorization remains largely unknown . Classical studies have reported preference for facelike over non-facelike patterns at birth ( Goren et al . , 1975; Johnson and Morton , 1991 ) . At a few months of age , differences in event-related potentials ( ERPs ) have been found between face stimuli and meaningless patterns ( Halit et al . , 2004; Kouider et al . , 2013 ) as well as between faces and exemplars of a single object category segmented from its natural background ( e . g . , toys , de Haan and Nelson , 1999; cars , Peykarjou and Hoehl , 2013; houses or cars , Gliga and Dehaene , 2007 ) . However , there is no evidence on the effectiveness of infant vision in segmenting faces in natural images and representing them as a distinct , generalized category , or on the developing neural systems that may achieve this process . Clarifying this issue is also important for understanding the origin of hemispheric lateralization for face-selective processes in the human brain . In human adults , areas of the ventral and lateral occipito-temporal cortex are more active when viewing faces vs a variety of non-face objects ( Sergent et al . , 1992; Puce et al . , 1995; Kanwisher et al . , 1997; Haxby et al . , 2000; Rossion et al . , 2012; Weiner and Grill-Spector , 2013 ) . This face-selective activation is typically larger in the right than the left hemisphere and , in right handed individuals at least , right unilateral brain lesions can lead to selective impairment in face recognition ( prosopagnosia: e . g . , Barton et al . , 2002; Busigny et al . , 2010; Hecaen and Angelergues , 1962; Sergent and Signoret , 1992 ) . According to a recent hypothesis , this right hemispheric dominance for face perception , which seems specific to humans ( e . g . , Tsao et al . , 2008 ) , is driven by the left hemispheric lateralization for words emerging during reading acquisition ( Dundas et al . , 2013 ) . Thus , according to this view , right hemispheric lateralization for faces should not be present in infancy . Up to now , infant ERP studies have not been able to provide evidence for right hemispheric lateralization of face-selective processes ( de Haan and Nelson , 1999; Gliga et al . , 2007; Peykarjou and Hoehl , 2013 ) and right hemispheric lateralization has only been observed when comparing faces to meaningless stimuli that differ in terms of low-level visual cues ( Tzourio-Mazoyer et al . , 2002; Kouider et al . , 2013 ) . We addressed these outstanding issues by means of a simple ‘frequency tagging’ or ‘fast periodic visual stimulation’ ( FPVS ) approach , providing robust electroencephalographic ( EEG ) responses—steady state visual evoked potentials ( SSVEPs , Regan , 1989; for a review see Norcia et al . , 2015 ) —over the left and right hemispheres of 4- to 6-month-old infants . This approach is ideal to study the infant brain because it is relatively immune from artifacts and provides high signal-to-noise ratio ( SNR ) responses in a few minutes only . Moreover , compared to other approaches such as ERPs to transient stimulation , the FPVS approach is objective and predictive because the response appears exactly at the periodic frequency of stimulation defined by the experimenter . So far , infants have been tested with this approach only in response to low-level visual stimuli ( i . e . , acuity , contrast sensitivity , spatial phase , orientation , or motion; e . g . , Braddick et al . , 1986; Norcia et al . , 1990 ) . A recent EEG study tested infants with segmented faces and objects in different stimulation streams , but without testing face vs object discrimination or generalization across diverse face views , and without providing evidence of hemispheric lateralization ( Farzin et al . , 2012 ) . Here , to achieve these goals , we isolated face-selective responses by means of a fast periodic oddball paradigm ( Heinrich et al . , 2009 ) recently adapted to characterize adults' individual face discrimination ( Liu-Shuang et al . , 2014 ) and face-selective responses in adults ( Rossion et al . , 2015 ) .
We recorded 32-electrode EEG in a group of 15 4- to 6-month-old infants ( 5 females , mean age = 155 days , range 125–197 days ) looking at complex images of various faces and objects presented one-by-one on a computer screen at a rapid frequency rate of 6 images/s ( i . e . , 6 Hz , stimulus onset asynchrony of 167 ms , Figure 1A; Video 1 ) , in sequences of 20 s . Infants viewed between 5 and 12 sequences ( i . e . , 100 s–240 s; eight sequences on average ) . 10 . 7554/eLife . 06564 . 003Figure 1 . ( A ) Examples of face ( F ) and object ( O ) stimuli presented during a 20-s sequence at 6 Hz ( i . e . , 120 images ) . Face stimuli , varying considerably in size , viewpoint , expression , gender , so on appeared as every fifth image , that is , at 1 . 2 Hz rate ( =6 Hz/5 ) . For copyright reasons , the face pictures displayed in the figure are different than those used in the actual experiment , but the degree of variability across images is respected . The full set of face pictures is available at http://face-categorization-lab . webnode . com/publications/ together with the paper reporting the original study performed in adults ( Rossion et al . , 2015 ) . ( B ) Stimuli were presented in the center of the screen by means of sinusoidal contrast modulation at a rate of 6 Hz ( i . e . , 6 images/s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 00310 . 7554/eLife . 06564 . 004Video 1 . 8 s excerpt of experiment 1 ( 20 s sequences ) showing faces at a rate of 1 image every 5 images , at a 6 Hz base rate . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 004 Thanks to the high temporal resolution of EEG and the high frequency resolution provided by the analysis ( 1/20 s = 0 . 05 Hz ) , responses occurring exactly at the fast 6 Hz rate were identified in the SNR spectrum , obtained by dividing each frequency bin by the 20 neighboring bins ( Rossion et al . , 2012; see ‘Materials and methods’ ) . On grand-averaged data , this high SNR response at 6 Hz ( averaged SNR = 8 . 87 at channel Oz ) focused over the medial occipital cortex , reflecting infants' visual system synchronization to the stimulus presentation rate ( Figure 2A ) . On these grand-averaged data , a Z-Score computed as the difference between amplitude at the frequency of interest and the mean amplitude of 20 surrounding bins divided by the standard deviation of the 20 surrounding bins ( e . g . , Rossion et al . , 2012; Liu-Shuang et al . , 2014; see also ‘Materials and methods’ ) was highly significant at electrode Oz ( Z = 52 . 9 , p < 0 . 00001 ) . To ensure that this effect was not driven by the data of a few infants , a t-test against 1 ( i . e . , signal above noise level ) was also performed using the individual SNR values at Oz ( range: 0 . 19–17 . 07; Figure 2B ) . This response was highly significant ( t ( 14 ) = 7 . 075 , p < 0 . 0001 ) . Moreover , the high frequency resolution of the approach provides many frequency bins to estimate the noise so that the Z-score procedure could be applied to each individual infant's data . At electrode Oz , a significant response was observed in every infant tested but one ( Z-score range of 14 infants: 6 . 10–35 . 46; not significant for 1 infant only ) . This observation indicates that the infant brain synchronizes strongly to the rapid 6 Hz visual presentations of multiple object categories . 10 . 7554/eLife . 06564 . 005Figure 2 . ( A ) Grand-averaged EEG signal-to-noise ratio ( SNR ) spectrum at a medial occipital electrode site ( channel Oz ) . The SNR is computed across the whole spectrum as the ratio of the amplitude ( in microvolts ) at each frequency bin and the 20 surrounding frequency bins ( Liu-Shuang et al . , 2014; see ‘Materials and methods’ ) . For EEG amplitude spectra . ( B ) The SNR response at 6 Hz on electrode Oz , showing above noise level ( >1 ) responses for all infants tested but one . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 005 Most interestingly , face stimuli were presented at a slower periodic rate in the stimulation sequence , that is , as every fifth stimulus ( Figure 1A ) . Hence , if the infant's brain discriminates between faces and non-face objects , another response is expected exactly at a rate of 6 Hz/5 = 1 . 2 Hz in the EEG spectrum . On grand-averaged data , a clear 1 . 2 Hz response emerged , with the largest response found over the right occipito-temporal channel P8 ( SNR = 2 . 56; i . e . , 156% signal increase; Figure 3A; Table 1 in Supplementary file 1A ) . This peak at 1 . 2 Hz was well above noise level at P8 ( Z = 12 . 16 , p < 0 . 001 ) even when correcting for multiple comparisons ( all electrode channels , see Table 1 in Supplementary file 1A for SNR and Z-scores at every channel at 1 . 2 Hz ) . Four other electrodes were associated with significant 1 . 2 Hz responses on grand-averaged data ( O1 , F3 , F7 , P7; see Table 1 in Supplementary file 1A ) but with much lower SNR values ( range: 1 . 14–1 . 47 ) . The subsequent analysis based on individual infant's data focused on electrode P8 . 10 . 7554/eLife . 06564 . 006Figure 3 . ( A ) Grand-averaged EEG SNR spectrum at the right hemisphere occipito-temporal channel P8 , showing a distinct peak exactly at the face stimulation frequency ( 1 . 2 Hz ) . ( B ) The SNR response of individual infants at 1 . 2 Hz , on electrode P8 . Color codes are congruent with Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 006 Considering the variance across infants' data for statistical tests , the response at P8 is highly significant ( i . e . , above noise level , or SNR = 1 , t ( 14 ) = 3 . 11 , p = 0 . 004; Figure 3B ) . For 12 infants out of 15 , the signal at 1 . 2 Hz is above noise level ( SNR range of all 15 infants at P8: 0 . 52–6 . 13 , Figure 3B ) . Using the Z-score approach for testing individual infants , the response at P8 is also significant for the individual data of 7 infants out of 15 ( ps < 0 . 05 , Z-score > 1 . 64 for signal vs noise computed over neighboring frequency bins , see ‘Materials and methods’ ) . The other 7 infants showed a significant 1 . 2 Hz face categorization response on at least one other electrode ( p < 0 . 05 ) , while none of the electrodes reached significance for one infant . Even though a 1 . 2 Hz response was also observed over the homologous left occipito-temporal channel P7 ( SNR of grand-averaged data = 1 . 47 , Z = 3 . 61; Table 1 in Supplementary file 1A ) , this response was significantly lower than that at P8 ( t ( 14 ) = 2 . 45 , p = 0 . 013 ) . A significant response at 1 . 2 Hz indicates that the infant brain generates a specific response to faces compared to the other object categories presented in the stimulation sequences ( i . e . , discrimination ) and that such a differential response is generated periodically , that is , for virtually every face presented in the sequence ( i . e . , generalization ) ( Figure 1A ) . Moreover , although the faces and objects are relatively well centered , the color images are embedded in their natural and diverse backgrounds . Hence , to be identified as distinct shapes , both the face and object stimuli have to be segmented from their background , a nontrivial accomplishment for the visual system ( Appelbaum et al . , 2006; Peterson , 2014 ) . Moreover , both the objects and faces substantially vary in size , color , lighting , and viewpoint , and the faces also vary in gender , age , ethnical origin , and expression . Thus , to generate a periodic discriminative 1 . 2 Hz response in the EEG , the infant brain has to categorize the face stimuli , namely to produce a response that is specific to face images and invariant to their differences ( Rossion et al . , 2015 ) . In theory , putative low-level visual cues differing between faces and objects cannot contribute to the periodic response unless they are systematically present in all or the large majority of face stimuli and if they differ systematically between faces and objects but not within non-face object categories . Given the naturalness and variability of the images used , this is highly unlikely . Thus , the constraint of periodicity provides an elegant way to identify a high-level face categorization response while preserving the natural aspect of the stimuli ( Rossion et al . , 2015 ) . Nevertheless , to ensure that low-level visual cues do not contribute to the infant face-selective response , we exposed another group of 10 4–6 months infants ( 4 females , mean age = 163 days ) to alternating 20-s sequences of phase-scrambled faces and objects ( e . g . , Sadr and Sinha , 2004; Rossion and Caharel , 2011 ) and of natural stimuli replicating exactly those used in the previous experiment . The phase-scrambled images contain the same low-level information ( i . e . , power spectrum ) as the natural images , but they are unrecognizable as faces or objects ( Video 2 ) . In this second experiment , infants performed 2 to 12 sequences in total , with no significant difference in the number of sequences by condition ( i . e . , 90 s; 4 . 5 sequences on average ) . On grand-averaged data , we again found a large EEG response at the base stimulation frequency ( 6 Hz ) over the medial occipital lobe for both conditions ( electrode Oz; SNR for natural images: 6 . 01; Z = 29 . 42 , p < 0 . 00001; SNR for scrambled images: 7 . 25; Z = 27 . 4 , p < 0 . 00001 ) . 10 . 7554/eLife . 06564 . 007Video 2 . 8 s excerpt of experiment 2 ( 20 s sequences ) showing scrambled faces at a rate of 1 image every 5 scrambled images , at a 6 Hz base rate . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 007 The response at channel Oz was significant for every infant in each of the conditions ( Z-score range of 10 infants for natural images: 1 . 66–29 . 58; for scrambled images: 2 . 81–27 . 76; SNR range for natural images: 1 . 9–13 . 65; for scrambled images: 2 . 57–13 . 55 ) . A comparison between the two conditions using the individual infants SNR values at 6 Hz did not reveal any difference ( t ( 9 ) = 1 . 103 , p = 0 . 3; Figure 4 ) , indicating that the synchronization of the visual system to the stimuli does not differ between conditions . 10 . 7554/eLife . 06564 . 008Figure 4 . Grand-averaged SNR at channel Oz in Experiment 2 . The SNR peak at the base stimulation frequency ( 6 Hz ) is highly significant and spread over the medial occipital lobe ( O1-Oz-O2 ) in both conditions , as indicated on the scalp topography . There was no significant difference between the 2 conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 008 On grand-averaged data , there was a significant response at the oddball ( 1 . 2 Hz ) face frequency at the right occipito-temporal electrode P8 for natural images ( mean SNR = 2 . 09 , Z = 2 . 09 , p < 0 . 05; Figure 5A ) . No other electrode was significant on grand-averaged data , which is based on a lower number of infants than in Experiment 1 ( 10 vs 15 ) and about half of the stimulation sequences . Critically , this response at P8 was absent for scrambled images ( mean SNR = 0 . 78 , Z = −0 . 8 , p > 0 . 05 ) . 10 . 7554/eLife . 06564 . 009Figure 5 . ( A ) Grand averaged EEG SNR spectrum at 1 . 2 Hz in experiment 2 , showing above noise-level ( >1 ) response for faces at channel P8 , as shown on the scalp map . On the right , individual SNR values at 1 . 2 Hz for this second experiment . ( B ) There was no distinct peak in the EEG spectrum at 1 . 2 Hz for corresponding phase-scrambled images , as displayed on the left . As in Figure 1 , for copyright reasons , the face pictures displayed in the figure are different than those used in the actual experiment , but the degree of variability across images is respected . The full set of face pictures is available at http://face-categorization-lab . webnode . com/publications/ together with the paper reporting the original study performed in adults ( Rossion et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 00910 . 7554/eLife . 06564 . 010Figure 5—figure supplement 1 . Face-selective responses at first and second harmonic for natural images but not phase-scrambled images . Top . Grandaveraged EEG spectrum ( in microvolts ) from 0 Hz to 5 Hz for experiment 2 ( originalimages in green , scrambled images in black ) . The peak at 1 . 2 Hz is visible only for the original images . Note also the smaller response at the second harmonic ( 2 . 4 Hz ) . Middle row . SNR transformed grandaveraged spectrum , showing the clear responses at 1 . 2 Hz and 2 . 4 Hz , well above 1 ( signal = noise level ) . Bottom . Topographical maps ( back view ) and SNR distribution across individuals for the original and scrambled images , for both harmonics . DOI: http://dx . doi . org/10 . 7554/eLife . 06564 . 010 For natural images , the 1 . 2 Hz response was above noise level ( i . e . , 1 ) for 9 infants out of 10 ( SNR range of all 10 infants: 0 . 82–3 . 98 ) and highly significant ( t ( 9 ) = 3 . 431 , p = 0 . 004; Figure 5A ) . It reached significance for 6 individual infants out of 10 ( ps < 0 . 05 , Z-score > 1 . 64 ) . The other 3 infants showed a significant 1 . 2 Hz face categorization response over at least one other electrode while none of the electrodes reached significance for the last infant . In contrast considering individual infants data as the source of variance , there was no significant response to phase-scrambled images at electrode P8 ( SNR range = 0 . 11–1 . 93; t ( 9 ) = 1 . 156 , p = 0 . 278; Figure 5B , see also Figure 5—figure supplement 1 for data in amplitude , also showing the second harmonic at 2 . 4 Hz ) . Hence , there was a significant difference at the oddball ( 1 . 2 Hz ) frequency between natural and scrambled images at P8 ( paired t-test: t ( 9 ) = 2 . 969 , p = 0 . 016 ) .
Collectively , the findings of Experiments 1 and 2 demonstrate that the infant right hemisphere discriminates natural photographs of faces from non-face objects of multiple categories and generalizes across face photographs despite their high physical variability . In both experiments , faces are temporally embedded in a rapid stimulation sequence of non-face objects , so that there is an inherent comparison , or contrast , without the need to perform a subtraction between conditions recorded at different times . That is , there is an oddball response only because the face is discriminated from all other object categories , activating a ( face- ) specific population of neurons at a rate of 1 . 2 Hz . Although this is unlikely , we cannot formally exclude at this stage that another visual category than faces would be represented by a distinct population of neurons and would therefore also elicit an oddball response of this amplitude at 4–6 months of age . However , to our knowledge , there is no other visual category that elicits such a large specific response , with a right hemisphere advantage , in the human adult brain . Moreover , the face is arguably the most frequent and socially relevant stimulus in the human ( infant ) visual environment , making it the best candidate for the early development of category-selective responses . Thanks to this original fast periodic visual stimulation ( FPVS ) approach , the infant's face categorization response identified here goes beyond previous observations of discrimination between segmented faces and non-face stimuli in ERPs ( de Haan and Nelson , 1999; Halit et al . , 2004; Gliga and Dehaene-Lambertz , 2007; Peykarjou and Hoehl , 2013 ) , near infrared spectroscopy responses ( NIRS; Kobayashi et al . , 2011 ) or positron emission tomography ( PET; Tzourio-Mazoyer et al . , 2002 ) activations obtained with a standard presentation mode ( i . e . , transient , slow , and non-periodic stimulation ) . Despite the great interest of these studies , it is fair to say that it is difficult to define sensitive ( i . e . , high SNR ) and objective face-selective responses in infants with a conventional stimulation mode as used in these studies , so that there is a lack of consistency across studies . Moreover , given time constraints , these studies used segmented stimuli rather than natural images , and could only compare faces to a limited number of categories . Hence , the face-selective responses obtained in previous studies could be due to systematic differences between categories in terms of a homogenous stimulus , such as contour for instance ( e . g . , all round faces vs rectangular pictures of cars ) . Finally , a significant contribution of low-level visual cues to faces vs objects responses could not be excluded from these studies , or precisely evaluated . Here , in Experiment 2 , removing shape information while preserving low-level visual differences in the power spectrum completely erased the 1 . 2 Hz face-selective response . In other words , the 1 . 2 Hz face-selective response identified here for natural face images cannot be attributed to low-level visual confounds , such as differences in power spectrum between faces and other object categories . Moreover , the face stimuli were always embedded within distinct natural backgrounds , suggesting that the infant's brain is able to perform complex figure-ground segregation . This is even more impressive considering the brief presentation duration of each face stimulus ( i . e . , 167 ms SOA , about 100 ms for stimulus duration above 20% contrast , see ‘Materials and methods’ ) and the rapid mode of stimulation where each stimulus interrupts the processing of the previous one . Considering the enormous amount of resources devoted to develop face segmentation algorithms in computer vision ( Scheirer et al . , 2014 ) , this is not a trivial accomplishment . Finding a dominant face-selective response over the right hemisphere in young infants has important implications for our understanding of hemispheric lateralization in humans . It demonstrates that the right hemispheric dominance for face-selective processes—typical of the adult brain ( Hecaen and Angelergues , 1962; Hillger and Koenig , 1991; Sergent et al . , 1992; Kanwisher et al . , 1997; Busigny et al . , 2010; Rossion , 2014; see Rossion et al . , 2015 with the present approach ) is already present in infancy , independently of low-level cues . This observation refutes the view that the right hemispheric lateralization for faces arises only after a few years of age , following and being driven by the left hemispheric lateralization for words that emerges during reading acquisition ( Dundas et al . , 2013 , 2014 ) . Rather , even if literacy can refine cortical organization for vision and language ( Dehaene et al . , 2010 ) , the right hemispheric face-selective response identified here in young infants indicates that the right lateralization for face perception is present well before reading acquisition ( see also Dehaene et al . , 2010 for right hemisphere lateralization in illiterate adults , and Cantlon et al . , 2011 for right lateralization in 4 years old children ) . Instead , our findings are in agreement with an early emergence of right lateralization for faces during development ( de Schonen and Mathivet , 1990 ) , a view so far based on evidence collected with face stimuli only ( de Schonen and Mathivet , 1990; Tzourio-Mazoyer et al . , 2002; Le Grand et al . , 2003 ) or by comparing faces to meaningless stimuli that also differ in terms of low-level visual cues ( Tzourio-Mazoyer et al . , 2002; Kouider et al . , 2013 ) . What is the origin of this early face-selective response ? Some authors have suggested a face-processing module pre-specified in the genome ( Farah et al . , 2000 ) , compatible with newborns' preferential looking behavior for face patterns at birth ( Goren et al . , 1975; Johnson and Morton , 1991; but see; Turati et al . , 2002 ) . However , infants are already extensively exposed to faces after a few months of life . Hence , face-selective responses observed here in 4–6 month-old infants may originate from a combination of initial biological constraints and of accumulation of visual experience with faces during early development . Neuroimaging studies in children show that the magnitude of face-selective neural responses is not adult-like at 7 years of age and keeps increasing until adolescence ( Golarai et al . , 2007 , 2010; Scherf et al . , 2007 ) , suggesting that face-selectivity continues to increase during development . Given its advantages in terms of sensitivity , implicit recording and objectivity ( i . e . , measuring brain responses at a known , exact rate of periodic stimulation ) , the FPVS approach used here with electroencephalography is well positioned to test this hypothesis and characterize the full human developmental course of face processing and natural visual scene categorization . | Putting names to faces can sometimes be challenging , but humans are generally extremely good at recognising faces . Computers , on the other hand , often find it difficult to categorize a face as a face . Indeed , a major challenge in face recognition arises because faces come in many different shapes and sizes . Moreover , both the lighting conditions and the orientation of the head can change , which makes the challenge even more difficult . Young infants also show a preference for pictures of human faces over nonsense images , which suggests that the ability to recognise faces is at least partly hard-wired . Neuroimaging studies have revealed that face recognition depends on activity in specific regions of the right hemisphere of the brain , and adults who sustain damage to these regions lose their face recognition skills . De Heering and Rossion have now provided the first evidence that the right hemisphere is specialized for distinguishing between natural images of faces and ‘non-face objects’ in infants as young as 4 to 6 months . By using scalp electrodes to record electrical activity in the brain as the infants viewed images on a screen , De Heering and Rossion showed that photographs of human faces triggered a distinct pattern of electrical activity in the right hemisphere: this pattern was clearly different to the patterns triggered by photographs of animals or objects . A consistent response was triggered by faces of different genders and expressions , and by faces presented from various viewpoints and under different lighting conditions . In a control experiment , De Heering and Rossion demonstrated that low-level visual features such as differences in luminance or contrast do not contribute to this selective response to faces . These results argue against the idea that face perception only becomes assigned to the right hemisphere of the brain when children learn to read ( that is , when language processing begins to occupy parts of the left hemisphere ) . By generating significant responses in a short period of time ( just five minutes or less ) , the protocol developed by De Heering and Rossion has the potential to prove very useful to researchers investigating developmental changes to the perception of visual images during childhood . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"neuroscience"
] | 2015 | Rapid categorization of natural face images in the infant right hemisphere |
Invasion of erythrocytes by Plasmodial merozoites is a composite process involving the interplay of several proteins . Among them , the Plasmodium falciparum Cysteine-Rich Protective Antigen ( PfCyRPA ) is a crucial component of a ternary complex , including Reticulocyte binding-like Homologous protein 5 ( PfRH5 ) and the RH5-interacting protein ( PfRipr ) , essential for erythrocyte invasion . Here , we present the crystal structures of PfCyRPA and its complex with the antigen-binding fragment of a parasite growth inhibitory antibody . PfCyRPA adopts a 6-bladed β-propeller structure with similarity to the classic sialidase fold , but it has no sialidase activity and fulfills a purely non-enzymatic function . Characterization of the epitope recognized by protective antibodies may facilitate design of peptidomimetics to focus vaccine responses on protective epitopes . Both in vitro and in vivo anti-PfCyRPA and anti-PfRH5 antibodies showed more potent parasite growth inhibitory activity in combination than on their own , supporting a combined delivery of PfCyRPA and PfRH5 in vaccines .
According to the World Health Organization 2015 Malaria Report ( who . int/malaria/publications/world_malaria_report/en ) , malaria is estimated to have caused 214 million clinical cases and 438 , 000 deaths in 2015 . The disease is transmitted by female Anopheles mosquitoes and caused by parasitic protozoans of the genus Plasmodium , of which P . falciparum and P . vivax are the most prevalent and P . falciparum is causing the most often fatal and medically most severe form of malaria . Debilitating clinical symptoms associated with the infection are caused by the multiplication of the asexual blood-stage parasites in erythrocytes . One of the most promising targets for malaria vaccine development is therefore at the stage where merozoites invade erythrocytes . Invasion of host erythrocytes by merozoites is a complex process , conceptually divisible into four phases: ( 1 ) initial recognition of and reversible attachment to the erythrocyte membrane by the merozoite; ( 2 ) junction formation leading to irreversible attachment of the merozoite , parasitophorous vacuole formation , and release of the Plasmodium rhoptry-microneme secretory organelles; ( 3 ) invagination of the erythrocyte membrane around the merozoite , accompanied by the shedding of the merozoite’s surface coat; ( 4 ) closing of the parasitophorous vacuole and resealing of the erythrocyte membrane mark the completion of merozoite invasion ( Pinder et al . , 2000 ) . The initial recognition and the active invasion of erythrocytes depend on specific molecular interactions between parasite ligands and receptors on the host erythrocyte membrane . Although several ligand-receptor interactions have already been identified , the entire network of molecular interactions involved in invasion is not yet fully disentangled . In addition , P . falciparum merozoite proteins are antigenically highly diverse and in part functionally redundant , to facilitate parasite escape from host immune surveillance and to ensure erythrocyte invasion via alternative pathways ( Cowman et al . , 2012 ) . Most efforts in malaria blood stage vaccine research and development have historically concentrated on immuno-dominant , polymorphic antigens that contribute to the invasion of red blood cells by merozoites . Despite major efforts , blood-stage vaccines based on merozoite surface antigens have so far shown limited efficacy in clinical trials ( reviewed in Halbroth and Draper , 2015 ) . Extensive antigenic polymorphism represents one major hurdle for the development of an effective blood-stage malaria vaccine ( Takala et al . , 2009; Dzikowski and Deitsch , 2009 ) . Therefore , the identification of new candidate antigens that are able to induce broad strain-transcending immunity and that are not susceptible to ‘vaccine resistance’ has become a recent research focus . Availability of pathogen genomes is facilitating the discovery of novel vaccine candidate antigens through ‘reverse vaccinology’ approaches ( Rappuoli , 2001; Donati and Rappuoli , 2013 ) . Sequencing and annotation of the P . falciparum genome ( Gardner et al . , 2002 ) has supported the identification of new blood-stage vaccine candidate antigens ( Conway , 2015; Proietti and Doolan , 2014 ) , among which the P . falciparum Cysteine-Rich Protective Antigen ( PfCyRPA ) has a number of noteworthy properties . While PfCyRPA is highly conserved among a plethora of P . falciparum isolates , it also is poorly immunogenic in the context of natural exposure ( Dreyer et al . , 2012 ) . Moreover , PfCyRPA-specific monoclonal antibodies ( mAb ) inhibit parasite growth both in vitro and in vivo by blocking merozoite invasion ( Dreyer et al . , 2012; Favuzza et al . , 2016 ) . PfCyRPA is a 42 . 8 kDa protein of 362 residues with a predicted N-terminal secretion signal . Orthologs of PfCyRPA have been found in the genomes of the human malaria parasite P . vivax and the primate pathogens P . knowlesi , P . cynomolgi , and P . reichenowi ( Figure 3—figure supplement 1 ) , but not in the sequenced genomes of other Plasmodium species . P . falciparum PfCyRPA shares on average 42% sequence identity with its orthologs , but within different P . falciparum isolates PfCyRPA is highly conserved: just 13 dimorphic amino acid positions ( highlighted in Figure 3—figure supplement 1 ) were found in 227 P . falciparum field isolates ( Manske et al . , 2012 ) , and only a single variant ( Arg399 instead of Ser399 ) was found at a frequency of greater than 2% . PfCyRPA is part of a multi-protein complex ( Reddy et al . , 2015; Volz et al . , 2016 ) including also the PfRH5-interacting protein PfRipr and the reticulocyte binding-like homologous protein PfRH5 , which binds to the erythrocyte receptor basigin ( Baum et al . , 2009; Crosnier et al . , 2011; Chen et al . , 2011b , 2014 ) . PfRH5 , PfCyRPA , and PfRipr colocalize during parasite invasion at the junction between merozoites and erythrocytes . The complex seems to be required both for triggering Ca2+ release and establishment of tight junctions ( Volz et al . , 2016 ) . While merozoites deficient in PfCyRPA or PfRH5 can still bind to erythrocytes , they do not attach irreversibly and cannot invade the host cells ( Volz et al . , 2016 ) . Like PfCyRPA ( Dreyer et al . , 2012 ) , PfRH5 induces invasion-blocking antibodies that are effective across common genetic variants ( Douglas et al . , 2011; Bustamante et al . , 2013; Douglas et al . , 2014 ) . ‘Structural vaccinology’ , a combination of immunological , structural , and bioinformatics approaches , is increasingly used for the design of improved vaccine antigens ( Dormitzer et al . , 2008; Cozzi et al . , 2013; Malito et al . , 2015 ) . To this end , the crystal structures of PfRH5 in complex with basigin and neutralizing inhibitory mAb have been determined ( Chen et al . , 2014; Wright et al . , 2014 ) . Here , we describe the crystal structure of the promising vaccine candidate PfCyRPA alone and in complex with the antigen-binding fragment ( Fab ) of the parasite growth inhibitory mAb c12 ( Dreyer et al . , 2012 ) . The structure of PfCyRPA represents a step toward elucidating its biological function . Furthermore , definition of the specific epitope–paratope interactions from the crystal structure of the PfCyRPA/c12 complex will support rational design of an epitope-focused PfCyRPA-based candidate vaccine .
The finding that PfCyRPA and PfRH5 form a complex essential for parasite invasion prompted us to investigate the fine specificities of previously generated parasite inhibitory and non-inhibitory anti-PfCyRPA mAbs . The 16 anti-PfCyRPA mAbs available for analysis ( Dreyer et al . , 2012; Favuzza et al . , 2016 ) , showed six distinctive reactivity patterns with seven overlapping recombinant protein fragments of PfCyRPA , assigning them to the epitope groups A – F , with groups A , B , C , and F comprising the parasite inhibitory and groups D and E the non-inhibitory mAbs ( Figure 1 ) . All mAbs bound to the full-length PfCyRPA ( without the signal sequence; fragment 1 ) . mAbs belonging to epitope group A exclusively bound this fragment , indicating that they recognize conformational epitopes not present in any of the shorter PfCyRPA sequence stretches . Lack of binding to fragments 2 and 7 indicates that the epitope may comprise sequences from both ends of the polypeptide chains . Epitope group B antibodies , including mAb c12 , bound only to fragments 1 , 2 , and 3 . Epitope group F mAbs bound to fragments 1 , 2 , and 5 , but not to fragment 3 , indicating that in contrast to group B , residues located in the sequence stretch between aa 181–251 are required for their binding . The single mAb c04 constitutes the epitope group C showing binding to fragment 7 ( only in IFA , not confirmed by Western blotting analysis ) . The non-inhibitory mAbs clustered into the distinct epitope groups D and E . 10 . 7554/eLife . 20383 . 003Figure 1 . Binding of anti-PfCyRPA mAbs to fragments of PfCyRPA . Binding of 16 mAbs to PfCyRPA fragments ( black bars ) expressed on the cell surface of HEK cells as assessed by Western blotting analysis and live-cell immunofluorescence staining . ( x ) indicates staining and ( – ) no staining; ( a ) indicates no reactivity in Western blot analysis of HEK cell lysates . Expression on the surface of the HEK cells has been demonstrated for all PfCyRPA fragments by immunofluorescence analysis using anti-Histidine tag HIS-6/9 mAb ( Figure 1—figure supplement 1 ) . For reference , the 17 residues constituting the epitope on PfCyRPA identified from the complex crystal structure with the Fab of mAb c12 is shown in all constructs as red bars . According to their reactivity pattern , anti-PfCyRPA mAbs were assigned to different epitope groups: A: c10 , SB2 . 5; B: c02 , c06 , c08 , c09 , c12 , SB3 . 7; C: c04; D: c05; E: c13 , SB3 . 9; F: SB1 . 6 , SB2 . 1 , SB2 . 3 , SB3 . 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 00310 . 7554/eLife . 20383 . 004Figure 1—figure supplement 1 . Cell-surface expression of PfCyRPA fragments on transiently transfected HEK cells . Fluorescence staining of HEK cells expressing PfCyRPA fragments on their surface after staining with ( A ) anti-His tag HIS-6/9 mAb or ( B ) anti-PfCyRPA c12 mAb and FITC-labelled anti-mouse IgG antibodies . Nuclei were stained with DAPI . Untransfected HEK cells served as negative control . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 004 Since it may prove useful to incorporate a combination of PfCyRPA and PfRH5 in a multivalent malaria vaccine , we investigated whether mAbs against these two vaccine candidate antigens have additive or synergistic effects . In a first step , we tested a combination of inhibitory mAbs against the two antigens in an in vitro parasite growth inhibition assay . Parasites were cultured for one cycle of merozoite invasion in the presence of the anti-PfCyRPA c12 mAb with or without the anti-PfRH5 BS1 . 2 mAb at concentrations of 500 , 250 , and 125 μg/mL . Either mAbs showed potent inhibitory activity , consistently reducing parasite growth of all four tested P . falciparum strains in a concentration-dependent manner and to the same extent as the well characterized inhibitory anti-MSP-1 mAb 12 . 10 ( Blackman et al . , 1990 ) ( Figure 2A and Figure 2—figure supplement 1 ) . When combining the anti-PfCyRPA c12 mAb with the anti-RH5 BS1 . 2 mAb , we found a significantly enhanced inhibitory activity: while mAbs c12 and BS1 . 2 at a concentration of 250 µg/mL inhibited growth by 21% ± 2 . 2% and 31% ± 4 . 6% , respectively , the combination of both mAbs ( 250 µg/mL each ) inhibited growth by 59% ± 1 . 4%; ( Figure 2A ) . The functional activity of both mAbs was not enhanced by the addition of a malaria-unrelated control mAb . 10 . 7554/eLife . 20383 . 005Figure 2 . Anti-PfCyRPA and anti-PfRH5 mAbs have both in vitro and in vivo an additive inhibitory effect on parasite growth . ( A ) Growth inhibition in vitro . Synchronized P . falciparum 3D7 blood-stage parasites were cultured for one cycle of merozoite invasion ( 48 hr ) in the presence of anti-PfCyRPA c12 mAb , anti-PfRH5 BS1 . 2 mAb , and their combinations . An isotype-matched , malaria-unrelated mAb ( NR4 . 2 ) ( Rose et al . , 2016 ) was used as negative control . Inhibitory and non-inhibitory anti-MSP-1 mAbs ( 12 . 10 and 2F10 , respectively ) were also included as reference ( Blackman et al . , 1990 , 1994 ) . Percent parasite growth inhibition was calculated against the parasitemia of PBS control wells . Each bar represents the mean of a triplicate experiment , and error bars indicate the standard deviation ( SD ) . Differences in parasite growth inhibition between mAbs c12 and BS1 . 2 alone and their combinations are statistically significant ( unpaired t test with Welch’s correction , 95% confidence interval , two-tailed p value ) . ( B ) Growth inhibition in vivo . NODscidIL2Rγnull mice received purified anti-PfCyRPA c12 mAb and/or anti-PfRH5 BS1 . 2 mAb by i . v . injections . Mice were then infected with P . falciparum 3D7 and parasitemia was monitored over 6 days . Values represent the mean parasitemia in human erythrocytes in peripheral blood of three mice per group . Error bars indicate the SD . PBS and an unrelated control mAb were used as negative control . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 00510 . 7554/eLife . 20383 . 006Figure 2—figure supplement 1 . Anti-PfCyRPA and anti-PfRH5 mAbs inhibit parasite growth of various P . falciparum strains . Synchronized P . falciparum 3D7 , 7G8 , K1 and D6 ( A , B , C , and D , respectively ) blood-stage parasites were cultured for one life cycle in the presence of anti-PfCyRPA c12 mAb , anti-PfRH5 BS1 . 2 mAb , and their combinations . An isotype-matched , malaria-unrelated control mAb ( NR4 . 2 mAb ) was used as negative control . Percent parasite growth inhibition was calculated against the parasitemia of PBS control wells . Each bar represents the mean of a triplicate experiment , and error bars indicate the standard deviation ( SD ) . Differences in parasite growth inhibition between mAbs c12 and BS1 . 2 alone and their combinations are statistically significant ( unpaired t test with Welch’s correction , 95% confidence interval , two-tailed p value ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 006 In a second step , the in vivo parasite inhibitory activity of the mAbs was evaluated in the P . falciparum SCID murine model that employs non-myelodepleted NODscidIL2Rγnull mice engrafted with human erythrocytes ( Dreyer et al . , 2012; Jiménez-Díaz et al . , 2009 ) . Groups of three mice received 2 . 5 or 0 . 5 mg of mAbs c12 or BS1 . 2 or a combination of both mAbs by i . v . injection . The control groups received either 2 . 5 mg of an isotype-matched malaria-unrelated mAb or the same volume of PBS without Ab . Mice were infected with parasitized erythrocytes 1 day after the antibody transfer and parasitemia was subsequently monitored ( Figure 2B ) . In the control groups , parasitemia reached 19 . 6% ± 0 . 8% on day 9 after mAb injection . Parasitemia in mice having received 2 . 5 mg c12 or BS1 . 2 mAb increased only marginally , reaching 2 . 2 ± 0 . 5 and 2% ± 0 . 3% on day nine after mAb injection , respectively . At the lower dose of 0 . 5 mg c12 and BS1 . 2 inhibited parasite growth to 10 . 1 ± 2 . 3 and 10 . 8% ± 6 . 9% parasitemia , respectively ( Figure 2B ) . In accord with the in vitro data , parasitemia decreased significantly ( p=0 . 0356; unpaired t test , 95% confidence interval , two-tailed ) and reached only 4 . 8% ± 1 . 9% on day 9 if mice received 0 . 5 mg of the anti-PfRH5 BS1 . 2 mAb in addition to 0 . 5 mg of the anti-PfCyRPA c12 mAb . These results demonstrated that anti-PfCyRPA and anti-PfRH5 antibodies have an additive parasite growth inhibitory effect , justifying the combination of both antigens in a subunit vaccine . While the structure of PfRH5 in complex with an inhibitory antibody has been determined , structural information for PfCyRPA is lacking . We therefore determined the crystal structures of PfCyRPA and its complex with the Fab of the growth inhibitory mAb c12 . The far-UV CD spectrum of PfCyRPA is consistent with an all-β structure connected by loop regions and the absence of α-helices ( Figure 3—figure supplement 2A ) . Mass spectrometric analysis of proteolytic fragments of PfCyRPA revealed at least four disulfide bonds that are sequential along the sequence ( Figure 3—figure supplement 2B ) . Intrinsic fluorescence showed that the two Trp residues present in PfCyRPA are buried in the native state . Also , the disulfide bonds seem to be buried , because addition of 50 mM reducing agent had no significant effect on the fluorescence of PfCyRPA ( Figure 3—figure supplement 2C ) . Taken together , these data are consistent with PfCyRPA forming a compact , disulfide-stabilized molecule of predominantly β-sheet structure . In order to crystallize PfCyRPA , we needed to pre-treat the protein with Actinase E ( Figure 4—figure supplement 1A ) . The crystal structure of PfCyRPA , determined to a resolution of 2 . 5 Å ( detailed in Supplementary file 1 ) , confirmed our biophysical analyses ( Figure 3 ) . PfCyRPA adopts a six-bladed β-propeller structure that buries the disulfide bonds and the Trp residues . Each blade of the propeller is constructed by a four-stranded anti-parallel β-sheet ( Figure 3B ) . The five disulfide bonds in PfCyRPA are located within blades 2–6 ( Figure 3A ) , stabilizing each individual blade . The first blade has no disulfide bond; it is formed by β-strands from the N- and C-terminal regions of PfCyRPA , potentially enabling conformational changes in PfCyRPA by opening and closing . A domain alignment search ( DALI; [Holm and Rosenström , 2010] ) for related structures revealed that PfCyRPA adopts a heavily modified sialidase/neuraminidase fold . The closest structural relative is the catalytic domain of Vibrio cholerae sialidase ( Moustafa et al . , 2004 ) ( Figure 3C ) . The two proteins have only 9% sequence identity and the structures have a large root mean square distance ( rmsd ) of 3 . 7 Å over 285 residues , clearly showing that while the overall fold is similar , the structures are very different with respect to the inclination of the blades ( Figure 3C ) and the length and conformations of the surface loops connecting the β-strands ( Figure 3D ) . PfCyRPA also contains a signature sequence motif for sialidases known as an Asp-box ( 201-SHDKGETW-208; conserved residues are underlined ) ( Roggentin et al . , 1989 ) , which serve structural roles in the β-propellers of sialidases . However , while bacterial sialidases contain between three to five Asp-boxes , PfCyRPA contains only a single one . 10 . 7554/eLife . 20383 . 007Figure 3 . PfCyRPA adopts the neuraminidase fold . ( A ) Structure-sequence relationship of PfCyRPA . Indicated are an Actinase E cleavage site at Asp189 ( red ) , a sialidase-typical Asp-box ( dotted underlined ) , the two Trp residues ( dotted underlined ) , and the sequential disulfide bonds ( connected by lines and in same color ) . β-strands are shown as arrows colored according to the blade they form . The epitope recognized by mAb c12 is underlined in bold . ( B ) Cross-eyed stereo view of the ribbon representation of a superposition of the two PfCyRPA molecules in the asymmetric unit with the blades numbered 1–6 from the N-terminus and colored individually . Blade one is made up of an N-terminal ( black ) and three C-terminal β-strands ( red ) . One protomer is shown with white , the other with black loop regions , which may differ substantially ( arrows in blade 5 ) . The Trp and Cys residues are drawn as stick models . ( C ) The same orientation of the catalytic domain of Vibrio cholerae sialidase ( PDB-ID 1w0o ) , the next structural homolog of PfCyRPA with a DALI score of 18 ( Z < 5 is structurally dissimilar ) . Sialic acid and residues in the Vibrio enzyme are displayed as balls and sticks . Structural Ca2+ ions are marked as magenta spheres . None of the residues necessary for metal ion binding , substrate binding , or catalysis is present in PfCyRPA . ( D ) Superposition of PfCyRPA with the V . cholerae sialidase . While both proteins are 6-bladed β-propellers , the blades have very different angles , extents , and loop lengths and conformations connecting the β-strands . The four Asp-boxes in the bacterial sialidase ( grey ) are colored black . PfCyRPA ( orange ) has only a single Asp-box connecting the third and fourth β-strands in blade 3 ( colored blue ) . Other β-strand connections are made by sequences unrelated to the Asp-box motif , in agreement with poor conservation of the Asp-box in other , e . g . viral , sialidases . The view in ( D ) is rotated by 180° about the horizontal axis compared to ( B ) and ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 00710 . 7554/eLife . 20383 . 008Figure 3—figure supplement 1 . Sequence alignment of CyRPA orthologs . Sequence alignment of CyRPA orthologs from P . falciparum ( P . fal . ) , P . vivax ( P . viv . ; PVX_090240 ) , P . knowlesi ( P . kno . ; PKNH_0515800 ) , P . cynomolgi ( P . cyn . ; PCYB_053730 ) , and P . reichenowi ( P . rei . ; PRCDC_0421000 ) . Full-length protein sequences ( PlasmoDB; plasmodb . org/plasmo ) were aligned with Clustal O ( 1 . 2 . 1 ) . Asterisks indicate conserved positions , colons indicate strong biophysical conservation , and periods indicate weak biophysical conservation . Ten of the twelve Cys residues are conserved ( same color code as in Figure 3 ) . The predicted PfCyRPA secretion signal sequence ( M1–C28 ) is indicated in boldface , and the predicted GPI-anchor motif ( I353-E362 ) is underlined . The N-glycosylation sites that are used in human cells ( N145 , N322 , and N338 ) are shown in lowercase and the non-synonymous SNPs ( D73 , D110 , V165 , P168 , R174 , F187 , D236 , N270 , V292 , N338 , R339 , L341 , N352 ) are colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 00810 . 7554/eLife . 20383 . 009Figure 3—figure supplement 2 . Biophysical analysis of PfCyRPA . ( A ) The CD spectrum of PfCyRPA shows a prominent minimum around 215 nm , consistent with a high content of β-secondary structure . The minimum is remote from the minima expected for helical ( 208 nm and 222 nm ) and random coil ( 200 nm ) structures . ( B ) Identification of disulfide bonds in PfCyRPA . The peptide fragments are labeled K ( LysC cleavage ) , D ( AspN cleavage ) , and T ( trypsin cleavage ) . A cut-off of 2% signal intensity was applied to separate highly populated from less populated fragments . Masses that correspond to two proteolytic peptides connected by disulfide bonds were repeatedly found ( parentheses ) when using different proteases . An equal sign ( = ) signifies a disulfide bond and a minus sign ( − ) signifies a skipped cleavage . For instance , the first disulfide bond between Cys48 and Cys64 was identified four times in the LysC hydrolysate: K2 = K3 corresponds to the mass of the second and third expected proteolytic fragment linked by a disulfide bond . K2 = K3–4 is the second expected LysC peptide ( K2 ) disulfide-linked to a larger peptide ( K3–K4 ) that still contains a LysC cleavage site but this site was not recognized by the enzyme . The same reasoning applies to K2 = K3–5 , where two LysC cleavage sites were not recognized . An asterisk ( * ) denotes a disulfide bond in a peptide . The K2-4* fragment has a mass corresponding to an un-cleaved and disulfide-linked peptide with two intact LysC cleavage sites that were not recognized . The last disulfide bond ion PfCyRPA between Cys303/327 was not identified by any protease , possibly due to low abundance of the proteolytic fragments containing it . ( C ) The fluorescence emission spectrum of native PfCyRPA ( black ) shows a maximum at 326 nm that shifts to 338 nm when heated to 70°C ( red ) or 355 nm when unfolded by 9 M urea ( blue ) . Addition of 50 mM DTT ( magenta ) has no significant effect on the fluorescence , suggesting that the disulfide bonds are not solvent-accessible . The unusual finding that the Trp fluorescence quantum yield of the unfolded state is higher than in the native state indicates that the excited state of Trp is efficiently quenched by nearby residues in the native structure ( Chen and Barkley , 1998 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 00910 . 7554/eLife . 20383 . 010Figure 3—figure supplement 3 . PfCyRPA lacks detectable sialidase activity in a functional assay . Sialidase activity was estimated using the Amplex Red Neuraminidase ( Sialidase ) assay kit ( Molecular Probes , Inc . ) . The assay utilizes Amplex Red to detect H2O2 generated by galactose oxidase oxidation of desialiated galactose , the end product of sialidase activity . The H2O2 in the presence of HRP reacts with Amplex Red reagent to generate resorufin , the red fluorescent oxidation product , which was detected at 570 nm using the Sunrise Absorbance Reader ( Tecan ) . PfCyRPA was mixed 1:1 ( v/v ) with the Amplex Red reagent/HRP/galactose oxidase/fetuin working solution ( 0 . 2 U/mL HRP , 4 U/mL galactose oxidase , 500 µg/mL fetuin ) , and the mixture was incubated at 37°C for 30 min in a light-protected container . Sialidase and H2O2 ( supplied with the kit ) were used as positive controls ( Ctrl+ ) ; a no-sialidase sample as negative control ( Ctrl− ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 010 Two molecules of PfCyRPA are present in the asymmetric unit and have conformational differences in several surface loops , suggesting possible flexibility of these loops in solution ( Figure 3B ) . From the point of peptidomimetics that could be derived from PfCyRPA for vaccine development , these surface loops connecting the blades on the back and front of the β-propeller are natural candidates . Although some of the loops , e . g . in blade 5 ( arrows in Figure 3B ) , have significant structural plasticity , it should be possible to stabilize them in a conformation suitable to raise an immune response . The sialidase fold of PfCyRPA and the presence of an Asp-box motif raised the question whether PfCyRPA exhibits sialidase activity . In view of the involvement of viral and microbial sialidases in the unmasking of cryptic host ligands , host cell adhesion , and invasion ( Chen et al . , 2011a; Lewis and Lewis , 2012; Matrosovich et al . , 2015 ) , sialidase activity could make sense for PfCyRPA-mediated invasion of erythrocytes . We detected no sialidase activity , however , when we tested recombinant PfCyRPA for neuraminidase activity using a colorimetric assay ( Figure 3—figure supplement 3 ) . The active site of sialidases usually contains a triad of Arg residues that bind to the substrate , a Glu/Tyr pair where the acid acts as a general base to activate the Tyr nucleophile , and a hydrophobic pocket with a conserved Trp that accommodates the acetyl group of sialic acid ( Buschiazzo and Alzari , 2008 ) . In addition , many sialidases bind to Ca2+ ions , of which one is close to the active site and is bound by three main-chain carbonyl groups and oxygen atoms in the side-chains of Asn , Thr , and Asp . PfCyRPA contains none of the residues necessary for catalysis , nor does it harbor a Ca2+ binding site , providing structural correlates of the absence of sialidase activity in PfCyRPA . PfCyRPA and sialidases may have evolved from a common ancestor , or PfCyRPA could have evolved from a genuine sialidase to adopt other functionalities . The parasite inhibitory anti-PfCyRPA c12 mAb binds tightly to PfCyRPA with a Kd of ca . 1 nM as determined by surface plasmon resonance analysis ( Dreyer et al . , 2012 ) . This mAb recognizes PfCyRPA independent of its glycosylation , as revealed by Western blot analyses ( Figure 4A ) . Because of these favorable properties , we chose c12 for epitope mapping by determining the structure of a PfCyRPA/c12 complex . In a first step , we determined the crystal structure of c12 in isolation ( Figure 4B and C ) . We obtained three different crystal forms containing a total of four crystallographically independent Fab molecules . Superposition of the structures showed very little structural plasticity of the CDR loops ( Figure 4C ) , suggesting that they retain their structures upon epitope binding . 10 . 7554/eLife . 20383 . 011Figure 4 . Recognition of PfCyRPA by the mAb c12 and structure of c12 . ( A ) Reducing SDS-PAGE of glycosylated ( left ) and non-glycosylated ( right ) PfCyRPA detected by Coomassie-staining ( blue ) and Western blotting with mAb c12 ( black ) . Recognition by c12 is independent of the glycosylation . ( B ) Overview of the c12 structure with a glycan located at heavy-chain Asn37 . mFo-DFc electron density for the glycan is shown as a red mesh drawn at the three rmsd level . The light and heavy chains are colored light pink and light blue , respectively . Heavy chain CDR1-3 are colored cyan , magenta , and yellow , and light chain CDR1-3 are drawn in dark blue , pink , and green . ( C ) Comparison of the four c12 structures shows little conformational variability of the CDR loops . The four c12 molecules superimpose onto their variable VHVL di-domains with an average rmsd of 0 . 35 Å , which reveals a minor spread of the elbow angles between 133 . 1° and 135 . 8° . The high structural congruence indicates that the CDR conformations are genuine and not dominated by crystal contacts . The view is from above on top of the CDR loops . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 01110 . 7554/eLife . 20383 . 012Figure 4—figure supplement 1 . Limited proteolysis of PfCyRPA and the PfCyRPA/c12 complex . ( A ) PfCyRPA . ( B ) PfCyRPA/c12 complex . Proteases are abbreviated as: αC: α-Chymotrypsin; TR: Trypsin; EL: Elastase; PA: Papain; SU: Subtilisin; EGC: Endoproteinase Glu-C; PK: Proteinase K; CL: Clostripain; PE: Pepsin; TH: Thermolysin; BR: Bromelain; AE: Actinase E . The two bands around 28 kDa in ( B ) correspond to the Fab . Actinase E affects PfCyRPA irrespective of bound c12 , yielding two major proteolysis products of apparent molecular weight 17 kDa and 20 kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 012 We then purified the PfCyRPA/c12 complex , analyzed it by limited proteolysis , and crystallized it . Actinase E treatment of the PfCyRPA/c12 complex resulted in the same proteolysis pattern as observed for PfCyRPA alone ( Figure 4—figure supplement 1B ) , suggesting that the epitope for mAb c12 is distant from the Actinase E recognition site at Asp189 . Actinase E treatment was not necessary to crystallize the PfCyRPA/c12 complex , however . We determined the crystal structure of the PfCyRPA/c12 complex at a resolution of 4 . 0 Å by molecular replacement using the individual high-resolution structures of c12 and PfCyRPA as search models ( Figure 5A ) . Despite the limited resolution , novel molecular features were visible in the electron density maps of the complex , providing confidence in the relative orientation of PfCyRPA and c12 ( Figure 5—figure supplement 1 ) . First , the electron density visible for PfCyRPA after molecular replacement with c12 was used as a search model for structure determination of PfCyRPA alone ( see Materials and methods ) . This strategy would have been impossible had the placement of c12 been wrong . Second , a loop region that is absent in the PfCyRPA search model due to Actinase E proteolysis ( see above ) exhibits omit electron density in the PfCyRPA/c12 complex ( Figure 4—figure supplement 1 ) . Third , consistent with the similar conformations of the CDR loops in the individual c12 Fab structures ( Figure 4 ) there are negligible conformational changes in the c12 CDR loops when bound to PfCyRPA . Similarly , the epitope conformation in PfCyRPA is very similar in the unbound and complexed form . Minor adjustments of side-chains were required during rebuilding of the complex structure . While the resolution of the PfCyRPA/c12 complex is limited and many side-chains lack clear electron density , as is typical for this resolution , knowledge of the relative orientation of PfCyRPA with respect to c12 in the complex is sufficient for designing peptidomimetics to target the immune response to the protective epitope . 10 . 7554/eLife . 20383 . 013Figure 5 . Structure of the PfCyRPA/c12 complex . ( A ) Overview showing that the majority of the interface is made by interactions between the light chain of c12 and blade 2 of PfCyRPA . ( B ) Details of the interface viewed from top onto the CDR loops . The light and heavy chain surfaces buried by PfCyRPA are colored pink and blue , respectively . Possible hydrogen bonds and van der Waals interactions are indicated by dashed green and black lines . The CDR loops are color-coded as in Figure 4 . The Asp66-Arg50 salt bridge is circled . ( C ) Close-up of ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 01310 . 7554/eLife . 20383 . 014Figure 5—figure supplement 1 . Electron density of the PfCyRPA/c12 complex . ( A ) Initial electron density at a resolution of 3 . 6 Å contoured at the one rmsd level for PfCyRPA after molecular replacement phasing of the PfCyRPA/c12 complex data with c12 as the search model . The light and heavy chains of c12 are colored pink and cyan , respectively . The final model of PfCyRPA ( blades colored in the same pattern as in Figure 3A ) is superimposed for reference but could not be traced using this map . This map fragment was cut out and used as starting model for molecular replacement of the PfCyRPA structure . ( B ) Zoom of the loop region 186-KFKDDNK-192 . This loop was cut at Asp189 by Actinase E . The loop opened and the sequence could not be completely traced in the isolated PfCyRPA structure ( yellow ) . Difference electron density at a resolution of 3 . 6 Å contoured at the 1 . 8 rmsd level shows that the loop is intact in the PfCyRPA/c12 complex and can be traced in the electron density maps . The refined model of the complex ( loop colored in cyan ) is superimposed for reference . ( C ) Region around Arg50 in the free c12 Fab structure 5ezj . 2mFo-DFc electron density is shown at the one rmsd level . Arg50 binds to nearby Asp52 . ( D ) This interaction is lost in the complex with PfCyRPA . Arg50 now binds to PfCyRPA residue Asp66 ( light blue ) . 2mFo-DFc electron density is shown at the 0 . 8 rmsd level . ( E ) Superposition of the structures from panels ( C ) and ( D ) shows the large movement of Arg50 in the free c12 Fab ( magenta ) versus its position in the complex with PfCyRPA ( pink ) . By contrast , the conformation of Tyr112 does not change much . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 01410 . 7554/eLife . 20383 . 015Figure 5—figure supplement 2 . Electron density of the PfCyRPA/c12 complex at the interface . The final model of the PfCyRPA/c12 complex is shown . PfCyRPA is at the top , shown as a grey ribbon representation . Fab c12 is at the bottom with the light and heavy chain colored magenta and blue , respectively . 2Fo-Fc electron density at 4 Å resolution is contoured at the one rmsd level at a radius of 22 Å around the center of the image . The panel on the left-hand side shows the electron density after molecular replacement ( A ) , while the panel on the right-hand side depicts the density after refinement ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 015 The epitope recognized by mAb c12 is a surface composed of blade two and part of blade 3 of PfCyRPA ( Figure 5A ) . The most frequent amino acid dimorphism of PfCyRPA at position 339 is thus located outside the epitope recognized by mAb c12 , consistent with the observation that mAb c12 binds to P . falciparum independent of the PfCyRPA variant they express ( Dreyer et al . , 2012 ) . mAb c12 buries a total surface area of 950 Å2 on PfCyRPA with the major contributor being the light chain , which buries 520 Å2 , while the heavy chain buries only 430 Å2 . The surface complementarity coefficient Sc of the complex is 0 . 67 , a typical value for antibody-antigen interactions ( Lawrence and Colman , 1993 ) ; a value of one would denote perfect complementarity . The light chain also forms more potential hydrogen bonds to PfCyRPA than the heavy chain ( Figure 5B ) . Seven hydrogen bonds are possible between PfCyRPA and the c12 light chain . Each side-chain of light chain CDR1 residues 50-RND-52 can form a hydrogen bond with the side-chains of PfCyRPA residues Asp66 , Asn45 , and Tyr78 , respectively . A particularly large contribution to complex stability may stem from a salt bridge between PfCyRPA residue Asp66 and CDR1 residue Arg50 , which by mutagenesis was found to be true ( see below ) , a further confirmation that the PfCyRPA/c12 complex structure is correct . In addition , the heavy chain contributes significantly to antigen binding . Tyr120 of heavy chain CDR3 can engage in two hydrogen bonds with the side-chains of PfCyRPA residues Asp92 in blade two and Tyr144 in blade 3 , thus establishing a mini-network that bridges blades 2 and 3 . Together with heavy chain CDR3 residue Tyr122 , Tyr120 also forms numerous van der Waals interactions with Asn91 , Lys95 , Glu96 , and Phe136 , Tyr144 , Asn145 , and Asn146 ( Figure 5B and C ) . A complete listing of the derived potential interactions is available in Supplementary file 2 , which due to the limited resolution of the complex structure must remain tentative . What is clear from the structure is that mAb c12 recognizes a discontinuous epitope that contains at least 17 residues distributed over four PfCyRPA sequence stretches ( marked in Figure 3A ) . The epitope extracted from the complex structure matches the binding pattern of mAb c12 to the PfCyRPA fragments in our epitope analysis ( Figure 1 ) : only fragments 1–3 containing all the seventeen interaction sites were recognized by mAb c12 . Furthermore , fragment 4 , lacking three out of these 17 residues ( Tyr144 , Asn145 , and Asn146 ) and fragments 5 and 6 , lacking four residues ( Asn45 , Val46 , Pro47 , and Asp66 ) , were not recognized by mAb c12 . The PfCyRPA epitope recognized by mAb c12 was further verified by mutagenesis . The crystal structure suggests that Asp66 forms a salt bridge with Arg50 in the CDR1 loop of the c12 light chain ( Supplementary file 2 and Figure 5—figure supplement 1 ) . When Asp66 was changed to Lys , a capture ELISA detected significantly weaker binding of the mAb c12 to the Asp66Lys variant of PfCyRPA ( Figure 6A ) . As expected , the Asp66Lys replacement did not affect binding of the control mAb SB1 . 6 ( Figure 6B ) , which belongs to the unrelated epitope group F and is expected to engage in crucial interactions with PfCyRPA residues located between residues 181 and 251 ( Figure 1 ) . 10 . 7554/eLife . 20383 . 016Figure 6 . Identification of Asp66 as a key contact residue for PfCyRPA/c12 interaction . mAb binding to purified wild-type PfCyRPA and an Asp66Lys single amino acid sequence variant was analyzed by capture-ELISA experiments . ELISA plates were coated with 10 µg/mL anti-PfCyRPA mAbs and then incubated with serial dilutions of wild-type PfCyRPA ( WT ) or an Asp66Lys variant ( D66K ) ; HRPO-labeled anti-Histidine tag mAbs were used as detection antibody . ( A ) When compared to wild-type PfCyRPA , the Asp66Lys single amino acid exchange strongly reduced the binding to mAb c12 mAb ( epitope group B; Figure 1 ) . ( B ) In contrast , the Asp66Lys amino acid exchange did not affect the binding of the mAb SB1 . 6 , which belongs to epitope group F . DOI: http://dx . doi . org/10 . 7554/eLife . 20383 . 016
PfCyRPA has been identified as a novel malaria blood-stage vaccine target in an endeavor to test predicted P . falciparum open reading frames for the capacity to elicit parasite-inhibitory mAbs ( Dreyer et al . , 2010 ) . PfCyRPA is stage-specifically expressed in late schizonts and elicits mAbs that inhibit parasite growth in vitro and in a P . falciparum experimental infection model based on NODscidIL2Rγnull mice engrafted with human erythrocytes ( Dreyer et al . , 2012 ) . Recent studies ( Reddy et al . , 2015 ) have placed PfCyRPA in a ternary complex with PfRH5 and PfRipr that as a whole is essential for erythrocyte invasion ( Volz et al . , 2016 ) . Like PfRH5 and PfRipr ( Baum et al . , 2009; Chen et al . , 2011b ) , PfCyRPA is refractory to genetic disruption ( Reddy et al . , 2015 ) , and loss of PfCyRPA or PFRH5 function in conditionally expressing mutants blocks parasite growth due to the inability of merozoites to invade erythrocytes ( Volz et al . , 2016 ) . Here , we show that antibodies against PfRH5 and PfCyRPA have additive parasite growth inhibitory activity . Inclusion of both antigens into a multivalent subunit vaccine therefore represents an attractive strategy to prevent emergence of escape mutants . While an ortholog of PfCyRPA is also present in the genome of the human parasite P . vivax , an ortholog of PfRH5 has only been found so far in the genome of P . reichenowi , a chimpanzee malaria parasite closely related to P . falciparum ( Otto et al . , 2014 ) . This may indicate a function of PfCyRPA beyond its role as a protein-binding platform in the ternary complex with PfRH5 and PfRipr . While it has been claimed that PfCyRPA anchors the ternary complex through a GPI anchor to the parasite membrane ( Reddy et al . , 2015 ) , recent results are not consistent with this suggestion ( Volz et al . , 2016 ) . This inconsistency suggests that additional proteins might be involved in anchoring the complex to the parasite membrane and that PfCyRPA has an essential function other than that of an anchor . Further functional studies are required to elucidate the molecular mechanisms how the ternary complex is involved in the triggering of Ca2+ release and the establishment of tight junctions between the merozoite and the erythrocytes ( Volz et al . , 2016 ) . In this context , it may be relevant that the five sequential disulfide bonds of PfCyRPA are in blades 1–5 , leaving blade 6 of the β-propeller structure without such a local stabilization . Blade six is constructed by β-strands from the N- and the C-terminus , and in principle could act as a gate to allow PfCyRPA to undergo substantial conformational changes . Structural vaccinology defines epitopes from antigen-antibody complex crystal structures and thus allows the design of surrogate antigens that elicit protective humoral immune responses . The approach may be of particular use for pathogens that cause chronic infections by presenting immuno-dominant antigens and epitopes to the immune system , which , however , only elicit non-protective antibody responses ( Liljeroos et al . , 2015; Loomis and Johnson , 2015; Robinson , 2013; Dormitzer et al . , 2012 ) . The structural studies of Neisseria meningitidis adhesin A and factor H binding protein ( Malito et al . , 2013 ) , as well as of the Staphylococcus aureus manganese transport protein ( Ahuja et al . , 2015 ) represent remarkable examples in which crystal structure determination of antigens elucidated both the molecular mechanism of their biological functions and their immunological viability as vaccine antigens . For the development of synthetic malaria subunit vaccines , conformationally stabilized and structurally optimized peptide antigens mimicking conserved protective epitopes ( Mueller et al . , 2003; Okitsu et al . , 2007; James et al . , 2006 ) may be more suitable components than entire recombinant proteins containing highly polymorphic immuno-dominant non-protective surface loops . Both PfCyRPA and PfRH5 elicit protective as well as non-protective antibodies ( Dreyer et al . , 2012; Douglas et al . , 2011 ) , making them candidates for structure-based approaches , that aim to focus the immune responses on protective epitopes . Even at the comparatively moderate resolution of 4 Å , the crystal structure of the protective mAb c12 bound to PfCyRPA provides an ideal starting point for the design of a conformationally stabilized PfCyRPA-derived peptide antigen to elicit primarily protective antibodies . Particle-based antigen delivery systems such as virosomes ( Cech et al . , 2011; Pluschke and Tamborrini , 2012 ) are highly suitable to support , by repetitive epitope display , the development of strong immune responses against peptidomimetics . The epitope recognized by mAb c12 is monomorphic and , like PfCyRPA in general , little immunogenic in the natural context . Focusing the vaccine-induced immune response on this PfCyRPA epitope may confer a strong protective effect . Malaria vaccine candidates based on single-antigens have been uniformly unsuccessful and an effective subunit vaccine will likely have to include antigens from more than one stage of the parasite life cycle ( pre-erythrocytic , liver stage , and blood stage components ) , as well as sexual blood-stage gametocyte or mosquito-stage parasite antigens as transmission-blocking components . The Plasmodium genome contains more than 5000 open reading frames of which only <0 . 5% have been tested as vaccine candidates , so careful selection of new targets for an optimized multi-valent vaccine is called for ( Proietti and Doolan , 2014; Dups et al . , 2014; Longley et al . , 2015; Wu et al . , 2015 ) . PfCyRPA represents one of the most promising new blood-stage vaccine candidate antigens identified with the support of –omics approaches , and there is renewed hope that an effective combination and formulation of such rationally selected antigens can lead to the development of a malaria subunit vaccine that offers high level strain-transcending protection ( Halbroth and Draper , 2015 ) . With the availability of structural analyses for both PfCyRPA and PfRH5 in complex with an inhibitory antibody , a structure-based approach is now possible . In view of the observed additive effect of antibodies , inclusion of immunogens representing both antigens in a subunit vaccine would make sense , thereby reducing the danger of selection of immune escape mutants .
Plasmids were amplified in E . coli strain Top10 ( Life Technologies ) grown in LB medium under 100 μg/mL ampicillin selection . The expression vector for secretion of recombinant PfCyRPA was generated by PCR using the plasmid BVM_PFD1130W_FLAG_GP_His as template ( Dreyer et al . , 2010 ) . The resulting expression vector pcDNA3 . 1_BVM_CyRPA ( 26-362 ) _His6 encodes the bee-venom melittin ( BVM ) signal sequence to secrete PfCyRPA into the cultivation medium , and a C-terminal His6-tag . Three asparagine residues ( N145 , N322 , and N338 ) were predicted to be potential sites for N-glycosylation of PfCyRPA in mammalian cells . Since Plasmodium proteins are not glycosylated ( Dieckmann-Schuppert et al . , 1992 ) , an expression vector coding for the PfCyRPA triple variant N145Q/N322Q/N338Q comprising residues 29–362 and containing no N-glycosylation sites was derived from the first vector by site-directed mutagenesis ( GenScript ) . The expression vector coding for the PfCyRPA Asp66Lys single amino acid variant ( D66K ) was also generated by site-directed mutagenesis ( GenScript ) resulting in the expression plasmid pcDNA3 . 1_BMV_CyRPA ( 29–362/D66K ) _His6 . Generation of anti-PfCyRPA mAbs has been described elsewhere ( Dreyer et al . , 2010; Favuzza et al . , 2016 ) . Anti-PfRH5 antibodies were generated using the same strategy as described by Dreyer et al . ( 2010 ) . Plasmodium falciparum strains 3D7 , K1 , 7G8 and D6 were obtained from the Malaria Research and Reference Reagent Resource Center ( MR4; Manassas , VA , USA ) ( MRA-102 , –159 , −154 and −285 , respectively ) . Parasites were cultured essentially as described previously ( Matile and Pink , 1990 ) . The culture medium was supplemented with 0 . 5% AlbuMAX ( Life Technologies ) as a substitute for human serum ( Dorn et al . , 1995 ) . Cultures were synchronized by sorbitol treatment ( Lambros and Vanderberg , 1979 ) . Erythrocytes for passages were obtained from the Swiss Red Cross ( Switzerland ) . In vitro growth inhibition assays with P . falciparum strains 3D7 , K1 , 7G8 and D6 were conducted essentially as described ( Persson et al . , 2006 ) . Each culture ( trophozoite stage parasites , 0 . 5% hematocrit , 2% parasitemia ) was set up in triplicate in 96-well flat-bottomed culture plates . After 48 hr of incubation ( one cycle of merozoite invasion ) , viable parasites were stained with hydroethidine and analyzed in a FACSscan flow cytometer ( Becton Dickinson ) using CellQuest software . A total of 50 , 000 cells per sample were analyzed . Percent inhibition was calculated from the mean parasitemia of triplicate test and control wells as:Percent inhibition ( % ) =control−test/ ( control/100 ) Parasitemia of control samples was also determined by counting GIEMSA stained parasites . All procedures involving living animals were performed in strict accordance with the Rules and Regulations for the Protection of Animal Rights ( Tierschutzverordnung ) of the Swiss Federal Food Safety and Veterinary Office . The protocol was granted ethical approval by the Veterinary Office of the county of Basel-Stadt , Switzerland ( Permit Numbers: 2375 and 2303 ) . Monoclonal antibodies were tested in the murine P . falciparum model essentially as described ( Dreyer et al . , 2012; Jiménez-Díaz et al . , 2009 ) . Human erythrocytes ( hE ) were administered daily ( 0 . 75 mL ) by the i . v . or i . p . route . Mice received a single dose of mAbs formulation by i . v . injection . The following day , mice were infected with 3 × 107 erythrocytes parasitized by P . falciparum PfNF540230/N3 , a strain developed at GlaxoSmithKline ( GSK ) for growth in hE engrafted mice ( Jiménez-Díaz et al . , 2009 ) . Parasitemia was monitored daily by flow cytometry over 6 days ( days 4-9 after mAb injection ) . FreeStyle 293 F cells ( Invitrogen , R790-07 ) , a variant of human embryonic kidney HEK cells , were cultured in suspension in serum-free medium ( FreeStyle 293 Expression Medium , Thermo Fisher Scientific , Waltham , MA ) at 37°C in a humidified incubator with 5% CO2 in volumes of 1 L shake flasks ( Corning; 120 rpm , 5 cm diameter ) or 10 L wave bioreactors ( Sartorius; 30 rpm , pH 7 . 2 , 30% dissolved oxygen ) . Cells were diluted 1:2 with fresh culture medium and transfected at 1 . 2·106 cells/mL with 0 . 4 mg/L expression plasmid using a riDOM-based transfection system ( Québatte et al . , 2014 ) . 72 hr post-transfection , cells were removed by filtration and the supernatant was concentrated with a 10K Pellicon three cassette ( Millipore ) . The His6-tagged recombinant proteins were purified by immobilized metal ion affinity chromatography on a HisTrap HP column ( 5 or 10 mL volume; GE-Healthcare ) equilibrated with 50 mM HEPES/NaOH pH 7 . 2 , 500 mM NaCl . After washing the column with the same buffer containing 20 , 40 , and 50 mM imidazole , the protein was eluted with a linear 50–500 mM imidazole gradient over 20 column volumes . The eluate was concentrated by ultra-filtration ( Amicon Ultra-4 Ultracel 10K ) and applied to a HiLoad 16/600 Superdex 200 gel permeation column ( GE-Healthcare ) equilibrated with 50 mM Tris/HCl pH 7 . 4 , 150 mM NaCl . Homogeneity of PfCyRPA was assessed by reversed-phase chromatography ( RP-HPLC ) on a Poroshell 300 SB-C8 1 × 75 mm2 column using a H2O +0 . 01% TFA / Acetonitrile + 0 . 08% TFA gradient , and was confirmed by LC/MS intact mass analysis . Protein yields were 17 and 10 mg/Lr of culture for the glycosylated and non-glycosylated PfCyRPA , respectively . Purified anti-PfCyRPA mAb c12 ( Dreyer et al . , 2012 ) was diluted to 0 . 5 mg/mL in 20 mM sodium phosphate pH 7 . 0 and cleaved overnight at 21°C with 0 . 01 mg/mL papain ( Sigma-Aldrich ) in a molar ratio of 1:20 . The reaction was stopped with 0 . 001 mg/mL E64 inhibitor ( Sigma-Aldrich ) and the concentrated ( Amicon Ultra-4 Ultracel 10K ) hydrolysate was applied to a 1 × 5 cm2 Toyopearl protein A ( Tosoh Biosciences ) column equilibrated in 20 mM sodium phosphate pH 7 . 0 . The flow-through containing the Fab was concentrated and chromatographed on a 21 . 5 × 60 cm2 TSKgel G3000SW column ( Tosoh Bioscience ) equilibrated with 20 mM bis-Tris propane/HCl pH 7 . 0 , 200 mM NaCl . The purity of the Fab was determined by RP-HPLC as described above . The complex of PfCyRPA with the Fab of mAb c12 ( abbreviated as c12 in the following ) was prepared with a 1 . 5-fold excess of PfCyRPA , which was incubated for 20 min at 21°C , concentrated as above and chromatographed via a Superdex 200 Increase 10/300 GL column ( GE-Healthcare ) equilibrated with 20 mM bis-Tris propane/HCl pH 7 . 0 , 200 mM NaCl . Complex-containing fractions were pooled and analyzed for homogeneity by asymmetric flow field-flow fractionation with static multi-angle light scattering ( SEC/AF4-MALS ) . For the expression of the Asp66Lys PfCyRPA variant and PfRH5 , FreeStyle 293 F cells were cultured in suspension in serum-free medium ( FreeStyle 293 Expression Medium , Thermo Fisher Scientific ) at 37°C in a humidified incubator with 5% CO2 in 125 mL shake flasks . Cells were diluted 1:2 with fresh culture medium and transfected at 106 cells/mL with 0 . 5 mg/L expression plasmid using the 293fectin transfection reagent ( Thermo Fisher Scientific ) . 72 hr post-transfection , cells were removed by centrifugation , and the His6-tagged recombinant Asp66Lys PfCyRPA variant was purified by immobilized metal ion affinity chromatography on a HisTrap HP column ( 1 mL volume; GE-Healthcare ) equilibrated with 50 mM Na-phosphate buffer pH 7 . 2 , 500 mM NaCl . After washing the column with the same buffer containing 20 mM imidazole , the protein was eluted ( isocratic elution ) with 500 mM imidazole over five column volumes . FreeStyle 293 F cells were tested and shown to be free of mycoplasma using MycoAlert Mycoplasma detection kit ( Lonza; LT07-318 ) . Identity of cells was confirmed using STR-PCR ( Qiagen; Investigator Idplex Pkus Kit , 381625 ) from genomic DNA purified using High Pure PCR Template Kit ( Roche Applied Science 11796828001 ) . The obtained profile was compared to the database available from DSMZ and found to match the HEK293 profile . 293 HEK cells ( ATCC , CRL-1573 ) expressing PfCyRPA fragments on the cell surface were generated essentially as described previously by Dreyer et al . ( 2012 ) . Briefly , DNA sequences coding for the fragments of PfCyRPA were amplified by PCR from the BVM_PFD1130W_FLAG_GP_His plasmid ( Dreyer et al . , 2010 ) . These expression vectors allow the anchoring of the protein of interest on the cell surface via the transmembrane domain of mouse glycophorin-A ( GP ) . In addition , they contain the secretion signal of bee-venom melittin ( BMV ) , a FLAG tag located extracellularly , and a His6-tag located in the cytosol . The 293 HEK cells were transfected with the different expression vectors using JetPEI transfection reagent ( PolyPlus ) according to the manufacturer’s instructions . Transient transfectants were harvested 48 hr post-transfection . HEK cell lysates were prepared at 107 cells/mL in RIPA-Buffer ( 1 % NP40 , 0 . 25% DOC , 10% glycerol , 2 mM EDTA , 137 mM NaCl , 20 mM Tris/HCl pH 8 . 0 , plus protease inhibitors ) and used for Western blot analysis as described in Favuzza et al . ( 2016 ) . 293 HEK cells were regularly tested by PCR with mycoplasma-specific primer GPO-1 ( 5'- ACTCCTACGGGAGGCAGCAGTA-3' ) and MGSO ( 5'-TGCACCATCTGTCACTCTGTTAACCTC-3' ) and shown to be mycoplasma-free . To locate the disulfide bonds in PfCyRPA , peptides derived from proteolysis under native conditions by the endo-proteases LysC , AspN , and trypsin were analyzed by UPLC-tandem mass spectrometry using a Dionex UltiMate3000 RRLC system ( Thermo Scientific ) coupled to a Synapt G2 HDMS mass spectrometer ( Waters ) with an electrospray ion source and working in resolution mode under default parameters . 10 µg PfCyRPA was hydrolyzed for several hours at 37°C with 0 . 3 µg of LysC , trypsin , or AspN . Peptides were separated on a Waters BEH130 C18 1 . 7 µm UPLC column ( 0 . 3 × 150 mm2 ) by a 5–45% gradient of acetonitrile in water containing 0 . 1% HCOOH at a flow rate of 10 µL/min over 90 min and eluted into the mass spectrometer . Data were analyzed using BiopharmaLynx software Ver . 1 . 3 ( Waters ) . The data derived from the proteolysis are summarized in Figure 3—figure supplement 2 . The secondary structure composition of PfCyRPA was estimated by far-UV circular dichroism ( CD ) spectroscopy on a Jasco J-815 spectropolarimeter using 10 µM PfCyRPA in phosphate buffered saline ( PBS ) at 20°C . A cell of 1 mm path length was used to record four spectra between 200 and 320 nm with a step size of 0 . 1 nm and an integration time of 1 s . The spectral average was corrected for buffer contributions . Intrinsic tryptophan fluorescence of a 5 μM PfCyRPA solution under native ( PBS ) and denaturing ( 70°C or PBS with 9 M urea ) conditions was measured on an ISS PC-1 photon counting spectrofluorimeter ( ISS , Inc . ) at 20°C . Fluorescence emission was excited at 295 nm with 4 . 8 nm band width and path length 1 mm . Emission was monitored in 2 nm steps between 300 and 400 nm with a band width of 16 nm , a path length 5 mm , and an integration time of and 1 s . All crystallization were done at 21°C in the sitting drop vapor diffusion setup using a Mosquito LCP crystallization robot ( TTP Labtech ) . If not stated otherwise crystals were cryo-protected with paraffin oil and vitrified in liquid N2 . A non-glycosylated variant of PfCyRPA ( residues Asp29–Glu362 , N145Q/N322Q/N338Q ) did not yield well-diffracting crystals unless pre-treated with proteases . 1 mg/mL PfCyRPA in 50 mM Tris/HCl pH 7 . 4 , 150 mM NaCl was diluted 1:1 with 10 mM HEPES/NaOH pH 7 . 5 , 500 mM NaCl , subjected for 16 hr to a panel of twelve proteases ( Hampton Research Proti-Ace Kit ) at a final concentration of 0 . 01 mg/mL , and analyzed by reducing SDS-PAGE ( Figure 4—figure supplement 1 ) . Of the proteases that changed the apparent molecular mass of PfCyRPA , subtilisin and proteinase K produced multiple bands , while Glu-C , elastase , and thermolysin produced a single band of slightly smaller apparent molecular mass . The thermolysin-treated PfCyRPA was subjected to crystallization , yielding triclinic crystals upon mixing of 120 nL 7 . 3 mg/mL PfCyRPA in 50 mM Tris/HCl pH 7 . 4 , 150 mM NaCl with 80 nL of precipitant consisting of 0 . 2 M LiOAc and 20% ( w/v ) PEG 3350 . These crystals are consistent with four molecules per unit cell but did not diffract X-rays beyond a resolution of 3 . 3 Å . Actinase E-treatment of PfCyRPA results in a different proteolysis pattern with two bands of apparent molecular mass 17 kDa and 20 kDa ( Figure 4—figure supplement 1 ) . Thin , plate-shaped monoclinic crystals diffracting to a maximum resolution of about 2 . 5 Å were obtained from mixing of 100 or 140 nL 15 mg/mL Actinase E-treated PfCyRPA in 50 mM HEPES/NaOH pH 7 . 5 , 150 mM NaCl with 100 nL and 60 nL precipitants consisting of 0–0 . 5 M MgCl2 and 20–25% PEG 3350 . Crystals of the same habit were also obtained from 7 . 6 mg/mL PfCyRPA mixed with 19% PEG 3350 , 0 . 3 M MgCl2 and additional 5 mM CaCl2 , SrCl2 , and BaCl2 . Data collected from crystals grown in the presence of the heavier cations were tested for anomalous and isomorphous information content for phasing , which turned out to be negative . Fab of mAb c12 crystallized from several conditions and its structure was determined in three crystal settings . Crystals were obtained by mixing of 80 nL 16 . 6 mg/mL c12 in 20 mM bis-Tris propane/HCl pH 7 . 0 , 200 mM NaCl with 120 nL of precipitants consisting of 1 . 6 M sodium citrate pH 6 . 5 ( hexagonal crystal form ) or 0 . 1 M HEPES/NaOH pH 7 . 0 , 20% w/v PEG 8000 ( monoclinic and orthorhombic crystal forms ) . Monoclinic crystals were cryo-protected with reservoir solution supplemented with 20% glycerol . Needle-shaped crystals of the PfCyRPA/c12 complex were obtained by mixing 120 nL of a 15 . 1 mg/mL solution in 50 mM Tris/HCl pH 7 . 4 , 150 mM NaCl with 80 nL reservoir consisting of 0 . 1 M Sodium citrate pH 5 . 5 , 17 % w/v PEG 5000 MME , 0 . 2 M NDSB-201 ( non-detergent sulfobetaine ) . Cryo-protection was achieved with reservoir solution supplemented with 20% ethylene glycol . Diffraction data were collected at Swiss Light Source beamline PX-II on a Pilatus 6M single photon counting detector using 1 Å radiation over a total range of at least 180° in fine slicing mode ( Δϕ = 0 . 25° ) . Data were indexed , integrated , and scaled with XDS ( Kabsch , 2010 ) , except for data from the hexagonal form of c12 , which was integrated with MOSFLM and scaled with AIMLESS ( The collaborative computational project number 4 , 1994 ) . The high-resolution limit was chosen as the shell where the correlation coefficient for half datasets CC1/2 dropped below 70% . Data sets were tested for internal symmetry by self-Patterson and self-rotation function analyses ( not shown ) . Twinning was excluded based on L-values and second moments . Data collection and also refinement statistics are collected in S1 Table . The presence of a short sequence motif in PfCyRPA that is characteristic for sialidases suggested that a six-bladed β-propeller might serve as a molecular replacement model . However , all attempts using several hundred β-propeller structures as search models both with and without loop regions and/or as poly-Ala models were unsuccessful . Sialidases often contain structural Ca2+ ions , which prompted co-crystallization of PfCyRPA with the earth alkali cations Ca2+ , Sr2+ , and Ba2+ for SIRAS phasing , but none of these provided useful derivatives . Attempts to phase the PfCyRPA diffraction data by sulfur-SAD were also unsuccessful . Packing density estimations indicated two molecules of PfCyRPA in the monoclinic asymmetric unit . The PfCyRPA data could be phased by molecular replacement using the unexplained electron density of the PfCyRPA/c12 complex ( see below ) as a search model . The volume of density was separated using a mask , placed in a large cubic cell , and back transformed to yield structure factor amplitudes , which were then used in a PHASER ( The collaborative computational project number 4 , 1994 ) molecular replacement search for two molecules . The solution had a non-random log-likelihood gain of LLG = 175 and the orientations of the two entities replicated the 180° two-fold NCS ( non-crystallographic symmetry ) estimated from the self-rotation function . Density averaging using PARROT ( The collaborative computational project number 4 , 1994 ) and the NCS operator defined by the two PHASER solutions resulted in manually interpretable electron density maps . The final model has a discontinuity at Asp189 , which is consistent with cleavage of a surface loop by Actinase E . A symmetry-related PfCyRPA occupies the space liberated by the cleaved loop , explaining the necessity of the proteolytic pre-treatment for crystal formation in this setting . The model was refined using automatically generated NCS restraints excluding diverging surface loop regions . 99 . 5% of all residues are in the favored regions of the Ramachandran plot . The hexagonal dataset of c12 ( Supplementary file 1 ) was phased by molecular replacement using a homology model generated from the c12 sequence in MOE ( Chemical Computing Group ) as the search model . Separate searches for the variable and constant parts of the Fab were performed in PHASER , which , as anticipated , placed the VHVL and CHCL domain boundaries close to each other to generate a complete Fab . The final model has 99 . 2% of all residues in the favored regions of the Ramachandran plot . The monoclinic and orthorhombic c12 datasets were phased using the refined hexagonal c12 structure as molecular replacement search model . Both structures have 98% of their main-chain torsion angles in the favored region of the Ramachandran plot . Elbow angles were calculated with PHENIX ( Adams et al . , 2010 ) . According to the Matthews parameter , the 4 Å dataset collected from the PfCyRPA/c12 crystal has enough space per asymmetric unit to harbor a single complex . Molecular replacement readily placed c12 , and initial maps calculated from this partial solution revealed additional electron density at the tips of the Fab ( Figure 5—figure supplement 1 ) , which occupied a volume large enough to host PfCyRPA . The density resembled a β-propeller but was not interpretable due to the limited resolution of the data . However , this density provided enough phasing power to solve the 2 . 5 Å PfCyRPA structure ( see above ) , showing that the placement of the Fab was correct . The same arrangement of VHVL and CHCL was obtained when using the VHVL and CHCL parts of the Fab as separate search models , indicating that the elbow angle of c12 does not change upon binding to PfCyRPA . The final model of PfCyRPA was used to complete the molecular replacement phasing of the PfCyRPA/c12 complex ( log-likelihood gain of >1260 ) , density for which is shown in Figure 5—figure supplement 2 . A loop region not present in the PfCyRPA model could be traced in the electron density of the complex ( Panel B in Figure 5—figure supplement 1 ) , establishing confidence in the correctness of the molecular replacement solution . The model of the complex was refined according to established protocols for lower resolution structures ( DeLaBarre and Brunger , 2006 ) , that is using simulated annealing , group ADP values , automatically assigned TLS definitions , secondary structure restraints , and external restraints provided by the higher resolution individual structures of PfCyRPA and c12 . No real-space refinement was applied during building and refinement of the PfCyRPA/c12 complex . Strong geometric restraints were applied throughout model building of the PfCyRPA/c12 complex . All models were built with COOT ( Emsley et al . , 2010 ) and refined with PHENIX ( Adams et al . , 2010 ) . | Malaria is one of the deadliest infectious diseases worldwide , killing over 400 , 000 people a year . About 200 million people are infected every year , placing a huge social and medical burden especially on developing countries . Microscopic parasites known as Plasmodium are responsible for causing this disease . Plasmodium parasites have a complex life cycle involving both mosquito and mammal hosts . This includes a stage where the parasites infect the mammal’s red blood cells , which causes the symptoms of the disease . In 2012 , a team of researchers discovered that a protein called CyRPA forms a group ( or ‘complex’ ) with several other proteins to allow the parasites to enter red blood cells . Developing a vaccine is one of the most promising approaches to prevent malaria . Vaccines help the body to recognise and fight an invading microbe by triggering an immune response that results in the production of proteins called antibodies , which can bind to specific molecules on the surface of the microbe . If the microbe later enters the body , these antibodies can be produced quickly to eliminate the microbe before it causes disease . However , efforts to develop a highly effective vaccine against malaria have so far been unsuccessful . Favuzza et al . – including some of the researchers involved in the 2012 work – used a technique called X-ray crystallography to investigate the three-dimensional structure of the CyRPA protein . The experiments show that an antibody is able to bind to a region of CyRPA – a designated ‘protective epitope’ – that is similar in the CyRPA proteins of all Plasmodium falciparum strains . These antibodies can prevent the parasite from entering the red blood cells , and vaccines containing CyRPA may therefore be effective at protecting individuals from malaria . The findings of Favuzza et al . also suggest that using CyRPA in combination with another protein in the complex called RH5 could make the vaccine more powerful as it would make it harder for the parasite to become resistant . The next step following on from this work is to design a vaccine containing protective CyRPA epitopes that triggers an immune response in mammals that is strong enough to reduce the numbers of parasites in the blood . A future challenge will be to develop a vaccine that combines several proteins involved in different stages of the parasite’s life cycle to provide full protection against malaria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2017 | Structure of the malaria vaccine candidate antigen CyRPA and its complex with a parasite invasion inhibitory antibody |
Exposure to early-life adversity ( ELA ) increases the risk for psychopathologies associated with amygdala-prefrontal cortex ( PFC ) circuits . While sex differences in vulnerability have been identified with a clear need for individualized intervention strategies , the neurobiological substrates of ELA-attributable differences remain unknown due to a paucity of translational investigations taking both development and sex into account . Male and female rats exposed to maternal separation ELA were analyzed with anterograde tracing from basolateral amygdala ( BLA ) to PFC to identify sex-specific innervation trajectories through juvenility ( PD28 ) and adolescence ( PD38;PD48 ) . Resting-state functional connectivity ( rsFC ) was assessed longitudinally ( PD28;PD48 ) in a separate cohort . All measures were related to anxiety-like behavior . ELA-exposed rats showed precocial maturation of BLA-PFC innervation , with females affected earlier than males . ELA also disrupted maturation of female rsFC , with enduring relationships between rsFC and anxiety-like behavior . This study is the first providing both anatomical and functional evidence for sex- and experience-dependent corticolimbic development .
Exposure to early life adversity ( ELA ) increases vulnerability to various psychiatric disorders across the lifespan ( Smyke et al . , 2007; McEwen , 2008; Maccari et al . , 2014; Callaghan and Tottenham , 2016; Hane and Fox , 2016; Krugers et al . , 2017 ) . Males and females appear to be affected differently by ELA , with females more prone to developing disorders including anxiety and depression ( Hammen et al . , 2000; Heim et al . , 2008; Davis and Pfaff , 2014 ) . Importantly , these sequelae often emerge later in childhood or adolescence , providing an opportunistic window for intervention before psychopathology takes hold . Thus , development of effective intervention strategies requires biological and developmental targets specific to individuals based on factors including sex and timing of stress exposure ( Lupien et al . , 2009 ) . Evidence suggests that ELA in humans leads to life-long changes in connectivity and/or functionality of limbic and cortical regions ( Choi et al . , 2009; VanTieghem and Tottenham , 2018 ) , with consequential deficits in emotion regulation and cognition ( Tyrka et al . , 2013 ) . Both human and animal studies highlight the importance of corticolimbic circuitry in affective behavior regulation , its disruption in mental disorders ( Herringa et al . , 2013; Bangasser and Valentino , 2014; Killgore et al . , 2014 ) , and alterations directly related to ELA ( Kaiser et al . , 2018 ) . Specifically , Tottenham and colleagues illustrated that normative developmental changes in task-based functional connectivity ( FC ) between amygdala and medial prefrontal cortex ( mPFC ) are accelerated after ELA with associated changes in anxiety ( Gee et al . , 2013a ) and are particularly evident in females ( Dickie and Armony , 2008 ) . Indeed , children institutionalized during the first two years of life in orphanages display precocial development of amygdala-mPFC FC , though the anatomical substrates for this accelerated connectivity remain unknown . In rats , the basolateral amygdala ( BLA ) sends inputs to the mPFC and modulates anxiety-related behaviors ( Felix-Ortiz et al . , 2016 ) , recall of emotionally salient information ( McGaugh , 2004 ) , decision-making ( St Onge et al . , 2012 ) , and goal-directed behavior ( Schoenbaum et al . , 2000 ) . Notably , two subpopulations of BLA neurons project to different regions of the mPFC ( Senn et al . , 2014 ) : projections to the dorsal ( prelimbic; PL ) region of the mPFC are active during threat-associated fear learning and expression , whereas projections from the BLA to the ventral ( infralimbic; IL ) region of the mPFC are active upon extinction of fear , or learning about safety signals . In typically developing male rats , BLA innervation of mPFC increases through adolescence ( Cunningham et al . , 2002 ) , likely contributing to healthy maturation of threat and safety appraisal . Notably , BLA-mPFC axonal innervation has not yet been evaluated in females , nor has any study examined the effects of ELA on BLA-derived innervation of PL or IL . Evidence in both the human and rodent literature supports the role of amygdala-PFC circuitry in the regulation of anxiety-like behaviors ( Likhtik et al . , 2014; Arruda-Carvalho and Clem , 2015; Fragale et al . , 2016 ) . However , the neurobiological mechanisms are ill-defined and therefore require investment in translational animal research to pinpoint neural contributions to affective pathology ( Shackman and Wager , 2019 ) . Emerging evidence in rodents suggests that perturbations of this circuit – particularly following various models of ELA – may be implicated in maladaptive maintenance of anxiety-like behaviors ( Chocyk et al . , 2013; Maccari et al . , 2014; Krugers et al . , 2017 ) . ELA via chronic stress has been shown to increase BLA-derived glutamatergic release within the PFC , a finding that can be recapitulated via BLA-PFC stimulation in typical mice and is associated with increased anxiety-like behavior ( Lowery-Gionta et al . , 2018 ) . This is in line with reports of altered excitatory:inhibitory balance within the BLA-PFC circuit that is associated with changes in affective behaviors ( Arruda-Carvalho and Clem , 2014 ) . For example , excitatory latencies of PFC neurons to amygdalar stimulation in adult rats were significantly longer in animals with a history of ELA ( Ishikawa et al . , 2015 ) . It is likely that ELA-induced changes in plasticity lead to increased BLA-derived excitatory signaling into the PFC , with insufficient reciprocal PFC-driven anxiolytic signals returning to the BLA . However , there is presently insufficient evidence to determine the neuroanatomical time-course of this developing circuit , and a lack of evidence evaluating its modulation by additional factors ( i . e . sex , development ) . Clinical evidence points to striking sex differences in the clinical time-course and symptomology of ELA effects ( Wainwright and Surtees , 2002; Martin et al . , 2014 ) . Moreover , several studies in animals over the past two decades have revealed sex-specific effects of ELA on anxiety-like behaviors ( Tractenberg et al . , 2016; Bonapersona et al . , 2019b ) and adolescent or adult corticolimbic measures ( Salzberg et al . , 2007; Holland et al . , 2014; Farrell et al . , 2016; Blaze and Roth , 2017; Bonapersona et al . , 2019a ) , however little is known about the interaction of sex and ELA on corticolimbic development . Notably , typically developing females display earlier maturation of the PFC ( Lenroot et al . , 2007; Lenroot and Giedd , 2010 ) . Identifying how ELA affects these sex-dependent trajectories is crucial to understanding sex differences in vulnerability and to developing individually targeted intervention strategies . Therefore , we used anterograde tracing to examine ELA effects on BLA-PFC innervation over development in male and female rats . We hypothesized that if heightened anxiety-like behaviors following ELA are associated with increased BLA-derived PFC innervation , then rats exposed to ELA via repeated isolation from dam and littermates will display anxiety-like behavior and increased BLA-PFC innervation that will be more robust in females . While task-based FC illustrates coordinated responsivity to anxiety-provoking stimuli ( Gee et al . , 2013a ) , resting-state FC ( rsFC ) is excellent for probing the functional integrity of the amygdala-PFC circuit independent of task demands ( Thomason et al . , 2011a; Thomason et al . , 2011b; Alarcón et al . , 2015; Gabard-Durnam et al . , 2016 ) . ELA effects on corticolimbic rsFC in humans are inconsistent , likely because of reliance on autobiographical questionnaires , different ages of measurement , and different ELA criteria . In rats , maternal separation results in early emergence of both adult-like fear learning based in fronto-amygdala circuitry ( Callaghan and Richardson , 2011 ) and early amygdala structural maturation ( Ono et al . , 2008 ) . rsFC has recently been utilized in rats to characterize ELA ( via limited bedding ) induced changes in BLA-PFC connectivity in preweanling males , revealing an association between decreased rsFC and adult fear behaviors ( Guadagno et al . , 2018 ) . Early or blunted maturation of corticolimbic connectivity likely has deleterious consequences because a sufficient degree of PFC immaturity during juvenility is critical for learning anxiolytic cues ( safety signals ) in adulthood ( Yang et al . , 2012 ) . Therefore , we provide a back-translation to examine whether ELA-exposed rats display accelerated maturation of amygdala-PFC connectivity to parallel humans with a history of adversity , with increased BLA-PFC innervation as a potential anatomical substrate driving sex-specific developmental effects . This multi-level approach aims to elucidate the neurobiological underpinnings of ELA-attributable vulnerability through the use of mechanistic neural tracing paired with a translational imaging investigation: an approach which enhances our ability to develop a cross-species understanding of ELA-associated pathology ( Fox and Shackman , 2019 ) .
In order to chart the trajectory of innervation from the BLA to the PFC , the anterograde tracer biotinylated dextran amine ( BDA ) was injected into the BLA of ELA-exposed and control ( CON ) males and females at PD21 , PD31 , or PD41 . Seven days following surgery ( PD28 , PD38 , or PD48 ) , animals were assessed for anxiety-like behavior in the elevated plus maze ( EPM ) and then sacrificed for quantification of BLA axonal terminals in the PFC . Results from all analysis of correlation between behavior on the EPM and connectivity measures are shown in Table 1 . Here we discuss significant correlations , which are illustrated in Figure 4A–D . Fisher’s r- to- z transformations revealed an impact of sex , but not rearing , on the strength of some relationships between innervation and behavior ( Table 1 ) . At PD28 , higher IL innervation in females was correlated with less time spent in the open arms ( R2 ( 14 ) =0 . 478; p=0 . 009 ) ( Figure 4B ) . At PD38 , males showed a similar relationship between less time in the open arms and higher PL innervation ( R2 ( 13 ) =0 . 266; p=0 . 049 ) ( Figure 4C ) and higher IL innervation ( R2 ( 13 ) =0 . 260; p=0 . 05 ) ( Figure 4D ) . No relationships were seen at PD48 ( Table 1 ) . A separate cohort of CON and ELA-exposed males and female rats were subjected to anxiety-like behavioral assessment in the EPM , as well as MRI scanning , at both PD28 and PD38 using a longitudinal design . rsFC was analyzed with the BLA as a seed , and correlation coefficients between the BLA and either PL or IL were compared between rearing groups and age within each sex . Results from all regression analyses of the relationship between behavior on the EPM and functional connectivity are shown in Tables 2 and 3 . Fisher’s r- to- z transformations revealed an impact of both sex and rearing on relationships between rsFC and behavior at PD28 . Significant correlations revealed that in PD28 females exposed to ELA , lower BLA-IL rsFC predicted less time spent in the open arms ( R2 ( 8 ) =0 . 852; p=0 . 001 ) ( Figure 6A ) ; notably , this relationship is juxtaposed with our finding that higher axonal innervation at the same age predicted less time spent in the open arms in all females ( see Figure 4B ) . Our longitudinal design further allowed the analysis of early rsFC with later adolescent behavior . In all females , lower BLA-IL rsFC at PD28 also predicted less time spent in the open arms twenty days later at PD48 ( R2 ( 7 ) =0 . 405; p=0 . 011 ) , suggesting an enduring and predictive relationship in females ( Figure 6B ) . This enduring relationship in females was also seen in a significant relationship between lower BLA-IL rsFC at PD48 and less time in the open arms at the same age ( Figure 6C; R2 ( 13 ) =0 . 303; p=0 . 041 ) .
This work identifies sex-specific neuroanatomical development and corresponding functional maturation of the BLA-PFC circuit following ELA . The data reveal newly uncovered aberrations to PFC innervation that can be interpreted in the context of atypical behavior and FC that has been observed in humans ( Gee et al . , 2013a; Philip et al . , 2013; Thomason et al . , 2015; Teicher et al . , 2016 ) and in animals ( Yan et al . , 2017; Johnson et al . , 2018 ) . These findings are the first description of sex-specific developmental trajectories following ELA , using a juxtaposition of rsFC as assessed in humans with an anatomically discrete measurement of monosynaptic innervation . We observed sex- and age-dependent effects on BLA innervation in PL and IL regions of the PFC following ELA in a rat model of caregiver deprivation . It is important to note here , especially given the current need for more transparency in reporting of ELA paradigms ( Kentner et al . , 2019; Brenhouse and Bath , 2019 ) that pregnant dams were shipped to our facility , therefore all animals in our study underwent shipping stress during gestation . It is thus possible that effects of maternal separation reported here are a consequence of both prenatal and postnatal stress , which is common amongst human instances of early life stress ( Pedersen et al . , 2018 ) . Our findings suggest that females may be particularly vulnerable to neuroanatomical consequences of early adversity , with innervation effects seen earlier in females compared to a later effect observed in males . This is in line with human studies describing sex-dependent effects of early adversity where female participants appear to exhibit more severe adolescent and later-life consequences , especially with regard to affective disorders ( Humphreys et al . , 2015 ) . Rodent studies investigating potential sex differences have also largely revealed female-specific increases in adolescent anxiety-like behavior following ELA ( e . g . Salberg et al . , 2019; Manzano-Nieves et al . , 2018; Jin et al . , 2018; Viola et al . , 2019; but see Bonapersona et al . ( 2019b ) for further examination ) , though some studies have failed to show any anxiety-like effects in adolescence ( e . g . , Doherty et al . , 2017 ) . Little , however , has been known to date regarding biological substrates for sex-specific consequences on anxiety-like behaviors . Our data indicate that BLA-IL innervation increases through adolescence in a bilaminar manner across development , with more dramatic increases following ELA in both males and females . Importantly , these main effects of age support previous work in male rats ( Cunningham et al . , 2002 ) and now characterize a similar trajectory in female rats . Interestingly , increased innervation to the IL in PD28 females and PD38 males occurred in IL2 , whereas previous work has shown that in wild-type adult mice , the BLA preferentially targets amygdala-projecting neurons in IL5 over IL2 ( Cheriyan et al . , 2016 ) . Since IL2 projection neurons do also project back to the BLA ( Gabbott et al . , 2005; Little and Carter , 2013 ) , it is possible that atypical hyper-innervation after maternal separation aberrantly targets layer 2 , which may drive the altered maturation of BLA-IL functional connectivity we observed in the present work . While IL innervation showed a developmental increase through adolescence , PL innervation appeared to reach adult-like levels in both sexes earlier than was previously captured ( Van Eden and Uylings , 1985; Bouwmeester et al . , 2002 ) . Following ELA , however , our data indicate that aberrant effects on innervation were apparent at PD28 in females , but not until PD38 in males; this finding was observed across both the PL and IL . However , while ELA conferred a similar age-dependent influx of BLA axonal innervation in both regions , the long-term neuroanatomical alterations were different depending on the region examined . In IL there was evidence of precocial maturation , such that ELA induced BLA-IL innervation that was comparable to more mature ( PD48 ) patterns in female juveniles ( PD28 ) and male early adolescents ( PD38 ) . Interestingly , in PL we saw a transient effect of ELA that was age- and sex-dependent . Indeed , within the PL we observed that both male and female rats showed an unexpected transient spike in innervation that appeared to resolve at the following developmental timepoint . It is possible that reversal of the innervation seen at earlier time points may be due to pruning mechanisms ( Koss et al . , 2014 ) that typically occur in adolescence , and these mechanisms may serve to mediate excess innervation ( Rakic et al . , 1994; Spear , 2000; Riccomagno and Kolodkin , 2015 ) . Prior work has shown a peak in BLA-PL connectivity , specifically at PD30 ( Pattwell et al . , 2016 ) , that may contribute to the present findings of transient ELA-exacerbated hyper-innervation . However , this does not explain why innervation to the PL spiked at PD28 , was reduced to CON levels at PD38 , and then spiked again by PD48 in female ELA . It is possible that the temporary ( one week ) isolation following surgery may have acted as a secondary stressor acting in conjunction with pubertal changes to produce a resurgence of this ELA-specific phenotype ( Andersen and Teicher , 2008; Tzanoulinou and Sandi , 2016 ) . Relatedly , surgery itself during peripubertal development compared to other time periods may have differentially interacted with ELA . Importantly , the ages at which ELA-exposed females and males first displayed increased BLA-IL innervation ( PD28 and PD38 , respectively ) were the ages at which higher innervation correlated with higher anxiety-like behavior ( less time spent in open arms of the EPM ) . Since monosynaptic input from the BLA to the medial PFC drives aversion to anxiogenic stimuli as measured in the EPM ( Felix-Ortiz et al . , 2016 ) , these data suggest that hypertrophic effects of ELA at the BLA-IL circuit lead to heightened anxiety-like behavior in ELA-exposed animals . However , we were surprised that group-wise comparisons only showed higher anxiety-like behaviors in ELA females at PD38 , but not at PD28 when innervation was increased ( see Figure 3 ) . Group comparisons also failed to reveal higher anxiety-like behaviors in ELA-exposed males at any age . Our group and others have previously observed increased anxiety-like behaviors in male and female rats following ELA ( e . g . , Holland et al . , 2014; Ganguly et al . , 2015; Jin et al . , 2018 ) , therefore further work will reveal whether more sensitive assays can more reliably demonstrate anxiety-like behavior in ELA-exposed rats . Here we demonstrate that , regardless of mean differences , ELA experience can forge a relationship between heightened innervation and anxiety-like behavior at distinct developmental time-points that could suggest mechanistic ties between early hypertrophy of inputs and behavior . Transient effects of ELA on innervation and behavior are notable because adolescent perturbations have been shown to have lasting consequences on later life function . Multiple lines of evidence from both clinical and animal studies suggest that physiological or experiential anomalies during critical periods of development can program the central nervous system for susceptibility or resilience to future environments ( Andersen , 2003; Nederhof and Schmidt , 2012 ) . One example is seen in an animal model utilizing prenatal treatment with the mitotoxin MAM , which produces adolescent-specific corticolimbic and mesolimbic dysfunctions that drive a hyperresponsivity to stress with anxiety-like and psychosis-like behavior . If conversion to the dysfunctional phenotype in adolescence is prevented by relieving stress in adolescence , MAM-treated animals will still be more susceptible to affective dysfunction in adulthood ( Gomes et al . , 2019 ) . Moreover , ELA reportedly leads to adolescent alterations in PFC NMDA receptor subunit composition that regulate anxiety-like behaviors in males ( Ganguly et al . , 2015 ) ; future work will reveal whether increased glutamatergic input from the BLA drives these receptor changes in the PFC . Since the PFC can serve reciprocally as a modulator of BLA activity ( Rosenkranz and Grace , 2002 ) , it is therefore possible that transient innervation changes lead to long-term receptor alterations and subsequent changes to the efficiency of BLA-PFC functional connectivity . Indeed , we observed an effect of rearing in female anxiety-like behavior and functional connectivity after the age when innervation to the PL and IL2 was increased ( PD28 ) . Together , the early increases in innervation after ELA reported here may perturb normal critical period maturation , leading to late-onset effects on functional connectivity and affective dysregulation . While the current work did not directly address the causal relationship between innervation and rsFC , results from the rsFC analyses in females supported the idea that early maturation of BLA-IL innervation disrupts later functional relationships between these regions ( Tottenham and Galván , 2016 ) . Specifically , we report here that females exposed to ELA displayed dampened maturation of BLA-IL rsFC compared to female CON; while no effects of rearing were apparent at PD28 , ELA female rats failed to display the increase in BLA-IL rsFC by PD48 that was observed in female CON . This is in line with previous work showing reduced BLA-mPFC connectivity following a limited bedding model ( Guadagno et al . , 2018 ) . Interestingly , increased [more mature] rsFC BLA-IL connectivity in female ELA at PD28 correlated with decreased anxiety-like behavior at that age , as well as 20 days later . This suggests that more mature connectivity in female juveniles may confer enduring behavioral resilience , which is consistent with previous reports that dampened connectivity is associated with increased anxiety in adolescence ( Kim et al . , 2011; Nooner et al . , 2013 ) . While ELA exposure in males appeared to result in lower juvenile ( PD28 ) rsFC between the BLA and the PL , no effects on the magnitude of change between PD28-PD48 were noted , therefore altered maturation per se was not observed in males . Interestingly , ELA affected male BLA-PL rsFC at an age that preceded any ELA-attributable increase in innervation , suggesting that the reciprocal connectivity between the two regions was altered without the influence of aberrant amygdalofugal innervation . ELA experience resulted in dampened maturation of BLA-IL rsFC in females , in contrast to the precocial maturation we observed in axonal innervation of the IL from the BLA , as well as the accelerated maturation of task-based FC reported following childhood maltreatment in humans ( Gee et al . , 2013a ) . While BLA-mPFC rsFC in humans has been observed to increase towards more positive connectivity through adolescent development ( Gabard-Durnam et al . , 2014 ) , task-based FC during fearful face presentations declines from positive connectivity to a negative connectivity ( Gee et al . , 2013b ) . This task-based FC was further found to reach mature ( more negative ) levels in previously institutionalized children ( Gee et al . , 2013a ) . In contrast , girls who express higher basal cortisol levels at 4 . 5 years display lower rsFC between the ventromedial PFC and the amygdala at 18 years ( Burghy et al . , 2012 ) , corroborating our current findings in PD48 females ( see Figure 5 ) . rsFC reflects an intrinsic alternate resonance between different brain areas connected at large scale , in contrast to the more isolated activation of the corresponding brain areas during a task ( Rasero et al . , 2018 ) . Therefore , it appears that early hyperinnervation to the PFC may alter the stability of the functional BLA-PFC network as it develops , leading to increased anxiety when innervation arises and disrupted maturation of rsFC; though how these effects in rats relate to task-based FC is currently unknown . It is also possible that early maturation of BLA-PFC projections may overwhelm the slow-maturing reciprocal projections to the BLA that dampen anxiety-related circuit activity in mature brains ( Arruda-Carvalho et al . , 2017; Selleck et al . , 2018 ) . The present findings support previous work indicating the distinct functional relationship between the BLA and PL and IL regions of mPFC ( Calhoon and Tye , 2015 ) , with PL connectivity related to anxiogenic effects ( Felix-Ortiz et al . , 2016 ) and IL connectivity promoting anxiolytic effects ( Maroun et al . , 2012 ) . Furthermore , these findings are in line with the idea that altered neuroanatomy and functionality of PL and IL , along with BLA , likely have reciprocal effects on one another , further contributing to the exacerbation of ELA-induced anxiety-like phenotypes in adolescence and early adulthood ( Likhtik and Paz , 2015; Zimmermann et al . , 2019 ) . Taken together , maternal separation in rats was found to disrupt normative development of both anatomical ( axonal innervation ) and functional ( rsFC ) connectivity within a circuit regulating emotional processing , with sex-specific effects and evidence of resilience in individuals with precocial maturation of rsFC . Future work will investigate the physiological consequences of increased innervation that underlies rsFC and behavioral effects , as well as the potential impact of puberty on increased innervation , since females also reportedly display accelerated puberty initiation following ELA ( Grassi-Oliveira et al . , 2016 ) . The present findings have implications for intervention based on experience , age , and sex – three functionally interactive factors that uniquely define risk in every individual .
For Studies 1 ( BLA-mPFC innervation ) and 2 ( rsFC ) , timed-pregnant Sprague-Dawley rats ( Charles River , Wilmington , MA ) arrived at gestational day 15 . Rats were housed under standard laboratory conditions in a 12 hr light/dark cycle ( lights on at 0700 hr ) temperature- and humidity-controlled vivarium with access to food and water ad libitum . Following birth ( postnatal day [PD]0 ) , litters were randomly assigned as either: CON , and left undisturbed with the exception of cage-changing twice/week and weighing ( PD9 , 11 , 15 , 20 ) ; or ELA via maternal and peer separation , as described previously ( Coley et al . , 2019; Farrell et al . , 2016; Ganguly et al . , 2019; Grassi-Oliveira et al . , 2016; Wieck et al . , 2013 ) and below . In preliminary power analyses , we achieved with group sizes of 7–9 subjects effect sizes of η2 = 0 . 08–0 . 27 and power of 1-β=0 . 80–0 . 98 for piloted anterograde tracing data , and effect sizes of η2 = ~ 0 . 12 and power of ~1-β=0 . 80 with group sizes of 22 subjects in behavior studies , therefore we aimed for n = 10 for all studies involving innervation , allowing behavioral studies to include animals that did not reach criteria for inclusion in innervation analyses in Study 1 . Group sizes were chosen for Study two based on preliminary studies showing that BOLD data from 7 to 8 rats/group yielded Cohen’s D values of 1 . 2–2 . 5 . On PD1 litters were culled to 10 ( + /- 2 ) pups , maintaining equal ratio of male and female whenever possible , with one rat per litter assigned to each experimental group ( i . e . age and sex ) to avoid litter effects . Pups were assigned pseudo-randomly to experimental groups , where an investigator not involved in the study assigned a number to each pup in a litter , and a separate investigator assigned one number from each sex to an experimental group . All measurements were conducted by experimenters blinded to experimental condition through coding of microscopic slides , video recordings of behavior , and rsFC files . Throughout all experiments , outliers were identified when subjects were visibly agitated or otherwise behaving atypically outside of experimentation and were excluded from all analyses . ELA pups were separated from dams and littermates in individual cups with home cage pine shavings in a circulating water bath ( 37°C ) from PD2-PD10 . At PD11-20 , when body temperature is self-regulated , pups were individually separated into cages . Pups were separated for 4 hr each day ( 0900 h-1300h ) during which time pups were deprived of maternal and littermate tactile stimulation and nursing , but not from maternal odor . ELA dams remained in their home cages but were deprived of their entire litters during separations . Pups were weaned at PD21 into same-sex mixed-litter pairs and left undisturbed until surgery/behavioral assessment for Study 1 ( either PD21/28 , 31/38 , or 41/48; Figure 7A ) . Separate cohorts were used for Study 2 – with treatment identical to Study 1 – and left undisturbed until behavioral assessment and rsFC ( PD28 , PD48 ) , with subjects imaged at both ages . Experiments were performed in accordance with the 1996 Guide for the Care and Use of Laboratory Animals ( NIH ) and with approval from Northeastern University’s Institutional Animal Care and Use Committee . MRI scanning was performed on male and female CON and ELA subjects at PD28 and PD48 in a Bruker BioSpec 7 . 0T/20 cm USR horizontal magnet ( Bruker , Billerica , MA ) with a 20 G/cm magnetic field gradient insert ( ID = 12 cm ) capable of 120µs rise time . Rats were anesthetized and maintained at 1–2% isoflurane and oxygen with a flow rate of 1 L/min throughout scanning , with breathing rate ( 40–50 breaths per minute ) carefully monitored by an investigator and anesthetic levels adjusted accordingly . Rats were scanned at 300 MHz using a quadrature transmit/receive volume coil built into the rat head holder and restraint system ( Animal Imaging Research , Holden , MA ) . rsFC was acquired by gradient-echo triple-shot echo-planar imaging ( EPI ) pulse sequence with the following parameters: matrix size = 96×96 × 20; repetition time ( TR ) /echo time ( TE ) = 3000/15msec; voxel size = 0 . 312×0 . 312 × 1 . 2 mm; slice thickness = 1 mm; volume = 200 . T2-weighted high-resolution anatomical scans were conducted using RARE pulse sequence with imaging parameters as follows: matrix size = 256×256 × 20; TR/TE = 4369/12msec; voxel size = 0 . 117×0 . 117 × 1 mm; slice thickness = 1 mm . | Having a traumatic childhood increases the risk a person will develop anxiety disorders later in life . Early life adversity affects men and women differently , but scientists do not yet know why . Learning more could help scientists develop better ways to prevent or treat anxiety disorders in men and women who experienced childhood trauma . Anxiety occurs when threat-detecting brain circuits turn on . These circuits begin working in infancy , and during childhood and adolescence , experiences shape the brain to hone the body’s responses to perceived threats . Two areas of the brain that are important hubs for anxiety-related brain circuits include the basolateral amygdala ( BLA ) and the prefrontal cortex ( PFC ) . Now , Honeycutt et al . show that rats that experience early life adversity develop stronger connections between the BLA and PFC , and these changes occur earlier in female rats . In the experiments , one group of rats was repeatedly separated from their mothers and littermates ( an early life trauma ) , while a second group was not . Honeycutt et al . examined the connections between the BLA and PFC in the two groups at three different time periods during their development: the juvenile stage , early adolescence , and late adolescence . The experiments showed stronger connections between the BLA and PFC begin to appear earlier in juvenile traumatized female rats . But these changes did not appear in their male counterparts until adolescence . Lastly , the rats that developed these strengthened BLA-PFC connections also behaved more anxiously later in life . This may mean that the ideal timing for interventions may be different for males and females . More work is needed to see if these results translate to humans and then to find the best times and methods to help people who experienced childhood trauma . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2020 | Altered corticolimbic connectivity reveals sex-specific adolescent outcomes in a rat model of early life adversity |
The localisation of oskar mRNA to the posterior of the Drosophila oocyte defines where the abdomen and germ cells form in the embryo . Kinesin 1 transports oskar mRNA to the oocyte posterior along a polarised microtubule cytoskeleton that grows from non-centrosomal microtubule organising centres ( ncMTOCs ) along the anterior/lateral cortex . Here , we show that the formation of this polarised microtubule network also requires the posterior regulation of microtubule growth . A missense mutation in the dynactin Arp1 subunit causes most oskar mRNA to localise in the posterior cytoplasm rather than cortically . oskar mRNA transport and anchoring are normal in this mutant , but the microtubules fail to reach the posterior pole . Thus , dynactin acts as an anti-catastrophe factor that extends microtubule growth posteriorly . Kinesin 1 transports dynactin to the oocyte posterior , creating a positive feedback loop that increases the length and persistence of the posterior microtubules that deliver oskar mRNA to the cortex .
Although most tissue culture cells organise a radial array of microtubules from their centrosomes , most differentiated cell-types lose or inactivate their centrosomes , but still create polarised microtubule arrays that play important roles in the establishment of cell polarity , intracellular trafficking and organising the internal architecture of the cell ( Bartolini and Gundersen , 2006 ) . For example , both Drosophila and vertebrate neurons polarise normally without functional centrosomes , and the latter can even regenerate their axons after centrosome ablation ( Stiess et al . , 2010; Nguyen et al . , 2011 ) . Thus , specialised cells , such as neurons , must use other mechanisms to nucleate microtubules and to organise them into the polarised microtubule arrays that underlie cell function . Non-centrosomal microtubules play a particularly important role in the Drosophila oocyte , where they form a polarised network at stage 9 of oogenesis that directs the localisation of the maternal determinants that define the anterior-posterior axis of the embryo ( Bastock and St Johnston , 2008 ) . The organisation of the oocyte microtubules ultimately depends on an unknown signal from the follicle cells that induces the formation of complementary cortical polarity domains: an anterior/lateral domain that is defined by the localisation of the Bazooka/Par-6/aPKC complex and a posterior domain that is marked by Par-1 ( González-Reyes et al . , 1995; Roth et al . , 1995; Shulman et al . , 2000; Tomancak et al . , 2000; Doerflinger et al . , 2006 ) . Par-1 then acts to exclude noncentrosomal microtubule organising centres from the posterior , so that the majority of microtubules grow with their minus ends anchored to the anterior/lateral cortex ( Doerflinger et al . , 2010; Nashchekin et al . , 2016 ) . This results in the formation of an anterior-posterior gradient of microtubules in the oocyte , with a weak orientation bias of 60% of the microtubules growing towards the posterior and 40% towards the anterior ( Parton et al . , 2011 ) . One of the key functions of the oocyte microtubule cytoskeleton is to direct the transport of oskar mRNA to the posterior of the oocyte ( Figure 1A ) , where it defines the site of assembly of the pole plasm , which contains the abdominal and germline determinants ( Ephrussi et al . , 1991; Kim-Ha et al . , 1991; Ephrussi and Lehmann , 1992 ) . oskar mRNA is transcribed in the nurse cells and is then transported along microtubules into the oocyte by the minus-end directed motor , dynein ( Clark et al . , 2007; Sanghavi et al . , 2013 ) . Once it enters the oocyte , oskar mRNA switches motors and is transported by the plus end directed motor protein , kinesin 1 , to the posterior pole , where it is translated and anchored to the cortex by Oskar protein ( Markussen et al . , 1995; Rongo et al . , 1995; Brendza et al . , 2000; Vanzo and Ephrussi , 2002 ) . Tracking of oskar mRNA particles in the oocyte reveals that they move in all directions with a similar net posterior bias to the growing plus ends of microtubules ( Zimyanin et al . , 2008; Parton et al . , 2011 ) . This suggests that oskar mRNA is a passive cargo of kinesin 1 and that its destination is determined solely by the arrangement of the microtubule cytoskeleton . In support of this view , computer simulations show that simply allowing microtubules to extend in random directions from ncMTOCs along the anterior and lateral cortex , but not the posterior , is sufficient to give the observed microtubule distribution in the oocyte and to account for the posterior localisation of oskar mRNA by kinesin 1 ( Khuc Trong et al . , 2015 ) . The computer simulations treat microtubules as static rods , and do not take into account the dynamic nature of the microtubules in the oocyte , in which the growing plus ends undergo catastrophes when they transition into rapid shrinking , especially on hitting solid surfaces , such as the cortex ( Zhao et al . , 2012 ) . Here , we show that just controlling the distribution of microtubule minus ends is not enough to explain the delivery of oskar mRNA to the posterior cortex when the microtubules are dynamic , and describe an additional layer of control , in which dynactin regulates the microtubule plus ends to increase their persistence specifically at the posterior of the oocyte .
We have previously identified factors that regulate oskar mRNA localisation by performing large genetic screens in germline clones for mutants that disrupt the posterior localisation of GFP-Staufen , a dsRNA-binding protein that associates with oskar RNA throughout oogenesis ( St Johnston et al . , 1991; Ramos et al . , 2000; Martin et al . , 2003 ) . One mutant from this screen ( 4D2 ) gave an unusual phenotype , in which both Staufen and oskar mRNA accumulate in the posterior cytoplasm rather than forming a tight crescent at the posterior cortex ( Figure 1B–E ) . This phenotype could reflect a failure to anchor oskar mRNA to the cortex or a defect in oskar mRNA translation , since Oskar protein anchors its own RNA ( Vanzo and Ephrussi , 2002 ) . Staining for Oskar protein revealed that the small amount of mRNA that is at the posterior cortex is translated , whereas the diffusely-localised cytoplasmic RNA is not ( Figure 2A and B ) . Furthermore , the Staufen and Oskar protein that contact posterior cortex remain anchored there at later stages , whereas the higher concentration of protein in the posterior cytoplasm disappears , presumably because it is washed away by cytoplasmic flows ( Figure 2C and D ) . Thus , the phenotype does not appear to be a consequence of an anchoring or translation defect , suggesting that the 4D2 mutation disrupts the delivery of oskar mRNPs to the posterior cortex . To test whether the 4D2 phenotype is specific to oskar mRNA , we examined whether other molecules that are transported to the posterior by Kinesin 1 are also affected . We first tested a fusion between the motor domain of the kinesin heavy chain and β-galactosidase ( Kin-βgal ) that behaves as a constitutively active motor ( Clark et al . , 1994 ) . In wild-type egg chambers , Kin-βgal forms a crescent at the posterior of the oocyte at stage 9 ( Figure 3A ) . In 4D2 germline clones , however , Kin-βgal localises to a diffuse cloud near the posterior cortex , showing a very similar defect to oskar mRNA ( Figure 3B ) . Kinesin 1 also transports the dynein/dynactin complex to the posterior , although the function of this localisation is not known ( Li et al . , 1994; Palacios and St Johnston , 2002 ) . In 4D2 germline clones , dynein and the dynactin component , p150Glued , show an identical posterior localisation phenotype to oskar mRNA ( Figure 3C–E ) . Thus , 4D2 causes a general defect in kinesin-dependent transport to the posterior cortex of the oocyte . As Kin-βgal and the dynein/dynactin complex are not anchored at the posterior , these observations also rule out the possibility that the phenotype arises from a lack of anchoring . Recombination mapping and fine scale deletion mapping showed that the 4D2 chromosome carries a single lethal mutation within the 66 kb region that is removed by Df ( 3R ) Exel6166 , but not by Df ( 3R ) ED5612 ( Figure 3—figure supplement 1 ) . Crosses to the lethal mutations in this interval revealed that 4D2 fails to complement the arp1c04425 and arp1G3709 alleles of the Drosophila dynactin component , Arp1 , indicating that 4D2 is a new arp1 allele . Consistent with this , arp14D2 contains a missense mutation that changes a highly conserved glutamate residue ( E53 ) to lysine ( Figure 4A ) . We also crossed arp14D2 to two hypomorphic , EMS-induced alleles , arp11 and arp12 ( Haghnia et al . , 2007 ) . Some of the transheterozygotes were viable and showed a similar diffuse posterior localisation of GFP-Staufen to arp14D2 homozygous mutant germline clones , confirming that the mutation in Arp1 is the cause of the phenotype ( Figure 3—figure supplement 2 ) . The dynactin complex acts as a cargo adaptor complex for dynein ( Holleran et al . , 2001; Muresan et al . , 2001; Zhang et al . , 2011 ) , enhances the processivity of dynein movement along microtubules ( McKenney et al . , 2014; Schlager et al . , 2014 ) and also binds to microtubule plus ends through p150Glued ( Akhmanova and Steinmetz , 2015; Duellberg et al . , 2014; Lazarus et al . , 2013 ) . Arp1 polymerises into a filament composed of eight copies of Arp1 and one β-actin subunit that forms the backbone of the dynactin complex ( Urnavicius et al . , 2015; Figure 4B ) . The E53 to K mutation in arp14D2 falls in the loop of subdomain two that mediates the interaction between Arp1 subunits . It is not predicted to affect protofilament formation , because it points away from the interaction interface ( Figure 4C ) . However , the E53K mutation should disrupt the interaction of Arp1E subunit with one of the extended regions of p50 dynamitin that anchor the p150Glued shoulder to the Arp1 rod ( Figure 4D and E ) . This suggests that the mutation does not disrupt the whole dynactin complex , but may alter the conformation or activity of the p150Glued/dynamitin/p24 shoulder domain . Early in oogenesis , dynein and dynactin are required for the transport of oocyte determinants into one cell of the 16-cell germline cyst , and null mutants in components of either complex therefore fail to form an oocyte ( McGrail and Hays , 1997; Liu et al . , 1999; Bolívar et al . , 2001; Mirouse et al . , 2006; Haghnia et al . , 2007 ) . The oocyte is specified normally in arp14D2 germline clones , however , as shown by the accumulation of Orb in one cell of the 16-cell germline cyst in region 3 of the germarium ( Figure 5A and B ) . By contrast , the null mutation , arp1c04425 , completely blocks oocyte determination to give rise to cysts with 16 nurse cells ( Figure 5C ) . This indicates that arp14D2 does not disrupt the function of the dynactin complex as an activator and cargo adaptor for dynein . Dynein and dynactin both fulfil multiple functions during the later stages of oogenesis: they are required to transport RNAs , such as bicoid , gurken and oskar from the nurse cells into the oocyte ( Clark et al . , 2007; Mische et al . , 2007 ) , they transport bicoid mRNA to the anterior of the oocyte ( Duncan and Warrior , 2002; Weil et al . , 2006; Weil et al . , 2008; Trovisco et al . , 2016 ) , anchor the nucleus at the dorsal/anterior corner of the oocyte ( Swan et al . , 1999; Lei and Warrior , 2000; Januschke et al . , 2002; Zhao et al . , 2012 ) , and localise and anchor gurken mRNA above the nucleus ( MacDougall et al . , 2003; Delanoue et al . , 2007 ) . Both bicoid and oskar mRNAs enter the oocyte normally in arp14D2 homozygous germline clones , and bicoid mRNA localises to the anterior cortex as in wild-type ( Figure 1D , Figure 5D–G ) . However , the oocyte nucleus is not anchored at the dorsal/anterior corner ( Figure 5D–E , H–K ) . gurken mRNA is also not localised , although this could be an indirect consequence of the failure in nuclear anchoring ( Figure 5H and I ) . The germline clone egg chambers complete oogenesis normally and are laid as eggs , but these never develop , probably because the nuclear localisation defect disrupts meiosis or pronuclear fusion . Thus , most dynein/dynactin-dependent RNA transport processes occur normally in the arp14D2 mutant . The defects in the delivery of oskar mRNA to the posterior cortex , the anchoring of the nucleus , and possibly also gurken mRNA localisation must therefore reflect a specific function of dynactin that is disrupted by this allele . To investigate which function of dynactin is affected by the arp14D2 mutation , we focused on the defect in the posterior localisation of oskar mRNA , as this process has already been extensively characterised . In principle , the failure to deliver oskar mRNA to the posterior cortex could reflect either a problem with mRNA transport or a defect in the organisation of the microtubules along which the RNA is transported . Both dynein/dynactin and oskar mRNA are transported to the posterior by kinesin 1 , and it is therefore possible that dynactin plays a role in coupling oskar mRNA to kinesin 1 . We therefore examined the movements of oskar mRNA in living oocytes using the MS2 system for fluorescently labelling RNA in vivo ( Bertrand et al . , 1998; Forrest and Gavis , 2003; Zimyanin et al . , 2008 ) . The speed and direction of movements as well as mobile fraction of oskar mRNA in arp14D2 homozygotes were very similar to wild-type indicating that dynactin is not required for the kinesin-dependent transport of oskar mRNA ( Table 1 ) . Two aspects of the movements were significantly different from normal , however . Firstly , the number of oskar mRNA movements was strongly reduced in the region from 0 to 15 μm from the posterior pole , where the movements are most frequent in wild-type ( Figure 6A ) . This provides further evidence that the diffuse localisation of oskar mRNA in the posterior cytoplasm in the mutant arises from a failure to transport the mRNA all of the way to the posterior cortex . Secondly , the posterior bias in the direction of oskar mRNA movements decreased closer to the posterior pole: the bias is 69% ±2 . 8 in the region 20–30 μm from the posterior pole; 63% ±3 . 1 , 10–20 μm from posterior and 60% ±3 . 9 , in the region 0–10 μm from posterior pole , whereas the directional bias increases towards the posterior in wild-type ( 63% ±2 . 9 , 20–30 μm from posterior; 66% ±2 . 3 , 10–20 μm from posterior and 71% ±2 . 5 , 0–10 μm from posterior ) ( Figure 6B; Parton et al . , 2011 ) . The arp14D2 mutant phenotype highlights an unresolved question in oskar mRNA localisation , which is how the RNA is actually delivered to the posterior cortex . The microtubules in the oocyte are highly dynamic and disappear within a few minutes of the addition of drugs , such as colchicine , that sequester free tubulin dimers . Thus , growing microtubules are likely to contact the cortex for only a few seconds before they undergo catastrophe and start shrinking towards the anterior . To deliver oskar mRNA to the posterior pole , oskar mRNA/Kinesin 1 complexes must therefore reach the plus end of the microtubule while it is still in contact with the posterior cortex ( Figure 6C left ) . One way that dynactin might contribute to this process is by preventing oskar mRNA/kinesin complexes from running off the ends of growing microtubules by tethering the complex to the growing plus ends through p150Glued ( Figure 6C right ) . In wild-type , multiple oskar mRNA/kinesin complexes could then track the growing plus ends and be deposited on the cortex when the microtubule reaches the posterior , whereas these complexes would fall off the end of the microtubule in the arp14D2 mutant . If this model is correct , one would expect a proportion of oskar mRNA particles to move at the speed of the growing microtubules in wild-type . However , there is no significant peak in the distribution of oskar mRNA velocities that corresponds to the speed of the growing microtubule plus ends , indicating that dynactin does not couple the RNA particles to the plus ends ( Figure 6D; Trovisco et al . , 2016 ) . Since kinesin 1 transports oskar mRNA with an average speed of 0 . 47 μm/sec , while the microtubules grow at 0 . 23 μm/sec , the oskar mRNA/kinesin 1 complexes will continually catch up with the growing plus ends and then fall off ( Figure 6D ) . The amount of oskar mRNA deposited on the posterior therefore primarily depends on the amount of time that the plus ends persist once they reach the posterior cortex . Since the arp14D2 mutant does not appear to affect the behaviour of oskar mRNA directly , we next examined the organisation of the microtubules . Anti-tubulin stainings of fixed egg chambers showed a slight reduction in the density of microtubules in arp14D2 homozygotes compared to wild-type , but no significant change in their overall arrangement ( data not shown ) . However , these stainings do not preserve the microtubules near the posterior of the oocyte . We therefore examined the behaviour of the growing microtubule plus ends by making movies of oocytes expressing the plus end tracking protein EB1 fused to GFP ( Parton et al . , 2011 ) . In wild-type , the growing plus ends slow down close to the posterior to a speed of 0 . 18 μm/sec , but this does not occur in arp14D2 homozygotes and the microtubules continue to grow at 0 . 24 μm/sec ( Figure 7A ) . More importantly , very few plus ends extend into the most posterior region of the oocyte ( Figure 7B and C ) . Quantifying the frequency of EB1 tracks in this region reveals that almost all microtubules stop in the region between 10 and 20 μm from the posterior cortex in the mutant , with only a very few growing to the posterior pole ( Figure 7D ) . To confirm that this was due to an increased catastrophe rate in the arp14D2 mutant , we measured the lifespan of EB1 comets in the posterior region ( Figure 7E ) . Comets can disappear either because the microtubule undergoes catastrophe or because its plus end moves out of the imaging plane . Since the probability of a microtubule moving out of the imaging plane increases the longer the microtubule grows , the apparent lifespan of longer comets is underestimated . Nevertheless , this analysis revealed a significant reduction in the lifespan of EB1 comets in arp14D2 homozygotes: the mean persistence time of comets in wild-type was 11 . 29 s ( SEM = 0 . 14 , n = 2058 ) compared to 8 . 94 s in arp14D2 homozygotes ( SEM = 0 . 39 , n = 211; p<0 . 0001 by the unpaired t test . ) . Thus , the arp14D2 mutation increases both the growth rate and the catastrophe rate of the microtubule plus ends near the posterior . This indicates that wild-type dynactin , which is highly concentrated in the posterior region , acts as an anti-catastrophe factor that allows the growing plus ends to extend all of the way to the posterior pole . These results suggest that the diffuse localisation of oskar mRNA in the posterior cytoplasm in the arp14D2 mutant is a consequence of the microtubules being too short . To examine whether this effect is sufficient to account for the mutant phenotype , we used the computer simulations of microtubule organisation , cytoplasmic flows and oskar mRNA transport to test the effects of varying the microtubule length parameter ( Khuc Trong et al . , 2015 ) . When the microtubules have a mean target length of 0 . 5 x the distance from the anterior to the posterior of the oocyte ( ~25 μm ) , the simulation reproduces the wild-type microtubule organisation and the robust posterior localisation of oskar mRNA ( Figure 7F , left ) . Shortening the microtubule mean target length by 30% to 17 . 5 μm still results in the posterior enrichment of oskar mRNA , but in a cloud in the posterior cytoplasm rather than at the cortex , exactly as observed in the arp14D2 mutant ( Figure 7F , right ) . Thus , the simulations support the view that the arp14D2 phenotype is caused by the reduced length of the posterior microtubules . These results raise the question of whether the kinesin-dependent transport of dynactin contributes to its anti-catastrophe function by delivering dynactin to the growing microtubule plus ends and concentrating it posteriorly . To investigate this issue , we examined the behaviour of growing microtubule plus ends labelled with EB1-GFP in germline clones of khc27 , a null mutant in the kinesin heavy chain . As seen in the arp14D2 mutant , the microtubules grow more rapidly in khc27 homozygous oocytes than in wild-type ( Figure 7—figure supplement 1A ) . Furthermore , the relative frequency of EB1 comets is significantly reduced in the region within 20 μm of the posterior cortex in the mutant ( p<0 . 0001 ) and the lifespan of the comets is also reduced from a median of 9 . 01 s in wild-type to 7 . 74 s in khc27 homozygotes ( p=0 . 0006 , Wilcoxon rank-sum test ) . Thus , a failure to transport dynactin to the growing microtubule plus ends near the posterior of the oocyte has a similar effect to the arp1 mutation , presumably because dynactin needs to be concentrated in this posterior region to efficiently bind the growing plus ends and suppress microtubule catastrophes . The final issue that we addressed is whether the anti-catastrophe function of dynactin continues when the microtubules reach the posterior cortex , as this would give more time for kinesin 1 to deliver oskar mRNA to its anchoring site . EB1 recognises the GTP-bound tubulin incorporated into the growing plus end , while increasing the hydrolysis of GTP to GDP ( Maurer et al . , 2011; Maurer et al . , 2012; Seetapun et al . , 2012; Maurer et al . , 2014; Zhang et al . , 2015; Duellberg et al . , 2016 ) . Catastrophe occurs when the EB1-binding GTP-tubulin cap of the microtubule is reduced below a critical threshold , triggering a rapid loss of tubulin dimers from this end . EB1 comets have been previously observed to slow down to 0 . 08 μm/sec on hitting the cortex of tissue culture cells and can persist there for a few seconds ( Straube and Merdes , 2007; van der Vaart et al . , 2013 ) . Thus , the duration of a ‘static’ EB1-GFP signal at the cortex provides a readout of how long the microtubule plus end persists before undergoing catastrophe . We plotted kymographs along regions of either the lateral or the posterior cortex and measured the lifetime of the static EB1 foci ( Figure 8A–C ) . These foci , which appear as vertical lines in the kymograph , persist significantly longer at the posterior cortex than at the lateral cortex . Quantifying this effect reveals that microtubules remain for an average of 15 s at the posterior , with a maximum of over 50 s , compared to a mean persistence time of only 8 s laterally ( p=4 . 1 × 10−11; Wilcoxon rank-sum test ) . Thus , dynactin may continue to protect the plus ends after they have reached the posterior cortex , providing more time for kinesin 1 to transport oskar mRNA to the posterior pole ( Figure 8D ) .
Because dynactin and dynein are required for oocyte determination , their functions later in oogenesis have been studied by over-expressing p50 dynamitin or shRNAs against the dynein heavy chain ( Duncan and Warrior , 2002; Januschke et al . , 2002; Sanghavi et al . , 2013 ) . Both treatments disrupt the localisation of bicoid mRNA and the anchoring of the oocyte nucleus , but only slightly reduce the amount of oskar mRNA that is correctly localised to the posterior cortex . This may be an indirect consequence of reduced transport of oskar mRNA from the nurse cells into the oocyte , as these treatments merely lower the levels of wild-type dynactin or dynein ( Clark et al . , 2007; Mische et al . , 2007 ) . Here , we report a very different phenotype when the entire germline is homozygous for the E53K mutation in the Arp1 subunit of dynactin: oocyte determination , mRNA transport into the oocyte and bicoid mRNA localisation are unaffected , but oskar mRNA is diffusely localised in the posterior cytoplasm , rather than at the posterior cortex . This defect is a consequence of an increased catastrophe rate of the microtubules growing towards the posterior , indicating a novel requirement for dynactin in extending microtubules posteriorly , so that kinesin 1 can deliver oskar mRNA to its cortical anchoring site . Dynactin has been shown to function as an anti-catastrophe factor in mammalian tissue culture cells and in neurons , where it promotes the growth of microtubules towards the axonal terminals ( Komarova et al . , 2002; Lazarus et al . , 2013 ) . This activity depends on the neuronal isoforms of p150Glued , which contain both an N-terminal CAP-Gly domain and an internal basic region . Drosophila p150Glued is not alternatively spliced and contains well-conserved CAP-Gly and basic domains , making it a good candidate for the subunit that mediates the anti-catastrophe function of Drosophila dynactin in the oocyte . The arp14D2 mutation is predicted to disrupt the interaction between the Arp1 rod and the dynactin shoulder domain that contains p150Glued , and this may disturb an allosteric interaction that is necessary for the correct conformation of the p150Glued N-terminus . The association of dynactin with growing plus ends is thought to depend on the interaction of p150Glued with the C-terminus of CLIP-170 ( CLIP-190 in Drosophila ) , which in turn binds to the tail of EB1 ( Duellberg et al . , 2014; Honnappa et al . , 2006; Lansbergen et al . , 2004 ) . However , CLIP-190 null mutations are viable and fertile and have no discernible effect on oskar mRNA localisation ( Dix et al . , 2013; data not shown ) . Thus dynactin must be able to associate with the growing plus ends independently of CLIP-190 , presumably through its interaction with EB1 ( Duellberg et al . , 2014; Askham et al . , 2002; Komarova et al . , 2002; Ligon et al . , 2003; Vaughan et al . , 2002 ) . Consistent with our results with arp14D2 , deletion of p150Glued N-terminus has no effect on dynein-dependent cargo transport , but affects microtubule organisation in Drosophila S2 cells ( Kim et al . , 2007 ) . Although it has been known for many years that dynein and dynactin are transported to the posterior of the oocyte by kinesin 1 , the functional significance of this localisation has remained unclear ( Li et al . , 1994; Palacios and St Johnston , 2002 ) . Our analysis reveals that kinesin-dependent localisation of dynactin creates a positive feedback loop that amplifies the directional bias in microtubule orientation posteriorly and extends microtubule growth to the posterior pole , both of which are essential for the final step in oskar mRNA localisation ( Figure 9 ) . The oocyte microtubules grow with a weak orientation bias towards the posterior that arises from the fact that they emanate from the anterior and lateral cortex , but are repressed posteriorly ( Khuc Trong et al . , 2015; Nashchekin et al . , 2016 ) . Plus end-directed transport of dynactin by kinesin 1 along these microtubules will therefore concentrate dynactin posteriorly , where it can associate with the growing plus ends and prevent them from undergoing catastrophes . As a consequence , the plus ends of the microtubules growing into this region extend further towards the posterior , amplifying the directional bias in microtubule orientation posteriorly . This effect explains why the orientation bias in oskar mRNA movements increases towards the posterior in wild-type oocytes , but decreases in arp14D2 mutants ( Figure 6B ) . The increased length and directional bias of the microtubules at the posterior allows kinesin 1 to transport dynactin even more posteriorly , generating a positive feedback loop that eventually results in a high posterior concentration of dynactin that promotes microtubule growth all of the way to the posterior cortex . This amplification loop therefore generates the posterior microtubules tracks for the kinesin-dependent transport of oskar mRNA to the posterior cortex . The dynamic nature of the oocyte microtubules means that in order to localise oskar mRNA , kinesin/oskar mRNA complexes must reach the end of a microtubule at the posterior cortex before the microtubule undergoes catastrophe . This is aided by the fact that the microtubules persist twice as long at the posterior cortex than at the lateral cortex , suggesting that dynactin continues to suppress catastrophes at the cortex . Microtubules undergo catastrophe when subjected to the pushing forces produced by growing against a barrier , such as the cortex ( Janson et al . , 2003 ) . The longer persistence at the posterior could therefore result from the slower growth of the dynactin-associated plus ends , which would increase the pushing force more slowly . In many cases where microtubule plus ends make sustained contacts with the cortex , such as during centrosome positioning and spindle orientation , cortical dynein captures the plus ends and anchors them to the cortex as they shrink ( Burakov et al . , 2003; Carminati and Stearns , 1997; di Pietro et al . , 2016; Hendricks et al . , 2012; Koonce et al . , 1999; Laan et al . , 2012; Nguyen-Ngoc et al . , 2007; Yamamoto et al . , 2001 ) . In budding yeast , for example , a kinesin transports dynein along astral microtubules to the cortex , where dynein is anchored and captures microtubule plus ends to pull on the spindle pole ( Caudron et al . , 2008; Farkasovsky and Küntzel , 2001; Heil-Chapdelaine et al . , 2000; Markus et al . , 2009; Sheeman et al . , 2003 ) . Since dynein is transported to the posterior of the oocyte by kinesin 1 , and becomes highly enriched in the posterior cytoplasm , like dynactin , it is tempting to speculate that dynein might further extend the time for oskar mRNA delivery by tethering shrinking microtubule plus ends to the cortex . However , the very high concentration of dynein in the posterior cytoplasm makes it difficult to tell whether it is specifically recruited to the cortex , and it is not currently possible to image the plus ends of microtubules that have lost their EB1 cap . Thus , investigation of this potential mechanism will require ways to visualise the plus ends of shrinking microtubules .
white1118 was used as a wild-type stock throughout . The following mutant alleles and transgenic lines were used: arp14D2-12 ( Martin et al . , 2003 ) , arp11and arp12 ( Haghnia et al . , 2007 ) , arp1c04425 ( Thibault et al . , 2004 ) , hsFLP kinβ-Gal ( Clark et al . , 1994 ) , hsFLP maternal α4tubulin::GFP-Staufen ( Martin et al . , 2003 ) , khc27 ( Brendza et al . , 2000 ) , oskMS2 ( Zimyanin et al . , 2008 ) , UAS EB1-GFP ( Jankovics and Brunner , 2006 ) . Germline clones were generated with FRT 82B ovoD , FRT 42B ovoD or FRT 82B GFP ( Bloomington Stock Center ) using the heat shock FLP/FRT system ( Chou and Perrimon , 1992 ) . For live imaging , ovaries were dissected and imaged in Voltalef oil 10S ( VWR International ) on either a widefield DeltaVision microscope ( Applied Precision , WA , USA ) equipped with a Photometrics 512 EMCCD camera ( Photometrics , AZ , USA ) and a 2x magnification tube fitted between the unit and the camera , or an Olympus Fluoview FV1000 confocal microscope ( Olympus , Japan ) or on an Olympus IX81 inverted microscope with a Yokogawa CSU22 spinning disk confocal imaging system using 100 × 1 . 4 NA Oil UPlanSApo objective lens ( Olympus , Japan ) . Fixed preparations were imaged using an Olympus Fluoview FV1000 confocal microscope ( 40 × 1 . 35 NA Oil UPlanSApo , 60 × 1 . 35 NA Oil UPlanSApo ) or Zeiss LSM510 Meta laser scanning confocal system ( Carl Zeiss Microimaging , Inc . ) with Plan-Neofloar 40x ( Oil ) NA1 . 3 and Plan-Apochromat 63x ( Oil ) NA1 . 4 objectives . Images were collected with the softWorXs software ( Applied Precision ) , Olympus Fluoview ( Olympus , Japan ) , MetaMorph Microscopy Automation and Image Analysis Software ( Molecular Devices , CA , USA , RRID:SCR_002368 ) or Zeiss LSM 510 AIM software ( Carl Zeiss Microimaging , Inc . ) and processed using Fiji ( Fiji , RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) . Moving oskar-MS2-GFP particles were tracked manually using the MTrackJ plugin for the Fiji image analysis software ( Fiji , RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) . The speed , direction of movement and the mobile fraction of oskar RNA particles were analysed as previously described ( Zimyanin et al . , 2008 ) . To visualize oskar-MS2-GFP particles moving near the posterior of the oocyte , we bleached the GFP signal from oskar RNA that was already localized and imaged particle movements after 10 min of recovery . We analysed at least 5 oocytes per sample type . EB1-GFP comets were tracked using plusTipTracker ( Matov et al . , 2010; Applegate et al . , 2011 ) . For each comet , the speed was calculated as the mean of its velocities at individual time points . Distance from the posterior was measured from its initial position . At least 5 oocytes per sample type were examined . To compute the time growing microtubules persist at the lateral or posterior cortex , individual EB1-GFP tracks were first extracted by the plusTipTracker package . Custom-written MATLAB code ( available at GitHub , https://github . com/MaxJakobs/Nieuwburg2017; copy archived at https://github . com/elifesciences-publications/Nieuwburg2017 ) then separated the tracks into those that extend to the oocyte cortex ( excluding those that reach the cortex at angle of <20° ) . The oocyte cortex was drawn by hand , tracing the fluorescent outline of the cell and all tracks that came within 1 μm of the hand-drawn boundary were considered as touching the cortex . The cortical dwell time for each track was calculated as the time between the EB1-GFP comet moving within 1 μm of the boundary and its disappearance . Data for 3 different oocytes was pooled , and a location-scaled t-distribution fitted to each posterior and lateral dataset . The significance of any differences in speed or relative frequency of oskar RNA particle and EB1-GFP comet movements was evaluated using t tests and Wilcoxon rank-sum tests . When the distribution was normal , the significance of the difference in the means was assessed using a standard Student’s t test for distributions with equal variance , or the Welch-corrected Student’s t test for distributions with unequal variance . When the distribution was non-normal , we compared medians using the non-parametric counterpart of the Student’s t test , the Wilcoxon rank-sum test . | Many cells are asymmetric or polarized , which allows them to perform the tasks necessary for an organism to live and grow . In these polarized cells , the top and bottom , left and right , and front and back parts are different from one another . To achieve this , cells actively move molecules to the locations in the cell where they are needed . One type of molecule that is often confined is messenger RNA , or mRNA for short , which carries a portion of the DNA code to other parts of the cell where it can be translated to make a specific protein . These mRNA molecules are transported by motor proteins , which run along tracks called microtubules that form a network throughout the cell . Each microtubule has a stable ‘minus’ end and a dynamic ‘plus’ end , which constantly grows or shrinks . Motor proteins can generally only transport their cargo in one specified direction . For example , the protein Kinesin-1 moves towards the plus end of the microtubule . In the fruit fly egg , many molecules are asymmetrically arranged , which later dictate how the larvae will develop . For example , a gene called oskar is necessary for the development of the back region of the fly embryo , and its mRNA is transported by Kinesin-1 along microtubules towards the plus ends , at the back end of the egg cell . However , it was unclear how the plus ends accumulate at the back of the egg cell in the first place . Now , Nieuwburg , Nashchekin et al . used live microscopy to watch how growing microtubules and oskar mRNA move in fruit fly egg cells . Comparing normal and mutant fruit flies revealed that a large protein complex called Dynactin stabilizes the microtubule plus ends at the back of the cell . This gives Kinesin-1 enough time to carry oskar mRNA along the length of a microtubule to its plus end . When one subunit of Dynactin was mutated , the microtubule plus ends became less stable and did not reach all the way to the back of the developing egg . With this mutation , oskar mRNA was still transported by Kinesin-1 but delivered to the wrong place . All together , these experiments provide evidence that the plus ends of microtubules must be controlled so that motor proteins can deliver their cargoes to the correct destination . Future work will determine exactly how Dynactin stabilizes microtubules and whether this is a general mechanism that can also set up polarized microtubule tracks in asymmetric cells , such as nerve cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2017 | Localised dynactin protects growing microtubules to deliver oskar mRNA to the posterior cortex of the Drosophila oocyte |
Although decades of studies have produced a generalized model for tetrapod limb development , urodeles deviate from anurans and amniotes in at least two key respects: their limbs exhibit preaxial skeletal differentiation and do not develop an apical ectodermal ridge ( AER ) . Here , we investigated how Sonic hedgehog ( Shh ) and Fibroblast growth factor ( Fgf ) signaling regulate limb development in the axolotl . We found that Shh-expressing cells contributed to the most posterior digit , and that inhibiting Shh-signaling inhibited Fgf8 expression , anteroposterior patterning , and distal cell proliferation . In addition to lack of a morphological AER , we found that salamander limbs also lack a molecular AER . We found that amniote and anuran AER-specific Fgfs and their cognate receptors were expressed entirely in the mesenchyme . Broad inhibition of Fgf-signaling demonstrated that this pathway regulates cell proliferation across all three limb axes , in contrast to anurans and amniotes where Fgf-signaling regulates cell survival and proximodistal patterning .
Limb development is an ideal model to investigate how cellular and molecular networks exhibit plasticity or resilience during tetrapod evolution . Since the turn of the twentieth century the limb has endured as a fundamental model to investigate morphogenesis , cellular growth , and differentiation during embryonic development . From studies spanning comparative embryology through developmental genetics , limb development has provided deep insight into mechanisms underlying pattern formation , genotype-phenotype relationships , and the complexity of molecular networks ( Swett , 1937; Wolpert , 1969; Zeller et al . , 2009 ) . Perhaps equally studied as a fully developed structure , the limb has also served as a model for evolutionary biologists seeking to reconstruct tetrapod ancestry and how micro- and macro-evolutionary changes might explain major events in vertebrate evolution such as the fin to limb transition ( Fröbisch and Shubin , 2011; Shubin and Alberch , 1986 ) . The two most important embryonic models for modern limb development studies have been chicken and mice . Chicken embryos were integral to our understanding of limb bud outgrowth and morphogenesis aided by the availability of genetic limb mutants and the ability to discern skeletal defects in the wing caused by surgical manipulation ( Saunders , 1948; Summerbell et al . , 1973 ) . With the advent of stable transgenesis , mice became the model of choice to investigate the molecular basis of limb development . Together , data from chicken , mice and frogs have been synthesized into contemporary models covering limb development in tetrapods ( Zeller et al . , 2009; Zuniga , 2015 ) . In these models , all vertebrate limb buds utilize the same molecular network ( e . g . Shh , Fgfs , Bmps , Wnts and retinoic acid ) governed by Hox genes and controlled by two major signaling centers; the zone of polarizing activity ( ZPA ) and the apical ectodermal ridge ( AER ) . And then there are the urodeles . With tetrapod monophyly solidly supported by molecular and morphological data ( Ahlberg and Milner , 1994; Marjanović and Laurin , 2013 ) , for scientists investigating the evolution and development of vertebrate limbs , salamanders and newts have longed proved problematic ( Holmgren , 1933 ) . Although salamander and newt embryos were used to uncover key principles regarding specification of the prospective limb field and establishment of the primary limb axes ( i . e . , proximal-distal , anterior-posterior and dorsal-ventral ) ( Harrison , 1918; Stocum and Fallon , 1982; Swett , 1937 ) , it was also observed that urodele limb development deviated from anurans and amniotes in at least two key respects: skeletal specification exhibited preaxial dominance ( anterior elements form before posterior elements ) rather than postaxial dominance ( Shubin and Alberch , 1986 ) and urodele limb buds did not form an apical ectodermal ridge ( AER ) ( Sturdee and Connock , 1975; Tank et al . , 1977 ) . In addition to these important developmental differences , adult urodeles differ from anurans and amniotes in their ability to completely regenerate an amputated limb . With these ideas in mind , we sought to investigate limb development in salamanders to determine whether morphological and molecular data support a unified model of limb development that includes or excludes urodeles .
In order to study axolotl limb outgrowth and axis specification , we first staged developing limbs on the basis of external morphology and skeletal chondrification at 20–21°C ( Figure 1A–B and Figure 1—figure supplement 1 ) . As the limb bud emerged from the flank , limb mesenchyme expanded directly above the body wall musculature and at no time during limb development did we observe an AER or thickening of the limb ectoderm ( Figure 1C ) . In agreement with previous work in salamanders ( Holmgren , 1933; Nye et al . , 2003; Shubin and Alberch , 1986 ) , we found that cartilage condensations of the limb skeleton formed proximal to distal , and within the zeugopod and autopod , anterior to posterior ( preaxial dominance ) ( Figure 1D ) . While this was strictly true for the radius and ulna , alcian blue staining always demonstrated digit I and II forming together between stages 47 and 48 ( Figure 1D ) . To further examine chondrogenesis in the limb , we identified axolotl Sox9 ( Supplementary file 1 ) and used its expression to analyze condensing mesenchymal cells prior to cartilage formation ( Wright et al . , 1995 ) ( Figure 1E and Figure 1—figure supplement 2 ) . We first observed Sox9 expression at stage 45 in a broad proximal area of the axolotl forelimb bud ( Figure 1E ) . Sox9 expression at stage 46 appeared in a centrally located domain corresponding to the future radius , ulna and carpals ( Figure 1D–E ) . However , in contrast to the Alcian blue staining pattern , Sox9 expression in the presumptive digits appeared sequentially II-I-III-IV and this result was consistent across 13/15 ( 86 . 66% ) limbs analyzed across stage 48 ( Figure 1E and Figure 1—figure supplement 2 ) . At stage 49 , Sox9 expression expanded to clearly mark digit I along with digit II ( Figure 1E and Figure 1—figure supplement 2 ) . Thus , although chondrification of the limb skeleton proceeds anterior to posterior in the zeugopod , digit specification as marked by Sox9 expression in the autopod exhibits a pre-pattern that first emerges along the central axis marked by digit II . Having established the spatiotemporal pattern of mesenchymal specification during skeletal formation , we next sought to investigate the expression of key genes involved in limb bud outgrowth and anterior-posterior patterning . For this , we optimized wholemount in situ hybridization such that we could accurately identify mesenchymal and ectodermal expression ( Figure 2—figure supplement 1 ) . During limb initiation in other tetrapods , Sonic hedgehog ( Shh ) is induced in the posterior limb mesenchyme ( Riddle et al . , 1993 ) . Using an antisense-probe ( ~1000 bp ) that was a gift from the Tanaka lab ( IMP , Austria ) , we determined that Shh was first expressed at stage 45 in a posteriorly restricted domain and persisted through stage 49 in a posterior-proximal position ( Figure 2A ) . Compared to anuran and other amniote forelimb buds where Shh expression precedes expression of Hoxd11 , the onset of Shh expression in axolotls appeared relatively late and after Hoxd11 expression ( Matsubara et al . , 2017; Riddle et al . , 1993 ) ( Figure 2—figure supplement 2 ) . While the posterior Shh expression pattern was consistent with previous reports using relatively short probes at several stages ( Bickelmann et al . , 2018; Imokawa and Yoshizato , 1997; Torok et al . , 1999 ) , we also observed Shh expression in an anteriorly restricted domain at stages 46 , 47 and 48 ( Figure 2A ) . To support our in situ data , we divided stage 46–49 limb buds into posterior and anterior halves and collected RNA ( Figure 2—figure supplement 3A–C ) . Using qRT-PCR , we observed Shh expression from the anterior and posterior compartments ( Figure 2—figure supplement 3D ) . We also isolated full-length Indian hedgehog ( Ihh ) and aligned it with Shh revealing that our RNA probe shared 42% similarity between sequences ( Supplementary file 1 ) . Furthermore , we generated an RNA probe to Ihh and found spatiotemporal expression domains distinct from Shh expression localized to areas of skeletal ossification ( Figure 2—figure supplement 2 ) . We next examined the spatiotemporal expression of the hedgehog receptor Patched 1 ( Ptch1 ) and effector molecule Gli1 , both of which are direct targets of Shh-signaling ( Figure 2A–B ) . Ptch1 expression was localized in two broad domains corresponding to the posterior and anterior domains of Shh expression ( Figure 2A ) . While we observed Gli1 expression in an anterior domain similar to Ptch1 expression , the posterior expression of Gli1 appeared to localize more closely with Ihh ( Figure 2B and Figure 2—figure supplement 2 ) . Lastly , we examined the expression of the repressor form of Gli3 which serves to restrict Shh to the posterior of the limb bud in mice and chickens and found a broad expression pattern across the anterior-posterior axis that excluded the presumptive ZPA at stages 45 , 46 and 47 ( Figure 2 and Figure 2—figure supplement 3D ) . Although Shh expression remains posterior in the forelimb buds of chicks and mice , it tracks distally into the future autopod where it maintains a close association with the AER ( Figure 2C–D ) . In contrast , Shh expression in axolotl forelimb buds did not appear in the autopod ( Figure 2A , C–D ) . Given this proximally restricted position of the Shh domain we asked whether Shh-expressing cells ( or those close by ) contribute to the digits . To track ZPA cells we injected DiI into the approximate position of Shh expression around stage 45 and monitored fluorescence till stage 53 ( Figure 2E ) . Although our injections labeled Shh-expressing cells and nearby cells as well , we only observed cells migrating along the posterior limb margin and contributing to digit IV ( Figure 2E ) . This data mimics labeling experiments in chick limbs where ZPA cells only give rise to the most posterior digit in the hindlimb and posterior margin of the forelimb ( Towers et al . , 2008; Towers et al . , 2011 ) . Thus , despite some late spatial expression differences , our data suggests a conserved role for Shh during forelimb development . Proximal-distal outgrowth of the limb bud and maintenance of Shh-signaling from the ZPA are regulated in tetrapods by the AER , and specifically by Fgf-signaling ( Lewandoski et al . , 2000; Mariani et al . , 2008; Niswander et al . , 1993; Saunders , 1948 ) . Several Fgfs are expressed in the anuran and amniote AER ( i . e . Fgf4 , 8 , 9 , 17 ) , but Fgf8 alone is required for cell survival and limb bud outgrowth ( Lewandoski et al . , 2000; Sun et al . , 2002 ) . Although an AER does not form during limb development in the direct developing frog Eleutherodactylus coqui ( Gross et al . , 2011 ) or the marsupial Monodelphis domestica ( Doroba and Sears , 2010 ) , AER-Fgfs are still restricted to , and expressed , in the ectoderm . Because salamanders lack a morphological AER , we asked if Fgf4 , 8 , 9 and 17 were expressed in the axolotl limb bud ectoderm . In contrast to anurans and amniotes , we found that Fgf8 , Fgf9 and Fgf17 were solely expressed in the mesenchyme ( Figure 3A–C and Supplementary file 1 ) . We could not consistently detect Fgf4 during limb development and , when we did , it was expressed at very low levels in the mesenchyme only ( Figure 3—figure supplement 1 ) . At stage 44 , we detected Fgf8 in a broad mesenchymal zone directly beneath the ectoderm ( Figure 3A ) . Fgf8 expression persisted in the distal mesenchyme until stage 47 when it segregated into symmetrical domains within the dorsal and ventral mesenchyme ( Figure 3A–B ) . Although Fgf8 appeared to exhibit an anteriorly restricted expression pattern at stage 44 , using qRT-PCR we found Fgf8 expression in both the anterior and posterior compartment at stages 46 , 47 and 49 ( Figure 2—figure supplement 3D ) . Fgf9 and Fgf17 showed distally restricted expression at stages 45–46 and Fgf9 appeared to have an additional proximal expression domain at stage 46 ( Figure 3C ) . Fgf17 appeared to have a posterior bias at stage 46 ( Figure 3C ) . Consistent with what is known for amniotes and anurans , we found that Fgf10 was broadly expressed in the distal mesenchyme at stages 45–46 ( Figure 3C and Supplementary file 1 ) . We also examined expression of Fgf receptors 1 and 2 ( FgfR1 and FgfR2 ) . FgfR1 was first expressed weakly at stage 44 and became more proximally restricted during stages 45–46 ( Figure 3D and Supplementary file 1 ) . At stage 46 , the proximal-anterior domain of FgfR1 overlaps with the Fgf9 domain ( Figure 3C–D ) . FgfR2 showed weak proximal expression at stage 44 and was later expressed during stages 45–46 in a domain proximal to Fgf8 , 9 and 17 ( Figure 3A–E and Supplementary file 1 ) . Lastly , we examined Gremlin1 expression in developing salamander limbs , and observed strong mesenchymal expression until digits began condensing at stage 48 ( Figure 3A ) . We also detected Gremlin1 staining at stage 49 in the area destined to become digits III and IV ( Figure 3A ) . Taken together , our expression analysis shows that key Fgf ligands normally expressed in the ectoderm during amniote and anuran limb development are instead , compartmentalized entirely in the limb mesenchyme ( Figure 3E ) . Our gene expression analysis suggested that Fgf ligands and their cognate receptors might be spatially segregated within the limb mesenchyme . To address this possibility , we analyzed single-cell RNA-seq ( scRNA-seq ) data from developing axolotl limbs that matched our limb stages 44 and 45 ( see Materials and methods ) ( Gerber et al . , 2018 ) . Single-cell data acquired from these developing limbs included only mesenchymal cells ( Gerber et al . , 2018 ) . Using principal component analysis , we first identified principal component 3 ( PC3 ) as a model for the proximodistal ( P-D ) axis during stage 45 based on known proximal and distal marker genes ( Figure 4A and Figure 4—figure supplement 1 ) . Specifically , we found that proximal ( Meis1 and Meis2 ) and distal markers ( Fgf8 and Hoxd11 ) were expressed in cells on the opposite ends of contributors to PC3 ( Figure 4A–B and Figure 4—figure supplement 1 ) . Moreover , Etv4 and Dups6 which are direct targets of Fgf8 were among the top 20 genes contributing to the distal end ( Hoxd11+ ) of PC3 ( Figure 4A–B and Figure 4—figure supplement 1 ) . Consistent with expression patterns revealed by in situ hybridization , Fgf ligands such as Fgf8 , 9 , 17 and 10 were all restricted to the distal end of PC3 , while Fgf receptors FgfR1-4 were restricted to the proximal end ( Figure 4A–B and Figure 4—figure supplement 1 ) . Despite seeing FgfR1 expressed broadly across the modeled proximal-distal axis , our analysis showed that FgfR1 was expressed to a greater extent in cells at the proximal end of PC3 ( Figure 4A–B ) . Statistical analysis confirmed that Fgf ligands ( n = 12 ) and Fgf receptors ( n = 5 ) were separated along PC3 ( Welch Two Sample t-test , T = −4 . 1588 , p=0 . 0038 ) . We also asked whether Fgf ligands and receptors might still be co-expressed in some cells and found that where FgfR1 was expressed in the distal portion of PC3 , some of these cells also expressed Fgf8 , Fgf9 and Fgf17 ( Figure 4—figure supplement 2 ) . However , we found few to no cells co-expressing FgfR2-4 and Fgf8 , 9 and 17 ( Figure 4—figure supplement 2 ) . Considered together with our spatiotemporal expression analysis , our scRNA-seq analysis supports compartmentalization of Fgf-signaling within the developing limb mesenchyme and largely points to cellular segregation of ligand-receptor interactions . In amniotes and anurans , a Shh-Grem-Fgf signaling loop regulates proximodistal outgrowth and maintains limb bud mesenchyme in a proliferative , undifferentiated , and multipotent state ( Globus and Vethamany-Globus , 1976; Reiter and Solursh , 1982; ten Berge et al . , 2008; Towers et al . , 2008 ) . To identify if a similar signaling loop exists in salamanders , we tested the functional requirement for Shh- and Fgf-signaling during axolotl limb development . First , in order to explore mesenchymal Fgf-signaling , we used the broad spectrum Fgf-receptor inhibitor ( SU5402 ) that selectively binds to the intracellular kinase domain thereby inhibiting downstream signaling ( Mohammadi et al . , 1997 ) . We treated axolotl embryos with SU5402 beginning prior to limb bud outgrowth from the flank ( at stage 39 ) and then harvested limbs at stages 45 , 46 or 54 ( Figure 5A ) . We administered SU5402 based on a dose-response study and selected a maximum dose that could be delivered continuously which was not toxic to the developing animals ( see Materials and methods ) . Limb buds harvested at stage 45 did not show a significant difference in limb size between treatment and control animals ( one-tailed student’s t-test , T = −1 . 68637 , p=0 . 0514 ) ( Figure 5—figure supplement 1A ) whereas those harvested at stage 46 exhibited significantly smaller limbs ( one-tailed student’s t-test , T = −8 . 7759 , p<0 . 0001 ) ( Figure 5B ) . Although we did find a small , but significant decrease in total animal length ( snout to tail-tip length ) following SU5402 treatment ( one-tailed student’s t-test , T = −4 . 52 , p=0 . 0007 ) ( Figure 5—figure supplement 1B ) , this did not account for the size differences between treatment and control limbs ( T = −8 . 1 , p<0 . 44 ) . To test the efficacy of Fgf-signaling inhibition , we determined expression of the Fgf-signaling targets Etv1 and Etv4 ( Kawakami et al . , 2003 ) ( Figure 5C–E ) . In response to Fgf inhibition , we were unable to detect Etv1 expression in the limb bud at stages 45 or 46 and Etv4 expression was barely detectable ( Figure 5C–D ) . qRT-PCR confirmed that both targets were significantly down-regulated at stage 46 ( Figure 5E ) . To determine if Fgf-signaling controls Shh-signaling , we quantified Shh expression in response to Fgf inhibition and found that it was reduced compared to control limbs ( Figure 5E ) . Given the relatively small number of Shh-expressing cells at these early time points , we asked if reduced Shh transcription translated into reduced pathway activity as assessed by Ptch1 expression ( Figure 5E–G ) . Using qRT-PCR and in situ , we found that Ptch1 was not reduced in treated limb buds suggesting that Shh pathway activity remained normal ( Figure 5E–G ) . Interestingly , while Gremlin1 is inhibited by AER-Fgf-signaling in amniotes ( Merino et al . , 1999; Verheyden and Sun , 2008 ) , we found that Gremlin1 expression was virtually eliminated in SU5042-treated limbs ( Figure 5E–G ) . Loss of AER-Fgf-signaling during mouse limb development does not affect cell proliferation , but produces smaller limbs and proximal truncation due to increased proximal cell death ( Mariani et al . , 2008; Sun et al . , 2002 ) . We used Lysotracker to mark dying cells ( Mariani et al . , 2008; Seifert et al . , 2009 ) and during normal limb development we did not detect any dying cells ( Figure 5—figure supplement 2 ) . Similarly , when we inhibited Fgf-signaling we did not detect dying cells in limb buds ( Figure 5—figure supplement 2 ) . We next assessed if smaller limbs might have resulted from alterations in cell proliferation ( Figure 5H–I ) . Inhibiting Fgf-signaling as above , we examined actively proliferating cells ( EdU+ ) in treated and control stage 46 limb buds using lightsheet microscopy ( Figure 5H–I ) . We calculated the proliferative population as a fraction of total limb volume and observed an 83% decrease in actively proliferating cells ( one-way ANOVA , Kruskal-Wallis test , p<0 . 05 ) ( Figure 5I ) . While the decrease in proliferating cells appeared proportionally across the anteroposterior axis , we noted that proliferating cells were nearly absent from the proximal limb bud and from a small distal domain in treated limbs ( Figure 5H ) . To determine the ultimate effect of Fgf inhibition on limb development , we assessed the effect of treating embryos from before limb outgrowth until all four digits appeared in control limbs at stage 54 ( Figure 5J–K ) . Analyzing the forelimb skeleton , we found that inhibiting Fgf-signaling led to smaller , but nearly complete limbs which generally lacked posterior digit IV ( Figure 5J–K ) . Specifically , 73 . 5% of SU5402-treated embryos exhibited this phenotype ( 25/34 ) while ~11% ( 4/34 ) had fewer than three digits ( Figure 5I ) . In those cases where more than one digit was lost , the missing digits were the next most posterior in sequence . These results reveal that Fgf-signaling regulates mesenchymal proliferation , but not cell survival during axolotl limb development . Thus , inhibiting Fgf-signaling leads to smaller limbs and loss of posterior digits , a result consistent with colchicine treatment of developing axolotl and Xenopus larva ( Alberch and Gale , 1983 ) . Lastly , Fgf-signaling regulates Gremlin1 expression , whereas Shh-signaling is relatively independent of Fgf-signaling . While our results above suggest that a Shh-Grem-Fgf signaling loop does occur during axolotl limb development , we sought to examine if Fgf-signaling was reliant on Shh-signaling . Using cyclopamine to inhibit Shh signal transduction , previous work showed that Shh-signaling controls anterior-posterior patterning of the zeugopod and autopod during axolotl limb development , a role consistent with ZPA function in other tetrapods ( Stopper and Wagner , 2007 ) . Cyclopamine-treated axolotl limbs phenocopied Shh-/- mouse limbs with significant proximodistal outgrowth , fusion of the radius/ulna , and almost complete loss of the autopod ( Chiang et al . , 2001 ) . To analyze the interaction of Shh- and Fgf-signaling during limb bud outgrowth , we treated stage 39 larvae with cyclopamine for 10 days and analyzed the limb buds at stages 45 and 46 ( Figure 6A ) . Limb buds harvested at stage 45 did not show a significant difference in limb size between treatment and control animals ( one-tailed student’s t-test , T = 1 . 43 , p=0 . 909 ) ( Figure 5—figure supplement 1C ) , whereas at stage 46 , limb buds from cyclopamine-treated larvae were significantly smaller compared to control treated limbs ( one-tailed student’s t-test , T = 8 . 36 , p<0 . 0001 ) and this effect was independent of ( T = 0 . 03 , p=0 . 975 ) a small , but significant decrease in animal length ( one-tailed student’s t-test , T = 3 . 87 , p<0 . 0016 ) ( Figure 6B and Figure 5—figure supplement 1D ) . Analyzing Ptch1 expression by qRT-PCR and in situ hybridization , we found that cyclopamine treatment efficiently inhibited hedgehog signaling in limb buds ( Figure 6C–E ) . Using qRT-PCR we saw a significant reduction in Fgf8 and Gremlin1 expression at stage 46 and using in situ we were unable to detect these genes at either stage 45 or 46 ( Figure 6B–D ) . These data place Fgf8 and Gremlin1 downstream of Shh-signaling during axolotl limb development . We also tested whether Shh-signaling controlled cell survival and cell proliferation . Similar to our results using SU5042 , when we inhibited Shh-signaling we did not observe cell death in developing limb buds ( Figure 5—figure supplement 2 ) . However , we did find that loss of Shh-signaling led to a 53% reduction in cell proliferation ( Figure 6F–G ) . Whereas Fgf-inhibition led to proportionally smaller limbs , Shh-inhibition led to a dramatic loss of cell proliferation in the distal limb bud ( one-way ANOVA , Kruskal-Wallis test , p<0 . 05 ) ( Figure 6F ) . Together , our findings place Fgf8 downstream of Shh-signaling and suggest that Shh-signaling also controls cell proliferation , although more specifically in the distal limb bud where digit progenitors reside .
Urodeles were among the first vertebrates used to study limb field specification and morphogenesis ( Harrison , 1918; Stocum and Fallon , 1982; Swett , 1937 ) . Although chick and mouse embryos largely replaced urodeles as model systems to study limb development , a perpetual fascination with understanding the molecular basis of limb regeneration has resurrected interest into amphibian limb development ( Gerber et al . , 2018; Keenan and Beck , 2016; Stocum , 1975 ) . Elucidating the cellular and molecular basis of limb development in urodeles and other amphibians also has important implications for our understanding of how limbs evolved ( Alberch and Gale , 1983; Fröbisch and Shubin , 2011; Stopper and Wagner , 2005 ) . Despite deep homology among tetrapod limbs , biologists have long recognized several unique aspects of urodele limb development ( Holmgren , 1933 ) . For example , although the limb field in urodeles is established in the gastrula , the limb bud does not emerge from the flank until much later in a free-swimming larva . Unlike amniotes , this situation demonstrates a level of autonomous development that is temporally de-coupled from limb specification and patterning of the main body axis ( Stocum and Fallon , 1982 ) . Retinoic acid ( RA ) generated from the somites in chicks and mice diffuses into the lateral plate mesoderm where it permits correct spatiotemporal induction of Tbx5 ( Cunningham et al . , 2013; Nishimoto et al . , 2015; Stephens and McNulty , 1981 ) . In contrast , during anuran limb bud development which occurs in larval animals ( similar to urodeles ) , retinoic acid ( RA ) appears to be generated autonomously in the forelimb bud with raldh2 expressed proximally and cyp26b distally ( McEwan et al . , 2011 ) . In addition to limb heterochrony , urodele limb buds do not form an apical ectodermal ridge ( AER ) ( Sturdee and Connock , 1975; Tank et al . , 1977 ) and exhibit preaxial dominance of the limb skeleton ( Shubin and Alberch , 1986 ) . Lastly , urodeles can regenerate an entire limb , something no other group of tetrapods can do as adults ( Table 1 ) . Thus , while these aspects of urodele limb development challenge our notion of an inclusive vertebrate limb development model ( Zeller et al . , 2009 ) they also beg the question; does the molecular machinery directing limb morphogenesis exhibit critical differences when compared to amniotes and anurans ? In this study , we examined several key aspects of salamander limb development as they relate to patterning and outgrowth of the tetrapod limb . In doing so , we considered salient features of salamander forelimb development as they compare to Xenopus , chickens and mice ( summarized in Table 1 ) . First , we confirm previous reports that axolotls lack a morphological AER . Second , using Sox9 expression , we show that digit specification occurs first along the metapterygial axis of the limb with digit II , and then proceeds postaxial with digits I , III and IV . However , using Alcian blue staining as a proxy for cartilage condensation , we show that digits I and II differentiate simultaneously and are followed in sequence by digits III and IV . Thirdly , we show that Shh is restricted posteriorly , and although Shh expression does not overlap with the autopod , ZPA cells contribute to digit IV . Together with cyclopamine experiments these data support Shh-signaling as a key mediator of anterior-posterior patterning . Fourthly , we focused on the cellular source of Fgf-signaling and find that , in contrast to anurans and amniotes where reciprocal signaling is compartmentalized between the limb ectoderm and mesenchyme , Fgf ligands ( Fgf8 , 9 , 17 , 10 ) and receptors ( FgfR1-4 ) are all expressed solely in the mesenchyme . By functionally testing the requirement for Fgf-signaling using a broad Fgf-receptor antagonist ( SU5042 ) , we demonstrate that Fgf-signaling regulates limb size by controlling cell proliferation across all three limb axes . Again , this stands in contrast to anurans and amniotes where Fgf-signaling regulates cell survival and cellular differentiation along the proximal-distal axis . Another key finding from these experiments is that Fgf-signaling regulates Gremlin1 , whereas Shh-signaling is maintained in the face of Fgf-inhibition . While these results strongly suggest that a Shh-Grem-Fgf signaling loop is not present during salamander limb development , they do show that Fgf8 expression is dependent on Shh-signaling . Together , our results show Shh-signaling from the ZPA has maintained its core function while de-coupling itself from Fgf-signaling and that Fgf-signaling has evolved to regulate cell proliferation in the limb . Our analysis of skeletogenesis shows that axolotl digits appear to be specified in a different order than they differentiate . Taking advantage of subtle variations in within animal and between animal staging , our analysis using Sox9 expression shows a 2>1>3>4 pattern of digit specification . However , when we observed digit differentiation using Alcian blue to label condensing cartilage we always found digit I and II appearing together followed in sequence by digits III and IV . This pattern of digit differentiation is consistent with that observed by other investigators using histological preparations ( Fröbisch , 2008; Shubin and Alberch , 1986 ) or Alcian blue ( Nye et al . , 2003 ) . These findings support independent molecular control of digit specification and differentiation and hint at the wide diversity in ontogenetic and heterochronic shifts that have occurred in the limb during urodele evolution ( Blanco and Alberch , 1992; Franssen et al . , 2005 ) . Similarly , Sox9 expression shows spatiotemporal variability across urodeles ( Kerney et al . , 2018 ) and this underscores the need for more comparative limb studies to better understand the ancestral condition . While preaxial vs . postaxial dominance of skeletal formation clearly separates urodele limb development from anurans and amniotes , it is likely this difference points to alterations in the upstream genetic control of digit specification involving Fgfs , Bmps , and retinoic acid ( Montero et al . , 2017; ten Berge et al . , 2008 ) . With respect to anterior-posterior patterning , our data show that the members of the hedgehog signaling pathway are expressed in a mesenchymal pattern consistent with other tetrapods studied to date . However , our results also reveal three important differences . First , our data reveals that Shh expression appears relatively late during limb bud outgrowth and after Fgf8 and Hoxd11 are expressed . Second , salamander forelimb buds exhibit a proximal-anterior domain of Shh expression that emerges as the axolotl limb bud begins to bend posteriorly; a domain which is not found during normal limb development in amniotes and anurans . Third , posterior Shh expression corresponding to the ZPA remains in a relatively small proximal domain ( Torok et al . , 1999 ) rather than the elongated expression domain found in other tetrapods that extends the proximal-distal length of the autopod ( Matsubara et al . , 2017; Riddle et al . , 1993; Shapiro et al . , 2003 ) . With respect to the first point , previous work has demonstrated that flank tissue surrounding the limb field plays an important role in specifying the anteroposterior axis and thus , the temporal appearance of Shh expression may be less important than its posterior induction ( Stocum and Fallon , 1982 ) . In chick limbs , Hoxd11 can be induced by Shh , but only in proximity to the AER or in the presence of AER-Fgfs ( Laufer et al . , 1994 ) and in Shh KO mice Hoxd11 is still expressed in the limb bud ( Chiang et al . , 2001 ) . While intriguing , the transient appearance of an anterior Shh domain is difficult to explain . We did not detect an anterior necrotic zone ( Figure 5—figure supplement 2 ) , and Shh inhibition did not lead to increased cell death in this region . Moreover , Shh inhibition produced phenotypes consistent with other amniote models of limb development ( Chiang et al . , 2001; Scherz et al . , 2007; Stopper and Wagner , 2007; Towers et al . , 2008; Vargas and Wagner , 2009; Zhu et al . , 2008 ) . When Shh-signaling is inhibited in axolotls with cyclopamine , zeugopodial and autopodial skeletal elements are lost in a posterior to anterior direction that is dependent on when cyclopamine is administered ( Stopper and Wagner , 2007 ) . A similar phenotype occurs in chick embryos where elegant work demonstrated that Shh-signaling controls digit progenitor specification and limb bud growth ( Towers et al . , 2008 ) . When coupled with the results of Shh inhibition ( Stopper and Wagner , 2007 ) , our observations that Shh-signaling controls distal cell proliferation and that Shh-expressing cells contribute to digit IV , our findings support a conserved role for Shh-signaling as it pertains to specifying digit progenitors during limb development across all tetrapods . Future experiments ectopically expressing Shh using beads or virus will serve as a means to test downstream targets of Shh-signaling , as will transgenic elimination of Shh during development . In amniotes and anurans , limb bud outgrowth and proximal-distal patterning are controlled by the AER via expression of specific Fgf ligands ( e . g . Fgf4 , 8 , 9 , 17 ) ( Cohn et al . , 1995; Lewandoski et al . , 2000; Mariani et al . , 2008; Ohuchi et al . , 1997; Sun et al . , 2002 ) . The AER maintains ZPA activity ( Vogel and Tickle , 1993 ) and AER-Fgfs and Fgf-regulated Etvs are essential in inducing and maintaining posterior expression of Shh in the ZPA ( Niswander , 2002; Zhang et al . , 2009 ) . Furthermore , Gremlin1 acts as an intermediary between these signaling centers where it relays signals from the ZPA to control Bmp-signaling and maintain the AER ( Zeller et al . , 2009; Zuniga , 2015 ) . In stark contrast to anurans and amniotes , our data represents the first instance of a tetrapod lacking Fgf ligand and receptor expression in the developing limb bud ectoderm . Although previous studies suggested that Fgf8 was expressed early in salamander limb bud ectoderm ( Han et al . , 2001 ) , a result consistent with studies in Monodelphis domestica ( Doroba and Sears , 2010 ) and Eleutherodactylus coqui ( Gross et al . , 2011 ) which lack an AER but exhibit ectodermal Fgf8 expression , our findings during development show this is not the case . Moreover , our expression patterns for Fgf8 are consistent with recent data from regenerating axolotl limbs where Fgf8 expression was found restricted to the mesenchyme ( Nacu et al . , 2016 ) . Consistent with our spatiotemporal expression results , analysis of scRNA-seq data from stage 45 axolotl limb buds ( Gerber et al . , 2018 ) showed similar patterns along a modeled proximodistal axis . Co-expression analysis of Fgf ligands and receptors at stage 45 provide evidence that some cells expressing FgfR1 also express several Fgf ligands . Given the relatively small number of cells analyzed in this dataset ( <200 ) , a more complete scRNA-seq analysis across multiple time points will help address the complexity of Fgf-signaling and whether mesenchymal cells secreting Fgf ligands respond in an autocrine fashion . Our experiments also call into question what role , if any , the ectoderm plays during urodele limb development . For instance , experiments in Pleurodeles waltl showed that early stage forelimb bud mesoderm could autonomously develop a fully formed limb when grafted under heterologous flank epidermis ( Lauthier , 1985 ) . A clear functional role of the ectoderm awaits genetic manipulation since manual removal of the ectoderm results in rapid regeneration . Our experiments using the Fgf receptor antagonist SU5402 sought to test if spatial re-location of Fgf ligands and receptors affected the established function of Fgf-signaling during anuran and amniote limb bud development . SU5402 inhibition ( beginning before limb bud outgrowth ) revealed that the primary function of Fgf-signaling in axolotl limb development is to regulate cell proliferation throughout the limb bud . It does not , however , appear to control cellular differentiation , cell survival or proximodistal patterning . Removal of the AER in chickens ( Saunders , 1948 ) leads to a stage dependent loss of skeletal elements in a proximal to distal direction and combined genetic deletion of Fgf8/Fgf4/Fgf9 in the mouse AER ( Mariani et al . , 2008 ) leads to a complete loss of the stylopod , zeugopod and autopod . In contrast , broadly inhibiting Fgf-signaling from the outset of axolotl limb development phenocopies urodele limbs treated with the mitotic inhibitor colchicine where in both cases , the most posterior digit fails to form in an otherwise smaller , but normal limb ( Alberch and Gale , 1983 ) . These results echo experiments in chickens where pharmacological inhibition of cell proliferation using trichostatin A , colchicine , or vinblastine leads to loss of anterior digits , which are the last to form ( Towers et al . , 2008 ) . Broadly inhibiting Fgf-signaling also shows that Shh operates independently of Fgf-signaling , while our cyclopamine experiments show that Shh-signaling regulates Fgf8 expression . In addition , our finding that Gremlin1 is regulated by Fgf-signaling suggests further rearrangement of genetic interactions observed in amniotes ( Lewandoski et al . , 2000; Sun et al . , 2002 ) . Together , these data show that while the role of Shh-signaling from the ZPA is conserved in salamander limb development , movement of amniote/anuran AER-Fgf ligands to the mesenchyme was accompanied by a change in how Fgf-signaling regulates limb development . Ultimately , our findings support similar molecular players being deployed during limb development across tetrapods but demonstrate that a divergent molecular program in urodeles resides predominantly in one cellular compartment: the limb mesenchyme . While it is tempting to speculate that mesenchymal compartmentalization of limb developmental signaling is somehow causally related to regenerative ability , available data suggests otherwise . Data from basal Actinopterygians ( paddlefish ) , Chondrichthyians ( catfish ) and anurans show ectodermal-mesenchymal segregation of the Shh- and Fgf-signaling ( Christen and Slack , 1998; Tulenko et al . , 2017 ) . Pectoral fin regeneration is an ancient feature of Actinopterygians , Sarcopterygians and Chondrichthyians suggesting that mesenchymal core signaling alone is not exclusive to regenerating species . On the flip side , anurans show subtle molecular differences during limb development compared to amniotes , especially along the dorsoventral axis ( Christen and Slack , 1998 ) . Thus , it is plausible that ectodermal-mesodermal compartmentalization was the first step toward developmental canalization that ultimately increased robustness in the limb program , specifically in the autopod . Future limb development studies using a diverse array of anurans and lungfish will help shed light on these questions as will continued studies in salamanders and newts .
Axolotls ( Ambystoma mexicanum ) were acquired from the Ambystoma Genetic Stock Center ( Lexington , KY ) and from our own laboratory colony . Chicken eggs ( University of Kentucky , Department of Animal Sciences ) were incubated to stage and mouse embryos were harvested from Swiss Webster mice ( ND4 , Envigo , Indianapolis , IN ) . All procedures were conducted in accordance with , and approved by , the University of Kentucky Institutional Animal Care and Use Committee ( IACUC Protocol: 2013–1174 ) . Axolotl embryos were kept at 20–21°C and reared in glass bowls ( 7 . 5’/6’/2 . 5’ dimensions ) in 800 ml 20% Holtfreter’s solution . 15–20 larvae were kept in a single bowl and were fed with brine shrimps from ~3 weeks post-fertilization . Larvae used for drug treatments , proliferation , cell death assays , qRT-PCR and DiI labeling were reared in 6-well plates . Larvae used for forelimb staging and area/snout to tail-tip measurements , Alcian blue-Alizarin red staining , in situ hybridization , histology and proliferation assays were anesthetized using 1x benzocaine ( Sigma ) and fixed overnight in 4% paraformaldehyde at 4°C . Larvae used for cell death assay were rinsed gently in Hanks BSS four times , five mins each on a rocker , fixed overnight in 4% PFA post-Lysotracker treatment . Developmental stages were referenced against previously reported post-hatching stages ( Nye et al . , 2003 ) and were extended as outlined in Figure 1 . Individual animals ( n = 35 ) were examined every day to assess limb stage . Larvae used for qRT-PCR were anesthetized using 1x benzocaine ( Sigma ) , limb tissue samples were snap frozen and stored at −80°C until further use . Chicken ( HH25 ) and mouse ( E11 . 5 ) embryos were harvested , fixed overnight in 4% paraformaldehyde at 4°C and processed for in situ hybridization experiments . Larvae were fixed overnight , washed three times , 10 mins each in PBT ( Phosphate Buffer Saline and 1% Tween 20 ) , dehydrated in graded ethanol series 25% , 50% , 75% and stored in 100% ethanol at −20°C until further use . Dehydrated larvae were washed three times , 10 mins each in 1x PBS ( Phosphate Buffer Saline ) , stained with 0 . 02% Alcian blue 8GX ( Sigma Aldrich ) in 70% ethanol and 30% glacial acetic acid for 3 hr to overnight . Stained larvae were rehydrated in graded ethanol series ( 100% , 75% , 50% and water 1 hr each ) and stained with 0 . 1% Alizarin Red ( Sigma Aldrich ) in 1%KOH overnight . Larvae were cleared in 1%KOH/glycerol series: 3KOH:1glycerol ( imaged when cleared ) , 1KOH:1glycerol ( 1 day ) and 1KOH:3glycerol ( stored at room temperature ) . Larvae were fixed overnight , washed twice in 1x PBS and stored in 70% ethanol at 4°C until further use . Larvae were then processed for paraffin embedding ( Histo5 Tissue Processor , Milestone ) and tissue samples were sectioned at 5 μm . H and E staining was done on deparaffinized and rehydrated sections . For bright-field visualization of limb buds , Mayer’s haematoxylin was used to counterstain the nuclei and coverslips were mounted with Cytoseal XYL ( ThermoFisher , Waltham , MA ) . Coding sequences for axolotl genes were obtained from NCBI , ( Bryant et al . , 2017 ) or www . ambystoma . org ( Supplementary file 1 ) . Axolotl sequences were aligned with human homologs to locate the 3’UTRs . Primers were designed against ( when possible ) or close to 3’UTRs of the axolotl sequences . Briefly , RNA was extracted using Trizol reagent from stage 31 larvae , stage 32 larvae or regenerating early forelimb buds from 5 to 10 cm juveniles . cDNA was synthesized from 1 to 0 . 5 μg RNA using SensiFast cDNA synthesis kit . Coding genes sequences were amplified out using Advantage HD polymerase kit and amplified products were ligated into pGEMT-Easy vector ( Promega ) via T-A cloning using manufacture’s protocol . Plasmids were transformed into Max Efficiency DH5α cells ( Invitrogen ) and blue-white colonies were obtained . Colony PCR was done to confirm insert sequence size and positive colonies were picked for plasmid mini-prep ( Qiagen ) . Plasmids were sent out for sequencing . Gene sequences and orientation of insertion into vector was verified and positive colonies were used for plasmid maxiprep ( Zymo Research ) . Plasmids containing Shh and Fgf8 genes for chicken and mouse were a gift from the Cohn lab , University of Florida . Plasmids containing Keratin5 and Keratin17 genes for axolotl were a gift from the Satoh lab , Okayama University . Plasmids containing Fgf8 and Gli3 genes for axolotl were a gift from the Tanaka lab , IMP , Austria . ~ 20 μg of plasmid was linearized using specific restriction enzymes to obtain the sense and antisense probe templates . Sense and antisense probes for axolotl genes were synthesized from 2 . 5 μg linearized plasmids using DIG RNA Labeling Kit ( SP6/T7 ) ( Roche ) . Sense and antisense probes for chicken and mouse genes were synthesized from 1 μg linearized plasmids using the same kit . In vitro transcription of probes was carried out for 3 hr to overnight at 37°C . Probes were treated with 1unit DNAse ( Promega , CAT#M610A ) for 1 min at 37°C and reaction was terminated with 2 μl DNAse stop solution ( Promega , CAT#M198A ) . Probes were purified and eluted in 50 μl of nuclease-free water using mini Quick Spin RNA Columns ( Roche ) or RNeasy MinElute Cleanup Kit ( Qiagen ) , run on 1% agarose gel to access quality and quantified using NanoDrop . Larvae/embryos fixed overnight were washed three times , 10 min each in PBT ( Phosphate Buffer Saline and 1% Tween 20 ) , dehydrated in graded methanol/PBT series 25% , 50% , 75% and stored in 100% at −20°C until further use . For all the experiments at least three larvae or embryos/stage/gene were used . Each axolotl larva was decapitated , and the bottom half of the trunk was amputated . Two axolotl larvae per stage were placed in a DNAse/RNAse free 2 ml tube and treated with 2 ml of each solution . For chicken and mouse , one embryo was placed in a similar 2 ml tube . Dehydrated larvae/embryos were rehydrated in a graded methanol/PBT series 75% , 50% , 25% , washed with PBT twice 5 min each , bleached with 6% H2O2/1x PBS for 1 hr , washed with PBT twice 5 min each . The above steps were done under ice-cold conditions . Larvae/embryos were permeabilized with 20 μg/ml Proteinase K ( Roche ) in PBS for 7–10 min ( 40 μg/ml Proteinase K was used for Sox9 and Ihh axolotl genes ) , washed with PBT twice for 5 min each , fixed with 0 . 2% Gluteraldehyde/4% paraformaldehyde and washed with PBT twice for 5 min each . The above steps were done at room temperature . Larvae/embryos were incubated overnight in hybridization buffer ( 5% Dextran sulphate , 2% blocking powder from Roche , 5X SSC , 0 . 1% TritonX , 0 . 1% CHAPS from Sigma Aldrich , 50% formamide , 1 mg/ml tRNA from Roche , 5 mM EDTA from Sigma and 50 μg/ml Heparin from Sigma ) at 65°C . The tubes were replaced with fresh hybridization buffer , 0 . 1–1 μg of probe was added into each vial and incubated at 65°C for 2 days . High stringency washes were done with 2X SSC/0 . 1% CHAPS thrice for 20 min each , 0 . 2X SSC/0 . 1% CHAPS 4 times for 25 mins each and with KTBT ( 15 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 10 mM KCl and 1% Tween 20 ) twice for 5 min each . Larvae were blocked with 20% goat serum in KTBT for 3 hr . Later , fresh blocking solution was added . An anti-Digoxigenin-AP , Fab fragment antibody ( Roche ) was added at 1:3000 dilution and incubated overnight at 4°C . Larvae were washed with KTBT 5 times for 1 hr each and then incubated in KTBT overnight at 4°C . Larvae were washed with NTMT ( 100 mM Tris-HCl pH 9 . 5 , 50 mM MgCl2 , 100 mM NaCl and 1% Tween 20 ) twice for 5 min each and incubated in NBT/BCIP ( Roche ) solution in NTMT ( BCIP-0 . 17 mg/ml , NBT-0 . 33 mg/ml , 10% DMF ) till a signal developed with minimum background staining . Limbs were photographed and larvae were washed with TE buffer ( 10 mM Tris HCl , pH 8 and 1 mM EDTA , pH eight made up in DEPC treated water ) 3 times for 10 mins each and fixed in 4% PFA until further use . Larvae from Keratin5 and Keratin17 in situ hybridizations were washed in TE buffer ( 10 mM Tris HCl pH 8 and 1 mM EDTA pH 8 ) 3 times for 10 min each , fixed in 4% PFA for 1 hr and washed with 1x PBS 3 times for 5 min each . Larvae were transferred into 2 ml vials with 30% sucrose in 1x PBS and placed on a rotor for 20 mins till they sank to the bottom . Larvae were then placed in OCT for 25 min and frozen for cryo-sections . Cryo-sections were taken at 5 µm thickness , dried overnight at 37°C , fixed with 4% PFA for 10 min , washed with 1x PBS 3 times for 5 min each , treated with Hoechst solution ( 1:10 , 000 dilution ) , air dried , sealed with ProLong Gold antifade reagent ( Invitrogen ) and imaged . For drug experiments , post-hatch axolotl larvae were reared in 6-well plates in either 1 . 5% DMSO , 45 µM SU5402 , 0 . 02% ethanol or 1 µg/ml cyclopamine till stage 46 ( see below or treatment details ) . Whole limbs were dissected from the body wall , immediately snap frozen and stored at −80°C until RNA extraction . Three replicates were used for each condition ( treatment/control ) and each replicate represented a pool of limbs ( both left and right ) from 10 to 20 animals . To validate the anterior versus the posterior expression of Shh , Fgf8 and Gli3 genes , larvae were reared in glass bowls ( 7 . 5’/6’/2 . 5’ dimensions ) in 800 ml 20% Holtfreter’s solution . Whole limbs were dissected from the body wall , further dissected into anterior and posterior halves/compartments ( Figure 2—figure supplement 3D ) , immediately snap frozen and stored at −80°C until RNA extraction . n = 2 or 3 was used and each set was pooled from 20 limbs ( both right and left ) . RNA was extracted using Trizol reagent ( Invitrogen ) . cDNA was synthesized from 1 to 0 . 5 μg RNA using SensiFast cDNA synthesis kit . The following primers were used for qRT-PCR: Tubulin alpha was used as the internal control/house-keeping gene for limbs obtained from larvae reared in 20% Holtfreter’s solution . GAPDH and RLP32 were used as the internal control/house-keeping genes for the DMSO/SU5402 and ethanol/cyclopamine experiments , respectively , since there was no significant fold change in the 2-Ct values ( Schmittgen and Livak , 2008 ) ( Supplementary file 2 ) . 2-ΔΔCt method was used to calculate the fold change values between control ( DMSO or ethanol ) and treatment ( SU5402 or cyclopamine ) groups ( Schmittgen and Livak , 2008 ) . Post-hatch axolotl larvae were reared in 6-well plates in 3 ml Holtfreter’s solution . DiI ( D-282 , molecular probes ) was dissolved in dimethyl formamide at 3 mg/ml concentration . Larvae between stages 45 and 46 were anesthetized in 1x benzocaine and approximately 5 nl ( 0 . 05–0 . 20 mm diameter ) of DiI was injected into the approximate position of the posterior Shh domain . A total of 34 limbs were analyzed for this experiment . Images were taken immediately to confirm the domain specific restriction of the injection and fluorescence was tracked every 2 days till all four digits formed completely . All images were taken post-anesthesia in 1x benzocaine . Single-cell RNA-seq data from embryonic limb buds ( Gerber et al . , 2018 , Table S7 ) were analyzed as follows: Gene expression ( log2 ( TPM ) ) from stages 40 and 44 ( as reported in Gerber ) , and which correspond to our stages 44 and 45 ( Prayag Murawala , personal communication ) were used as input for principal component analysis . The top twenty loadings of the positive and negative axes of the first five principal components were inspected to identify principal components that segregated developmental axes markers on separate ends . No reliable anterior-posterior , or dorsal-ventral markers segregated in this manner in the current dataset . However , proximal-distal markers ( Meis2 and Hoxd11 ) were segregated on opposite ends of principle component three in the PCA of cells from stage 44 , but not 40 . In order to orient PC3 so that Hoxd11 expression is on the right and Meis2 on the left ( i . e . , conventional proximal-distal orientation ) , PC3 values were multiplied by -1 . To determine co-expression of Fgf ligands and receptors , expression was defined as the presence of one or more transcript per million . Axolotl larvae at pre-limb bud stage 39 were reared in 6-well plates and treated with 3 ml solutions for all drug experiments . A working stock of 3 mM SU5402 ( Sigma ) was made in DMSO and then diluted to 45 μM in 3 ml 20% Holtfreter’s solution per well . Control larvae were treated with an equivalent amount of DMSO ( 1 . 5% ) in 20% Holtfreter’s solution . Larvae were kept in the dark and the solution was changed every three days . At three weeks post fertilization each larva was fed 20–30 brine shrimp every day . Cyclopamine treatments were performed as previously described ( Stopper and Wagner , 2007 ) . A working stock of 5 mg/ml of cyclopamine was made in 100% ethanol and 0 . 6 μl from this stock was added into 3 ml 20% Holtfreter’s solution per well ( 1 μg /ml final concentration ) . An equal amount of 100% ethanol ( 0 . 02% ) was added into control wells . Larvae were kept in dark and solutions were replenished every two days . At three weeks post fertilization each larva was fed 20–30 brine shrimps every day . Limb area was measured for stage 45 and stage 46 larvae that were reared in 1 . 5% DMSO , 45 μM SU5402 , 0 . 02% ethanol or 1 μg/ml cyclopamine . Snout to tail-tip lengths were measured for stage 46 larvae that were reared in 1 . 5% DMSO , 45 μM SU5402 , 0 . 02% ethanol or 1 μg/ml cyclopamine . All measurements were made using Fiji software ( NIH ) after calibrations . Post-hatch larvae were reared in 6-well plates in 3 ml of either of the solutions: 20% Holtfreter’s solution , 1 . 5% DMSO , 45 μM SU5402 , 0 . 02% ethanol or 1 μg/ml cyclopamine . Larvae were additionally treated with 0 . 1 mg/ml of EdU at stage 46 for 24 hr , fixed overnight in 4% PFA , washed with 1x PBS twice for 5 min , dehydrated in 1x PBS/methanol series ( 25% , 50% , 75% and 100% methanol , 5 min each ) and stored in 100% methanol at −20°C until further use . For EdU staining , all steps were performed at room temperature unless mentioned otherwise . Larvae were rehydrated backwards through the methanol series starting at 100% methanol and ending at 100% 1x PBS . This was followed by PBT washes for 5 min twice and 2 . 5% trypsin ( Gibco ) treatment for 10 min . The larval limbs were checked for clarity at this point . Larvae were washed with water for 5 min twice , treated with 20 μg/ml of proteinase K in PBT for 7–10 min , washed with water for 5 min twice , fixed in 100% acetone at −20°C for 10 min , washed with water for 5 min once , washed with PBT for 5 min once , incubated in fresh click reaction solution ( 1x TRIS buffer saline , 4 mM CuSO4 in 1x TRIS buffer saline , 2 μl Alexa-flour-594 Azide ( Life technologies ) , 1 mM sodium ascorbate in 1x TRIS buffer saline ) for 30 min on a rocker in the dark , washed with 1x PSB for 5 min thrice , incubated in DAPI ( 1:1000 dilution ) for 30 mins , washed with 1x PSB for 5 min thrice , checked for fluorescence under a stereomicroscope and stored at 4°C in the dark till lightsheet imaging . Post-hatch larvae were reared in 6-well plates in 3 ml of either of the solutions: 20% Holtfreter’s solution , 1 . 5% DMSO , 45 μM SU5402 , 0 . 02% ethanol or 1 μg /ml cyclopamine . Larvae were transferred into 24-well plates and treated with 200 μl of 5 μM LysoTracker Red DND-99 ( molecular probes ) in Hanks BSS for 45 min to 1 hr at 20–21°C . Larvae were rinsed gently in Hanks BSS four times , 5 min each on a rocker , fixed overnight in 4% PFA , rinsed once in Hanks BSS for 10 min and dehydrated through a methanol series in Hanks BSS ( 50% , 75% , 80% , 100% , 5 min each step ) to eliminate background staining and stored in 100% methanol at −20°C until imaging . Whole mount images for limb staging/size measurements , Alcian blue-Alizarin red staining , in situ hybridization , DiI experiments and apoptosis assays were taken on an SZX10 light microscope ( Olympus , Tokyo , Japan ) using a DP73 CCD camera ( Olympus ) . Figure 1—figure supplement 1 depicts all different forelimb stages to scale . Forelimb images are presented unscaled for in situ hybridizations except where treatment/control limbs are presented . Images for H and E staining and cryo-sections ( from Keratin5 and Keratin17 wholemount hybridizations ) were taken on a BX53 microscope ( Olympus , Tokyo , Japan ) with DP80 CCD camera . Both the microscopes were equipped with CellSense software ( CellSense V 1 . 12 , Olympus corporation ) . EdU stained stage 46 larvae were imaged using a Zeiss Lightsheet Z . 1 ( College of Arts and Science Imaging Centre , University of Kentucky ) . Larvae were embedded in 1% low melting agarose ( Sigma ) dissolved in 1x PBS , mounted in a glass capillary ( 2 . 15 mm inner diameter ) and the glass capillary was placed in a chamber ( specific for the 20x objective lens ) filled with 1x PBS . Imaging was done with Zen software ( Zeiss ) and samples were excited using 561 nm and 488 nm lasers . 20x objective was used and images of all the limbs were taken at 0 . 75 zoom . Either right , left or both lightsheets were used based on the orientation of the limbs . Image processing and analysis was done with Arivis vision4D software ( Arivis ) . The czi extension files from Zen software was imported into the Arivis vision4D software . An object mask was hand drawn at each z-planes based on the DAPI signal to outline the limb and create a limb mask for total limb volume calculations . The following pipeline was made to calculate red cell aggregate volume and total limb volume: Input ROI ( current image set , all channels , scaling 50% ) , object mask ( add the hand drawn object ) , denoising filter ( mean , radius = 10 ) , Intensity filter ( radius = 9 , K = 0 , offset = −10 , binary - false ) , denoising filter ( median , radius = 10 ) and result storage ( intensity threshold , range specified , allow holes while quantification ) . Volume values in μm3 and voxel counts were given as outputs . All statistical analyses were made using JMP ( version Pro 12 . 10 , SAS Institute Inc ) . Box plots for forelimb staging were made using graph builder in JMP . The vertical line within the box represents the median number of larvae found at a specific stage and days post fertilization . The ends of the box represent the 25th and 75th quartiles and the whiskers on either side represent the interquartile range . For qRT-PCR data the 2-ΔΔCt method was used to calculate fold changes of genes between control ( DMSO or ethanol ) and treatment ( SU5402 or cyclopamine ) groups . Calculations for mean Ct values , ΔCt values for experimental and control groups , ΔΔCt values between experimental and control groups , 2-Ct and 2-ΔΔCt fold change values were made using Microsoft Excel and detailed in Supplementary file 2 . Student’s t-test was used to calculate the significant changes in relative gene expressions between control and treatment groups . For limb size and snout to tail-tip measurements , student’s t-test was used to calculate the significant changes between control and treatment groups . Differences were considered significant if p<0 . 05 . To analyze whether the decrease in limb size post drug treatment was due to the decrease in overall body length ( snout to tail-tip length ) we used a fit model ANOVA with treatment , overall body length and treatment*overall body length interaction . To evaluate co-expression of Fgf ligands and receptors along the limb proximal-distal axis from single-cell RNAseq data , we utilized the fact that contribution to principle component three followed a normal distribution according to a Shapiro-Wilks normality test ( W = 0 . 95551 , p-value<2 . 2e-16 ) . A Welch Two Sample t-test test was conducted to test the hypothesis that two populations ( Fgf ligands and receptors ) have the same mean contribution to principle component 3 . For lightsheet data , red cell aggregate volume/limb volume ( % ) was calculated in Microsoft Excel . Comparisons within control and treatment groups were determined by one-way ANOVA ( Kruskal-Wallis tests ) . | Salamanders are a group of amphibians that are well-known for their ability to regenerate lost limbs and other body parts . At the turn of the twentieth century , researchers used salamander embryos as models to understand the basic concepts of how limbs develop in other four-limbed animals , including amphibians , mammals and birds , which are collectively known as “tetrapods” . However , the salamander’s amazing powers of regeneration made it difficult to carry out certain experiments , so researchers switched to using the embryos of other tetrapods – namely chickens and mice – instead . Studies in chickens , later confirmed in mice and frogs , established that there are two major signaling centers that control how the limbs of tetrapod embryos form and grow: a small group of cells known as the “zone of polarizing activity” within a structure called the “limb bud mesenchyme”; and an overlying , thin ridge of cells called the “apical ectodermal ridge” . Both of these centers release potent signaling molecules that act on cells in the limbs . The cells in the zone of polarizing activity produce a molecule often called Sonic hedgehog , or Shh for short . The apical ectodermal ridge produces another group of signals commonly known as fibroblast growth factors , or simply Fgfs . Several older studies reported that salamander embryos do not have an apical ectodermal ridge suggesting that these amphibian’s limbs may form differently to other tetrapods . Yet , contemporary models in developmental biology treated salamander limbs like those of chicks and mice . To address this apparent discrepancy , Purushothaman et al . studied how the forelimbs develop in a salamander known as the axolotl . The experiments showed that , along with lacking an apical ectodermal ridge , axolotls did not produce fibroblast growth factors normally found in this tissue . Instead , these factors were only found in the limb bud mesenchyme . Purushothaman et al . also found that fibroblast growth factors played a different role in axolotls than previously reported in chick , frog and mouse embryos . On the other hand , the pattern and function of Shh activity in the axolotl limb bud was similar to that previously observed in chicks and mice . These findings show that not all limbs develop in the same way and open up questions for evolutionary biologists regarding the evolution of limbs . Future studies that examine limb development in other animals that regenerate tissues , such as other amphibians and lungfish , will help answer these questions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"evolutionary",
"biology"
] | 2019 | Fgf-signaling is compartmentalized within the mesenchyme and controls proliferation during salamander limb development |
Circadian oscillations emerge from transcriptional and post-translational feedback loops . An important step in generating rhythmicity is the translocation of clock components into the nucleus , which is regulated in many cases by kinases . In mammals , the kinase promoting the nuclear import of the key clock component Period 2 ( PER2 ) is unknown . Here , we show that the cyclin-dependent kinase 5 ( CDK5 ) regulates the mammalian circadian clock involving phosphorylation of PER2 . Knock-down of Cdk5 in the suprachiasmatic nuclei ( SCN ) , the main coordinator site of the mammalian circadian system , shortened the free-running period in mice . CDK5 phosphorylated PER2 at serine residue 394 ( S394 ) in a diurnal fashion . This phosphorylation facilitated interaction with Cryptochrome 1 ( CRY1 ) and nuclear entry of the PER2-CRY1 complex . Taken together , we found that CDK5 drives nuclear entry of PER2 , which is critical for establishing an adequate circadian period of the molecular circadian cycle . Of note is that CDK5 may not exclusively phosphorylate PER2 , but in addition may regulate other proteins that are involved in the clock mechanism . Taken together , it appears that CDK5 is critically involved in the regulation of the circadian clock and may represent a link to various diseases affected by a derailed circadian clock .
The circadian clock , prevalent in most organisms , is an evolutionary adaptation to the daily light-dark cycle generated by the sun and the earth’s rotation around its own axis ( Rosbash , 2009 ) . This clock allows organisms to organize physiology and behavior over the 24 hr time scale in order to adapt and thus optimize , body function to predictably recurring daily events . Malfunctioning or disruption of the circadian clock in humans results in various pathologies including obesity , cancer , and neurological disorders ( Roenneberg and Merrow , 2016 ) . In order to maintain phase synchronicity with the environmental light-dark cycle , the suprachiasmatic nuclei ( SCN ) , a bipartite brain structure located in the ventral part of the hypothalamus above the optic chiasm , receive light information from the retina . The SCN convert this information into humoral and neuronal signals to set the phase of all circadian oscillators in the body ( Dibner et al . , 2010 ) . In order to measure the length of one day , organisms have developed cell-based molecular mechanisms relying on feedback loops involving a set of clock genes . The existence of such loops was suggested by the analysis of Drosophila having various mutations in their period ( per ) gene ( Hardin et al . , 1990 ) . Further studies completed the picture of intertwined transcriptional feedback loops at the heart of the Drosophila circadian oscillator ( Darlington et al . , 1998 ) . Every day , per accumulates to a certain concentration upon which it enters into the nucleus together with timeless ( tim ) . This protein complex inhibits transcriptional activation mediated by dClock and cycle acting on the expression of per and tim . After the degradation of the inhibitor complex , the repression is relieved and a new circadian cycle starts . To fine-tune the period of the circadian oscillator , kinases regulate the accumulation and nuclear entry of per and tim . The kinase double-time ( dbt ) phosphorylates per to destabilize it and to prevent its transport into the nucleus ( Kloss et al . , 1998; Price et al . , 1998 ) . On the other hand , the kinase shaggy ( shg ) phosphorylates tim to stabilize the heterodimer and to promote its nuclear translocation ( Martinek et al . , 2001 ) . Many other kinases and phosphatases are necessary to complete the Drosophila circadian cycle and to adjust its phase to the external light-dark rhythm ( Garbe et al . , 2013 ) . The circadian oscillator of mammals is arranged very similarly to the one of Drosophila , with some modifications ( Dibner et al . , 2010; Takahashi , 2017 ) . For instance , the function of Drosophila tim to escort per into the nucleus was replaced by the Cryptochromes ( Cry ) in the mammalian system ( van der Horst et al . , 1999 ) . Furthermore , the first mutation to affect the mammalian circadian oscillator , Tau , was later mapped to Casein kinase Iε ( CK1ε ) , which is the Drosophila dbt orthologue ( Lowrey et al . , 2000 ) . One of the sites phosphorylated by CK1ε within human PER2 is mutated in the Familial Advanced Sleep Phase Syndrome ( FASPS ) ( Toh et al . , 2001 ) . This mutation and also the Tau mutation were subsequently introduced into the mouse genome to prove their functional relevance ( Meng et al . , 2008; Xu et al . , 2007 ) . However , a kinase similar to the function of shg in Drosophila , which stabilizes and promotes the import of PER proteins into the nucleus of mammals ( Hirano et al . , 2017 ) , has not been identified . Interestingly , PER2 contains over 20 potential phosphorylation sites ( Vanselow et al . , 2006 ) , indicating that mammalian PER and specifically PER2 are highly regulated at the post-translational level . This degree of phosphorylation is probably contributing to the precise rhythmicity of PER2 , which stands out as a crucial feature of the core clock ( Chong et al . , 2012 ) . Among the plethora of kinases identified that phosphorylate mammalian clock proteins , cyclin-dependent kinase 5 ( CDK5 ) was found to target CLOCK ( Kwak et al . , 2013 ) . CDK5 is a proline-directed serine-threonine kinase belonging to the Cdc2/Cdk1 family that is controlled by the neural specific activators p35 , p39 ( Tang et al . , 1995; Tsai et al . , 1994 ) , and cyclin I ( Brinkkoetter et al . , 2009 ) . CDK5 regulates various neuronal processes such as neurogenesis , neuronal migration , and axon guidance ( Kawauchi , 2014 ) . Outside of the nervous system CDK5 regulates vesicular transport , apoptosis , cell adhesion , and migration in many cell types ( Contreras-Vallejos et al . , 2012 ) . It has been proposed that CDK5 modulates the brain reward system ( Benavides et al . , 2007; Bibb et al . , 2001 ) and that it is consequently linked to psychiatric diseases ( Engmann et al . , 2011; Zhu et al . , 2012 ) . Interestingly , the clock components PER2 and CLOCK have been associated with the same processes ( Abarca et al . , 2002; Hampp et al . , 2008; Roybal et al . , 2007 ) . However , it is unknown whether CDK5 plays an important role in the central oscillator of the circadian clock . In this study , we wanted to identify proteins promoting the nuclear transport of PER2 with focus on kinase ( s ) acting similarly to shg . Using a genetic synthetic lethal dosage screen in yeast , we observed a genetic interaction between Per2 and PHO85 , which encodes a cyclin-dependent protein kinase that is orthologous to CDK5 in mammals . Subsequent experiments in mice demonstrated that silencing of Cdk5 in the SCN shortened the clock period . Our study identified CDK5 as a critical protein kinase in the regulation of the circadian clock and in particular as an important regulator of the crucial clock component PER2 .
In order to gain insight into the regulation of PER2 function in mice , we initially tried to identify genes that genetically interact with Per2 in budding yeast by using a variation of the Synthetic Genetic Array ( SGA ) method ( Tong , 2001 ) . To this end , we carried out a synthetic dosage lethality ( SDL ) screen , which is based on the concept that a high dosage of a given protein ( i . e . PER2 in this case ) may have negligible effect on growth in wild-type cells ( as we found to be the case for PER2; Figure 1A ) , but may compromise growth in mutants that have defects in pathway components or in functionally related processes ( Measday et al . , 2005; Sopko et al . , 2006 ) . Of note , SDL screens have been instrumental in the past to specifically predict the relationship between protein kinases and their targets ( Sharifpoor et al . , 2012 ) . Our search in a yeast knockout collection ( encompassing 4857 individual deletion strains ) for mutants that exhibited significantly reduced growth when combined with increased dosage of PER2 ( see Materials and methods for further details ) allowed us to isolate three mutants , namely eap1∆ , gnd1∆ , and pho85∆ ( Figure 1A ) . Among these , the strain lacking the cyclin-dependent protein kinase Pho85 was most dramatically compromised for growth in the presence of high doses of PER2 . Hence , Pho85 antagonizes the growth-inhibitory effect of PER2 in yeast , which indicates that the Pho85-orthologous CDK5 may potentially act upstream of PER2 in mammalian cells . The protein kinase CDK5 is mostly expressed in the brain and has previously been implicated in phosphorylation of mammalian CLOCK ( Kwak et al . , 2013 ) . However , the functional relevance of CDK5 for the clock mechanism has never been tested . Therefore , we investigated whether CDK5 affected the functioning of the circadian clock . First , we assessed whether CDK5 displayed time of day-dependent expression and activity in the SCN , the master clock of the circadian system . We collected SCN samples every 4 hr starting from ZT0 until ZT20 ( ZT0 = light on , ZT12 = light off ) , and performed western blots on total extracts using specific antibodies ( Figure 1B ) . The immunoblot against CRY1 showed a diurnal profile of this protein with a peak during the late-night phase , confirming that the mice were entrained properly to the light-dark cycle . In contrast , the CDK5 accumulation profile seemed to be unaffected by the time of day ( Figure 1B ) . Next , we investigated whether CDK5 kinase activity displayed a diurnal profile . While CDK5 levels did not change significantly over one day ( Figure 1B ) , we observed that histone-H1 , a known CDK5 target ( Peterson et al . , 2010 ) , was phosphorylated by this kinase in a time of day-dependent manner , with the highest levels of CDK5 activity observed at ZT12 to ZT20 , that is during the dark phase ( Figure 1C ) . Phosphorylation of histone-H1 was specifically blocked by roscovitine , a CDK5 inhibitor ( Hsu et al . , 2013 ) , whereas LiCl , a Gsk3β inhibitor , did not affect this phosphorylation ( Figure 1D ) , suggesting a CDK5-specific phosphorylation . Altogether , these data demonstrated that CDK5 kinase activity ( but not protein accumulation ) was diurnal in the SCN . Since CDK5 activity displayed a diurnal profile in the SCN , we tested whether knock-down of CDK5 in the master clock of the SCN changed circadian behavior in mice . To this end , we tested various shRNAs against Cdk5 in NIH 3T3 fibroblast cells ( Figure 2—figure supplement 1 ) and subsequently injected into the SCN region adeno-associated viral particles containing expression vectors for either a scrambled set of shRNA or a Cdk5-specific shRNA ( variant D , Figure 2—figure supplement 1 ) . After recovery from the procedure the animals were transferred into cages containing a running-wheel in order to assess their activity profiles . The control animals expressing the scrambled set of shRNA displayed normal activity in the light-dark ( LD ) cycle with precise onset of activity at the beginning of the dark phase ( ZT12 ) . This onset of activity was less precise in mice with a Cdk5 knock-down ( shCdk5 ) but comparable to animals with a deletion mutation in the clock gene Per2 , designated as Per2Brdm1 ( Figure 2A , Figure 2—figure supplement 2 ) . In constant darkness ( DD ) , χ2-periodogram analysis revealed a normal average free-running period for the scramble control mice , whereas for shCdk5 and Per2Brdm1 , the period was significantly shortened ( Figure 2B ) . In one case , the shCdk5 animals became arrhythmic ( Figure 2C ) , again comparable to Per2Brdm1 mice that eventually became arrhythmic in DD as well ( Zheng et al . , 1999 ) . The total wheel-running activity was significantly reduced in shCdk5 and Per2Brdm1 mice under DD as well as under LD conditions when compared with the scrambled control animals ( Figure 2—figure supplement 3 ) . The reduction of activity in the mutants under LD conditions is confined to the dark phase , but comparable between all three genotypes in the light phase ( Figure 2—figure supplement 4 ) . These results indicate that the period of the clock is affected by the lack of Cdk5 expression in the SCN . Interestingly , period in Per2Brdm1 mutant and wild-type shCdk5 knocked-down mice was not significantly different ( Figure 2B ) , suggesting that CDK5 activity is linked to PER2 as indicated by our SDL screen ( Figure 1 ) . In order to test the contribution of Cdk5 , we knocked down Cdk5 in Per2Brdm1 mutant mice . This even further shortened period in Per2Brdm1 mutant animals compared to scramble control Per2Brdm1 animals ( Figure 2D , E , Figure 2—figure supplement 5 ) . This effect , however , was not simply additive ( Δ wt versus wt KD ≈ 0 . 8 hr; Δ Per2Brdm1 versus Per2Brdm1 KD ≈ 0 . 6 hr , Figure 2F ) . Additionally , Δ wt KD versus Per2Brdm1 KD ≈ 0 . 8 hr , indicating that the difference in genetic background plays an important role . Overall , our observations suggest that Cdk5 may affect period partially via PER2 but also via additional factors ( e . g . CLOCK , Kwak et al . , 2013 ) . Taken together , it appears that CDK5 is a main regulator of the circadian clock mechanism . In order to confirm that the different phenotypes were associated with the accumulation levels of CDK5 in control and Cdk5-silenced mice , we performed immunofluorescence assays on coronal sections of the SCN . Sections were stained with DAPI ( blue ) in order to label nuclei , with GFP antibody ( green ) in order to show cells infected by the virus , and with CDK5 antibody ( red ) in order to compare protein accumulation between the two strains . Scramble as well as shCdk5 mice expressed GFP in the SCN , indicating that the two different viruses infected cells in this brain region ( Figure 3A , Figure 3—figure supplements 1–2 ) . The expression of Cdk5 was efficiently suppressed in the SCN by the shCdk5 but not by the scrambled shRNA ( Figure 3A , Figure 3—figure supplements 1–2 ) , indicating that the behavioral phenotypes observed are due to efficient knock-down of Cdk5 . The Cdk5 shRNAs was expressed in the SCN ( the injection site ) and to some extent also dorsal to the SCN but not in distant brain regions ( i . e . the piriform cortex ) as confirmed by lack of the GFP signal outside of the targeted region ( Figure 3A ) . Surprisingly , the phenotypes of shCdk5 and Per2Brdm1 mice showed considerable similarity , implicating that the levels of PER2 accumulation might be similar in these two different mouse strains . In order to test whether Cdk5 knock-down affected PER2 , we stained with DAPI ( blue ) and immunostained with anti-PER2 ( red ) SCN sections obtained from control , shCdk5 and Per2Brdm1 mice perfused at ZT12 . PER2 was observed in the SCN of scramble controls , but was strongly reduced in shCdk5 and almost absent in Per2Brdm1 animals ( Figure 3B , Figure 3—figure supplements 3–4 ) . These data suggested that CDK5 is a main regulator of the core circadian clock in the SCN and may alter PER2 accumulation and potentially other proteins involved in clock regulation . A study in Drosophila has shown that several kinases , including cyclin-dependent kinases , phosphorylate specific sites on per to maintain the circadian period ( Garbe et al . , 2013 ) . Therefore , we aimed to understand whether a molecular interaction exists between CDK5 and PER2 , as suggested by our SDL screen ( Figure 1 ) . We transfected cells with Per2 and Cdk5 expression vectors and tested whether the two proteins co-immunoprecipitated . We observed that immunoprecipitation with an anti-CDK5 antibody pulled down PER2 protein in two different cell lines ( Figure 4A , Figure 4—figure supplement 1 ) . Similar interactions were observed when cells were transfected with expression constructs resulting in PER2 and CDK5 proteins fused to short amino-acid tags of viral protein 5 ( V5 ) and haemaglutinine ( HA ) fused to them , respectively ( Figure 4B ) . Interestingly , interaction between PER2-V5 and CDK5-HA was reduced when roscovitine , which inhibits interaction of CDK5 with its targets ( Hsu et al . , 2013 ) , was added to the cells ( Figure 4B ) . This suggested that active CDK5 protein interacted better with PER2 than CDK5 in its inhibited form . In order to test whether this interaction could be observed in tissue , we prepared total brain extracts at ZT12 , when kinase activity of CDK5 was high ( Figure 1C ) . At two different salt concentrations , we could pull-down PER2 and CDK5 using either anti-CDK5 or anti-PER2 antibodies ( Figure 4—figure supplement 2 ) . The specificity of the signals was confirmed by using brain extracts from Per2-/- mice ( Chavan et al . , 2016 ) that completely lack PER2 protein ( Figure 4C ) . Next , we wanted to investigate whether the interaction between the two proteins is time of day-dependent in the SCN . Total extracts of SCN tissue at ZT0 , 4 , 8 , 12 , 16 and 20 were prepared and immunoprecipitation with an anti-CDK5 antibody pulled down PER2 at ZT8 , 12 , and 16 , with the strongest signals at ZT12 and ZT16 ( Figure 4D ) . Taken together , these observations suggested that the interaction between CDK5 and PER2 can occur in brain tissue and that in the SCN this interaction was time of day-dependent . This observation was confirmed on SCN tissue sections , where we observed PER2 expression at ZT12 but less at ZT0 with co-localization of CDK5 restricted to ZT12 ( Figure 4—figure supplement 3 ) . Next , we tested in which subcellular compartment the interaction between CDK5 and PER2 takes place . We prepared nuclear and cytoplasmic extracts from total brain tissue and performed immunoprecipitation using an anti-CDK5 antibody . PER2 could only be observed in the cytoplasmic but not the nuclear fraction ( Figure 4E ) . This was supported by the observation that the two proteins were co-localized only in the cytoplasm in SCN tissue ( Figure 4F , yellow color ) . Furthermore , we evaluated with which part of PER2 the CDK5 protein interacts . We tested whether deletions in the PAS-domain of PER2 , a known domain for protein interactions ( Ponting and Aravind , 1997 ) , influenced CDK5 binding . No significant effect of deletions of the PAS-A and PAS-B domains on the interaction was observed ( Figure 4—figure supplement 4 ) . Next , we generated expression vectors coding either for the N-terminal ( 1-576 ) or the C-terminal part ( 577–1257 ) of PER2 fused to GST ( Figure 4—figure supplement 5 ) . The recombinant forms of PER2 and histidine-tagged CDK5 were produced in bacteria . A pull-down assay with these proteins showed that the C-terminal but not the N-terminal half of the PER2 protein was pulled-down by CDK5 , suggesting that CDK5 binds to the C-terminal part of PER2 ( Figure 4G ) . This does , however , not exclude weak interactions of the CDK5 protein with the N-terminal half in vivo . Taken together , our data suggest a physical interaction of PER2 and CDK5 in the cytoplasm . In order to understand whether CDK5 phosphorylates the PER2 protein we overexpressed the N-terminal and C-terminal parts of PER2 fused to GST in bacteria ( Figure 5—figure supplement 1 ) and performed an in vitro kinase assay with the recombinant proteins . Recombinant CDK5/p35 protein complex along with γ-32P labeled ATP resulted in phosphorylation of the N-terminal part of the PER2 protein with a main signal at around 120 kD ( Figure 5A , Figure 5—figure supplement 2 , 32P panels ) . Addition of roscovitine abolished phosphorylation of PER2 whereas LiCl had no effect ( Figure 5—figure supplement 3 ) . Interestingly , no phosphorylation of the C-terminal part of PER2 was observed , only a signal corresponding to the auto-phosphorylation of CDK5 was detected at around 60 kD ( Figure 5A , 32P panel ) . Next , we aimed to identify the phosphorylation site ( s ) in the N-terminal part of PER2 using the recombinant protein , which was phosphorylated by CDK5/p35 in vitro . Mass spectrometry revealed several phosphorylation sites at serine and threonine residues , respectively ( Supplementary file 1 ) . One of the serine residues of PER2 was located within a CDK5 consensus sequence and had the highest probability score for being phosphorylated ( Figure 5B ) . The serine residue at position 394 ( S394 ) of PER2 is located at the end of the PAS domain and within the deletion of the mutated PER2 of Per2Brdm1 mice ( Zheng et al . , 1999 ) . This suggested that CDK5/p35 phosphorylates S394 and that this phosphorylation is of functional relevance . Mutations of this serine to aspartic acid ( S394D ) or glycine ( S394G ) reduced phosphorylation by CDK5/p35 significantly ( Figure 5C ) , confirming that CDK5/p35 phosphorylated S394 . Next , we produced a monoclonal antibody against the phosphorylated serine at 394 of PER2 ( P-S394-PER2 ) ( Figure 5—figure supplements 4–6 ) . With this antibody we detected the phosphorylated N-terminal fragment of PER2 in presence of CDK5/p35 but not when S394 was mutated to glycine ( S394G ) or when CDK5 was inhibited by roscovitine ( Figure 5D ) , confirming S394 phosphorylation by CDK5/p35 . In order to determine whether PER2 phosphorylation at S394 is time of day-dependent , we collected SCN tissue every 4 hr . The P-S394-PER2 specific antibody detected highest phosphorylation at ZT12 with weaker or no phosphorylation at other time points indicating that S394 is phosphorylated in a time of day-dependent manner ( Figure 5E ) . Fractionation of wild-type brain cellular extracts prepared at ZT12 into nuclear and cytoplasmic parts showed phosphorylated S394 predominantly in the cytoplasm with little or no signal in the nucleus when labeled with the P-S394-PER2 antibody ( Figure 5F ) . Total PER2 was observed in both cellular compartments with higher levels in the nucleus ( Figure 5F ) . This suggested that phosphorylation of S394 of PER2 happens predominantly in the cytoplasm and that this phosphorylation is either removed or occluded when PER2 enters the nucleus . To evaluate the function of CDK5-driven PER2 phosphorylation , we wanted to determine whether CDK5 affects PER2 stability . We treated NIH 3T3 cells with roscovitine and DMSO as control and determined endogenous levels of PER2 . We observed that roscovitine treatment of cells reduced PER2 levels , suggesting that CDK5 can affect protein stability ( Figure 6A ) . In order to challenge this observation , we deleted Cdk5 in NIH 3T3 cells using the CRISPR/Cas9 method ( Figure 6—figure supplements 1–3 ) . We observed that deletion of Cdk5 led to reduced amounts of PER2 ( Figure 6B ) , consistent with the data shown in Figure 6A . These observations support the notion that phosphorylation by CDK5 affects PER2 abundance . In order to monitor PER2 stability , we knocked down Cdk5 using the shRNA D ( Figure 2—figure supplement 1 ) . We observed that increasing amounts of shCdk5 dampened PER2 levels proportionally to the decreasing CDK5 levels ( Figure 6C ) . In order to determine whether CDK5 modulates degradation of PER2 , we blocked protein synthesis using cycloheximide . Under conditions that partially knocked down Cdk5 ( at a concentration of 2 . 7 µM of shCdk5 , Figure 6C ) , we measured PER2 and CDK5 protein levels over 6 hr after cycloheximide treatment . We found that degradation of PER2 was faster when Cdk5 was knocked down compared with unspecific shRNA treatment ( shCdk5 t1/2=4 h , scr t1/2=11 h ) ( Figure 6D ) , indicating that reduction of Cdk5 accelerated PER2 degradation . Next , we investigated whether PER2 degradation involved the proteasome . Cells were treated with epoxomycin , a proteasome inhibitor , or with the solvent DMSO . In line with our previous experiments , shCdk5 treatment efficiently knocked down CDK5 and reduced PER2 levels compared with scrambled shRNA treatment . Addition of epoxomycin , but not DMSO , significantly increased PER2 levels despite absence of CDK5 ( Figure 6E ) , indicating that PER2 degradation involved the proteasome . Residual amounts of CDK5 in the cells still may phosphorylate PER2 and direct it into the nucleus . Therefore , we wanted to see whether PER2 could be detected in nuclear extracts of shCdk5 knocked down cells . In line with our previous observations we did not detect PER2 in nuclear extract ( Figure 6F ) , supporting the idea that PER2 needed to be phosphorylated by CDK5 in order to enter the nucleus . Data from immunofluorescence experiments on SCN sections were in line with this hypothesis . PER2 was only detected in nuclei when CDK5 was available ( Figure 6G , arrowheads , Figure 6—figure supplement 4 ) , but not when shCdk5 was expressed in SCN cells ( Figure 6G , white arrow , Figure 6—figure supplement 4 ) . It has been described that nuclear entry of PER2 involves CRY1 ( Kume et al . , 1999; Ollinger et al . , 2014 ) . In addition , CRY1-mediated hetero-dimerization stabilizes PER2 by inhibiting its own ubiquitination ( Yagita et al . , 2000 ) . Therefore , we tested the interaction potential of wild-type PER2 and the S394G PER2 mutation with CRY1 by overexpressing the two PER variants in NIH 3T3 cells . Immunoprecipitation of wild-type PER2 pulled down CRY1; however , the S394G PER2 mutation was significantly less efficient in doing so ( Figure 6H ) . The small amounts of CRY1 detected may be bound to endogenous PER2 that is present in the cells . In summary , these experiments suggested that CDK5 affects PER2 stability , interaction with CRY1 , and nuclear localization .
Not only do kinases play a crucial role in signal transduction in response to extracellular stimuli , but they also regulate cycling processes such as the cell cycle and circadian rhythms . Most cyclin dependent kinases ( CDKs ) regulate the cell cycle , with few exceptions such as the cyclin dependent kinase 5 ( CDK5 ) . This kinase is ubiquitously expressed and its function is vital in post-mitotic neurons , where other CDKs are not active . Although CDK5 is not implicated in cell cycle progression , it can aberrantly activate components of the cell cycle when it is dysregulated in post-mitotic neurons , leading to cell death ( Chang et al . , 2012 ) . Interestingly , cell death is affected by the clock component PER2 as well ( Magnone et al . , 2014 ) , suggesting that both , CDK5 and PER2 act in the same pathway , or that their pathways cross at a critical point during the regulation of cell death . The synthetic dosage lethal screen that we performed in yeast supports this notion , as expression of PER2 in yeast lacking Cdk5 strongly and significantly compromised growth ( Figure 1A ) . The kinase CDK5 displays many effects that ensure proper brain function and development . Mice deficient for Cdk5 are perinatal lethal ( Gilmore et al . , 1998; Ohshima et al . , 1996 ) . CDK5 influences cortical neuron migration , cerebellar development , synapse formation and plasticity ( Kawauchi , 2014 ) . Here , we identified a new role for this kinase , that is the regulation of the circadian clock in vivo . Previously , CDK5 had been identified to phosphorylate CLOCK and thereby regulate CLOCK stability and cellular distribution in cells ( Kwak et al . , 2013 ) . In the SCN , however , NPAS2 may replace the function of CLOCK ( Debruyne et al . , 2006; DeBruyne et al . , 2007 ) and therefore phosphorylation of CLOCK by CDK5 may play a minor role in the SCN . Hence , to unravel the novel function of CDK5 in the circadian oscillator , we had to restrict ourselves to the use of SCN tissue and whole animals . CDK5 activity , but not its protein accumulation , displays a diurnal profile in the SCN with high activity during the night and low activity during the day ( Figure 1C ) . The activity displayed a typical on/off profile similar to other CDKs . This finding raises the question how this diurnal activity of CDK5 may be achieved . On one hand , ATP accumulation , which is required for phosphorylation , peaks during the night in the SCN ( Yamazaki et al . , 1994 ) . On the other hand , CDK5 activity is regulated by cofactors . Depending on its cofactor , CDK5 in the brain phosphorylates targets involved in neurodegenerative diseases ( e . g . Tau , MAP1B ) , neuronal migration ( e . g . DCX ) , and synaptic signaling ( e . g . Cav2 . 2 , Dynamin1 , NR2A , DARPP-32 ) ( Kawauchi , 2014 ) . The most obvious candidates to regulate its time-dependent activity would be cyclins D1 and E , which inhibit CDK5 , or cyclin I , which activates it . Alternatively , other known CDK5 regulators such as p35 may be involved ( Shah and Lahiri , 2014 ) . Most likely , positive and negative feedback loops of other kinases and phosphatases are necessary to generate the on/off profile , although the components involved in this mechanism are probably different from the ones known for CDKs that regulate the cell cycle . Interestingly , CK1 phosphorylates and activates CDK5 in vitro ( Sharma et al . , 1999 ) and CDK5 is thought to phosphorylate and inhibit CK1δ in vitro ( Ianes et al . , 2016; Eng et al . , 2017 ) potentially establishing a feedback loop between the two kinases . However , additional research is needed to determine the precise mechanism of diurnal on/off activation of CDK5 . As evidenced in Figure 2 , Cdk5 knock-down affects circadian clock period at the behavioral level . The shortening of period in mice with knocked-down Cdk5 is comparable to mice containing a mutation of the Per2 gene ( Per2Brdm1 mutant mice , Zheng et al . , 1999 ) . Interestingly , however , knock-down of Cdk5 in Per2Brdm1 mutant mice leads to further shortening of circadian period . This suggests that Cdk5 may affect period either independently of Per2 or , while PER2 may still be important , CDK5 regulates other proteins important for clock function . Since the difference between the period in control versus Cdk5 knock-down ( 0 . 8 hr , black and gray bars , Figure 2F ) is not the same as in Per2Brdm1 mutant versus its Cdk5 knock-down ( 0 . 6 hr , red and rose bars , Figure 2F ) the second possibility is more likely . Moreover , wt KD versus Per2Brdm1 KD show a difference in period ( Figure 2F , gray and rose bars ) , suggesting that the difference in genotype plays an important role . From a dynamic perspective , it is possible that lack of PER2 protein will unmask Cdk5 targets that otherwise would be phosphorylated less efficiently or not at all . For example , the PER2 site that is phosphorylated by CDK5 ( PFDYSPIR ) is very similar in PER1 ( PFDHSPIR ) . If PER1 would be phosphorylated by CDK5 at this site at the same rate as PER2 , then PER1 as well as PER2 should be absent in the nucleus of SCN cells . This would correspond to a PER1/PER2 double knock-out , which become immediately arrhythmic when subjected to constant darkness ( Zheng et al . , 2001 ) . This is not the phenotype we observe in the Cdk5 knock-down mice and hence it is unlikely that CDK5 affects PER1 in the same manner as it affects PER2 . However , in the absence of PER2 the dynamics may change and PER1 may become a better target for CDK5 and influence period . This view is consistent with the observation that knock-down of Cdk5 in Per2Brdm1 mutant mice can shorten period ( Figure 2D , E ) . CDK5 binds to the C-terminal half of PER2 ( Figure 4G ) and phosphorylates it at S394 ( Figure 5 ) , which is located in the PAC domain of the N-terminal half of the protein . Hence , the binding and phosphorylation sites are far apart , suggesting a structure of PER2 allowing proximity of the CDK5 binding and phosphorylation domains . We cannot exclude weak binding of CDK5 to the N-terminal half of PER2 , because phosphorylation at S394 occurs in vitro even in the absence of the C-terminal half of the PER2 protein ( Figure 5A ) . This may be due to the fact that the N-terminal half is overexpressed in vitro , which strongly increases the probability of phosphorylation by CDK5 even in the absence of the C-terminal binding domain . It is also known that p35 ( which is used in the in vitro kinase assay to activate CDK5 ) can increase the interaction between CDK5 and its targets ( Hsu et al . , 2013 ) . In SCN tissue PER2 phosphorylation at S394 appears to be time of day-dependent , with highest levels at ZT12 and ZT16 ( Figure 5E ) when CDK5 activity is high ( Figure 1C ) . Compared with total PER2 protein the S394 phosphorylated form appears to be slightly advanced in its phase . The difference in phase is probably even larger than it appears here , because the polyclonal antibody that detects total PER2 also detects the phosphorylated S394 PER2 variant . This is especially important in the rise of the signal detected , which appears to be identical in Figure 5E . Probably the steep increase between ZT8 and ZT12 represents the S394 phosphorylated forms in both curves . In contrast , the decrease in PER2 levels differs between total PER2 and P-S394-PER2 form . Consistent with previous studies total PER2 peaks in the nucleus at ZT16 in the SCN ( Nam et al . , 2014 ) when P-S394-PER2 is not detected anymore . This highlights that additional post-translational modifications of PER2 exist ( Toh et al . , 2001; Vanselow et al . , 2006 ) and that P-S394-PER2 disappears faster compared with other modified forms . Probably , P-S394-PER2 plays a role in PER2 dynamics in terms of shuttling from the cytoplasm to the nucleus , because P-S394-PER2 can only be observed in the cytoplasmic and not the nuclear fraction ( Figure 5F ) . The phosphorylation of PER2 by CDK5 may therefore be critical for the assembly of a macromolecular complex in the cytoplasm ( Aryal et al . , 2017 ) , which then enters the nucleus . The difference in the decline between PER2 and its S394 phosphorylated form in the SCN may suggest a role of the S394 phosphorylation not only for nuclear transport but also for PER2 protein stability . The earlier decline of the P-S394-PER2 signal compared with total PER2 ( Figure 5F ) might suggest that the S394 phosphorylated form is less stable . Apparently , the opposite is the case , as shown in Figure 6 . Pharmacological inhibition of CDK5 ( Figure 6A ) , CRISPR/Cas9-mediated knock-out of Cdk5 ( Figure 6B ) , and shRNA-mediated knock-down of Cdk5 ( Figure 6C ) all led to reduced levels of PER2 in cells . The half-life of PER2 is clearly increased in the presence of CDK5 , rising from about 4 hr to 11 hr , indicating that phosphorylation at S394 has a stabilizing function . This is in accordance with previous results that described almost absent levels of PER2 in the Per2Brdm1 mutant mice ( Zheng et al . , 1999 ) . This mouse strain expresses a PER2 lacking 87 amino acids in the PAS and PAC domains , where the S394 and the CDK5 consensus sequence are localized . CDK5 cannot phosphorylate this mutant PER2 and therefore the protein is less stable . As a consequence , the formation of the macromolecular complex responsible for nuclear transport of PER2 is disturbed . This results in a temporal change of BMAL1/CLOCK/NPAS2 activity , shortening the clock period . Accordingly , Per2Brdm1 mutant mice display a short period or no circadian period in constant darkness ( Zheng et al . , 1999 ) , similar to the phenotype observed for the CDK5 knock-down mice ( Figure 2B ) . PER2 stability is affected by CK1δ/ε , which phosphorylate PER2 at several sites and regulate degradation of PER2 via the proteasome ( Eide et al . , 2005; Xu et al . , 2007; Narasimamurthy et al . , 2018 ) . This effect is similar to the action of dbt on Drosophila per . Interestingly , CDK5 can phosphorylate CK1δ to reduce its activity ( Ianes et al . , 2016 ) . This phosphorylation could cross-regulate the activities of both kinds of kinases to fine-tune the amount of PER2 . This is evidenced by the observation , that knock-down of Cdk5 in Per2Brdm1 mutant mice further shortens period in these animals ( Figure 2D , E ) . The mammalian orthologue of shg , Gsk3β , does not phosphorylate the mammalian Tim but the nuclear receptor NR1D1 ( Mukherji et al . , 2015 ) . This change in substrate may be related to the shift in function of the CRYs to replace Tim in the mammalian circadian oscillator . Similar to shg , CDK5 phosphorylation of PER2 increases its half-life ( Figure 6D ) . Lack of CDK5 , and therefore lack of phosphorylation at S394 of PER2 , leads to proteasomal degradation of PER2 as evidenced by epoxomycin treatment , which inhibits the proteasome and reduces the decline of PER2 levels in the cell ( Figure 6E ) . This is consistent with a recent report that describes the ubiquitin ligase MDM2 as controlling PER2 degradation via the proteasome ( Liu et al . , 2018 ) . However , it is not clear whether it is the phosphorylation at S394 per se or the capacity to participate in a macromolecular complex to enter the nucleus that stabilizes PER2 . In any case , this phosphorylation appears to be essential for nuclear entry of PER2 ( Figure 6F , G ) . A recent report showed that mammalian PER represses and de-represses transcription by displacing BMAL1-CLOCK from promoters in a CRY-dependent manner ( Chiou et al . , 2016 ) . Our data support these findings . PER2 containing a S394G mutation , which abolishes CDK5-mediated phosphorylation , displayed reduced interaction potential with CRY1 ( Figure 6H ) . Because CRY1 is involved in nuclear transport of PER2 ( Kume et al . , 1999; Ollinger et al . , 2014; Yagita et al . , 2000 ) , lack of interaction with the S394G mutant form of PER2 leaves this protein in the cytoplasm , unable to enter the nucleus ( Figure 6G ) . The present data are also in agreement with previous experiments in which we investigated the role of protein phosphatase 1 ( PP1 ) and its effects on the circadian clock ( Schmutz et al . , 2011 ) . Expression of a specific PP1 inhibitor in the brain lengthened circadian period and increased PER2 levels and its nuclear accumulation in neurons . These effects are all opposite to what we observe when PER2 is not phosphorylated at S394 . Therefore , it could be speculated that PP1 is involved in the dephosphorylation of P-S394 , thereby counterbalancing phosphorylation of this site by CDK5 . Taken together , our results indicate that CDK5 potentially affects several proteins that regulate circadian clock period . In particular , we find that CDK5 phosphorylates PER2 at S394 . This phosphorylation appears to be important for PER2 to bind efficiently to CRY1 in order to allow entry of PER2 into the nucleus . Inhibition of CDK5 in cells leads to degradation of PER2 in the proteasome ( Figure 7 ) . Inhibition of CDK5 in vivo inhibits nuclear entry of PER2 and shortens period to a similar extent as observed in Per2Brdm1 mutant mice , which express a barely detectable level of protein lacking 87 amino acids including S394 . Taken together , CDK5 regulates the circadian clock and influences PER2 nuclear transport via phosphorylation . Because PER2 is involved in several physiologically relevant pathways in addition to clock regulation ( Albrecht et al . , 2007 ) , PER2 may mediate several biological functions that were previously linked to CDK5 , such as the regulation of the brain reward system ( Benavides et al . , 2007; Bibb et al . , 2001 ) and psychiatric diseases ( Engmann et al . , 2011; Zhu et al . , 2012 ) .
All mice were housed with food and water ad libidum in transparent plastic cages ( 267 mm long ×207 mm wide ×140 mm high; Techniplast Makrolon type 2 1264C001 ) with a stainless-steel wire lid ( Techniplast 1264C116 ) , kept in light- and soundproof ventilated chambers . All mice were entrained to a 12:12 hr LD cycle , and the time of day was expressed as Zeitgeber time ( ZT; ZT0 lights on , ZT12 lights off ) . Two- to four-month-old males were used for the experiments . Housing as well as experimental procedures were performed in accordance with the guidelines of the Schweizer Tierschutzgesetz and the declaration of Helsinki . The state veterinarian of the Canton of Fribourg approved the protocol . The floxed Per2 mice ( Chavan et al . , 2016 ) are available at the European Mouse Mutant Archive ( EMMA ) strain ID EM:10599 , B6;129P2-Per2tm1Ual/Biat . The SDL screen was essentially performed as described earlier ( Measday et al . , 2005; Tong , 2001 ) . Briefly , the bait strain Y2454 ( MATα mfa1Δ::MFA1pr-HIS3 , can1Δ , his3Δ1 , leu2Δ0 , ura3∆0 , lys2Δ0 ) carrying the plasmid YCplF2-mPer2 ( that drives expression of PER2 from the galactose-inducible GAL1 promoter ) was inoculated into 50 mL glucose-containing synthetic dropout medium lacking leucine ( SD-Leu ) and grown at 30°C overnight with shaking . Cells were then centrifuged , resuspended in 20 mL of the supernatant , poured into a sterile rectangular petri dish , spotted in a 96-well format on rectangular SD-Leu plates ( coined ‘bait plates’ hereafter ) using a Biomek 2000 robot ( Beckman Coulter , USA ) , and then grown at 30°C for 3 days . In parallel , the gene deletion array in the strain BY4741 ( MATa his3∆1 , leu2∆0 , met15∆0 , ura3∆0 ) was spotted from the storage plates onto fresh G418-containing YPD plates ( 96-well format ) and also grown at 30°C for 3 days . For the mating procedure ( overnight at 30°C ) , colonies from bait plates were ( robotically ) spotted onto plates containing YPD ( plus adenine ) and the colonies from the gene deletion array plates were ( each separately and in duplicate ) spotted on top of them . The next day , the colonies were transferred to G418-containing SD plates lacking lysine , methionine , and leucine ( SD-Lys/Met/Leu/+G418 ) to select for diploids that harbour the YCplF2-mPer2 plasmid . Diploids were then spotted onto plates containing sporulation medium ( 10 g L−1 potassium acetate , 1 g L−1 yeast extract , 0 . 1 g L−1 glucose , 2% w/v agar , supplemented with uracil , histidine , and G418 ) and incubated at 24°C . After 9 days , tetrads were observed and the colonies were transferred to canavanine-containing SD plates lacking arginine and histidine ( SD-Arg/His/+canavanine ) to select for MATa haploids . Following growth at 30°C for three days , a second haploid selection was carried out by spotting the colonies on SD-Arg/His/Leu/+canavanine plates ( to select for MATa haploids containing the YCplF2-mPer2 plasmid ) . Following growth at 30°C for 2 days , a third haploid selection was carried out by spotting cells on SD-Arg/His/Leu/+canavanine/+G418 plates ( to select for MATa haploids containing the YCplF2-mPer2 plasmid as well as the respective gene deletions of the yeast knockout collection ) . Following incubation at 30°C for 5 days , colonies were then spotted in parallel onto SD-Arg/His/Leu/+G418 plates and on SD-Raf/Gal-Arg/His/Leu/+G418 plates ( containing 1% raffinose and 2% galactose as carbon sources ) to induce expression of PER2 . Both types of plates were incubated at 30°C for 4 days and photographed every day . Strains that grew significantly less well on SD-Raf/Gal-Arg/His/Leu/+G418 than on SD-Arg/His/Leu/+G418 included eap1∆ , gnd1∆ , and pho85∆ . In control experiments , the respective original yeast knockout collection mutants were transformed in parallel with the YCplF2-mPer2 or the empty YCplF2 plasmid ( Foreman and Davis , 1994 ) , selected on SD-Leu plates , grown overnight in liquid SD-Leu , spotted ( 10-fold serial dilutions ) on SD-Raf/Gal-Leu plates , and grown for 3 days at 30° ( Figure 1A ) . Please note that all media containing G418 were made with glutamate ( 1 g L−1 ) instead of ammonium sulfate as nitrogen source , as recommended in Tong ( 2001 ) . Adeno Associate Viruses ( AAVs ) were produced in the Viral Vector Facility ( ETH Zurich ) . Plasmids used for the production are available on the VVF web site . Two constructs were produced . ssAAV-9/2-hSyn1-chI[mouse ( shCdk5 ) ]-EGFP-WPRE-SV40p ( A ) carried the shRNA against Cdk5 ( shD , see Figure 2—figure supplement 1 and Supplementary file 2 ) which knocked down only neuronal Cdk5 . ssAAV-9/2-hSyn1-chI[1x ( shNS ) ]-EGFP-WPRE-SV40p ( A ) was the scrambled control . Stereotaxic injections were performed on 8-week-old mice under isofluorene anaesthesia using a stereotaxic apparatus ( Stoelting ) . The brain was exposed by craniotomy and the Bregma was used as reference point for all coordinates . AAVs were injected bilaterally into the SCN ( Bregma: anterior-posterior ( AP ) − 0 . 40 mm; medial-lateral ( ML ) ±0 . 00 mm; dorsal-ventral ( DV ) – 5 . 5 mm , angle + /- 3° ) using a hydraulic manipulator ( Narishige: MO-10 one-axis oil hydraulic micromanipulator , http://products . narishige-group . com/group1/MO-10/electro/english . html ) at a rate of 40 nL/min through a pulled glass pipette ( Drummond , 10 µl glass micropipet; Cat number: 5-000-1001-X10 ) . The pipette was first raised 0 . 1 mm to allow spread of the AAVs , and later withdrawn 5 min after the end of the injection . After surgery , mice were allowed to recover for 2 weeks and entrained to LD 12:12 prior to behavior and molecular investigations . Analysis of locomotor activity parameters was done by monitoring wheel-running activity , as described in Jud et al . ( 2005 ) , and calculated using the ClockLab software ( Actimetrics ) . Briefly , for the analysis of free-running rhythms , animals were entrained to LD 12:12 and subsequently released into constant darkness ( DD ) . Internal period length ( τ ) was determined from a regression line drawn through the activity onsets of ten days of stable rhythmicity under constant conditions . Total and daytime activity , as well as activity distribution profiles , was calculated using the respective inbuilt functions of the ClockLab software ( Acquisition Version 3 . 208 , Analysis version 6 . 0 . 36 ) . Numbers of animals used in the behavioral studies are indicated in the corresponding figure legends . Animals used for the immunohistochemistry were killed at appropriate ZTs indicated in the corresponding figure legends . Brains were perfused with 0 . 9% NaCl and 4% PFA . Perfused brains were cryoprotected by 30% sucrose solution and sectioned ( 40 µm , coronal ) using a cryostat . Sections chosen for staining were placed in 24-well plates ( two sections per well ) , washed three times with 1x TBS ( 0 . 1 M Tris/0 . 15 M NaCl ) and 2x SSC ( 0 . 3 M NaCl/0 . 03 M tri-Na-citrate pH 7 ) . Antigen retrieval was performed with 50% formamide/2x SSC by heating to 65°C for 50 min . Then , sections were washed twice in 2x SSC and three times in 1x TBS pH 7 . 5 , before blocking them for 1 . 5 hr in 10% fetal bovine serum ( Gibco ) /0 . 1% Triton X-100/1x TBS at RT . After the blocking , the primary antibodies , rabbit anti-PER2-1 1:200 ( Alpha Diagnostic , Lot numb . 869900A1 . 2-L ) , mouse anti-Cdk5 clone 2H6 1:20 ( Origene , Lot numb . A001 ) , and rabbit anti-GFP 1:500 ( abcam ab6556 ) diluted in 1% FBS/0 . 1% Triton X-100/1x TBS , were added to the sections and incubated overnight at 4°C . The next day , sections were washed with 1x TBS and incubated with the appropriate fluorescent secondary antibodies diluted 1:500 in 1% FBS/0 . 1% Triton X-100/1x TBS for 3 hr at RT . ( Alexa Fluor 488-AffiniPure Donkey Anti-Rabbit IgG ( H+L ) no . 711–545–152 , Lot: 132876 , Alexa Fluor647-AffiniPure Donkey Anti-Mouse IgG ( H+L ) no . 715–605–150 , Lot: 131725 , Alexa Fluor647-AffiniPure Donkey Anti-Rabbit IgG ( H+L ) no . 711–602–152 , Lot: 136317 and all from Jackson Immuno Research ) . Tissue sections were stained with DAPI ( 1:5000 in PBS; Roche ) for 15 min . Finally , the tissue sections were washed again twice in 1x TBS and mounted on glass microscope slides . Fluorescent images were taken by using a confocal microscope ( Leica TCS SP5 ) , and images were taken with a magnification of 40x or 63x . Images were processed with the Leica Application Suite Advanced Fluorescence 2 . 7 . 3 . 9723 according to the study by Schnell et al . ( 2014 ) . Immunostained sections were quantified using ImageJ version 1 . 49 . Background was subtracted and the detected signal was divided by the area of measurement . An average value obtained from three independent areas for every section was used . The signal coming from sections obtained from silenced mice was quantified as relative amount to the scramble , which was set to 1 . Statistical analysis was performed on three animals per treatment . NIH3T3 mouse fibroblast cells ( ATCCRCRL-1658 ) were maintained in Dulbecco's modified Eagle's medium ( DMEM ) , containing 10% fetal calf serum ( FCS ) and 100 U/mL penicillin-streptomycin at 37°C in a humidified atmosphere containing 5% CO2 . Cdk5 KO cells were produced applying the CRISPR/Cas9 technique according to the manufacturer’s protocol of the company ( Origene , SKU # KN303042 ) . The following plasmids used were previously described: pSCT-1 , pSCT-1mPer2 , pSCT-1 mPer-V5 , pSCT1 ΔPasA mPer2 -V5 , pSCT1 ΔPasB mPer2 -V5 ( Langmesser et al . , 2008 ) ( Schmutz et al . , 2010 ) . pSCT-1 Cdk5-HA , pet-15b Cdk5-HIS , Gex-4T Per2 1–576 , pGex-4T Per2 577–1256 were produced for this paper . The full-length cDNA ( or partial fragments ) encoding PER2 and the full-length Cdk5 were previously sub-cloned in the TOPO vector according to the manufacturer’s protocol ( Catalog numbers pCR2 . 1-TOPO vector: K4500-01 ) . They were subsequently transferred into the plasmid pSCT-1 using appropriate restriction sites . pGex-4T Per2 1–576 S394G , S394D , pSCT-1 mPer2 S394G were obtained using site-specific mutagenesis according to the manufacturer’s protocol on the requested codon carrying the interested amino acid of interest ( Agilent Catalog # 200518 ) . For accession numbers , vectors , mutations , and primers sources , see Supplementary file 2 . NIH 3T3 cells were transfected in 10 cm dishes at about 70% of their total confluency using linear polyethylenimine ( LINPEI25; Polysciences Europe ) . The amounts of expression vectors were adjusted to yield comparable levels of expressed protein . Thirty hours after transfection , protein extracts were prepared . With regard to immunoprecipitation , each antibody mentioned in the paper was used in the conditions indicated by the respective manufacturer . The next day , samples were captured with 50 µL at 50% ( w/v ) of protein-A agarose beads ( Roche ) at 50% ( w/v ) and the reaction was kept at 4°C for 3 hr on a rotary shaker . Prior to use , beads were washed three times with the appropriate protein buffer and resuspended in the same buffer ( 50% w/v ) . The beads were collected by centrifugation and washed three times with NP-40 buffer ( 100 mM Tris-HCl pH7 . 5 , 150 mM NaCl , 2 mM EDTA , 0 . 1% NP-40 ) . After the final wash , beads were resuspendend in 2% SDS , 10% glycerol , 63 mM Trish-HCL pH 6 . 8 and proteins were eluted for 15 min at RT . Laemmli buffer was finally added , samples were boiled for 5 min at 95° C and finally loaded onto 10% SDS-PAGE gels ( Laemmli , 1970 ) . Medium was aspirated from cell plates , which were washed two times with 1x PBS ( 137 mM NaCl , 7 . 97 mM Na2HPO4 × 12 H2O , 2 . 68 mM KCl , 1 . 47 mM KH2PO4 ) . Then PBS was added again and plates were kept 5 min at 37°C . NHI3T3 or HEK cells were detached and collected in tubes and washed two times with 1x PBS . After the last washing , pellets were frozen in liquid N2 , resuspended in Ripa buffer ( 50 mM Tris-HCl pH7 . 4 , 1% NP-40 , 0 . 5% Na-deoxycholate , 0 . 1% SDS , 150 mM NaCl , 2 mM EDTA , 50 mM NaF ) with freshly added protease or phosphatase inhibitors , and homogenized by using a pellet pestle . After that samples were centrifuged for 15 min at 16 , 100 g at 4°C . The supernatant was collected in new tubes and pellet discarded . Total brain or isolated SCN tissue was frozen in liquid N2 , and resuspended in lysis buffer ( 50 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 0 . 25% SDS , 0 . 25% sodium deoxycholate , 1 mM EDTA ) and homogenized by using a pellet pestle . Subsequently , samples were kept on ice for 30 min and vortexed five times for 30 s each time . The samples were centrifuged for 20 min at 12 , 000 rpm at 4°C . The supernatant was collected in new tubes and the pellet discarded . Tissues or cells were resuspended in 100 mM Tris-HCl pH 8 . 8/10 mM DTT and homogenized with a disposable pellet pestle . After 10 min incubation on ice , the samples were centrifuged at 2500 g for 2 min at 4°C and the supernatant discarded . After adding 90 μL of completed cytoplasmic lysis buffer ( 10 mM EDTA , 1 mM EGTA , 10 mM Hepes pH 6 . 8 , 0 . 2% Triton X-100 , protease inhibitor cocktail ( Roche ) , NaF , PMSF , ß-glycerophosphate ) , the pellet was resuspended by vortexing , followed by centrifugation at 5200 rpm for 2 min at 4°C . The supernatant transferred into a fresh 1 . 5 mL tube was the CYTOPLASMIC EXTRACT . The pellet was washed three times with cytoplasmic lysis buffer and resuspended in 45 μL 1x NDB ( 20% glycerol , 20 mM Hepes pH 7 . 6 , 0 . 2 mM EDTA , 2 mM DTT ) containing 2x proteinase and phosphatase inhibitors followed by adding 1 vol of 2x NUN ( 2 M Urea , 600 mM NaCl , 2% NP-40 , 50 mM Hepes pH 7 . 6 ) . After vortexing the samples were incubated 30 min on ice , centrifuged 30 min at 13 , 000 rpm at 4°C and the supernatant that was transferred into a fresh tube was the NUCLEAR EXTRACT . A protein amount corresponding to between 400 and 800 µg of total extract was diluted with the appropriate protein lysis buffer in a final volume of 250 µL and immunoprecipitated using the indicated antibody ( ratio 1:50 ) and the reaction was kept at 4°C overnight on a rotary shaker . The day after , samples were captured with 50 µL of 50% ( w/v ) protein-A agarose beads ( Roche ) and the reaction was kept at 4°C for 3 hr on a rotary shaker . Prior to use , beads were washed three times with the appropriate protein buffer and resuspended in the same buffer ( 50% w/v ) . The beads were collected by centrifugation and washed three times with NP-40 buffer ( 100 mM Tris-HCl pH7 . 5 , 150 mM NaCl , 2 mM EDTA , 0 . 1% NP-40 ) . After the final wash , beads were resuspendend in 2% SDS 10% , glycerol , 63 mM Trish-HCL pH 6 . 8 and proteins were eluted for 15 min at RT . Laemmli buffer was finally added , samples were boiled 5 min at 95° C and loaded onto 10% SDS-PAGE gels . GST-fused recombinant Per2 proteins were expressed in the E . coli Rosetta strain [plasmids: GST-Per2 N-M ( 1-576 ) , GST-Per2 M-C ( 577-1256 ) ] . Proteins were induced with 1 mM IPTG ( Sigma-Aldrich ) for 3 hr at 30°C . Subsequently , proteins were extracted in an appropriate GST lysis buffer ( 50 mM Tris-Cl pH 7 . 5 , 150 mM NaCl , 5% glycerol ) adjusted to 0 . 1% Triton X-100 and purified to homogeneity with glutathione-agarose beads for 2 hr at 4°C . The beads were then incubated overnight at 4°C and washed with GST lysis buffer adjusted with 1 mM DTT . Subsequently , elution with 10 mM reduced glutathione took place for 15 min at room temperature . Elution was stopped by adding Laemmli buffer and samples were loaded onto the gel after 5 min at 95°C and WB was performed using anti-GST ( Sigma no . 06–332 ) and anti-HA antibodies ( Roche no . 11867423001 ) for immunoblotting . The CRISPR/Cas9 Cdk5 cell line was produced starting from NIH3T3 cells using a kit provided by Origene ( www . origene . com ) . The knock-out cell line was produced according to the manufacturer’s protocol . Briefly , cells at 80% of confluency were co-transfected with a donor vector containing the homologous arms and functional cassette , and the guide vector containing the sequence that targets the Cdk5 gene . In parallel , a scrambled negative guide was also co-transfected with a donor vector . 48 hr after transfection the cells were split 1:10 and grown for 3 days . Cells were split another seven times ( this time is necessary to eliminate the episomal form of donor vector , in order to have only integrated forms ) . Then , single colonies were produced and clones were analyzed by PCR in order to find those clones that did not express Cdk5 RNA . Positive clones were re-amplified . PCR primers for genomic Cdk5: FW: 5’-tgtgagtaccacctcctctgcaa-3’ RW: 5’-ttaaacaggccaggcccc-3’ About 24 hr after seeding cells , different shRNA Cdk5 plasmids ( Origene TL515615 A/B/C/D Cdk5 shRNA ) were transfected to knock down Cdk5 according to the manufacturer’s instructions . The knock-down efficiency was assessed at 48 hr post transduction by western blotting . Scrambled shRNA plasmid ( Origene TR30021 ) was used as a negative control . NIH3T3 cells were treated with 100 µM cycloheximide 48 hr after transfection with the indicated vectors , and cells were harvested 0 , 3 , and 6 hr after treatment . About 48 hr after transfection with either scrambled or shCdk5 , cells where Cdk5 was silenced were treated for 12 hr with either DMSO ( vehicle ) or epoxomicin ( Sigma-Aldrich ) at a final concentration of 0 . 2 µM . Samples were collected , and proteins extracted followed by western blotting . Recombinant GST-fused PER2 protein fragments were expressed and purified from the BL21 Rosetta strain of E . coli according to the manufacturer’s protocol described before , using glutathione-sepharose affinity chromatography ( GE Healthcare ) . Each purified protein ( 1 µg ) was incubated in the presence or absence of recombinant Cdk5/p35 ( the purified recombinant N-terminal His6-tagged human Cdk5 and N-terminal GST-tagged human p25 were purchased from Millipore ) . Reactions were carried out in a reaction buffer ( 30 mM Hepes , pH 7 . 2 , 10 mM MgCl2 , and 1 mM DTT ) containing [γ-32P] ATP ( 10 Ci ) at room temperature for 1 hr and then terminated by adding SDS sample buffer and boiling for 10 min . Samples were subjected to SDS-PAGE , stained by Coomassie Brilliant Blue , and dried , and then phosphorylated proteins were detected by autoradiography . CDK5 was immunoprecipitated from SCN samples at different ZTs ( circa 600 µg of protein extract ) ( Figure 8 ) . After immunoprecipitation at 4°C overnight with 2x Protein A agarose ( Sigma-Aldrich ) , samples were diluted in washing buffer and split in two halves . One half of the IP was used for an in vitro kinase assay . Briefly , 1 µg of histone H1 ( Sigma-Aldrich ) was added to the immunoprecipitated CDK5 and assays were carried out in reaction buffer ( 30 mM Hepes , pH 7 . 2 , 10 mM MgCl2 , and 1 mM DTT ) containing [γ-32P] ATP ( 10 Ci ) at room temperature for 1 hr and then terminated by adding SDS sample buffer and boiling for 5 min . Samples were subjected to 15% SDS-PAGE , stained by Coomassie Brilliant Blue , and dried , and then phosphorylated histone H1 was detected by autoradiography . The other half of the IP was used for Western blotting to determine the total amount of CDK5 immunoprecipitated from the SCN samples . To quantify the kinase activity at each time point , the following formula was used: ( [32P] H1/total H1 for each reaction ) /CDK5 IP protein . Filter-aided in vitro kinase assays and mass spectrometry analyses were performed essentially as described ( Hatakeyama et al . , 2019 ) . Briefly , recombinant Cdk5/p35 ( Millipore ) was incubated with the GST-fused PER2 protein fragment . On 10 kDa MW-cutoff filters ( PALL ) samples were incubated in kinase buffer containing 50 mM Hepes , pH 7 . 4 , 150 mM NaCl , 0 . 625 mM DTT , Phostop tablets ( Roche ) , 6 . 25 mM MgCl2 , and 1 . 8 mM ATP at 30°C for 1 hr . Samples without ATP were used as negative control . Assays were quenched by 8 M urea and 1 mM DTT . Protein digestion for MS analysis was performed overnight ( Wiśniewski et al . , 2009 ) . Phosphopeptides were enriched by metal oxide affinity enrichment using titanium dioxide ( GL Sciences Inc , Tokyo , Japan ) ( Zarei et al . , 2016 ) . LC-MS/MS measurements were performed on a QExactive Plus mass spectrometer coupled to an EasyLC 1000 nanoflow-HPLC . Peptides were separated on fused silica HPLC-column tip ( I . D . 75 µm , New Objective , self-packed with ReproSil-Pur 120 C18-AQ , 1 . 9 µm [Dr . Maisch , Ammerbuch , Germany] to a length of 20 cm ) using a gradient of A ( 0 . 1% formic acid in water ) and B ( 0 . 1% formic acid in 80% acetonitrile in water ) : loading of sample with 0% B with a flow rate of 600 nL min-1; separation ramp from 5–30% B within 85 min with a flow rate of 250 nL min-1 . NanoESI spray voltage was set to 2 . 3 kV and ion-transfer tube temperature to 250°C; no sheath and auxiliary gas was used . Mass spectrometers were operated in the data-dependent mode; after each MS scan ( mass range m/z = 370–1750; resolution: 70 , 000 ) a maximum of 10 MS/MS scans were performed using a normalized collision energy of 25% , a target value of 1000 and a resolution of 17 , 500 . The MS raw files were analyzed using MaxQuant Software version 1 . 4 . 1 . 2 ( Cox and Mann , 2008 ) for peak detection , quantification and peptide identification using a full-length Uniprot Mouse database ( April , 2016 ) and common contaminants such as keratins and enzymes used for digestion as reference . Carbamidomethylcysteine was set as fixed modification and protein amino-terminal acetylation , serine- , threonine- and tyrosine-phosphorylation , and oxidation of methionine were set as variable modifications . The MS/MS tolerance was set to 20 ppm and three missed cleavages were allowed using trypsin/P as enzyme specificity . Peptide , site and protein FDR based on a forwards-reverse database were set to 0 . 01 , minimum peptide length was set to 7 , and minimum number of peptides for identification of proteins was set to one , which must be unique . The ‘match-between-run’ option was used with a time window of 1 min . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD012068 ( project name: Cyclin dependent kinase 5 ( CDK5 ) regulates the circadian clock; project accession: PXD012068 ) . We raised in mouse a specific monoclonal antibody recognizing the phosphorylated form of serine 394 of PER2 in collaboration with GenScript Company . The sequence used for the immunogen preparation was: FDY {pSer} PIRFRTRNGEC . 3 Balb/c mice and 3 C57 mice were immunized with conventional strategies and antisera obtained from those animals were used for the first control experiment performed by in vitro kinase assay ( Figure 5—figure supplement 3 ) . The positive antiserum was used for the cell fusions . Subsequently , a screening with 16 96-well plates ( from 2 × 10E4 clones ) was performed by indirect ELISA , primary screening with phospho-peptide , then counter-screening with non-phospho-peptide . The obtained supernatants were tested by in vitro kinase assay in order to screen which one was better recognized the phospho-form of PER2 S394 ( Figure 5—figure supplement 4 ) . Finally , five selected positive primary clones selected were subcloned by limiting dilution and tested as final antibody ( Figure 5—figure supplement 5 ) . Statistical analysis of all experiments was performed using GraphPad Prism6 software . Depending on the type of data , either an unpaired t-test , or one- or two-way ANOVA with Bonferroni or Tukey’s post-hoc test was performed . Values considered significantly different are highlighted . [p<0 . 05 ( * ) , p<0 . 01 ( ** ) , or p<0 . 001 ( *** ) ] . | Anyone who has crossed multiple time zones on a long flight will be familiar with jet lag , and that feeling of wanting to sleep at lunchtime and eat in the middle of the night . Many bodily processes , including appetite and wakefulness , roughly follow a 24-hour cycle . These cycles are known as circadian rhythms , from the Latin ‘circa diem’ meaning about a day . An area of the brain called the suprachiasmatic nucleus ( SCN ) coordinates circadian rhythms . It acts as a master clock by generating a 24-hour signal for the rest of the body to follow . Jet lag occurs when this internal circadian rhythm becomes out of sync with the local day-night cycle . Although jet lag can be uncomfortable , it tends to disappear over the course of a few days . This is because exposure to daylight in our new location resets the SCN master clock , enabling us to adapt to a new time zone . But evidence suggests that long-term disruption of circadian rhythms , for example as a result of shift work , may have lasting harmful effects . These include an increased risk of degenerative brain disorders such as Parkinson's disease and Alzheimer's disease . Brenna et al . now identify a molecular mechanism that could explain this link . A key component of the SCN master clock is a protein called Period2 ( PER2 ) . Levels of PER2 rise and fall over each 24-hour period , helping the brain keep track of time . Brenna et al . show that PER2 interacts with CDK5 , a protein that helps regulate brain development and that has been implicated in Parkinson's disease and Alzheimer's disease . Reducing CDK5 levels in mice shortened their circadian rhythms by several hours . It also altered the animals’ behavioral patterns over a 24-hour period . Deleting the gene for PER2 had a similar effect , suggesting that CDK5 helps regulate PER2 . Future studies should investigate the molecular links between CDK5 , circadian rhythms and processes such as neurodegeneration . The results would provide clues to whether manipulating the circadian clock could help prevent or treat neurological disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"neuroscience"
] | 2019 | Cyclin-dependent kinase 5 (CDK5) regulates the circadian clock |
The postnatal neurodevelopmental disorder Rett syndrome , caused by mutations in MECP2 , produces a diverse array of symptoms , including loss of language , motor , and social skills and the development of hand stereotypies , anxiety , tremor , ataxia , respiratory dysrhythmias , and seizures . Surprisingly , despite the diversity of these features , we have found that deleting Mecp2 only from GABAergic inhibitory neurons in mice replicates most of this phenotype . Here we show that genetically restoring Mecp2 expression only in GABAergic neurons of male Mecp2 null mice enhanced inhibitory signaling , extended lifespan , and rescued ataxia , apraxia , and social abnormalities but did not rescue tremor or anxiety . Female Mecp2+/- mice showed a less dramatic but still substantial rescue . These findings highlight the critical regulatory role of GABAergic neurons in certain behaviors and suggest that modulating the excitatory/inhibitory balance through GABAergic neurons could prove a viable therapeutic option in Rett syndrome .
Maintaining a proper ratio of excitation and inhibition throughout the brain is critical to normal neurological function . Alterations in this excitatory/inhibitory balance are postulated to underlie a number of neuropsychiatric disorders , such as autism ( Vattikuti and Chow , 2010; Gogolla et al . , 2009; Rubenstein and Merzenich , 2003 ) , schizophrenia ( Belforte et al . , 2010 ) , and Rett syndrome ( Dani et al . , 2005 ) . Much of the work on this balance has focused on excitatory neurons , which make up the majority of the brain’s neuronal population , yet it has become increasingly clear that inhibitory neurons , which predominantly release the inhibitory neurotransmitter gamma-aminobutyric acid ( GABA ) , play important roles in the proper function of entire circuits ( Xue et al . , 2014 ) and in the behaviors regulated by these circuits ( Yizhar et al . , 2011 ) . Dysfunctional GABAergic signaling has been implicated in multiple neurological and neuropsychiatric disorders ( Siniatchkin and Koepp , 2009; Pizzarelli and Cherubini , 2011 ) , including Rett syndrome ( Chao et al . , 2010 ) . In fact , deletion of Mecp2 solely in GABAergic neurons is sufficient to reproduce the majority of the Rett-like features of the constitutive Mecp2-null mouse , including ataxia , stereotyped behaviors , seizures , breathing abnormalities , and premature death ( Chao et al . , 2010 ) . Rett syndrome ( RTT ) is caused by mutations in the X-linked gene encoding methyl CpG-binding protein 2 ( MeCP2 ) ( Amir et al . , 1999 ) , a protein that is highly expressed throughout the brain and involved in chromatin modulation ( Baker et al . , 2013 ) . RTT is notable for the variety of different behaviors it affects: disease-causing MECP2 mutations cause a rapidly fatal neonatal encephalopathy in human males but allow females to develop normally until around one year of age , at which point they regress developmentally , losing acquired language , social , and motor milestones , and instead develop ataxia , seizures , hand stereotypies , learning and memory deficits , and respiratory abnormalities ( Trevathan and Naidu , 1988 ) . Mecp2 knockout mice replicate all these features of the human disease ( Chen et al . , 2001; Guy et al . , 2001 ) , with male Mecp2-null mice developing symptoms and dying within the first five months of life . Mecp2-heterozygous females develop progressive neurological dysfunction with a later onset , because random X chromosome inactivation preserves expression of wildtype Mecp2 in more than half of their cells ( Samaco et al . , 2013; Young and Zoghbi , 2004 ) . While Mecp2-mutant male and female mice develop profound disabilities , RTT is not a neurodegenerative disorder . Mecp2-null neurons are rendered partly dysfunctional but are still present , active , and properly located developmentally ( Chao et al . , 2007; 2010 ) ; this makes the RTT mouse model ideal for interrogating how certain cell types affect behavior by bypassing the confound of widespread cell death and modeling non-neurodegenerative neuropsychiatric disorders more accurately . This system has already been used to demonstrate the critical role for GABAergic signaling in the pathogenesis of RTT ( Chao et al . , 2010 ) , as well as revealing non-overlapping roles for parvalbumin ( PV ) - and somatostatin ( SOM ) -expressing GABAergic neurons in behavior ( Ito-Ishida et al . , 2015 ) , a role in feeding and aggression for Sim1-positive hypothalamic neurons ( Fyffe et al . , 2008 ) , and a critical role for proper glutamatergic function in anxiety , tremor , and acoustic startle response ( see companion paper Meng et al . , 2016 ) . Conversely , rescue of Mecp2 expression in certain cell types can indicate whether targeting that cell type has therapeutic potential and can give insight into the interplay between multiple circuits in the diseased brain . Therefore , we genetically re-expressed Mecp2 solely in GABAergic neurons in a mouse otherwise null for Mecp2 , restoring Mecp2 expression and function to inhibitory neurons , and observed profound improvement in both male and female mice .
Our lab previously generated a mouse line in which Cre expression is driven by the Slc32a1 promoter ( Slc32a1-Cre , referred to as Viaat-Cre henceforth ) , limiting Cre expression to GABAergic and glycinergic neurons ( Chao et al . , 2010 ) . To confirm that Viaat-Cre was not expressed in glial cells or microglia ( Lioy et al . , 2011; Derecki et al . , 2012; Wang et al . , 2015 ) , we crossed Viaat-Cre male mice with female ROSA-LSL-YFP transgenics and co-stained for glial fibrillary acidic protein ( GFAP ) , a glial marker , and ionized calcium binding adapter molecule 1 ( Iba1 ) , a marker of microglia . As expected , we observed no colocalization ( Figure 1—figure supplement 1A , B ) . We also confirmed that there were no YFP+ cells in bone marrow isolated from Viaat-Cre;ROSA-YFP mice ( Figure 1—figure supplement 1C ) . To re-express Mecp2 specifically in Viaat-expressing cells , we crossed male Viaat-Cre mice with females carrying a Mecp2 allele with a floxed STOP cassette between the second and third exons ( Mecp2lox-Stop/Y ) ( Figure 1A ) . Without recombination , only a truncated , non-functioning version of Mecp2 is expressed , and the resulting mice phenocopy the traditional Mecp2 null ( Guy et al . , 2007 ) . In the presence of the Viaat-Cre allele , however , the STOP cassette is excised and Mecp2 is transcribed normally in GABAergic neurons , as indicated by colocalization with the GABAergic neuron marker glutamic acid decarboxylase 67 ( GAD67 , Figure 1A left panel ) ; CamKII-positive excitatory neurons ( Figure 1A right panel ) remain null for MeCP2 . We characterized males of each of the four genotypes resulting from this cross ( wildtype , Viaat-Cre , Mecp2lox-Stop/Y , and Viaat-Cre;Mecp2lox-Stop/Y ) for body weight , survival , and the presence of hind limb clasping , tremor , and breathing abnormalities . 10 . 7554/eLife . 14198 . 003Figure 1 . Conditional restoration of Mecp2 expression solely in GABAergic neurons normalized body weight and appearance and significantly extended lifespan . ( A ) Schematic of breeding and mice generated ( left column ) . Photos of each genotype show male Viaat-Cre; Mecp2lox-Stop/Y mice were indistinguishable from wildtype mice at 4 months of age ( second column ) . Cell images show expression of MeCP2 ( green ) was limited to GAD67-positive GABAergic neurons ( red; third column , arrowheads ) and was not found in CamKII-expressing excitatory neurons ( red , right column , arrowheads ) . ( B ) Male rescue mice maintained similar body weights to wildtype males throughout life . ( C ) Male rescue mouse lifespan was significantly extended , with ~40% surviving past 1 year of age . n = 33–47 per genotype . Scale bar is 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 00310 . 7554/eLife . 14198 . 004Figure 1—figure supplement 1 . Viaat-Cre was not expressed in glia , microglia , or bone marrow . ( A–B ) Viaat-Cre; ROSA-YFP did not colocalize with glial marker GFAP ( A ) or activated microglia marker Iba1 ( B ) in the cortex . Scale bars = 40 μm . ( C ) No YFP+ cells were detected in the bone marrow of Viaat-Cre;ROSA-YFP mice . DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 00410 . 7554/eLife . 14198 . 005Figure 1—figure supplement 2 . Male rescue mice maintained more of their body weight up to the point of death and showed improved EEG activity . ( A ) Male rescue mice lose less body weight from their peak body weight until death than Mecp2lox-STOP/Y littermates . ( B ) Abnormal EEG activity has been previously reported in Mecp2-/Y mice ( Chao et al . , 2010 ) and was observed in two of the three recorded Mecp2lox-STOP/Y male mice ( third trace ) . Five rescue mice ( fifth trace ) had similar EEG traces to wildtype and Viaat-Cre controls ( first and second trace , respectively ) , while one rescue animal displayed abnormal activity ( fourth trace ) . Wildtype = 5 , Viaat-Cre = 2 , Mecp2lox-STOP/Y = 3 , Rescue = 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 005 The Viaat-Cre;Mecp2lox-Stop/Y male mice ( referred to hereafter as the 'rescue mice' ) appeared grossly indistinguishable from their wildtype and Viaat-Cre littermates ( Figure 1A ) . While Mecp2lox-Stop/Y mice became obese by seven weeks of life , the body weight of the rescue mice paralleled that of their wildtype and Viaat-Cre littermates ( Figure 1B ) . Male rescue mice lived significantly longer than their Mecp2lox-Stop/Y counterparts: no Mecp2lox-Stop/Y male mouse survived past 30 weeks of life , but 47% of the rescue mice survived past one year ( 22 out of 47 total rescue mice recorded ) ( Figure 1C ) . Two rescue mice reached two years of age , which to our knowledge have never been reported with any manipulation of this model . Rescue mice also maintained their body weight throughout life , unlike Mecp2lox-Stop/Y males , which tended to lose a significant percentage of their body weight in the weeks preceding death ( Figure 1—figure supplement 2A ) . We did not observe behavioral seizures in any rescue mouse at any age; using electroencephalogram ( EEG ) recordings we found that five rescue mice had normal activity patterns while one exhibited abnormal electrographic discharges ( Figure 1—figure supplement 2B ) . Mecp2-null male mice develop pronounced ataxia early in life ( Chen et al . , 2001; Guy et al . , 2001 ) , as do the Mecp2lox-Stop/Y mice ( Guy et al . , 2007 ) and the GABAergic-specific Mecp2 knockout mice ( Chao et al . , 2010 ) . We therefore assessed motor coordination at six weeks of age using the accelerating rotarod and grip strength assays . The Mecp2lox-Stop/Y mice had a short latency to fall off the rotarod and impaired grip , as expected , but the male rescue mice performed nearly as well as their wildtype and Viaat-Cre littermates in both assays ( Figure 2A , B ) . At nine weeks of age , the male rescue mice were not more active than the nulls in the open field assay ( OFA ) , ( Figure 2C ) . Mecp2lox-Stop/Y mice had noticeable deficits in purposeful forepaw movements: they were unable to build nests at 8 weeks of age or to bury marbles at 9 weeks of age ( Figure 2D–E ) . This apraxia-like behavior was completely eliminated in the rescue mice ( Figure 2D–E ) . The rescue mice also never developed hindlimb clasping , a sign of motor dysfunction exhibited by both the Mecp2lox-Stop/Y mice ( Guy et al . , 2007 ) and the GABAergic-specific Mecp2 knockout ( Chao et al . , 2010 ) . Rescue mice also did not display the hypersociability of the Mecp2lox-Stop/Y male mice in the partition test at 8 weeks of age ( Figure 2F ) . 10 . 7554/eLife . 14198 . 006Figure 2 . Male rescue mice exhibited significant behavioral rescue . ( A–B ) Male rescue mice were indistinguishable from wildtype and Viaat-Cre controls at 6 weeks of age on the rotarod ( n = 23–28 per genotype ) ( A ) and grip strength ( B ) assays ( n = 20–26 per genotype ) . ( C ) Rescue mice trended toward improved locomotor activity in the open field assay at 9 weeks of age ( n = 8–15 per genotype ) . ( D–E ) Apraxia at 9 weeks of age was reversed in rescue mice as indicated by marble burying ( n = 9–15 ) ( D ) or , at 8 weeks , by nest building ( E , n = 23–27 per genotype ) . ( F ) Rescue mice had similar sociability to wildtype mice in the partition test at 8 weeks of age . * represents significance of genotype ( by color ) compared to Mecp2lox-Stop/Y . ( G ) Acoustic startle response was unchanged in rescue mice at 8 weeks of age ( n = 16–21 ) . ( H−I ) Rescue mice were similar in anxiety behavior to Mecp2lox-Stop/Y in the elevated plus maze test ( G ) but were similar to wildtype in the light dark box assay ( I ) . Error bars show SEM . *p<0 . 05 **p<0 . 01 ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 006 Not all features of the Mecp2lox-Stop/Y mice were rescued , however . Rescue mice developed tremor and labored breathing similar in timing and intensity to the Mecp2lox-Stop/Y cohort , suggesting that these symptoms might not be solely dependent on Mecp2's role in inhibitory signaling . This indicates that the development of tremor is particularly dependent on proper excitatory network activity; indeed , loss of Mecp2 from GABAergic neurons does not result in tremor ( Chao et al . , 2010 ) , while glutamatergic conditional knockout mice do develop a tremor which is not observed in glutamatergic conditional rescue mice ( see companion paper Meng et al . , 2016 ) . Breathing , however , seems to be affected by dysfunction in both excitatory and inhibitory networks , as GABAergic conditional Mecp2-knockout mice do develop abnormal breathing patterns ( Chao et al . , 2010 ) . In addition , the rescue mice shared the acoustic startle deficit observed in the Mecp2lox-Stop/Y mice ( Figure 2G ) , with no differences in prepulse inhibition at any decibel level noted ( data not shown ) . Interestingly , rescue mice also showed a partial benefit to anxiety: Mecp2lox-Stop/Y and rescue mice performed similarly in the elevated plus maze test ( Figure 2H ) , but the rescue mice were indistinguishable from wildtype and Viaat-Cre control mice in the light-dark box test Figure 2I . Taken together , these findings indicate that regulation of startle and tremor is more reliant on proper excitatory signaling than on inhibition , while anxiety may be partially mitigated by improved GABAergic function that in turn somewhat normalizes glutamatergic circuit function . As the majority of the rescue male mice outlived their Mecp2lox-Stop/Y littermates , we were able to assess whether these mice maintained the observed behavioral rescue later in life . We tested wildtype , Viaat-Cre , and rescue mice at 30 weeks of age for motor deficits by the open field assay , accelerating rotarod , and grip strength meter . By 30 weeks , all Mecp2lox-Stop/Y mice had died , while over 60% of the rescue mice were still alive ( Figure 1C ) . The surviving rescue mice appeared grossly similar to the wildtype and Viaat-Cre male mice ( Figure 3A ) but displayed an increased breathing rate and noticeable tremor . The older rescue mice tended to move less than their wildtype and Viaat-Cre counterparts ( Figure 3B ) , were slightly less coordinated on the rotarod ( Figure 3C ) , and slightly weaker in their grip strength ( Figure 3D ) . Rotarod performance and grip strength were impaired only when compared to those of the Viaat-Cre control . This apparent deficit , however , was likely due to the variable health of the rescue mice , some of which were more ill than others at 30 weeks . Regardless , as a group the rescue mice were able to perform near wildtype levels , and overall the behavioral rescue produced by Mecp2 reexpression in GABAergic neurons persisted later in life . 10 . 7554/eLife . 14198 . 007Figure 3 . Rescue persisted in male mice at 30 weeks of age . ( A ) Male rescue mice at 30 weeks of age continued to appear indistinguishable from wildtype and Viaat-Cre littermates . ( B ) Locomotion in the open field assay decreased when compared to wildtype but not Viaat-Cre . ( C–D ) Rotarod performance was maintained at wildtype levels , as was grip strength ( D ) . n = 20–25 per genotype . Error bars show SEM . *p<0 . 05DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 007 Loss of MeCP2 leads to decreases in GABA concentrations ( Chao et al . , 2010 ) and widespread gene expression changes ( Chahrour et al . , 2008 ) ; normalization of these deficits may underlie the behavioral rescue we observe . Indeed , we found that GABA concentrations in the striata of 8-week-old Mecp2lox-Stop/Y mice were significantly lower than in wildtype but were restored to wildtype levels in the rescue mice ( Figure 4A ) . We then assessed changes in the expression of genes known to be involved in neuronal GABAergic function using tissue from the cerebella of 8-week-old mice . Interestingly , levels of Gad1 , Gad2 , and Vgat were all decreased in the Mecp2lox-Stop/Y but were identical to wildtype and Viaat-Cre levels in the rescue mice ( Figure 4B ) . We next assessed a slate of genes that are expressed in inhibitory neurons and known to be altered in Mecp2-null brains and observed the expected downregulation of Nxph4 , Gabra3 , Kcng4 , Opn3 , and Scg2 as well as upregulation of A1593442 , Robo2 , Rassf8 , and Cabp7 in the Mecp2lox-Stop/Y cerebellum . Tissues from rescue mice , however , showed reversal of these expression patterns , with the exception of Robo2 , which trended to normalization ( Figure 4C ) . Re-expression of Mecp2 in GABAergic neurons thus normalizes several transcriptional changes related to inhibitory neuron function . 10 . 7554/eLife . 14198 . 008Figure 4 . Expression of Mecp2 in GABAergic neurons normalizes levels of GABA and other MeCP2 responsive genes . ( A ) GABA concentrations were restored to wildtype levels in striatum from rescue mice ( n = 4–5 per genotype ) . ( B ) RNA expression of genes related to GABAergic neuronal function , particularly gad1 and gad2 , were normalized in rescue mouse cerebella . ( C ) Genes known to be downregulated or upregulated in Mecp2-null cerebellum were normalized in the rescue mice . Error bars show SEM . *p< 0 . 05 **p<0 . 01 ***p<0 . 001 ****p<0 . 0001 . n = 6–7 per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 008 To understand the cellular mechanism of the observed behavioral rescue , we recorded miniature inhibitory postsynaptic currents ( mIPSCs ) and miniature excitatory postsynaptic currents ( mEPSCs ) from 5–6-week old wildtype , Viaat-Cre , Mecp2lox-Stop/Y and rescue male mice ( n = 3 mice per genotype ) . We prepared cortical slices and conducted whole cell recordings on pyramidal neurons in layer II/III of the somatosensory cortex , a region that showed alterations in these currents in the GABAergic conditional knockout mouse ( Chao et al . , 2010 ) . We found that the frequency and amplitude of mIPSCs in the rescue mice showed a strong trend to increase when compared to the other genotypes , while the decay rate trended lower ( Figure 5A ) , although none of these changes reached statistical significance when analyzed using a one-way ANOVA with a Tukey posthoc test ( Supplementary file 1 ) . Taken together , these data suggest a trend to increased presynaptic GABA release with Mecp2 reexpression in GABAergic neurons . Furthermore , rescue mice had similar numbers of vesicular GABA transporter ( VGAT ) -positive puncta to wildtype and Viaat-Cre controls in the hippocampal CA1 , while Mecp2lox-Stop/Y mice had significantly fewer VGAT+ synapses , indicating that reexpression of Mecp2 in GABAergic neurons normalizes inhibitory synapse number ( Figure 5—figure supplement 1A ) . In contrast , the frequency and amplitude of mEPSCs in rescue mice were lower than those of wildtype and Viaat-Cre controls but similar to those of Mecp2lox-Stop/Y mice ( Figure 5B ) , indicating that Mecp2 reexpression in GABAergic neurons failed to rescue mEPSCs . Vesicular glutamate transporter ( Vglut1 ) -positive puncta numbers in the CA1 were not significantly different across genotypes , although there was a strong trend to fewer Vglut1+ puncta in the rescue mice ( Figure 5—figure supplement 1A ) . Expression of Mecp2 solely in GABAergic neurons thus appears to mildly enhance inhibitory neuronal signaling but has no effect on excitatory synaptic activity . 10 . 7554/eLife . 14198 . 009Figure 5 . Inhibitory signaling in rescue mice showed some improvement while excitatory signaling was unchanged . ( A–B ) Sample traces of mIPSC ( A , Wildtype = 11 , Viaat-Cre = 17 , Mecp2lox-STOP/Y = 14 , Rescue = 11 ) and mEPSC ( B , Wildtype = 17 , Viaat-Cre = 16 , Mecp2lox-STOP/Y = 18 , Rescue = 17 ) from pyramidal cells of the somatosensory cortex of wildtype , Viaat-Cre , Mecp2lox-STOP/Y , and Rescue male mice , including cumulative distributions of amplitude and interval , grand average minis , and summaries of frequency , amplitude , decay , and average charge . Error bars show SEM . *p<0 . 05DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 00910 . 7554/eLife . 14198 . 010Figure 5—figure supplement 1 . Inhibitory synapse numbers and spontaneous action potential firing are normalized in rescue mice . ( A ) VGAT+ inhibitory synapses are decreased in the CA1 of Mecp2lox-STOP/Y mice but are normalized in the rescue; Vglut1+ excitatory synapses are not significantly changed in any genotype . ( B ) Left: Representative traces of spontaneous action potential firing from layer V cortical pyramidal neurons . Right: Summary of spontaneous firing rates across genotypes ( wildtype n = 15; Viaat-Cre = 12; Mecp2-/Y = 20; Rescue = 17 ) . ( C ) Left: Representative traces of intrinsic neuronal excitability . Right: Summary of intrinsic neuronal excitability across genotypes . Error bars show SEM . *p<0 . 05DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 010 To determine if this enhancement has any effect on neuronal firing activity , we recorded spontaneous action potentials from layer V pyramidal neurons of the somatosensory cortex of 6–8 week old male wildtype , Viaat-Cre , Mecp2-null , and rescue mice . While Mecp2-null mice exhibited significantly fewer spontaneous action potentials than wildtype animals , rescue mice showed normalized activity ( Figure 5—figure supplement 1B ) . Intrinsic neuronal excitability , assessed by intracellular injection of step currents , did not vary among genotypes ( Figure 5—figure supplement 1C ) . It is thus clear that reexpression of Mecp2 in GABAergic neurons normalizes neuronal firing activity in the cortex; taken together with the improved inhibitory input , this normalization could feasibly be due to normalized upstream disinhibition . Most mouse models of RTT rely on male mice to avoid the confounding influence of X chromosome inactivation . Given that RTT primarily affects females , however , we wanted to determine whether the reexpression of Mecp2 in GABAergic neurons could exert similar benefits in female Mecp2lox-Stop/+ mice . We characterized females generated from the same cross that produced the Mecp2lox-Stop/Y male mice ( Figure 1A ) . All genotypes from this cross lived normal lifespans , with no premature death noted in either Mecp2lox-Stop/+ or rescue females . By 15 weeks of age , the female Mecp2lox-Stop/+ mice , which lack functional MeCP2 in roughly half their cells , were noticeably heavier than their wildtype and Viaat-Cre female littermates and developed a readily observable tremor and labored breathing ( Figure 6A , B ) . 10 . 7554/eLife . 14198 . 011Figure 6 . Female rescue mice exhibited a partial but sustained functional recovery . ( A ) Female rescue mice appeared grossly normal at 23 weeks of age . ( B ) Female rescue mice body weight was partially rescued ( n = 27–37 per genotype ) . ( C ) Female rescue mice performed better at 9 weeks of age on the accelerating rotarod than Mecp2lox-STOP/+ females , but not to wildtype levels ( n = 14–16 per genotype ) . ( D ) Acoustic startle response at 9 weeks of age was not rescued ( n = 14–17 per genotype ) . ( E–G ) At 30 weeks of age , female rescue mice showed partial rescue of grip strength ( E ) and complete rescue of hypersociability in the partition test ( F ) and nesting ability ( n = 13–18 per genotype ) ( G ) . ( H–K ) Rescue females exhibited a partial but sustained rescue of locomotion in the open field ( H ) , rotarod ( I ) , footslip ( J ) , and grip strength ( K ) assays ( n = 20–35 per genotype ) . Error bars show SEM . *p<0 . 05 **p<0 . 01 ***p<0 . 001 ****p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 01110 . 7554/eLife . 14198 . 012Figure 6—figure supplement 1 . Female rescue mice show partial rescue particularly at late ages . ( A–C ) At 9 weeks of age , no differences in locomotion were noted ( A , n = 11–17 per genotype ) . ( B ) 18-week-old female mice of all genotypes showed similar levels of motor activity in the open field assay . ( C ) Mecp2lox-Stop/+ females exhibited more footslips , whereas rescue females were more coordinated ( n = 8–18 per genotype ) . ( D ) Even at 30 weeks of age , females of all genotypes exhibited similar levels of motor activity ( though less than younger mice; n = 14–17 per genotype ) . ( E–F ) Mecp2lox-Stop/+ females were unable to remain balanced on a dowel at 30 weeks ( E , n = 14–18 per genotype ) or 13 months ( F , n = 25–35 per genotype ) of age , but rescue mice performed significantly better at both time points . Error bars show SEM . *p<0 . 05 **p<0 . 01 ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14198 . 012 We assessed behavior in the female mice at later ages than the male mice , as the effects of Mecp2 deletion are delayed in the female Mecp2 heterozygous mice ( Samaco et al . , 2013 ) . Rotarod performance was impaired in the female Mecp2lox-Stop/+ mice by 9 weeks of age ( Figure 6C ) , although there was no difference in locomotion in the open field assay ( Figure 6—figure supplement 1A ) . In contrast , the rescue females were only slightly heavier than wildtype females ( Figure 6A , B ) , and they performed as well as wildtype on the rotarod at 9 weeks of age ( Figure 6C ) . We also observed rescue of ataxic phenotypes at 18 weeks of age; while there was no significant difference in locomotion in the open field , the female rescue mice were no different from wildtype females in the footslip assay ( Figure 6—figure supplement 1B , C ) . Like the male rescue mice , however , the startle deficit of the female rescue mice at nine weeks was unaffected by reexpression of Mecp2 in inhibitory neurons ( Figure 6D ) , and we observed no difference in prepulse inhibition ( data not shown ) ; in contrast , this startle deficit was rescued in the glutamatergic neuron rescue female mice ( see companion paper Meng et al . , 2016 ) . The male Mecp2lox-Stop/Y phenotype is severe , killing the mouse within the first five months of life , but female Mecp2lox-Stop/+ mice commonly live past the one-year mark . This mimics the human disease , as some female RTT patients live full lifespans with considerable supportive care . To determine if the behavioral rescue observed at younger ages was sustained later in life , we assessed behavior in females at 30 weeks and 13 months of age . At both ages , Mecp2lox-Stop/+ females displayed a pronounced tremor and hindlimb clasping when suspended . At 30 weeks of age , female Mecp2lox-Stop/+ mice had similar locomotion to the other genotypes ( Figure 6—figure supplement 1D ) , but they had poor grip strength ( Figure 6E ) and impaired dowel walk performance ( Figure 6—figure supplement 1E ) . In addition , they were hypersocial when presented with a novel partner mouse in the partition test ( Figure 6F ) and were slower to build nests ( Figure 6G ) . In contrast , the female rescue mice performed similarly to wildtype mice in each of these assays , although the difference from the Mecp2lox-Stop/+ mice was not statistically significant at early ages because of the latter’s relatively mild impairment ( Figure 6E–G , Figure 6—figure supplement 1D–E ) . At 13 months of age , female rescue mice moved noticeably more than the Mecp2lox-Stop/+ females ( Figure 6H ) , were less ataxic by rotarod ( Figure 6I ) , footslip ( Figure 6J ) , and dowel walk ( Figure 6—figure supplement 1F ) , and had better grip strength ( Figure 6K ) . On each behavioral metric the performance of female rescue mice at this late age was worse than wildtype but still better than that of Mecp2lox-Stop/+ mice; interestingly , reexpression of Mecp2 solely in glutamatergic neurons resulted in a near complete reversal of RTT-like phenotypes ( see companion paper Meng et al . , 2016 ) , possibly because restoring glutamatergic function in a brain with approximately half of the GABAergic neurons retaining Mecp2 expression due to random X inactivation more closely approximated a normal excitatory-inhibitory balance . Notably , the GABAergic rescue mice had an observable tremor , but they never developed the hindlimb-clasping characteristic of the Mecp2lox-Stop/+ mice . The mitigation of the RTT-like phenotype observed in female rescue mice thus persisted well into their second year of life .
Perturbations in the balance between excitatory and inhibitory signaling have been hypothesized to underlie several neuropsychiatric disorders , including autism ( Vattikuti and Chow , 2010; Gogolla et al . , 2009; Rubenstein and Merzenich , 2003 ) , schizophrenia ( Belforte et al . , 2010 ) , and RTT ( Dani et al . , 2005 ) . Artificial elevation of the ratio between excitatory and inhibitory signaling leads to social deficits in mice , an effect that is rescued by an increase of inhibitory circuit activity ( Yizhar et al . , 2011 ) . Here , we present evidence that increasing inhibitory activity alone can reverse the majority of symptoms of a mouse model of RTT . Using existing , well-characterized mouse lines , we restored full-length Mecp2 transcription solely in GABAergic neurons and enhanced inhibitory function in an otherwise Mecp2-null brain . This was sufficient to extend lifespan significantly and rescue body weight , ataxia , and apraxia in both male and female rescue mice . At the time of this writing , there have been several conditional knockouts of Mecp2 that target different neuronal subtypes , each replicating a few aspects of the Mecp2 deletion phenotype: Sim1 conditional knockout mice are hyperphagic and become obese ( Fyffe et al . , 2008 ) ; mice with Mecp2 deleted from dopaminergic neurons develop some motor deficits ( Samaco et al . , 2009 ) ; and conditional knockout of Mecp2 from basal forebrain cholinergic neurons may result in decreased anxiety and impaired social interaction ( Zhang et al . , 2016 ) . Deletion of Mecp2 from two subtypes of inhibitory neurons , PV- and SOM-expressing neurons , also shortened lifespan and replicated a subset of the phenotypes of global Mecp2 loss: PV-conditional knockout mice lacked critical period plasticity in the visual cortex ( He et al . , 2014 ) and developed motor , learning and memory , and social interaction deficits , while SOM-conditional knockout mice exhibited stereotyped behaviors and seizures ( Ito-Ishida et al . , 2015 ) ; interestingly , reexpression of Mecp2 in these individual populations prolongs life and improves disease scores and auditory information processing ( Goffin et al . , 2014 ) . Of all the conditional knockouts , however , the GABAergic conditional knockout mouse most closely replicates the effect of global Mecp2 deletion , with premature lethality , motor dysfunction , learning and memory deficits , and stereotypies . Only the anxiety and the persistent tremor of the Mecp2-null mouse is not reproduced ( Chao et al . , 2010 ) —behaviors that are reproduced instead by the glutamatergic conditional knockout , along with motor incoordination , acoustic startle deficits , and early lethality ( see companion paper Meng et al . , 2016 ) . In parallel , our rescue of Mecp2 expression solely in GABAergic neurons in otherwise Mecp2-null male mice completely reverses hypersociability , ataxia , and apraxia , as well as partially normalizing anxiety , but has no effect on acoustic startle responses or tremor . Meng et al’s glutamatergic rescue of Mecp2 expression in the same Mecp2-null male mice completely normalizes anxiety , tremor , and acoustic startle deficits , as well as reversing hypersociability and extending lifespan , but does not affect motor coordination . This extraordinary complementarity makes it clear that while certain behaviors may be mainly dependent on the proper function of only excitatory neurons ( ie acoustic startle response , tremor ) or only inhibitory neurons ( ataxia ) , other behaviors are dependent on both populations functioning properly ( social behavior ) . However , caution must be exercised in assessing the exact contributions of each population , as the Viaat-Cre allele also expresses in glycinergic neurons , which are important for breathing and proper brainstem function ( Rahman et al . , 2015 ) . Furthermore , small populations of neurons in the lateral and medial habenula , the auditory cortex , and possibly in the CA3 release both GABA and glutamate ( Shabel et al . , 2014; Root et al . , 2014; Safiulina , 2006; Noh et al . , 2010; Uchigashima et al . , 2007 ) , which further complicates the picture . However , inhibitory neurons are found throughout the brain and , due to their exquisite sensitivity to the surrounding circuitry , can influence the behavior of large populations of surrounding neurons ( Xue et al . , 2014; Bonifazi et al . , 2009 ) , while glycinergic and GABA/glutamate-coreleasing neurons are far smaller in number and restricted to discrete regions of the brain . It is thus reasonable to attribute the bulk of the behavioral rescue to the improved inhibitory function we observe in the rescue mice . Thus , the work presented here not only indicates that some behavioral aspects of RTT pathogenesis are modularly regulated by specific populations of neurons but also emphasizes the need for a complete understanding of the interplay between excitatory and inhibitory neurons in behavioral outputs . It is telling that neither glutamatergic nor GABAergic rescue of Mecp2 completely rescued premature lethality , although both rescue mice displayed similarly extended lifespans ( Figure 1C , also ) . A normal lifespan in the male mice is likely particularly dependent on the function of both neuron types . In addition , there may also be a role for glia in RTT pathogenesis , as these cells are critical for neurotransmitter recycling ( Marcaggi and Attwell , 2004 ) . Notably , reactivation of Mecp2 transcription solely in GFAP-positive astrocytes rescues the respiratory abnormalities of the Mecp2-null mouse and extends survival ( Lioy et al . , 2011 ) , whereas our GABAergic rescue mouse showed no clear improvement in breathing . It is therefore conceivable that normal respiration and normal lifespan depend on an interplay between proper excitation , inhibition , and astrocyte function . Differences between male and female mice in the GABAergic rescue responses are worth some consideration . In the male rescue mice , inhibitory cells express Mecp2 , but all excitatory neurons are null for Mecp2 . When Mecp2 expression is restored in inhibitory neurons in males , the increase in GABA signaling may effectively counteract much of the deficit resulting from the loss of Mecp2 in all excitatory neurons , possibly explaining the profound rescue of disease-related symptoms in the male mice . In the female rescue mice , however , random X-inactivation should allow roughly 50% of the excitatory and inhibitory neurons to retain functional Mecp2 ( Young and Zoghbi , 2004 ) , resulting in a heterogeneity of excitatory neuron activity within the brain of female mice that may neutralize much of the benefit gained by normalizing inhibitory neuron function . This idea is supported by the work of Meng et al . , who show that rescue of Mecp2 expression solely in excitatory neurons confers greater behavioral rescue in females than males . These glutamatergic rescue females benefit from a fully normalized excitatory neuron population as well as normal functioning from approximately 50% of inhibitory neurons , resulting in a strong rescue that is less impeded by heterogeneity . Males , however , are still left with dysfunctional inhibitory circuits and thus experience less benefit from restoration of MeCP2 levels in excitatory neurons . It is clear that inhibitory neurons are a valuable target for future pharmacological studies; future therapies that boost GABAergic function in both wildtype and Mecp2-null cells might prove more effective than a treatment targeted to the null cells alone . However , the work presented here and the work of Meng et al . also indicate that to achieve complete reversal of RTT symptoms , it may be necessary to pharmacologically elevate both inhibition and excitation . Lastly , the discovery that genetic restoration of Mecp2 expression solely in GABAergic neurons improves several phenotypes in both male and female mice throughout their lifespans renders this pathway a viable route for modifying disease course , even if only in the first few months to years of the disease . Pharmacological therapies enhancing GABAergic signaling might confer noticeable benefit to RTT syndrome patients , buying time until additional therapies can be added to modulate other pathways such as glutamatergic signaling . However , our findings suggest great care is warranted in the choice of disease model for any future drug trial , as the Mecp2-null male and Mecp2-heterozygous female mice responded very differently to genetic rescue of Mecp2 expression . For a therapeutic to be truly translatable to human RTT patients , it must be tested in both models .
All mouse care and manipulation was approved by the Baylor College of Medicine Institutional Animal Care and Use Committee ( IACUC , Protocol AN-1013 ) . Mice were housed in an AAALAS-certified Level 3 facility on a 14 hr light cycle . Male FVB mice carrying the Viaat-Cre transgenic line ( Chao et al . , 2010 ) were mated with 129S6SvEvTac females heterozygous for the Mecp2lox-Stop allele ( Guy et al . , 2007 ) , resulting in male and female F1 hybrid offspring . For the reporter experiments , Viaat-Cre males were crossed with female C57Bl/6 ROSA-YFP mice . For the spontaneous firing experiments , Mecp2-/Y male mice from the same hybrid background were used . After weaning , all mice were group housed ( 2–5 mice per cage ) as a mix of genotypes . All mice included in the survival curve were weighed weekly and scored according to the 6-category disease scoring scale , as previously described ( Guy et al . , 2007 ) ( Wildtype = 41 , Viaat-Cre = 37 , Mecp2lox-Stop = 33 , Viaat-Cre; Mecp2lox-Stop = 47 ) . The investigator remained blind to the genotypes of all tested mice during phenotypic characterization and behavioral testing . Male mice used for behavioral assays were divided into 2 cohorts , each being assessed by different assays at different time points . All behavioral assays were conducted during the light cycle , generally in the afternoon . For Cohort 1 , mice ( n = 21–24 per genotype ) were tested at 6 weeks of age on OFA , grip strength , and rotarod; at 8 weeks for partition test with nesting assessment , PPI , and conditioned fear , and at 30 weeks by OFA , grip strength , and rotarod . Cohort 2 ( n = 9–15 per genotype ) was tested at 9 weeks for OFA and marble burying . For females , Cohort 1 ( n = 8–18 per genotype ) was tested at 18 weeks for OFA , grip strength , footslip , and partition with nesting assessment . Cohort 2 ( n = 11–17 per genotype ) was tested at 9 weeks for OFA , rotarod , PPI , and conditioned fear , at 30 weeks for OFA , grip strength , rotarod , partition with nesting assessment , and marble burying . Both female cohorts were tested at 1 year for OFA , grip strength , rotarod , footslip , and conditioned fear . All mice were assessed weekly for body weight and disease score , which included movement , gait , hindlimb clasping , tremor , breathing , and general condition ( Guy et al . , 2007 ) . The investigator was blinded to all genotypes after completion of data collection . Mice were habituated for 30 min in the test room lit at 200 lux with white noise playing at 60 dB . Each mouse was placed singly in the open field apparatus ( OmniTech Electronics , Columbus , OH ) and allowed to move freely for 30 min . Locomotion parameters and zones were recorded using the Fusion activity monitoring software . Data is shown as mean ± standard error of mean and was analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated for 30 min in the test room lit at 200 lux with white noise playing at 60 dB . The elevated plus maze is a plus sign-shaped maze with two opposite arms enclosed by walls and two opposite arms open without walls . The entire maze is elevated above the floor . Mice were placed singly at the intersection of the four arms and allowed to move freely for 10 min . Activity was recorded by a suspended digital camera and recorded by the ANY-maze software ( Stoelting Co . , Wood Dale , IL ) . Data is shown as mean ± standard error of mean . Time and distance in the open arm were each analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated for 30 min in the test room lit at 200 lux with white noise playing at 60 dB . Mice were placed singly in the light side of the light dark apparatus ( Omnitech Electronics ) and allowed to move freely for 10 min . Locomotion parameters and zones were recorded using Fusion activity monitoring software . Data is shown as mean ± standard error of mean . Time in Light was analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated in the test room for 30 min . Each mouse was allowed to grab the bar of a digital grip strength meter ( Columbus Instruments , Columbus , OH ) with both forepaws while being held by the tail and then pulled away from the meter with a constant slow force until the forepaws released . The grip ( in kg of force ) was recorded and the procedure repeated twice for a total of three pulls , which were averaged for the final result . Data is shown as mean ± standard error of mean . Grip strength was analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated in the test room for 30 min . The mouse was placed on a 0 . 635 cm diameter dowel with all four paws allowed to grip the dowel . Latency to fall and number of side touches were recorded during the two minute test . Data is shown as mean ± standard error of mean and was analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were placed on the rotating cylinder of an accelerating rotarod apparatus ( Ugo Basile , Varese , Italy ) and allowed to move freely as the rotation increased from 5 rpm to 40 rpm over a five-minute period . Latency to fall was recorded when the mouse fell from the rod or when the mouse had ridden the rotating rod for two revolutions without regaining control . Data is shown as mean ± standard error of mean . Latency to fall was analyzed by two-way ANOVA with Bonferroni’s post hoc analysis . Mice were habituated in the test room for 30 min . Each mouse was placed in a footslip chamber consisting of a plexiglass box with a floor of parallel-positioned rods and allowed to move freely for 10 min . Movement was recorded by a suspended digital camera , while footslips were recorded using ANY-maze software ( Stoelting Co . ) . At the completion of the test , mice were removed to their original home cage . Total footslips were normalized to the distance traveled for data analysis . Data is shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated for 30 min in the test room . A standard mouse housing cage was 50% filled with clean bedding material , and 20 black glass marbles were placed in a 4x5 grid pattern on the surface of the bedding . Mice were placed singly into the prepared cage for 30 min . After the mouse was removed , the number of buried marbles were counted , with a marble considered buried if 75% of its surface was covered with bedding . Data is shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were single-housed for 48 hr on one side of a standard mouse housing cage . The cage was divided across its width by a divider with holes small enough to allow scent but no physical interaction . The test mouse was provided with a KimWipe folded in fourths as nesting material . At 24 hr and 48 hr of single-housing , the KimWipe was assessed for nesting score , as described previously ( Chao et al . , 2010 ) . At least 16 hr before the partition tests , a novel age- and gender-matched partner mouse of a different background was placed on the opposite side of the partition . On the day of the test , the cage was placed on a well-lit flat surface . All nesting material and water bottles were removed from both sides of the cage , and the test mice were observed for 5 min while interaction time with the now-familiar partner mouse was recorded . Interactions involved the test mouse smelling , chewing , or actively exploring the partition . At the end of the first test ( Familiar 1 ) , the familiar partner mouse was replaced by a novel mouse of the same age , gender , and strain , and test mouse interactions were recorded for five minutes ( Novel ) . The novel mouse was then removed and the familiar partner mouse returned to the cage , followed by observation for another 5 min ( Familiar 2 ) . At the completion of the partition test , test mice were returned to their original home cage . Data is shown as mean ± standard error of mean . Interaction times were analyzed by two-way ANOVA with Bonferroni’s post hoc analysis , and nesting scores were analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated for 30 min outside the test room . Each mouse was placed singly in SR-LAB PPI apparatus ( San Diego Instruments , San Diego , CA ) , which consisted of a Plexiglass tube-shaped holder in a sound-insulated lighted box , with 70 dB white noise and allowed to habituate for 5 min . The mouse was presented with eight types of stimulus , each presented six times in pseudo-random order with a 10–20 ms intertrial period: no sound , a 40 ms 120 db startle burst , three 20 ms prepulse sounds of 74 , 78 , and 82 dB each presented alone , and a combination of each of the three prepulse intensities presented 100 ms before the 120 dB startle burst . After the test , mice were returned to their home cage . The acoustic startle response was recorded every 1 ms during the 65 ms period following the onset of the startle stimulus and was calculated as the average response to the 120 db startle burst normalized to body weight . Percent prepulse inhibition was calculated using the following formula: ( 1- ( averaged startle response to prepulse before startle stimulus/averaged response to startle stimulus ) ) x 100 . Data are shown as mean ± standard error of mean . Percent prepulse inhibition was analyzed by two-way ANOVA with Bonferroni’s post hoc analysis , and acoustic startle response was analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice ( 5–6 weeks old , n = 2–3 ) were perfused transcardially first with 1XPBS and then by 4% PFA . The brains were removed and postfixed overnight in 4% PFA , followed by storage in 30% sucrose until sinking . For sectioning , fixed brains were embedded in TissueTek and sectioned by cryostat at 25 um . Sections were maintained in 1XPBS with sodium azide at 4°C until stained using a free-floating protocol . In short , sections were washed once in 1XPBS and then blocked for 1 hr at room temperature ( RT ) in 3% normal goat serum ( NGS ) with 0 . 3% TritonX in 1XPBS . Primary antibodies were diluted in 1XPBS with 1 . 5% NGS with 0 . 3% Tween-20 in 1XPBS , and sections were stained for 48 hr shaking at 4°C in a ~300 μl working volume . Primary antibodies used were: chicken anti-GFP ( Abcam , Cambridge UK , Cat# ab13970 ) 1:2000; mouse-anti-GFAP ( Sigma , St . Louis MO , Cat# G3893 ) 1:1000; rabbit-anti-Iba1 ( Wako , Richmond VA , Cat#019–19741 ) 1:1000; rabbit anti-Mecp2 ( Cell Signaling , Danvers MA , Cat#3456S ) 1:200; mouse anti-GAD67 ( Chemicon , Temecula CA , Cat# MAB5406 ) 1:1000; mouse anti-CamKII ( Abcam Cat#ab22609 ) 1:50; guinea pig ant-VGAT 1:1000 ( Frontier , Ishikari Japan , Cat#VGAT-GP-Af1000 ) ; and rabbit anti-Vglut1 1:1000 ( Frontier Cat#VgluT1-Rb-Af500 ) . After three ten-minute washes in 1XPBS , sections were incubated overnight in secondary antibodies at 1:500 dilution as follows: goat anti-chicken Alexa 488 ( Invitrogen , Carlsbad CA , Cat# A-11039 ) ; horse anti-mouse Texas Red ( Vector , Burlingame CA , Cat# TI-2000 ) ; goat anti-rabbit Texas Red ( Vector Cat# TI-1000 ) ; goat anti-rabbit Alexa 488 ( Invitrogen Cat# A11034 ) ; goat anti-mouse Alexa 555 ( Invitrogen Cat# A-21422 ) ; goat anti-rabbit Alexa 555 ( Invitrogen Cat#A-21429 ) ; and goat anti-guinea pig Alexa 555 ( Invitrogen A-21435 ) . Sections were then washed three times with 1XPBS at 10 min each wash , mounted on charged glass slides with Vectashield ( Vector ) mounting medium with DAPI ( H-1200 ) , coverslipped , and imaged using a confocal microscope . For synaptic puncta counts , six images were captured from each animal’s CA1 using the 63X objective and maintaining the same imaging settings for each section . Images were counted for absolute puncta numbers using the Spots module of Imaris v8 . 0 ( Bitplane , Zurich , Switzerland ) and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Viaat-Cre male mice were crossed with females carrying the ROSA-YFP allele . 3-week-old male pups were anesthetized with isoflurane , sacked by decapitation , and the brains and bone marrow isolated . Brains were homogenized by mincing and digestion in Dulbecco’s Modified Eagle Media ( DMEM ) with trypsin and 0 . 1 mg DNAseI for 10 min at 37°C . The tissue was then run through a 22G1/2 needle 5 times and centrifuged for 10 min at 2200 rpm at 4°C . The pellet was washed with 10 mL DMEM and spun for 10 min at 1000 rpm at 4°C . The pellet was then resuspended in 1 mL Hank’s Balanced Salt Solution ( HBSS ) and filtered through a 40 uM filter before being FACS sorted . Bone marrow was isolated from the tibia and femur and run through an 18G needle several times to dissociate the tissue and then suspended in HBSS with 10% FBS and 10mM HEPES . The cells were centrifuged five minutes at 2200 rpm at 4°C , then resuspended in HBSS and sorted . For both brain and bone marrow , 100 , 000 events were sorted . Acute fresh cortical slices were prepared from 5- to 6-week-old mice as previously described ( Lu et al . , 2009 ) . Coronal slices ( 250–350 μm thick ) containing somatosensory cortex were cut with a vibratome ( Leica Microsystems Inc . , Buffalo Grove , IL ) in a chamber filled with chilled ( 2–5°C ) cutting solution containing ( in mM ) 110 choline-chloride , 25 NaHCO3 , 25 D-glucose , 11 . 6 sodium ascorbate , 7 MgSO4 , 3 . 1 sodium pyruvate , 2 . 5 KCl , 1 . 25 NaH2PO4 and 0 . 5 CaCl2 . The slices were then incubated in artificial cerebrospinal fluid ( ACSF , in mM ) containing 119 NaCl , 26 . 2 NaHCO3 , 11 D-glucose , 3 KCl , 2 CaCl2 , 1 MgSO4 , 1 . 25 NaH2PO4 at the room temperature . The solutions were bubbled with 95% O2 and 5% CO2 . Whole-cell recording was made from pyramidal neurons in the layer II/III/V region of somatosensory cortex by using a patch-clamp amplifier ( MultiClamp 700B , Molecular Devices , Union City , CA ) under infrared differential interference contrast optics . Microelectrodes were made from borosilicate glass capillaries and had a resistance of 2 . 5–5 MΩ . Data was acquired with a digitizer ( DigiData 1440A , Molecular Devices ) . The analysis software pClamp 10 ( Molecular Devices ) and Minianalysis 6 . 0 . 3 ( Synaptosoft Inc . , Decatur , GA ) were used for data analysis . Miniature EPSCs ( mEPSCs , holding at −70mV ) were recorded in voltage-clamp mode in the presence of 100 mM Picrotoxin and 0 . 5 mM Tetrodotoxin ( TTX ) . The intrapipette solution contained ( in mM ) 140 potassium gluconate , 5 KCl , 10 HEPES , 0 . 2 EGTA , 2 MgCl2 , 4 MgATP , 0 . 3 Na2GTP and 10 Na2-phosphocreatine , pH 7 . 2 ( with KOH ) . Spontaneous firings were recorded in current-clamp mode ( holding at −60 mV ) in modified ACSF ( in mM: 126 NaCl , 25 NaHCO3 , 14 D-glucose , 3 . 5 KCl , 1 CaCl2 , 0 . 5 MgCl2 , 1 NaH2PO4 ) . For recording mIPSCs , neurons were also held at −70 mV in voltage-clamp mode in the presence of 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX , 10 mM ) , D-2-amino-5-phosphonopentanoic acid ( AP5 , 25 mM ) and TTX ( 0 . 5 mM ) . The glass pipettes were filled with high-Cl- intrapipette solution containing ( in mM ) 145 KCl , 10 HEPES , 2 MgCl2 , 4 MgATP , 0 . 3 Na2GTP and 10 Na2-phosphocreatine , pH 7 . 2 ( with KOH ) . Data were discarded when the change in the series resistance was >20% during the course of the experiment or the rest membrane potential was about −60 mV . The whole-cell recording was performed at 30 ( ±1 ) °C with the help of an automatic temperature controller ( Warner Instruments , Hamden , CT ) . Summary data are presented as mean ± S . E . M . and analyzed by one-way ANOVA with Tukey’s post hoc analysis . 8-week-old male mice ( n = 6–7 per genotype ) were deeply anesthetized with isofluorane , quickly decapitated , and their cerebella dissected and immediately frozen by liquid nitrogen . The frozen tissue was placed in 1 ml Purezol ( Bio-Rad , Hercules , CA ) on ice and immediately homogenized with a Polytron homogenizer . The dissolved sample was diluted to 2x using Purezol , and 1 ml of the solution was processed by Aurum Total RNA Fatty and Fibrous Tissue kit ( Bio-Rad ) to collect RNA . DNA within the sample was removed with on-column DNase digestion which was included in the kit . First-strand cDNA was synthesized using M-MLV reverse transcriptase ( Life Technologies , Carlsbad , CA ) . RT-qPCR was performed using Bio-Rad CFX96 Real-Time system . The primer sets were designed to amplify target genes using UCSC genome browser and Primer 3 . The same primer sets from He et al . were used to amplify Gad1 , Gad2 , Pv , Vgat , Gephrin , Gabra1 , Kv3 . 1 , and Vglut1 ( He et al . , 2014 ) . To search for other candidate genes , we referred to the microarray results obtained from the cerebellum of Mecp2-null ( Mecp2-/y ) and overexpressing mice ( Tg3 ) ( Ben-Shachar et al . , 2009 ) . The top 20 genes altered in opposite directions in the two lines were selected from the microarray results . We checked expression pattern of these genes using the Allen Brain Atlas , and we chose those genes that were clearly expressed in inhibitory neurons in the cerebellum ( i . e . , either Purkinje cells , stellate cells , or Golgi cells ) as candidates . All RT-qPCR reactions were conducted in duplicate , and relative proportions of the cDNA were determined based on the threshold cycle ( Ct ) . The results were averaged for each sample and normalized to the value of Gapdh . Relative expression levels of the target genes were determined by normalizing the fold expression level in each sample to the average of the littermate WT controls . Statistical analysis were performed on delta Ct values ( i . e . , [average Ct of the target gene] – [average Ct of Gapdh] ) , by one-way ANOVA followed by Tukey’s post hoc test . The following primers were designed in this study: Gapdh: Forward 5’- ggagattgttgccatcaacga-3’ , Reverse 5’- tgaagacaccagtagactccacgac-3’ Nxph4: Forward 5’-gtgagcacccctactttgga-3’ , Reverse 5’-aaggctgtttttctccacca-3’ Gabra3: Forward 5’-ccaccatctccaagtctgct-3’ , Reverse 5’-agatgatgcgggaaattttg-3’ Kcng4: Forward 5’-caggaggaacctcagtcagc-3’ , Reverse 5’-gagcccatggatatgtggac-3’ Opn3: Forward 5’-tggtatccctgttcggagtc-3’ , Reverse 5’-tcataggccagcacagtgag-3’ Scg2: Forward 5’-gaaatgatcagggctttgga-3’ , Reverse 5’-ctctctgccaagtggctttc-3’ Ai593442: Forward 5’-gacattgacaccgaagcaaa-3’ , Reverse 5’-cagctgaaggtgaggaaagg-3’ Robo2: Forward 5’-agcagtccactgccactctt-3’ , Reverse 5’-gttgtggaggtggggtattg-3’ Rassf8: Forward 5’-ggtcaatgaggaggaggtga-3’ , Reverse 5’-ctgctccttgtcctgtagcc-3’ Cirbp: Forward 5’-cgaagtggtggtggtaaagg-3’ , Reverse 5’-ccagcctggtcaactctgat-3’ Cabp7: Forward 5’-atggcactgactttgacacg-3’ , Reverse 5’-tggctctcctcctctgtcat-3’ All other primers used have been previously published ( He et al . , 2014 ) . 8-week-old male mice ( n = 4–5 per genotype ) were deeply anesthetized with isofluorane , quickly decapitated , and their striata dissected into cold 1X PBS and snap frozen . Tissue was homogenized in cold 1X PBS with a polytron and centrifuged for 5 min at 13 . 2 krpm at 4°C . The supernatant was collected for HPLC analysis on an amino acid analyzer ( Mayne et al . , 2001 ) . The methods were modified from previous publications ( Sztainberg et al . , 2015 ) . Adult male mice at 7–8 weeks of age were anesthetized with isoflurane and mounted in a stereotaxic frame . Under aseptic conditions , each mouse was surgically implanted with 4 silver wire ( 127 µm diameter , A-M Systems , Sequim , WA ) recording electrodes aimed at the subdural space of the left/right frontal cortex ( A2 . 0 , LR1 . 8 ) and left/right parietal cortex ( P3 . 8 , LR1 . 8 ) , respectively ( Paxinos and Franklin , 2001 ) . The reference electrode was positioned in the occipital region of the skull . All electrode wires were attached to a miniature connector ( Harwin Connector , Portsmouth , Hampshire , UK ) that was secured on the skull by dental cement . After 2 weeks of post-surgical recovery , cortical EEG activities ( filtered between 0 . 1 Hz and 1 kHz , sampled at 2 kHz ) were recorded for 1 hr per day over 4 days . Traces were analyzed by eye for abnormal electrographic signals . Sample sizes for all analyses were determined based on previous experience ( Chao et al . , 2010; Chahrour et al . , 2008; Samaco et al . , 2009 , 2013 ) . All statistical analyses were conducted using Graphpad Prism software . Details of the analysis , including individual p values , are reported in Supplementary file 1 . | Rett syndrome is a childhood brain disorder that mainly affects girls and causes symptoms including anxiety , tremors , uncoordinated movements and breathing difficulties . Rett syndrome is caused by mutations in a gene called MECP2 , which is found on the X chromosome . Males with MECP2 mutations are rare but have more severe symptoms and die young . Many researchers who study Rett syndrome use mice as a model of the disorder . In particular , male mice with the mouse equivalent of the human MECP2 gene switched off in every cell in the body ( also known as Mecp2-null mice ) show many of the features of Rett syndrome and die at a young age . The MECP2 gene is important for healthy brain activity . The brain contains two major types of neurons: excitatory neurons , which encourage other neurons to be active; and inhibitory neurons , which stop or dampen the activity of other neurons . In 2010 , researchers reported that mice lacking Mecp2 in only their inhibitory neurons develop most of the same problems as those mice with no Mecp2 at all . This discovery led Ure et al . – including a researcher involved in the 2010 study – to ask if activating Mecp2 in the same neurons in otherwise Mecp2-null mice was enough to prevent some of their Rett syndrome-like symptoms . The experiments showed that male mice that only have Mecp2 activated in their inhibitory neurons lived several months longer than male Mecp2-null mice . These male “rescue mice” also moved normally and had a normal body weight , though they still experienced anxiety , tremors and breathing difficulties . Female mice represent a better model of human Rett syndrome patients , and Ure et al . found that female rescue mice showed smaller improvements than the males . These data suggest that when a brain is missing Mecp2 everywhere , as in male Mecp2-null mice , turning on Mecp2 in inhibitory neurons can make the brain network nearly normal and prevent most Rett-syndrome-like symptoms . However , the brains of female rescue mice contain both normal cells and cells with mutated Mecp2 . This mixture of normal and abnormal cells appears to cause abnormalities that cannot be overcome by rescuing just the activity of the inhibitory neurons . These findings also highlight the importance of doing future studies in female mice to better understand the development of Rett syndrome . The next challenge is to test different ways of activating the inhibitory neurons in the female mouse brain , for example by using drugs that target these neurons . It is hoped these methods will help researchers to refine a path toward potential new treatments for Rett syndrome patients . Finally , in a related study , Meng et al . asked how deleting or activating Mecp2 only in the excitatory neurons of mice affected Rett-syndrome-like symptoms . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Restoration of Mecp2 expression in GABAergic neurons is sufficient to rescue multiple disease features in a mouse model of Rett syndrome |
Salivary glands , such as submandibular glands ( SMGs ) , are composed of branched epithelial ductal networks that terminate in acini that together produce , transport and secrete saliva . Here , we show that the transcriptional regulator Yap , a key effector of the Hippo pathway , is required for the proper patterning and morphogenesis of SMG epithelium . Epithelial deletion of Yap in developing SMGs results in the loss of ductal structures , arising from reduced expression of the EGF family member Epiregulin , which we show is required for the expansion of Krt5/Krt14-positive ductal progenitors . We further show that epithelial deletion of the Lats1 and Lats2 genes , which encode kinases that restrict nuclear Yap localization , results in morphogenesis defects accompanied by an expansion of Krt5/Krt14-positive cells . Collectively , our data indicate that Yap-induced Epiregulin signaling promotes the identity of SMG ductal progenitors and that removal of nuclear Yap by Lats1/2-mediated signaling is critical for proper ductal maturation .
The mammalian epithelial branching program is a highly dynamic and organized process that leads to the formation of branched networks of tubule structures that terminate in acini with specialized functions . Understanding how epithelial progenitor cells pattern into ducts and acini , and how this is coordinated with ongoing tissue morphogenesis , is one of the central questions in epithelial development . The submandibular gland ( SMG ) offers a model to study the molecular mechanisms directing epithelial branching morphogenesis and patterning , with distinct synchronized processes of cell proliferation , clefting , differentiation , migration and apoptosis occurring rapidly during embryogenesis ( Hauser and Hoffman , 2015; Mattingly et al . , 2015 ) . The developing SMG epithelium communicates with neighboring mesenchymal , neuronal and endothelial cells to direct reiterative rounds of bud and duct formation that mature into epithelial domains that mediate the production , transportation , and secretion of saliva ( Knosp et al . , 2015; Knox et al . , 2010; Lombaert et al . , 2013; Patel et al . , 2011; Steinberg et al . , 2005; Wells et al . , 2014 ) . Following initial bud formation , it is thought that specification of distinct multipotent progenitor populations give rise to the specialized cell populations that compose the acinar and ductal domains . For example , multi-potent populations of Cytokeratin-5 ( Krt5 , K5 ) - and Cytokeratin-14 ( Krt14 , K14 ) -positive progenitors are thought to govern the formation of epithelial cells that give rise to the mature ductal structures ( Knox et al . , 2010; Lombaert et al . , 2013 ) . Although numerous molecular studies have focused on understanding the biology of SMG progenitors , much remains unclear about the intrinsic signals that define their identity and/or control their differentiation . Recent studies have provided evidence that the transcriptional regulator Yap plays essential roles in stem cell biology and that these roles are essential for the development of branching organs , such as the kidney , lung , pancreas , and mammary gland ( Varelas , 2014 ) . Ectopic expression of Yap has been shown to drive the expansion of progenitor populations in several tissues , while conditional deletion of Yap in organ-specific stem cells can lead to the inhibition of stem cell specification or the induction of premature differentiation ( Mahoney et al . , 2014; Panciera et al . , 2016; Zhao et al . , 2014 ) . In particular , the dynamics of Yap localization is implicated in controlling the balance of specification , self-renewal , and differentiation of various stem cell populations . Yap localization is controlled by a multitude of signals that include those mediated by the Hippo pathway ( Meng et al . , 2016 ) . The Hippo pathway is comprised of a series of kinase-mediated signaling events that result in the phosphorylation and activation of the Lats1 and Lats2 kinases ( herein together referred to as Lats1/2 ) . Activated Lats1/2 redundantly direct the phosphorylation of Yap on conserved serine residues , the best characterized of which is S112 in mouse Yap ( S127 in human Yap ) , which restricts the nuclear accumulation and transcriptional activity of Yap . Loss of Lats1/2-mediated regulation of Yap activity leads to defective organ patterning and function ( Heallen et al . , 2011 , 2013; Reginensi et al . , 2016; Yi et al . , 2016 ) , highlighting the importance of Hippo pathway signaling in development . Here , we used genetic approaches to examine the roles of Hippo-Yap signaling in SMG epithelial development . We found that embryonic deletion of Yap in developing SMGs resulted in severe morphogenesis defects , which notably did not arise from aberrant cell proliferation or apoptosis , but rather from defects in progenitor patterning . We show that Yap is required for the specification of Krt5/Krt14-positive ductal progenitor cells , and that Yap does so , in part , by controlling the expression of the epidermal growth factor family member Epiregulin ( Ereg ) . Treatment of ex vivo cultured SMGs with Ereg was sufficient to expand Krt5/Krt14-positive cells and rescue cell fate specification defects observed in Yap-deleted SMGs . Conversely , we found that deletion of the Hippo kinases Lats1/2 resulted in massive expansion of Krt5/Krt14-positive ductal cells in developing SMG epithelium , and that this phenotype could be blunted by EGFR ( ErbB-1 ) inhibition . These findings demonstrate that Yap is a critical regulator of ductal progenitor cell identity in SMG epithelium and that proper control of Yap localization by Lats1/2 is essential for the maturation of SMG ducts . Our study therefore identifies novel essential effectors of SMG development and provides important insight into early patterning events that are coupled with branching morphogenesis .
To gain insight into the role ( s ) of Yap during SMG development we examined the levels and localization of Yap in early branching SMGs using immunofluorescence ( IF ) microscopy . We closely monitored the distribution of Yap with respect to known markers of early patterning , including Krt14 , which labels multipotent progenitors that can give rise to ductal epithelial cells ( Knox et al . , 2010; Nedvetsky et al . , 2014 ) ( illustrated in Figure 1A ) . We found that Yap was prominently expressed in E13 . 5 SMG epithelium , and distinct Yap localization differences were apparent in various cell populations of the branching gland . A cell layer at the peripheral edge of each end-bud showed some cells with nuclear Yap localization , whereas all cells immediately adjacent and extending away from the edge of the bud showed cytoplasmic Yap localization ( Figure 1B–C ) . Yap also showed very prominent nuclear localization in cells transitioning proximally towards the newly developing ductal regions ( Figure 1B–C ) , overlapping precisely with Krt14-positive ductal progenitors ( Figure 1D ) . 10 . 7554/eLife . 23499 . 003Figure 1 . Nuclear Yap marks distinct populations of developing SMG ductal epithelial cells . ( A ) Illustration depicting the early developing SMG epithelium , with the positioning of relevant multipotent progenitor cells highlighted . ( B , C ) Two different magnified IF microscopy images of the bud-duct transition zone showing the localization of Yap ( green ) in E13 . 5 mouse SMG epithelium . ( D ) Images from IF microscopy analysis of Yap ( green ) together with Krt14 ( K14 , white ) shows prominent nuclear Yap in Krt14-positive cells . ( E ) Illustration depicting the maturing stratified SMG ductal epithelium with relevant cell populations highlighted , and a description of our observed localization pattern for Yap . ( F ) Images from IF microscopy analysis of total Yap ( green ) and phospho-S112 Yap ( red ) in E15 . 5 mouse SMGs . In luminal cells , Yap phosphorylation levels are elevated and Yap is excluded from the nucleus , while many basal cells exhibit prominent nuclear Yap localization . ( G ) Images from IF microscopy analysis of E18 . 5 mouse SMG ducts for Yap ( green ) and Krt5 ( K5 , red ) showing prominent nuclear Yap localization in a subset of Krt5-positive basal cells and cytoplasmic Yap in Krt5-negative luminal cells . White arrows highlight prominent nuclear Yap localization in the basal cells in ( F ) and ( G ) . DAPI was used to mark the nuclei ( blue ) in all images , and for clarity the basal surface of the epithelium is outlined with a white dotted line . Scale bar = 20 µm . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 003 As the ductal epithelium of the SMG starts to mature , cells begin stratifying into luminal and basal layers . Krt5-positive basal-positioned cells possess stem cell activity and are believed to play important roles in adult SMG injury repair ( Knox et al . , 2013 ) ( illustrated in Figure 1E ) . The large majority of maturing ductal epithelial cells in E15 . 5 and in differentiated E18 . 5 SMGs exhibited cytoplasmic Yap localization , particularly cells positioned at the luminal layer ( Figure 1F–G ) . Cytoplasmic Yap localization correlated with increased Serine-112 phosphorylation of Yap ( pS112-Yap ) ( Figure 1F ) , which is a site phosphorylated by the Hippo pathway kinases Lats1/2 and promotes cytoplasmic Yap localization ( Dong et al . , 2007 ) . While few in number , some ductal epithelial cells in E15 . 5 and E18 . 5 SMGs exhibited prominent nuclear Yap localization , and these cells were generally positioned in the basal layer of Krt5-positive epithelial cells . However , not all Krt5-positive cells showed nuclear Yap localization , suggesting that nuclear Yap marks a distinct sub-population of cells with this marker or that we captured a snapshot of dynamic Yap localization occurring in these cells ( Figure 1F–G ) . To assess the importance of Yap in SMG development , we sought to conditionally delete Yap in the epithelium of embryonic SMGs . Cre recombinase driven by the promoter of the Shh ( sonic hedgehog ) gene ( Shh-Cre ) ( Harris et al . , 2006 ) has previously been used to target the SMG epithelium ( Knosp et al . , 2015 ) , which prompted us to test the efficiency of this model for use in our studies . We started by crossing Shh-Cre mice with Rosa26-loxP-STOP-loxP-EYFP mice ( Srinivas et al . , 2001 ) , allowing us to mark Shh-expressing cells with Enhanced Yellow Fluorescent Protein ( EYFP ) . All Shh-Cre-positive/EYFP-positive SMGs that we examined showed robust EYFP signal marking the entire SMG epithelium ( Figure 2A ) , indicating that the SMG epithelium originates from Shh-expressing cells and therefore this Cre model could be used to conditionally target loxP-flanked genes in the SMG epithelium . Accordingly , we found that crossing Shh-Cre mice with Yap-loxP/loxP mice led to the efficient deletion of Yap in the developing SMG epithelium ( herein called Yap-cnull SMGs ) ( Figure 2B ) , and that this led to striking branching defects . E13 . 5 Yap-cnull SMGs lacked developed clefts and ducts ( Figure 2C–D ) , and E15 . 5 Yap-cnull glands showed severely disorganized bud-like structures and complete absence of ductal trees ( Figure 2C–E ) . Interestingly , these phenotypes did not appear to result from global increases in apoptosis ( Figure 2F ) or defects in overall epithelial proliferation ( Figure 2G ) . 10 . 7554/eLife . 23499 . 004Figure 2 . Deletion of Yap in developing SMG epithelium results in severe branching defects and defective ductal domain patterning . ( A ) Lineage tracing using Shh-Cre; ROSA26-lox-STOP-lox-EYFP reporter mouse . E15 . 5 mouse SMGs of the indicated genotypes were dissected and immediately imaged on an inverted microscope in dark-field for the left panels and for fluorescent EYFP signal in the right panels . In the furthest right panels , the SMGs were compressed under a coverslip to highlight the EYFP-positive epithelial branches . Scale = 200 μm ( B ) Images from IF microscopy imaging of E15 . 5 Shh-Cre-Yap-null ( Yap-cnull ) SMGs for total Yap ( green ) and phospho-S112 Yap ( red ) showing efficient deletion of Yap in the SMG epithelium ( outlined by a dotted white line ) . ( C ) Phase-contrast images of E13 . 5 WT and Yap-cnull SMGs showing severe morphogenesis defects in Yap-cnull SMGs . Scale = 100 μm . ( D ) Quantitation of bud number from E13 . 5 WT and Yap-cnull SMGs ( n = 23 ) . ( E ) Phase-contrast images of E15 . 5 WT and Yap-cnull SMGs indicating a disorganized bud structure and lack of ductal trees in Yap-deficient SMGs . Note that the image from the WT SMG is stitched together from two images . ( F ) Images from IF microscopy analysis of Cleaved-caspase 3 in E15 . 5 WT and Yap-cnull SMGs showing no apparent defect in apoptosis . Note , that the Cleaved-caspase 3 antibody activity was validated in parallel slides containing positive cells . Scale = 20 μm . ( G ) Images from IF microscopy analysis of E15 . 5 WT and Yap-cnull SMGs for Yap ( green ) , Ki-67 ( red ) , or PCNA ( white ) showing no apparent proliferation defects in Yap-deleted epithelium . Scale = 10 μm . DAPI was used to mark the nuclei ( blue ) . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 004 We hypothesized that the morphogenesis defects associated with Yap deletion may originate from compromised epithelial patterning , prompting us to examine the distribution of progenitor markers following Yap deletion . We first examined fixed E15 . 5 SMGs , which revealed an almost complete absence of Krt14-positive ductal progenitors ( Figure 3A ) . To understand the dynamics of progenitor patterning , we isolated E13 . 5 SMGs from wild-type and Yap-cnull embryos and cultured them ex vivo for 24 hr . After 24 hr of explant culture , WT SMGs exhibited extensive branching , with expected robust Krt14 and Krt5 expression in developing ductal epithelial regions , and Krt19 expression marking maturing cells ( Figure 3B–E ) . Yap-cnull SMGs failed to branch after 24 hr of culture and exhibited an almost complete absence of Krt5 , Krt14 , or Krt19-expressing cells , except for a small segment of the most proximal region ( Figure 3B–E ) . The lack of these markers suggested that Yap-cnull SMGs fail to specify ductal progenitors and consequently ductal epithelium . Parasympathetic nerve innervation plays a crucial role in the growth and regulation of SMG ductal progenitors ( Knosp et al . , 2015; Knox et al . , 2010; Nedvetsky et al . , 2014 ) , which prompted us to examine the distribution of the parasympathetic nerve in Yap-cnull SMGs by staining for the nerve marker TuJ1 . Despite being present in the Yap-cnull SMGs , parasympathetic nerve innervation was diminished and unorganized , suggesting that signaling crosstalk with the nerve may be compromised ( Figure 3F–G ) . The structural organization of the actin cytoskeleton was also severely disrupted in Yap-cnull SMGs ( Figure 3G ) , indicating defective polarization that is required for proper epithelial maturation . Interestingly , markers associated with the developing bud domains , such as Sox10 , were abundant in Yap-cnull SMGs ( Figure 4A–C ) , suggesting that the early development of the bud epithelium is unaffected following the loss of Yap . Taken together , our analyses indicated that the deletion of Yap leads to a loss of an early ductal progenitor population , resulting in branching morphogenesis defects . 10 . 7554/eLife . 23499 . 005Figure 3 . Yap is required for SMG ductal epithelial patterning . ( A ) IF analysis of Yap ( green ) and Krt14 ( K14 , white ) in E15 . 5 WT and Yap-cnull SMGs . Scale = 10 µm . ( B–G ) E13 . 5 WT and Yap-cnull SMGs were dissected and cultured for 24 hr ex vivo and then examined by microscopy . ( B ) Phase-contrast images of E13 . 5 WT and Yap-cnull SMGs at the time of dissection and 24 hr after culture . The same SMGs ( each column is one SMG ) were analyzed by IF for ( C ) Krt14 ( K14 , magenta ) , ( D ) Krt5 ( K5 , green ) , ( E ) Krt19 ( K19 , red ) , ( F ) TuJ1 ( yellow ) , and ( G ) F-actin ( white , Phalloidin ) . DAPI was used to mark the nuclei ( blue ) . Scale = 100 μm . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 00510 . 7554/eLife . 23499 . 006Figure 4 . A relative increase in the number of cells expressing the bud marker Sox10 is observed in Yap-cnull SMGs . E13 . 5 WT and Yap-cnull SMGs were dissected and cultured for 24 hr ex vivo and then examined by microscopy . ( A ) Phase-contrast images were taken of E13 . 5 WT and Yap-cnull SMGs at the time of dissection and 24 hr after culture . The same SMGs ( each column is one SMG ) were analyzed by IF for Sox10 ( green ) , TuJ1 ( yellow ) , and F-actin ( white ) . DAPI was used to mark the nuclei ( blue ) . Scale = 100 µm . ( B ) Images from IF microscopy analysis of Sox10 in E13 . 5 ( zoomed in from ( A ) ) and E15 . 5 WT and Yap-cnull SMGs . DAPI was used to mark the nuclei ( blue ) . Scale = 100 μm . ( C ) qPCR analysis of Sox10 expression in E15 . 5 WT and Yap-cnull SMGs . The average of three experiments is shown +S . E . M . [one sample t-test: **p<0 . 001] . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 006 To gain insight into how Yap directs SMG development , we isolated RNA from three wild type and three Yap-cnull E15 . 5 SMGs ( across three litters ) and analyzed global gene expression using microarrays . Differential expression analysis using a stringent cutoff ( FDRq <0 . 01 , and fold change of twofold or greater ) revealed 105 genes that differed between the Yap-cnull and wild-type SMGs , one of which expectedly was Yap ( Figure 5—source data 1 ) . Hierarchical clustering of these genes showed that the expression profiles from replicate samples clearly clustered next to each other , with two major clusters of genes showing either reduced or increased expression in Yap-cnull SMGs ( Figure 5A ) . Consistent with IF microscopy , Krt5 and Krt14 expression in ductal SMG progenitors was significantly reduced in the absence of Yap , which we further validated by quantitative real-time PCR ( qPCR ) ( Figure 5B ) . Several canonical target genes of Yap , such as Ctgf and Cyr61 , were among genes significantly reduced in Yap-cnull SMGs , validating that our data reflects Yap-driven gene expression . 10 . 7554/eLife . 23499 . 007Figure 5 . Global gene expression analysis of Yap-cnull SMGs indicates the aberrant regulation of genes encoding secreted factors and cell fate regulators . ( A ) Cluster analysis of microarray-generated gene expression data from E15 . 5 WT vs . Yap-cnull SMGs showing genes with a > 2-fold-change and FDR q ( filtered ) <0 . 01 . Red depicts increased and blue depicts decreased gene expression . Relevant genes are highlighted in blue including canonical Yap targets ( Ctgf , Cyr61 ) as well as relevant SMG epithelial markers ( Krt14 , Krt5 , and Kit ) . ( B ) qPCR analysis of Yap , Krt14 , and Krt5 expression in E15 . 5 WT and Yap-cnull SMGs . The average of three SMGs from different litters is shown +S . E . M . ( one sample t-test: ***p<0 . 0001 ) . ( C ) DAVID pathway analysis of genes in ( A ) that are reduced and induced in Yap-cnull SMGs . ( D ) GSEA of significantly downregulated genes in Yap-cnull SMGs shows enrichment for negative regulation of cell differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 00710 . 7554/eLife . 23499 . 008Figure 5—source data 1 . Genes differentially expressed in Yap-cnull vs WT ( FDRq <0 . 01; fold change >2 ) . Differential expression analysis was performed on microarray data obtained from wild type and Yap-cnull E15 . 5 SMGs . Shown below are genes with a FDRq < 0 . 01 that exhibit a fold change of twofold or greater when data from Yap-cnull was compared wild type SMGs . Genes showing reduced expression in the Yap-cnull vs . WT ( DOWN ) are shown in the left two columns , and genes showing increased expression in the Yap-cnull vs . WT ( UP ) are shown in the right two columns . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 00810 . 7554/eLife . 23499 . 009Figure 5—source data 2 . Gene Set Enrichment Analysis ( GSEA ) of genes differentially expressed in Yap-null versus WT SMGs . GSEA ( version 2 . 2 . 1 ) was used to identify biological terms , pathways and processes that are coordinately up- or down-regulated within each pairwise comparison . The Entrez Gene identifiers of the human homologs of the genes interrogated by the array were ranked according to the t statistic computed between the Yap-cnull and wild-type groups . Mouse genes with multiple human homologs ( or vice versa ) were removed prior to ranking , so that the ranked list represents only those human genes that match exactly one mouse gene . This ranked list was then used to perform pre-ranked GSEA analyses ( default parameters with random seed 1234 ) using the Entrez Gene versions of the Hallmark , Biocarta , KEGG , Reactome , Gene Ontology ( GO ) , and transcription factor and microRNA motif gene sets obtained from the Molecular Signatures Database ( MSigDB ) , version 5 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 009 Functional annotation clustering of the differentially expressed genes in Yap-cnull SMGs using DAVID Bioinformatics Resources ( Huang et al . , 2009 ) revealed enrichment of several interesting clusters of genes ( Figure 5C ) . Yap-cnull SMGs were enriched in genes encoding secreted factors , suggesting that Yap plays an important role in altering the microenvironment of SMG epithelium . Enriched among the repressed genes were genes linked to Hippo signaling and those encoding factors with EGF-like domains . Additionally , genes encoding factors associated with the control of stem cell pluripotency were repressed in Yap-cnull SMGs . Conversely , genes linked to promoting cell differentiation were induced in Yap-cnull SMGs . Similar enrichment for genes encoding secreted factors or stem cell regulators was obtained when differentially expressed genes were examined by Gene Set Enrichment Analysis ( Figure 5—source data 2 ) ( Subramanian et al . , 2005 ) . This analysis revealed a significant negative correlation between genes repressed in Yap-cnull SMGs and genes involved in the negative regulation of cell differentiation ( i . e . genes normally induced by Yap in a wild type setting positively correlate with genes that prevent cell differentiation ) ( Figure 5D ) . Thus , Yap-mediated transcription has a role in preventing cell differentiation and that genes regulated by Yap likely playing an important role in controlling the specification of multipotent ductal progenitors . We next interrogated genes differentially expressed in Yap-cnull SMGs for growth factors that may potentially be important for altering the microenvironment that directs ductal progenitor specification . Epiregulin ( Ereg ) , an ErbB receptor ligand that has been implicated in cell fate control in other contexts ( Gregorieff et al . , 2015 ) , was significantly repressed in Yap-cnull SMGs , which we confirmed by qPCR ( Figure 6A ) . To gain insight into the relevance of Ereg expression , we examined developing SMGs using RNA in situ analysis , which showed prominent Ereg expression in the epithelium of buds , particularly in the developing ductal progenitor regions , as well as in distinct populations of basal cells of the ductal epithelium ( Figure 6B ) . Ereg expression levels , however , were completely lost in Yap-cnull SMG ( Figure 6B ) . Interestingly , co-analysis of Ereg levels ( RNA in situ ) and Yap localization ( IF ) indicated that cells exhibiting high levels of nuclear Yap also express Ereg , suggesting that nuclear Yap activity promotes Ereg expression , consistent with our microarray analysis ( Figure 6C ) . 10 . 7554/eLife . 23499 . 010Figure 6 . Yap-induced Epiregulin ( Ereg ) expression directs ductal progenitor specification . ( A ) qPCR analysis of Ereg expression in E15 . 5 WT vs . Yap-cnull SMGs . The average of three SMGs from different litters is shown +S . E . M . [one sample t-test: ***p<0 . 0001] . ( B ) In situ hybridization of Ereg mRNA in E13 . 5 WT ( duct and bud ) and Yap-cnull ( bud ) SMGs . ( C ) Combined in situ hybridization for Ereg mRNA ( red ) and IF for Yap ( green ) in E13 . 5 WT SMGs . ( D , E ) E13 . 5 WT and Yap-cnull SMGs were dissected and cultured for 24 hr in the presence or absence of 0 . 5 μg/mL of exogenous Ereg protein and analyzed by phase-contrast and IF for Krt14 ( K14 , magenta ) , Krt5 ( K5 , green ) , Krt19 ( K19 , red ) , TuJ1 ( yellow ) , and F-actin ( white , Phalloidin ) . DAPI was used to mark the nuclei ( blue ) . Scale = 100 μm . ( F ) qPCR analysis of Yap , Krt5 , and Krt14 expression in the conditions of ( D ) and ( E ) . The average of three SMGs from different litters is shown +S . E . M . [one sample t-test: ***p<0 . 0001] . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 010 To test whether Ereg plays roles in SMG patterning , we treated WT SMG explants with purified exogenous Ereg for 24 hr in ex vivo culture conditions . Ereg treatment led to a morphological thickening and enlargement of the ductal domain that was accompanied by a reduction in the bud domain ( Figure 6D ) . IF microscopy analysis showed a striking enrichment in Krt5- and Krt14-positive cells , with these cells composing almost the entire SMG epithelium , including distal epithelial regions ( Figure 6D ) . Comparative staining with Tuj1 showed a more broadly distributed nerve in the Ereg-treated glands , and F-actin cytoskeletal analysis indicated that along with the expansion of ductal progenitors in response to Ereg , the cytoskeletal organization was disrupted ( Figure 6D ) . Interestingly , treatment of Yap-cnull SMG explants with exogenous Ereg partially rescued some of the observed defects . Most notably , we observed the Ereg treatment led to the emergence of Krt5/Krt14-positive cells , with a subset of these cells also expressing Krt19 ( Figure 6E ) , all of which were normally absent in Yap-cnull SMGs . Analysis of the F-actin cytoskeleton of Ereg-treated Yap-cnull SMGs suggested a partial rescue in the organization of duct and bud domains , as some regions exhibited ductal-like F-actin organization and buds showed a slight clefting morphology ( Figure 6E ) . Moreover , slight broadening of nerve innervation was also apparent in the Ereg-treated Yap-cnull SMGs . To quantify the effects on cell fate , we performed qPCR analysis on parallel SMG explant cultures , and found that treatment of WT and Yap-cnull SMGs with exogenous Ereg led to increased Krt5 and Krt14 expression comparable to control glands without treatment ( Figure 6F ) . These data indicated that Ereg stimulation of SMG epithelium was capable of rescuing some of the patterning defects observed in Yap-cnull epithelium , and that this modestly rescued the associated morphogenesis defects . To explore the role ( s ) of endogenous Ereg in SMG epithelial patterning , we transfected E13 . 5 WT SMGs with either control siRNA or siRNA targeting Ereg , and then cultured the developing SMGs ex vivo for 24 hr . Analysis of RNA isolated from Ereg-depleted SMGs by qPCR indicated a striking deficiency of Krt5 and Krt14 expression , accompanied by a small but significant reduction in Yap expression ( Figure 7A ) . Microscopy analysis of Ereg-depleted SMGs showed branching defects along with a dramatic loss of Krt5-and Krt14-positive cells ( Figure 7B ) , as well a reduction in cells marked with Krt19 ( Figure 7B ) . Depletion of Ereg also resulted in diminished nerve innervation and growth ( Figure 7B ) . Collectively , these observations offer evidence suggesting Ereg is required for the maintenance of a ductal progenitor niche . 10 . 7554/eLife . 23499 . 011Figure 7 . Epiregulin knockdown results in the loss of Krt5- and Krt14-positive ductal progenitors accompanied by a disruption of SMG branching . ( A ) qPCR analysis of Ereg , Krt5 , Krt14 , and Yap expression in E13 . 5 WT SMGs treated with control siRNA or siRNA targeting Ereg . The average of three SMGs from different litters is shown +S . E . M . [one sample t-test: *p<0 . 01 , **p<0 . 001 , ***p<0 . 0001] . ( B ) E13 . 5 WT SMGs were dissected and cultured for 24 hr in the presence of control siRNA or Epiregulin siRNA and analyzed by phase-contrast and IF for Krt14 ( K14 , magenta ) , Krt5 ( K5 , green ) , Krt19 ( K19 , red ) , TuJ1 ( yellow ) , and F-actin ( white , Phalloidin ) . DAPI was used to mark the nuclei ( blue ) . Scale = 100 μm . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 011 The Hippo pathway is the primary signaling pathway that controls Yap localization and activity in development and has been implicated in the control of branching morphogenesis ( Reginensi et al . , 2016 ) . The Lats1 and Lats2 kinases ( Lats1/2 ) are known to phosphorylate Yap , which restricts nuclear Yap localization and activity ( Meng et al . , 2016 ) . Given that we observed Yap phosphorylation that correlated with its cytoplasmic localization in maturing SMG ductal epithelium , we next set out to characterize the potential relationship with the Lats1/2 kinases . We started by examining Lats1/2 activity and localization in E15 . 5 SMGs by IF microscopy using an antibody that recognizes the phosphorylated-active forms of Lats1/2 ( p-Lats1/2 ) ( Szymaniak et al . , 2015 ) . We found that active Lats1/2 was absent in developing bud epithelium and abundant in mature luminal ductal SMG cells ( Figure 8A ) . P-Lats1/2 was localized to the apical domain of the majority of luminal ductal cells , correlating with cells that exhibit cytoplasmic Yap localization . We have shown previously that inhibition of Lats2 with siRNA in E13 . 5 SMG explant cultures gives rise to branching defects ( Enger et al . , 2013 ) , suggesting that p-Lats1/2 in the ductal epithelium plays an important role . To define how Lats1/2 activity impacts SMG development , we used the Shh-Cre recombinase to conditionally delete the Lats1 and the Lats2 genes in the epithelium of developing mouse SMGs ( herein referred to as Lats1/2-cnull ) ( Heallen et al . , 2011 ) . Deletion of either Lats1 or Lats2 alone did not show any observable morphological defects in SMG development ( data not shown ) . However , deletion of both Lats1 and Lats2 led to severe morphogenesis defects , with an almost complete lack of branching at early developmental time points ( Figure 8B–D ) . E13 . 5 Lats1/2-cnull SMGs displayed an enormous augmentation of the ductal domain at the expense of the distal bud , which continued to expand to a very large size in ex vivo cultures . Similarly , E15 . 5 Lats1/2-cnull SMGs exhibited a large ductal structure with a severely enlarged major/primary ductal structure ( Figure 8D ) . IF microscopy analysis showed that almost all the epithelial cells within Lats1/2-cnull SMGs exhibited strong nuclear Yap localization and that pS112-Yap levels were completely absent ( Figure 8E ) . To investigate the effect of Lats1/2 deletion on the entire ductal tree , we analyzed E18 . 5 WT and Lats1/2-cnull SMGs for Krt5 , Krt14 , and Yap by IF microscopy . While WT ducts only express these markers in basal cells , the major ducts in the Lats1/2-cnull SMGs were hyperplastic and did not form an observable lumen , with all cells staining positive for Krt5 and Krt14 ( Figure 8F ) . Minor ducts in the Lats1/2-cnull SMGs were markedly larger and did not exhibit stereotypical branching and budding features as compared to the WT ( Figure 8G ) . The minor ducts in Lats1/2 SMGs were composed predominantly of Krt5/Krt14-positive cells , while equivalent WT regions showed positive staining in few basal-positioned cells ( Figure 8G ) . Gene expression analysis of RNA isolated from Lats1/2-cnull SMGs confirmed these large increases in Krt5 and Krt14 expression ( Figure 8H ) . 10 . 7554/eLife . 23499 . 012Figure 8 . Deletion of Lats1/2 in developing SMG epithelium leads to aberrant nuclear Yap localization and severe branching morphogenesis and patterning defects . ( A ) IF microscopy analysis of E15 . 5 SMGs for phospho-Lats1/2 ( green ) and E-cadherin ( red ) , indicating active Lats1/2 in luminal cells of the ductal epithelium . Scale = 10 µm . ( B ) Phase-contrast images of E13 . 5 WT and Shh-Cre-Lats1/2 null ( Lats1/2-cnull ) SMGs . Scale = 100 μm ( C ) Quantitation of bud number from E13 . 5 WT and Lats1/2-cnull SMGs . n = 21 . ( D ) Phase-contrast images of E15 . 5 WT and Lats1/2-cnull SMGs highlighting the severe ductal expansion phenotype . Note that each respective image is stitched together from two images . ( E ) Images from IF microscopy analysis of E15 . 5 WT and Lats1/2-cnull SMGs for Yap ( green ) , phospho-S112 Yap ( red ) , and DAPI ( blue ) . Scale = 10 µm . ( F ) Images from IF microscopy analysis of E15 . 5 WT and Lats1/2-cnull primary/major ( 1° ) ducts for Krt14 ( K14 , green ) and Krt5 ( K5 , red ) . Scale = 10 µm . ( G ) Images from IF microscopy analysis of E15 . 5 WT and Lats1/2-cnull minor ( 2° ) ducts for Krt14 ( K14 , green ) and Krt5 ( K5 , red ) . Scale = 10 µm . ( H ) qPCR analysis of Lats1 , Lats2 , Krt5 , and Krt14 expression in E15 . 5 WT vs . Lats1/2-cnull SMGs . The average of three experiments is shown +S . E . M . [one sample t-test: **p<0 . 001; ***p<0 . 0001] . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 012 Our observations suggested that inappropriate nuclear Yap activation resulting from Lats1/2 deletion promotes the expansion of Krt5/Krt14-positive ductal SMG progenitors . We therefore examined whether Ereg levels were altered in Lats1/2-cnull SMGs and observed very high expression of Ereg compared to WT SMGs , as measured by qPCR ( Figure 9A ) and RNA in situ hybridization ( Figure 9B ) . To examine whether the expansion of cells expressing ductal epithelial progenitor markers within Lats1/2-cnull SMGs relied on an Ereg-mediated mechanism , we tested the effects of EGFR inhibition in wild type and Lats1/2-cnull developing SMGs ex vivo . Wild type SMGs treated with the EGFR inhibitor AG1478 exhibited branching morphogenesis defects , which were consistent with observations from EGFR-deleted mice ( Häärä et al . , 2009; Jaskoll and Melnick , 1999 ) . IF analysis of the EGFR-inhibited SMGs showed diminished patterning of Krt5 and Krt14 cells , reduced levels of mature Krt19 cells , as well as defective organization of the F-actin cytoskeleton and parasympathetic innervation ( Figure 9C ) , all of which are phenotypes that resembled Yap-cnull SMGs . Lats1/2-cnull SMGs grown ex vivo for 24 hr showed an expansion of ductal structures composed of Krt5 and Krt14-positive cells , with a remarkable enlargement of the major ductal region ( Figure 9D ) . Lats1/2-cnull glands also exhibited defective nerve innervation and F-actin cytoskeletal organization , indicating severely compromised patterning and extracellular signaling ( Figure 9D ) . Notably , some cells within Lats1/2-cnull SMGs also expressed Krt19 , many of which were also positive for Krt5 and Krt14 , suggesting that these cells may be in an aberrant primitive state normally not observed in wild type SMGs ( Figure 9D ) . Treatment of Lats1/2-cnull SMGs with the EGFR inhibitor AG1478 led to a loss of Krt5 and Krt14-positive cells , indicating that EGFR activation drives the expansion of these ductal progenitor-like cells ( Figure 9D ) . Further , gene expression analysis by qPCR validated the observed differences in Krt5 and Krt14 expression in our IF microscopy experiments , with AG1478 treatment of either WT and Lats1/2-cnull SMGs leading to an almost complete loss of the expression of these ductal progenitor markers ( Figure 9E ) . 10 . 7554/eLife . 23499 . 013Figure 9 . EGFR inhibition blunts the patterning defects observed in Lats1/2-cnull SMG epithelium . ( A ) qPCR validation of Ereg expression levels in E15 . 5 WT vs . Lats1/2-cnull SMGs . The average of three experiments is shown +S . E . M . [one sample t-test: ***p<0 . 0001] . ( B ) In situ hybridization of Ereg mRNA in E13 . 5 WT and Lats1/2-cnull SMGs . ( C , D ) E13 . 5 WT and Lats1/2-cnull SMGs were dissected and cultured for 24 hr in the presence or absence of 10 μM EGFR inhibitor AG-1478 and analyzed by phase-contrast and IF for Krt14 ( K14 , magenta ) , Krt5 ( K5 , green ) , Krt19 ( K19 , red ) , TuJ1 ( yellow ) , and F-actin ( white , Phalloidin ) . DAPI was used to mark the nuclei ( blue ) . Scale = 100 μm . ( E ) qPCR analysis of Lats1 , Lats2 , Krt14 , and Krt5 expression in the conditions of ( C ) and ( D ) . The average of three experiments is shown +S . E . M . [one sample t-test: **p<0 . 001; ***p<0 . 0001] . All images represent observations made from a minimum of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 013 Collectively , our observations suggest that nuclear Yap plays an essential role in the development of ductal epithelial SMG progenitors ( see model in Figure 10 ) and that Lats1/2-mediated removal of Yap from the nucleus is required for the maturation of ductal structures . 10 . 7554/eLife . 23499 . 014Figure 10 . Illustration depicting the roles of Yap in submandibular gland epithelial development . We propose nuclear Yap specifies the identity of ductal progenitors , in part by promoting the expression of Ereg and subsequent activation of EGFR signaling . Activated Lats1/2 in the maturing ductal structures promotes the phosphorylation of Yap , which directs the removal of nuclear Yap to support ductal epithelial differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 014
We present data describing an essential role for the Hippo pathway effector Yap in the development of the salivary gland . We show that the deletion of Yap in developing SMG epithelium leads to severe patterning and morphogenesis defects . Our data suggest that these defects arise , in large part , from the loss of nuclear Yap transcriptional activity , which defines signals important for instructing the specification of ductal epithelial progenitors . This conclusion is based on our observations that Yap-cnull SMG epithelium fails to develop ductal domains and shows severely reduced numbers of Krt5 and Krt14-positive cells , which are markers that have been associated with ductal progenitors ( Knox et al . , 2010; Lombaert et al . , 2013 ) . We show that developing SMG epithelium displays distinct localization patterns for Yap in various regions , with nuclear Yap most pronounced in the cells of the developing bud , cells developing into ductal areas , as well as in subpopulations of basal epithelial cells that line the SMG ducts , which interestingly are all regions that exhibit cytoskeletal tension ( Harunaga et al . , 2011 ) . In contrast , cytoplasmic Yap is observed in regions associated with luminal differentiation of the ducts , suggesting that the removal of nuclear Yap is required for SMG epithelial differentiation . Consistent with this premise , deletion of the Hippo pathway Lats1/2 kinases in developing SMG epithelium resulted in aberrant nuclear Yap localization and expansion of Krt5 and Krt14-positive cells in regions that normally undergo ductal maturation . Hippo pathway-mediated restriction of nuclear Yap activity has classically been associated with the governance of organ size via a growth-restricting mechanism that relies on precise coordination between proliferation and apoptosis ( Pan , 2010 ) . As such , Yap and upstream Hippo pathway effectors have emerged as influential oncogenes and tumor-suppressors , respectively . As evidenced here , dynamic Hippo signaling during epithelial morphogenesis extends beyond the view that Hippo signaling controls organ size only by directing cell growth . Rather , our observations from Yap- and Lats1/2-cnull SMGs indicate that Hippo pathway activity is primarily involved in directing a balance in the specification , renewal , and differentiation of ductal progenitors . Similar roles for Hippo pathway-mediated Yap signaling have been described in other branching organs , including the lung , kidney and pancreas ( Gao et al . , 2013; George et al . , 2012; Mahoney et al . , 2014; Reginensi et al . , 2013 ) . Interestingly , however , although Yap plays key roles in progenitor control across different organs , the mechanism ( s ) by which Yap exerts these functions appears to be context-dependent , likely relating to the distinct signaling network that is organized in each respective organ . The identification of a role for Ereg in SMG patterning and morphogenesis resembles its role in other systems . In the intestine , a Yap-Ereg network that involves neighboring stromal cells is central for promoting epithelial cell survival and inducing a regenerative program ( Gregorieff et al . , 2015 ) . However , unlike the intestine , our data suggest that in the SMG , Yap promotes the expression of Ereg in the epithelium to sustain a progenitor niche . Accordingly , we observed elevated Ereg expression in the newly forming ductal progenitor regions and in distinct regions of basal ductal epithelium , precisely the sites with the highest levels of nuclear Yap . We further found that depletion of Ereg or small molecule-mediated inhibition of EGFR , a major receptor for Ereg , results in epithelial branching and maturation defects that resemble those in Yap-cnull SMGs . These observations paralleled those observed following the deletion of EGFR ( Häärä et al . , 2009; Jaskoll and Melnick , 1999 ) and are consistent with recent observations that heparin binding-EGF treatment of SMGs can promote the expansion of Krt5 progenitors ( Knox et al . , 2010 ) . Importantly , we show that ex vivo treatment of developing SMG organ cultures with Ereg expands Krt5 and Krt14-positive cells even in the absence of Yap , indicating that Ereg signaling is sufficient to overcome cell fate defects associated with Yap deletion . Together , these data support a model ( illustrated in Figure 10 ) in which nuclear Yap-induced Ereg defines a niche of EGFR signaling that is central in promoting the identity of ductal progenitors . Signals transduced by the acetylcholine ( Ach ) /muscarinic ( M ) receptor 1 in the parasympathetic submandibular ganglion have been shown to increase EGFR protein expression and increase the expansion of Krt5 progenitors in the SMG ( Knox et al . , 2010 ) . Therefore , it is likely that signals emanating from the nerve may crosstalk with Yap , or Yap-regulated signals may communicate with the nerve to control ductal progenitor specification and maturation . Consistent with this idea , we observed diminished and disorganized parasympathetic innervation in both the Yap-cnull and Lats1/2-cnull SMGs . Recent work has shown that Neuregulin1 ( Nrg1 ) , a neuregulin family ligand that can activate EGFR , promotes the expression of Wnt ligands that signal to direct parasympathetic innervation ( Knosp et al . , 2015 ) . The Hippo signaling pathway is highly interconnected with the Wnt pathway ( Azzolin et al . , 2014; Varelas et al . , 2010a; Heallen et al . , 2011 ) . Thus , it is tempting to speculate that Yap-nerve signaling crosstalk plays a key role in ductal progenitor patterning and may be linked to signals that direct epithelial branching . Our observations implicate the dynamics of Yap localization in the control of ductal epithelial maturation . Several studies have shown that the developing ductal epithelium acquires specialized polarity cues as it matures . Given that polarity cues are tightly integrated with the control of Yap localization and activity ( Szymaniak et al . , 2015; Varelas et al . , 2010b ) , it is likely that epithelial polarity-mediated regulation of Yap directs the differentiation of the ductal epithelium . Indeed , recent work in the developing lung epithelium shows that polarity-regulating proteins , such as the Crumbs transmembrane proteins that direct apical domain specification , promote interactions between the Lats1/2 kinases and Yap ( Szymaniak et al . , 2015 ) . The dynamics of these polarity proteins , which may be controlled by the mechanical microenvironment , direct the localization of Yap and control cell differentiation . For example , basal stem cell cells in the lung epithelium lack aspects of epithelial polarity and exhibit high levels of nuclear Yap that promotes the basal stem cell identity ( Szymaniak et al . , 2015; Zhao et al . , 2014 ) . Notably , we observed a subset of basal epithelial cells in maturing SMG ducts with prominent nuclear Yap localization . These basal cells , a subset of which are marked by Krt5 in adult SMGs , are thought to possess stem cell activity that can repair adult SMG epithelium upon damage ( Knox et al . , 2013 ) . Therefore , nuclear Yap may contribute to the identity of these stem cells as it does in other organs , such as the skin and the lung ( Silvis et al . , 2011; Szymaniak et al . , 2015; Zhao et al . , 2014 ) . However , nuclear Yap in the ductal basal cells did not perfectly correlate with Krt5 expression , suggesting that the Krt5-positive population is composed of distinct subpopulations , as proposed previously ( Lombaert and Hoffman , 2010 ) , or that Yap localization is dynamic in these cells and our observations captured only a snapshot . Our in situ analysis of Ereg expression in maturing ducts also suggested that only a sub-population of basal ductal cells express Ereg . Thus , it is possible that a Yap-Ereg signaling niche specifies a unique basal progenitor identity . Such a progenitor niche may be dysregulated in salivary gland carcinomas , as Krt5-positive populations are frequently amplified in these tumors . Many studies have implicated Yap as a pro-tumorigenic factor ( Harvey et al . , 2013 ) , and thus knowledge into the developmental mechanisms , such as that provided by our study , may offer insight into the etiology of salivary gland diseases . One such disease may be Sjogren’s Syndrome , a debilitating autoimmune disease that affects salivary secretion , as aberrant Yap localization has been observed in the salivary gland epithelium of patients ( Enger et al . , 2013 ) . Similarly , Krt5-expressing cell populations have been observed to be expanded in Sjogren’s Syndrome epithelium ( Gervais et al . , 2015 ) . Thus , dysregulated Yap-mediated cell fate control may be linked to this diseased state . In summary , our observations indicate that Yap is required for directing ductal SMG epithelial progenitor patterning , with nuclear Yap inducing the expression of Ereg , which drives signals for the specification of ductal epithelial progenitors; removal of Yap from the nucleus is therefore a requirement for the maturation of ductal structures . Importantly , our work highlights similarities in Yap signaling with other branching organs while also uncovering significant differences . Thus , given the importance of Yap signaling in organ development and disease , understanding this context is an important challenge for the future .
Developmental studies on wild-type mice were performed using C57BL/6 mice . The Yap-loxP/loxP mice ( Reginensi et al . , 2013 ) were generously provided by Dr . Jeff Wrana ( LTRI , Toronto , CN ) in an ICR/S129 mixed background and bred with C57BL/6 Shh-gfpcre ( Jackson Laboratories , #005622 ) mice to generate Yap-cnull animals . The Lats1-loxP/loxP and Lats2-loxP/loxP mice ( Heallen et al . , 2013; Heallen et al . , 2011 ) were generously provided by Dr . Randy Johnson ( MD Anderson , Houston , TX ) and bred to the Shh-gfpcre mice to generate Lats1/2-cnull animals . All animal experiments were done in accordance with protocols approved by the Institutional Animal Care and Use Committee at Boston University . SMG explants were cultured as previously described ( Steinberg et al . , 2005 ) . Briefly , freshly dissected E13 . 5 SMGs were plated on nucleopore filters for 24 hr and were either lysed for RNA or fixed for immunostaining . When indicated , explants were treated with DMSO or PBS as a control , 0 . 5 μg/mL Epiregulin ( Fisher; 1068EP050 ) or 10 μM AG1478 ( Sigma; T4182 ) . For Ereg knockdown , freshly dissected E13 . 5 SMGs were transfected with siRNA targeting murine Ereg ( CTCAAGTGCAGATTACAAA ) or control siRNA ( Qiagen , 1027310 ) delivered by Lipofectamine RNAiMax transfection reagent ( Thermo Scientific , 13778030 ) and cultured for 24 hr before processing . E15 . 5 WT and Yap-cnull SMGs were dissected and the epithelium was separated from the mesenchyme as described ( Rebustini and Hoffman , 2009 ) . RNA was extracted from the epithelium using either the RNeasy kit or the miRNeasy Micro kit ( QIAGEN ) . The amount of isolated RNA was normalized and automated sample amplification and biotin labeling were carried out using the NuGEN Ovation RNA Amplification system V2 , Ovation WB reagent and Encore Biotin module according to manufacturer’s recommendations using an Arrayplex automated liquid handler ( Beckman Coulter ) . Two micrograms of biotin-labeled sscDNA probe were hybridized to a HT_MG-430_PM Affymetrix Array Plate with modified conditions as previously described ( Allaire et al . , 2013 ) . Washing and staining of the hybridized arrays were completed as described in the GeneChip Expression analysis technical manual for HT plate arrays using the Genechip Array Station ( Affymetrix ) with modifications as previously described ( Allaire et al . , 2013 ) . The processed GeneChip plate arrays were scanned using a GeneTitan scanner ( Affymetrix ) . Raw CEL files ( available at Gene Expression Omnibus ( GEO ) Series GSE90480 ) were normalized to produce gene-level expression values using the implementation of the Robust Multiarray Average ( RMA ) in the affy R package ( version 1 . 36 . 1 ) and an Entrez-Gene-specific probeset mapping ( 17 . 0 . 0 ) from the Molecular and Behavioral Neuroscience Institute ( Brainarray ) at the University of Michigan . Differential expression was assessed using the moderated t test implemented in the limma R package ( version 3 . 14 . 4 ) . Correction for multiple hypothesis testing was accomplished using the Benjamini-Hochberg false discovery rate ( FDR ) . Human homologs of mouse genes were identified using HomoloGene ( version 68 ) . All microarray analyses were performed using the R environment for statistical computing ( version 2 . 15 . 1 ) . GSEA ( version 2 . 2 . 1 ) was performed in a pre-ranked manner ( default parameters with random seed 1234 ) using a list of Entrez Gene identifiers of the human homologs of the genes interrogated by the array , ranked according to the t statistic computed between the Yap-cnull and wild-type groups . Mouse genes with multiple human homologs ( or vice versa ) were removed prior to ranking , so that the ranked list represents only those human genes that match exactly one mouse gene . Biocarta , KEGG , Reactome , Gene Ontology ( GO ) , and transcription factor and microRNA motif gene sets ( in human Entrez Gene ID space ) were obtained from the Molecular Signatures Database ( MSigDB ) , version 5 . 0 . For quantitative real-time PCR , RNA was extracted using the RNeasy kit or miRNeasy Micro Kit ( QIAGEN ) and reverse-transcribed using iScript enzymes ( BioRad ) . Reactions were prepared using Fast SYBR Green Master Mix ( Applied Biosystems; 4385612 ) and carried out in a ViiA7 Real-Time PCR System ( Applied Biosystems ) and analyzed using the ddCT method . Primer sequences are listed in the Table 1 . Statistics were performed using Prism 7 ( Graphpad ) . 10 . 7554/eLife . 23499 . 015Table 1 . qPCR primer sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 015TargetDirectionSequenceProduct sizeGAPDHForwardTGTTCCTACCCCCAATGTGT137 bpGAPDHReverseGGTCCTCAGTGTAGCCCAAG137 bpYapForwardAATGTGGACCTTGGCACACT106 bpYapReverseACTCCACGTCCAAGATTTCG106 bpLats1ForwardGCGATGTCTAGCCCATTCTC135 bpLats1ReverseGGTTGTCCCACCAACATTTC135 bpLats2ForwardACAGAGACGCAGCTGAAGGT101 bpLats2ReverseCACAGCTTCGTGATGAGGTC101 bpKrt5ForwardGGAGCAGATCAAGACCCTCA145 bpKrt5ReverseCGGATCCAGGTTCTGCTTTA145 bpKrt14ForwardAGCGGCAAGAGTGAGATTTCT106 bpKrt14ReverseCCTCCAGGTTATTCTCCAGGG106 bpSox10ForwardGACCAGTACCCTCACCTCCA83 bpSox10ReverseCGCTTGTCACTTTCGTTCAG83 bpEregForwardTTCTCATCATAACCGCTGGA102 bpEregReverseCCCCTGAGGTCACTCTCTCA102 bp Embryonic SMGs were dissected and fixed in 4% PFA ( Electron Microscopy Sciences; 15710 ) for 15 min to an hour and processed for paraffin embedding . Staining was performed using a standard dewaxing and hydration protocol , followed by a microwave-assisted antigen retrieval step using a low-pH buffer ( Vector Labs; H-3300 ) . Primary antibodies and dilutions used are described in Table 2 . Secondary antibodies used were conjugated to Alexa Fluor-488 , –555 , −568 , –594 , or −647 fluorophores ( Life Technologies ) . Slides were mounted using ProLong Gold Antifade Reagent ( Life Technologies; P36930 ) , and images were captured using a confocal microscope system ( Carl Zeiss; LSM710 ) or an inverted epi-fluorescent microscope ( Carl Zeiss; Axio Observer . D1 ) . 10 . 7554/eLife . 23499 . 016Table 2 . Antibodies used . DOI: http://dx . doi . org/10 . 7554/eLife . 23499 . 016AntigenSpeciesCompanyCat#DilutionLot#YapMouseSanta Cruz1011991/100B2713 and A0512YapRabbitCSTD8H1X1/1001Phospho-YapRabbitCST13008S1/1001 and 2Phospho-LATS1/2RabbitAssay Bio TechA81251/100118125Krt5ChickenBiolegend9059011/300D16CF00791Krt5RabbitBiolegend9055011/300D15LF02531Krt14Mouseabcamab78001/100GR185613-1Krt19RatDSHBTROMA-III-c1/10011/12/2015Sox10GoatSanta Cruz173421/100F0315TuJ1MouseR and D SystemsBAM11951/500HVS0215121Phalloidinn/aAlexa FluorA222871/1000866764Krt8RatDSHBTROMA-1-c1/50012/31/2014Ki-67MouseBD5506091/10067176Ki-67Rabbitabcamab166671/100GR86024-1PCNAMouseCST2586P1/1005Cleaved Caspase 3RabbitCST9661S1/10043 In situ hybridization was performed as previously described ( Mahoney et al . , 2014 ) and adapted for paraffin-embedded tissue slides . The DIG-labeled Ereg probe ( Gregorieff et al . , 2015 ) was generously supplied by Dr . Jeffrey Wrana ( Lunenfeld-Tanenbaum Research Institute , Toronto , Canada ) . For the combined in situ/IF experiments , hybridization was carried out as above , but DIG was detected using an antibody conjugated to HRP ( DIG-POD ) ( Roche ) , which was then amplified with tyramide ( Perkin Elmer ) , and visualized with streptavidin-594 ( Life Technologies ) . Primary incubation with the Yap antibody was carried out when the DIG-POD antibody was applied , and incubation with the secondary antibody was carried out before tyramide amplification . Embryonic SMGs were dissected and fixed in 4% PFA as above or with an ice-cold 1:1 acetone/methanol mixture at −20°C for 15 min , depending on the antibodies to be used . In general , TuJ1/Phalloidin worked in PFA while the keratin antibodies worked in acetone/methanol . SMGs were permeabilized in 0 . 5% Triton-X 100 ( American Bioanalytical; AB02025 ) for 30 min to an hour at 4°C , and blocked in the same mix plus 10% donkey serum ( Fisher; 50-588-37 ) for up to 2 hr at 4°C . Primary and secondary antibody incubations were in the serum mixture , and either at 4°C overnight or 1 hr at room temperature . After both primary and secondary incubations , SMGs were washed extensively in 0 . 5% Triton-X 100 ( at least 3 , up to six changes; 10 min each ) and post-fixed with 4% PFA for 30 min . Counterstaining was performed using Hoechst for 15 min . SMGs were immediately imaged by placing them on a coverslip . For increased resolution , SMGs were compressed between two coverslips before imaging , and imaged using an epi-fluorescent microscope ( Carl Zeiss; Axio Observer . D1 ) . | Our mouths are continually bathed by saliva – a thick , clear liquid that helps us to swallow and digest our food and protects us against infections . Saliva is produced by and released from salivary glands , which are organs that contain a branched network of tubes . Salivary glands can only properly develop if immature cells known as stem cells , which give rise to the mature cells in the organ , are controlled . Despite their importance for development of salivary glands , little has been known about the signals that control these stem cells . Szymaniak et al . have now discovered new regulators of the salivary gland stem cells in mice , including essential roles in the regulation of these cells by a protein known as Yap . The Yap protein is controlled by a set of proteins that together are known as the Hippo pathway . Szymaniak et al . found that when the gene for Yap was deleted in mice very few stem cells were made , and the transport tubes of the salivary tubes failed to develop . Conversely , when the Hippo pathway was disrupted in mice there were too many stem cells because they could not properly develop into the mature cells , leading to incorrect transport tube development . . These results indicate that Yap is essential for controlling the stem cells of the salivary glands , and offer important insight into the signals that control how the salivary glands develop . The next step will be to investigate whether the Hippo pathway or Yap are affected in diseases of the salivary gland , which often show incorrect numbers of stem cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2017 | The Hippo pathway effector YAP is an essential regulator of ductal progenitor patterning in the mouse submandibular gland |
The architecture of corticobasal ganglia pathways allows for many routes to inhibit a planned action: the hyperdirect pathway performs fast action cancellation and the indirect pathway competitively constrains execution signals from the direct pathway . We present a novel model , principled off of basal ganglia circuitry , that differentiates control dynamics of reactive stopping from intrinsic no-go decisions . Using a nested diffusion model , we show how reactive braking depends on the state of an execution process . In contrast , no-go decisions are best captured by a failure of the execution process to reach the decision threshold due to increasing constraints on the drift rate . This model accounts for both behavioral and functional MRI ( fMRI ) responses during inhibitory control tasks better than alternative models . The advantage of this framework is that it allows for incorporating the effects of context in reactive and proactive control into a single unifying parameter , while distinguishing action cancellation from no-go decisions .
When at bat , a baseball player can choose not to swing at the incoming pitch in one of two ways: he can cancel the swing reactively based on external cues , for example , of the estimated position of the ball based on sensory signals , or he can cancel it proactively using internal expectations , for example , a strategy of never swinging at the first pitch ( Aron , 2011; Botvinick et al . , 2001; Braver et al . , 2001 ) . These are generally considered to be separable control signals , with proactive suppression being a slower , action-specific process ( Aron et al . , 2011 ) and reactive inhibition being a fast global suppression mechanism ( Aron and Poldrack , 2006; Aron et al . , 2014 ) . However , it has been argued that a purely reactive form of stopping is unlikely to generalize to many real-world scenarios and , in some cases , response inhibition may result from a combination of proactive and reactive efforts ( Aron , 2011 ) . Indeed , recent evidence suggests that these processes interact , with proactive signals modulating the efficacy of reactive stopping ( Aron , 2011; Chen et al . , 2010; Forstmann et al . , 2008; Irlbacher et al . , 2014; Jahfari et al . , 2012; Stuphorn and Emeric , 2012; van Belle et al . , 2014; van Maanen et al . , 2015; Zandbelt et al . , 2013; Zandbelt and Vink , 2010 ) . Yet the nature of this interaction remains largely unclear . An interaction between proactive and reactive decisions makes sense given the shared circuitry associated with both forms of control ( Figure 1 ) . Proactive signals have been most commonly associated with activity in rostral prefrontal and striatal areas ( Cai et al . , 2011; Majid et al . , 2013; van Belle et al . , 2014 ) , suggesting a possible competition of facilitation and suppression signals relegated through the direct and indirect pathways , respectively ( Wei et al . , 2015 ) . In contrast , reactive braking has been linked to premotor and prefrontal areas in the right hemisphere , as well as the subthalamic nucleus , that together compose the hyperdirect pathway ( Aron et al . , 2014; Cavanagh et al . , 2011b; Rodriguez et al . , 2006; Swann et al . , 2012 ) . While in primates the control signals of the direct , indirect , and hyperdirect pathways all project to the internal segment of the globus pallidus ( GPi ) , the primary output of the basal ganglia , it remains unclear whether these signals converge on the same pool of neurons or run in parallel of each other . At the cellular level , successful reactive inhibition depends on the relative timing of descending execution signals from the striatum and fast-braking signals from the hyperdirect pathway at the output of the basal ganglia ( Schmidt et al . , 2013 ) , suggesting that these two forms of control may , in fact , converge before they continue on to the thalamus . 10 . 7554/eLife . 08723 . 003Figure 1 . Conceptual framework of stopping and deciding not to go as separate but dependent processes . ( A ) The organization of corticobasal ganglia pathways . Execution signals are relayed via the direct pathway ( green connections ) that result in a disinhibition of thalamic signals to cortex form the internal segment of the globus pallidus ( GPi ) . Thalamic output can be suppressed by one of two major pathways . The indirect pathway ( blue connections ) increases inhibition of the thalamus via cortical signals terminating on striatal nuclei and relegated through the external segment of the globus pallidus ( GPe ) . The hyperdirect pathway ( red connections ) quickly increase thalamic suppression via direct projections from the cortex to the subthalamic nucleus ( STN ) . The question mark highlights the uncertainty as to whether the hyperdirect signals terminate on the same GPi neurons as the direct and indirect pathways . ( B ) Parameter structure of the general drift-diffusion process used in all model simulations and fits . For the reactive task ( see Figure 1 ) , we compared three models , a dependent process model ( C ) , a traditional independent race model ( D ) , and interactive race model ( E ) . See main text for details on these models . For both reactive and proactive tasks , we compared three possible influences of context on the execution process: drift rate modulation ( F ) , boundary shift modulation ( G ) , and onset delay modulation ( H ) . See main text for descriptions of each model . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 00310 . 7554/eLife . 08723 . 004Figure 2 . Reactive stopping task and behavioral results . ( A ) Timeline of a go ( left ) and stop trial ( right ) in the reactive experiment . ( B ) Mean observed probability of stopping in the baseline ( dark , solid ) and caution ( light , dotted ) conditions; dots reflect individual subject means . As the stop-signal delay ( SSD ) increased , the probability of correctly stopping decreased . Inset in B shows the mean point of subjective equality ( PSE ) for both conditions . ( C ) Mean response times ( RT ) for the baseline and caution conditions . All error bars reflect the 95% confidence interval , * indicates p <0 . 05 , *** indicates p <0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 004 Here we use knowledge of the circuit-level organization of corticobasal ganglia networks to elucidate the dynamic interactions between proactive and reactive inhibitory control . Using both behavioral measures and functional MRI ( fMRI ) to confirm predictions from a novel computational model , we evaluate two hypotheses: 1 ) reactive stopping depends on the state of the execution signal ( i . e . , the two processes are not independent ) ; 2 ) proactive no-go decisions occur by modulating the execution process so that it fails to reach its decision boundary , rather than by active cancellation of the decision itself .
The architecture of corticobasal ganglia pathways ( Figure 1A ) , along with recent electrophysiological ( Schmidt et al . , 2013 ) and neuroimaging ( Jahfari et al . , 2011; Jahfari et al . , 2012; Jahfari et al . , 2010; Smittenaar et al . , 2013; Zandbelt et al . , 2013 ) evidence , suggests that the efficiency of fast hyperdirect braking depends on the state of the descending execution process that originates in striatal pathways . To capture this , we designed a novel decision model with nested drifting diffusion processes ( Figure 1B , see “Materials and methods” section ) . The motor decision is modeled as a single variable ( hereafter , referred to as the execution process ) that is assumed to reflect the differential activity between the direct ( facilitation ) and indirect ( suppression ) pathways; that is , as facilitatory systems are recruited more than the suppressive systems , the execution process accumulates toward the decision boundary . The cancellation decision ( hereafter , referred to as the braking process ) is modeled as a latent competing signal , assumed to originate from the hyperdirect ( brake ) pathway and whose starting state depends on the state of the facilitation versus suppression competition ( Figure 1C ) . In this dependent process model ( DPM ) , a response is realized only if the execution process reaches its decision boundary before the braking process reaches its decision boundary and before the trial time window has expired . Otherwise the model does not produce a response . Also , since it has been shown that the physiological dynamics of inhibitory control networks are best reflected by models where the race to the decision boundary has dynamic , nonlinear properties ( Ratcliff and Frank , 2012 ) , our model also includes a dynamic gain that dynamically accelerates the execution process as it approaches the time boundary ( i . e . , the deadline for making a response ) . We should note that others have proposed similar nested process models for inhibitory control decisions . Indeed , our DPM is conceptually similar to the interactive race model ( Boucher et al . , 2007 ) , whereby a response is suppressed by the onset of a braking signal that directly inhibits the execution process , preventing it from reaching the action threshold . The interactive race and DPMs make fundamentally different assumptions about the nature of the interaction between the execution and braking processes . The predictions of both the dependent process and interactive race models can be contrasted against the traditional independent race model ( Logan et al . , 1984; Logan et al . , 2014 ) where the outcome of the decision is determined by a parallel race between the execution and braking processes . To evaluate the different reactive stopping models we tested a group of subjects ( N = 60 , see Materials and methods section ) on a modified version of the stop-signal task that allows for precise timing of the target response time ( RT ) ( Coxon et al . , 2012; Lappin and Eriksen , 1966; Zandbelt et al . , 2013 ) . The task parameters were specifically designed to allow for a competition between proactive and reactive decisions , rather than adapted on a trial-by-trial basis in order to isolate only reactive stopping decisions . On each trial , subjects saw a blue bar ‘fill’ upwards toward a target line on the screen at a constant rate ( Figure 2A , see ‘Materials and methods’ section ) . On go trials , the bar would intersect the target line at 500 ms after trial onset . Participants were instructed to stop the bar by pressing a key . The closer the bar was stopped to the target line , the higher the financial reward on that trial . In a subset of trials ( stop trials; 45% trials ) , the bar would stop before it intersected with the target ( i . e . , the stop signal ) and participants were instructed to not make a response on these trials . Successful decisions were rewarded and incorrect decisions were penalized . All participants were run in two versions of the experiment: a baseline condition ( baseline ) and a condition where the feedback signals were more restrictive on accuracy to the target and higher penalties for failing to stop when required while higher rewards were given for successful stops ( caution ) . Similar manipulations on feedback have been shown to modulate stopping performance in the stop-signal task Leotti and Wager , 2010; Shenoy and Yu , 2011 ) . As expected , in both conditions it was harder for participants to withhold their responses when the stop signal was delivered later in the trial ( Figure 2B ) . In addition , changing the structure of the feedback promoted more cautious responding , resulting in more successful stopping at later stop signal delays ( SSDs ) and slower RTs on go trials . Using both the RT distribution and the stop probability curve in the baseline condition , we fit the experimental data to our DPM , the interactive race model , and the traditional independent race model ( see ‘Materials and methods’ section ) . The DPM provided a better fit to both the correct and incorrect RT distributions ( Figure 3A ) than either the interactive or independent race models . In addition , the DPM fell within the 95% confidence interval of the mean probability of stopping in each SSD condition ( Figure 3B ) , whereas the independent and interactive race models overpredicted the probability of stopping at the longest and shortest delays , respectively . Goodness of fit , assessed using both the Akaike information criterion ( AIC ) and Bayesian information criterion ( BIC ) , showed that the DPM was a substantially better fit to the data than the other two models ( Figure 3C; Table 1 ) , suggesting that the race between braking and execution signals may be biased depending on the state of the action selection decision . 10 . 7554/eLife . 08723 . 005Figure 3 . Comparison of reactive stopping models . Fits of the three reactive models ( Figure 1C–E ) to behavioral data in the baseline condition , shown against ( A ) the histogram of RTs for correct ( top ) and incorrect ( i . e . , responses made on stop trials; bottom ) trials and ( B ) the stop probability curve . Overall the dependent process model ( DPM; green ) explains speed and accuracy in the behavioral data ( gray histograms in A , black line in B ) better than the independent race model ( Ind-RM; red ) and interactive race model ( Int-RM; blue ) . ( C ) Bars show the calculated AIC for each model . Error bars on behavioral data in B show the 95% confidence interval of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 00510 . 7554/eLife . 08723 . 006Table 1 . Reactive model parameter estimates and fit statistics . In the top panel , best fit parameter estimates for boundary height ( a ) , onset delay ( tr ) , execution drift rate ( ve ) , braking drift rate ( vb ) , dynamic bias gain ( xb ) are listed for each of the candidates reactive stopping models . Additionally , the stop signal onset delay ( sso ) was estimated for the interactive race model but was not included in the other models . The lower panel contains parameter estimates and fit statistics for the candidate models of contextual modulation between baseline and caution conditions of the reactive task . Parameters that were left free to vary between conditions contain two values , one estimate for the baseline condition and another estimate for the caution condition ( see Model fitting for details regarding acquisition of constant parameters and optimization across conditions ) . In both panels , the last three columns show the χ2 as an absolute index of how well each model fit the data as well as the Akaike information criterion ( AIC ) , and Bayesian information criterion ( BIC ) as complexity penalized goodness-of-fit measures . Lower values in all three measures imply a better fit to the data . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 006Modela tr ve vb sso xb χ2AICBICDPM0 . 5340 . 1741 . 266−0 . 9900 . 8780 . 0028−122 . 40−128 . 018Ind-RM0 . 2500 . 3381 . 1271 . 2691 . 520 . 0075−106 . 652−112 . 270Int-RM0 . 4450 . 2201 . 1953 . 0230 . 1971 . 4740 . 0069−104 . 379−111 . 815Drift0 . 5360 . 178B: 1 . 289C: 1 . 243−0 . 9840 . 8770 . 0051−273 . 459−273 . 301Onset0 . 531B:0 . 171C: 0 . 1801 . 236−0 . 9600 . 8930 . 0054−271 . 651−271 . 492Drift andonset0 . 538B: 0 . 173C: 0 . 178B: 1 . 269C: 1 . 247−0 . 9890 . 8580 . 0063−260 . 793−263 . 941BoundB: 0 . 525C: 0 . 5510 . 1781 . 268-0 . 9840 . 8780 . 0057−269 . 994−269 . 835DPM: dependent process model; Ind-RM: independent race model; Int-RM: interactive race model . Contextual factors , such asreward contingencies ( Leotti and Wager , 2010 ) and cued expectations of stopping probability ( Jahfari et al . , 2012; Smittenaar et al . , 2013; Zandbelt et al . , 2013 ) can influence stopping performance by modulating the speed of the execution decision . We see this same pattern in our caution manipulation , where the magnitude of feedback provides larger rewards for correct stopping and the window for reward on go trials is narrowed ( see “Materials and methods” section ) . Overall , in the caution condition subjects shifted their stop curve later in time , producing more accurate stopping performance ( Figure 2B; t ( 59 ) = −2 . 18 , p=0 . 03 ) by delaying their execution decision , resulting in longer go trial RTs ( Figure 2C t ( 59 ) = −0 . 53 , p <0 . 001 ) . Thus , our feedback manipulation was effective at inducing more conservative decisions . This contextual effect on the execution process can manifest in several ways . Context can modify the rate of drift ( Drift Modulation; Figure 1F; Hanks et al . , 2014; Standage et al . , 2011; Zhang and Rowe , 2014 ) , shift the onset time at which the execution process begins to accumulate ( Onset Modulation; Figure 1G; Pouget et al . , 2011 ) , change the distance to the threshold ( Boundary Modulation; Figure 1H; Boehm et al . , 2014; Cavanagh et al . , 2011a; Forstmann et al . , 2012; Jahfari et al . , 2012; Wiecki and Frank , 2012 ) , or modulate a combination of parameters ( Heitz and Schall , 2012 ) , such as both drift and onset modulation together . As mentioned above , if direct and indirect pathways converge at the GPi , this would predict that the execution process is driven by an active competition between facilitation and suppression signals , which is naturally implemented in diffusion models with the drift rate parameter which captures a moment-by-moment competition between a hypothesis and its alternative . To evaluate these possibilities , we fit four versions of our DPM in which either the drift , onset , onset and drift , or boundary height parameters were left free when fitting the baseline and caution conditions separately . Qualitatively all models appeared to fit the stop probability curves quite well ( Figure 4—figure supplement 1 ) . Where the different models varied was in their ability to capture subtle variation in the RT distributions . As a result , the drift modulation model provided the best overall fit of speed and accuracy in the reactive task , followed by the onset modulation and boundary models , respectively ( Figure 4; Table 1 ) . Although the drift modulation model best explained contextual effects on reactive stopping performance , the fact that the onset and boundary modulation models also had very good fits suggests that this manipulation may not cleanly be able to distinguish between the different models . We return to this issue in the next section . 10 . 7554/eLife . 08723 . 007Figure 4 . Comparison of modulation models in reactive task . Goodness-of-fit measures for execution modulation models in reactive task . Bars show the estimated Akaike information criterion ( AIC ) and Bayesian information criterion ( BIC ) for the drift , onset , combined drift and onset , and boundary modulation models . The model with the lowest score , in this case the drift modulation model , is preferred . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 00710 . 7554/eLife . 08723 . 008Figure 4—figure supplement 1 . Model predictions of behavior in reactive task . Example fits for modulation models on the execution process inthe baseline and caution conditions of the reactive task . Predicted RT distributions and stop accuracy are shown for the ( A ) drift-rate modulation model ( green ) , ( B ) onset-modulation model ( blue ) , ( C ) combined drift and onset modulation model ( yellow ) , and ( D ) boundary modulation model ( red ) . In each panel , RT distributions in the baseline ( top ) and caution ( bottom ) conditions are shown for correct ( left ) and error ( right ) trials . Empirical ( gray , histograms ) and model-predicted ( color , probability density ) RT distributions both reflect 1000 samples from a Gaussian kernel density estimate ( bandwidth of . 01 ) based on the empirical and model simulated RT quantiles . The stop accuracy curve is plotted with model predictions ( colored x’s ) overlaid on empirical values ( gray line , circles ) for each of the five SSD conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 008 The influence of contextual factors on decision dynamics , like demands for control as shown above , as well as evidence that execution uncertainty slows reaction times ( Jahfari et al . , 2012; Zandbelt et al . , 2013; Zandbelt and Vink , 2010 ) , both point to a novel mechanism of no-go decisions that distinguishes them from reactive stops . If the execution signal is determined by a dynamic competition between direct ( facilitation ) and indirect ( suppression ) signals ( Figure 1A; see also [Wei et al . , 2015] ) , then making a decision to not execute an action may reflect a scenario where the indirect pathways provide enough active suppression of the direct pathway so that no response is triggered . In other words , a proactive no-go decision occurs when the system fails to accrue enough evidence to reach the execution decision boundary . To evaluate this hypothesis we tested the same group of subjects in a novel proactive inhibition task ( see also Zandbelt et al . , 2013 ) . The overall design was largely similar to the reactive paradigm with the following major changes . First , only one SSD was used . This occurred at 450 ms and is too late for it to be used as a reactive stopping cue . Second , on each trial the color of the rising bar indicated the probability that the stop signal would be delivered ( Figure 5A ) : a purely red bar indicates 0% go probability ( i . e . , stop signal would be presented on every trial ) , purely blue bar indicates 100% go probability ( i . e . , stop signal would never be presented ) , and shades of purple indicate mixture probabilities . Rather than rely on the very late reactive stop signal , subjects were told to use the proactive cue to decide whether to execute the response or withhold it . Since subjects were asked to make an informed guess , all our analysis collapsed across both correct and incorrect trials . 10 . 7554/eLife . 08723 . 009Figure 5 . Proactive no-go decision task and behavioral results . ( A ) Timeline of correct high ( left ) and low ( right ) go probability trials in the proactive task . The low probability example shows a trial in which a stop signal was presented . ( B ) Mean probability of making a no-go decision; dots reflect individual subject means and ( C ) mean reaction time plotted as a function of go trial probability in the baseline ( dark , solid ) and caution ( light , dotted ) conditions . Inset in B shows the mean point of subjective equality ( PSE ) for both conditions . All error bars reflect the 95% confidence interval; n . s . indicates a nonsignificant comparison at α of 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 00910 . 7554/eLife . 08723 . 010Figure 5—figure supplement 1 . Subject-wise correlation of reactive and proactive control . Correlation between the average PSEs in the reactive stopping curves and proactive no-go curves collapsed across baseline and caution conditions . Each point represents a single subject . PSE: point of subjective equality . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 010 Participants were able to use the colored cue to make accurate go and no-go decisions ( Figure 5B ) . There was a slight bias toward making a go decision across trial types; however , the probability of making a no-go decision scaled monotonically with the cued probability of a go trial . Also , as expected , participants were faster with higher certainty of a go trial ( Figure 5C ) . It is worth noting that all RTs in this proactive task were much faster than the responses in the reactive task ( Figure 2 ) , confirming that subjects weren’t relying on the delayed stop signal to make a reactive decision . Thus , this second experiment was able to get subjects to make a fundamentally different decision than in the reactive task , while keeping most other experimental conditions ( e . g . , sensory signals ) constant . Unlike the reactive task , we did not observe a significant influence of demands for control on either the decision curve ( F[3 . 43 , 205 . 8] = 0 . 76 , p=0 . 53 ) or response times ( F[2 . 28 , 132 . 32] = 0 . 81 , p=0 . 46 ) in the proactive task . If our hypothesis is correct that reactive stopping ability depends on the current state of proactive control , then there should be a mild degree of shared variance between an individual’s reactive stopping ability and their proactive decisions . To test this we correlated the points of subjective equality ( PSE ) across subjects , between the reactive ( Figure 2B ) and proactive ( Figure 5B ) conditions . The PSE reflects the point at which the psychometric curve crosses 50% chance ( see “Materials and methods” section ) . Collapsing across both the baseline and caution conditions , we found a small but significant correlation in a subject’s reactive control abilities and their proactive decisions ( Figure 5—figure supplement 1 , r = 0 . 35 , p <0 . 006 , r2 = 0 . 06 ) . Thus , overall reactive task performance explains ∼6% of the variance in the proactive condition , which is consistent with a model where these two abilities reflect separable upstream control signals that eventually converge on common output pathways ( Figure 1A ) . One advantage of our proactive task is that it can better distinguish the mechanisms of modulation on the execution process . Therefore , we fit the same four models that were fit against the reactive task ( Figure 1H ) , leaving the specific model parameters free across each go trial probability condition ( see “Materials and methods” section ) . Since there were no differences between the baseline and caution conditions , all data was collapsed together to increase statistical power of the model fits . We found that the drift modulation model was able to capture the proactive no-go decision dynamics significantly better than the other models based on the AIC and BIC scores ( Figure 6; Table 2 ) and general patterns of speed and response proportions ( Figure 6—figure supplement 1 ) . This means that as the go trial probability decreased , the drift rate of the execution process decreased to the point that the decision boundary was often not intersected by the end of the trial . Thus , we can reliably capture the dynamics of a no-go decision as a modulation of competing facilitatory and suppressive signals that results in a failure to accumulate enough evidence to make a decision . This provides a fundamental distinction between the control mechanisms of stopping and deciding not to go . 10 . 7554/eLife . 08723 . 011Figure 6 . Comparison of modulation models in proactive task . Goodness-of-fit measures for execution process modulation models in proactive task . Same plotting conventions as in Figure 4 . As with the reactive experiment , the drift modulation model provided better fits to the proactive task data than the other models . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 01110 . 7554/eLife . 08723 . 012Figure 6—figure supplement 1 . Model predictions of behavior in proactive task . Predicted RT distributions and no-go probability curves in the proactive task for the drift modulation ( green , A ) , onset modulation ( blue , B ) , onset and drift modulation ( yellow , C ) , and boundary modulation models ( red , D ) . Empirical ( gray ) and kernel density estimates ( same conventions as in Figure 4 ) for the RT distributions are shown for the high and low go trial probability conditions in the left and middle columns , respectively . Right column shows no-go probability curves for each go trial probability condition . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 01210 . 7554/eLife . 08723 . 013Table 2 . Proactive model parameter estimates and fit statistics . Best fit parameter estimates for boundary height ( a ) , onset delay ( tr ) , execution drift rate ( ve ) , and dynamic bias gain ( xb ) are listed for each of the proactive modulation models . For each model , free parameter ( s ) are named in the shaded column , followed by fitted estimates in each go trial probability condition in columns P0–P100 , ( see Model fitting for details regarding acquisition of constant parameters and optimization across conditions ) . The last three columns show the χ2 as an absolute index of how well each model fit the data as well as the Akaike Information Criterion ( AIC ) , and Bayesian Information Criterion ( BIC ) as complexity penalized goodness-of-fit measures . Lower values in all three measures imply a better fit to the data . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 013Modela tr ve xb P0P20P40P60P80P100cXχ2AICBICDrift0 . 4870 . 2921 . 563ve 1 . 4111 . 5621 . 6831 . 7611 . 8801 . 9250 . 0022−122 . 65−130 . 08Onset0 . 6281 . 420 . 641tr 0 . 1820 . 1610 . 1340 . 1170 . 0840 . 0760 . 0095−99 . 38−106 . 81Drift and onset0 . 061 . 468ve 0 . 8310 . 9700 . 9680 . 9790 . 9321 . 0790 . 0033−104 . 65−122 . 77tr 0 . 5150 . 5060 . 4920 . 4790 . 4510 . 463Bound0 . 2720 . 9140 . 913a 0 . 3790 . 3440 . 3050 . 2810 . 2460 . 2360 . 0099−98 . 65−106 . 09 Because the assumptions of our DPM it makes very specific predictions about the physiological dynamics of go versus no-go decisions within corticobasal ganglia networks , specifically in regions where these pathways are known to interact , that is , striatum and thalamus . Previous fMRI studies have reliably shown that the magnitude of activation in cortical decision-making areas approximates the cumulative sum of evidence over time as the decision process approaches its threshold ( Basten et al . , 2010; Ho et al . , 2009 ) . While the temporal and spatial imprecision of fMRI precludes direct comparison of neural and computational mechanisms ( Simen , 2012 ) , the overall magnitude of the signal can be taken as a proxy for the duration and distance-to-bound of the decision process ( Forstmann et al . , 2010 ) . By integrating the execution process over time we compared model-predicted activation levels for both go and no-go decisions ( see “Materials and methods” section ) with the task-evoked blood oxygenation level dependent ( BOLD ) response measured by fMRI in corresponding conditions of our proactive task . Simulations were run for the drift , onset , and boundary modulation models of proactive control ( see Figure 7A–C ) . 10 . 7554/eLife . 08723 . 014Figure 7 . Simulated interaction between trial outcome and response expectation on BOLD activation . Time-course of BOLD and mean activity in the proactive task predicted by the ( A ) drift modulation , ( B ) onset modulation , and ( C ) boundary modulation models . Time courses ( left column ) reflect the rise of the execution process ( θe ) on ‘go’ trials ( left panels; green lines ) and ‘no-go’ trials ( right panels; red lines ) . Lighter colors reflect lower go probability conditions ( go , <50%; no-go , 0% ) and darker colors represent higher go probability conditions ( go , 100%; no-go , >50% ) . The arrows in A and B show the pressure on the execution process with decreasing go trial certainty . The arrow and bars in C illustrate the shift in boundary height between high ( bottom bar ) and low ( top bar ) go trial probability conditions . Mean simulated BOLD responses ( right column ) are calculated as the cumulative sum of execution process across the full time course . This summation is reflected in the filled area under the curve in the traces in the left column . BOLD: blood oxygenation level dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 014 Based on our simulations , the drift modulation model predicts a greater BOLD response when an action is executed ( i . e . , 'go' ) versus trials where the action is withheld ( i . e . , 'no-go' ) and , more importantly , that the BOLD response should increase with the go trial probability during 'no-go' decisions , but taper off when a response is executed . The main effect of trial outcome ( i . e . , 'go' greater than 'no-go' ) is due to the fact that , on average , withheld responses result from lower levels of evidence accumulation than executed responses ( i . e . , trials in which the accumulated evidence reaches the threshold ) . Within ‘no-go’ decision trials , the increase in drift-rate with higher go-trial certainty leads to higher levels of sub-threshold evidence accumulation at the end of the trial . On the other hand , the increase in drift-rate with go-trial certainty leads to a faster rise-to-threshold on 'go' decision trials and thus produces slightly less summed activity than executed responses in lower go-trial probability conditions . In contrast with the drift modulation model , the onset modulation model predicts that in addition to a main effect of trial outcome , increasing the probability of a go-trial should increase the magnitude of the BOLD response for both 'no-go' and 'go' decision trials . This is due to the fact that there is overall more time for the process to accumulate with earlier onsets , that is , higher go-trial probabilities . Thus , in high go-trial probability conditions both ‘go’ and ‘no-go’ decisions arise from longer cumulative processing time than in low go-trial probability conditions , and therefore predict greater activation . This model also predicts the effect of go-trial probability should be roughly equivalent for withheld and executed responses , which is a contrasting prediction from the drift modulation model which predicts a greater modulation wtih go-trial probability on 'no-go' decision trials . Finally , the boundary modulation model predicts that increasing go-trial probability will decrease the BOLD response , along with a general decrease in BOLD signal during 'go' decisions . This is because conditions with lower boundary heights ( i . e . , greater certainty of a go-trial ) , allow less time for the process to be active and a shorter distance to travel and thus predict lower levels of summed activity . To evaluate these different model predictions , we measured task-related hemodynamic responses using the fMRI BOLD contrast . Healthy subjects ( N=20 ) performed a modified version of the proactive task for the MRI environment ( see “Materials and methods” section ) . In order to assure that no covert actions were elicited during trials in which a key press was not recorded , we used a modified EMG over the first dorsal interosseous ( FDI ) muscle of the responding hand to record muscle activity ( Figure 8—figure supplement 1A ) so as to identify finger movements on all trials . Viable EMG data was available for 14 subjects and in these subjects suprathreshold muscle activity was only observed in <1% ( mean = 0 . 71% , max 4% ) of the trials where no key press was detected . The number of missed key presses was equally distributed across go trial probability conditions . For the initial analysis these error responses were corrected in these subjects using the EMG data to assure that no-go trials were uncontaminated by covert actions in these subjects . For the purpose of testing the predictions of the drift , boundary and onset modulation models , we used a general linear model ( GLM ) to fit task-related responses during ‘go’ and ‘no-go’ trials separately and a parametric regressor for each condition to assess the effect of go trial probability on the evoked BOLD response ( see “Materials and methods” section ) . Using a random effects analysis on all subjects , we isolated regions whose activity was modulated more during no-go trials than during trials where a response was executed , as predicted by the drift modulation model . After adjusting for cluster size ( k > 40 ) and multiple comparisons ( q < 0 . 05 ) we identified 24 clusters with differential modulation during no-go responses than go responses ( Supplementary file 1 ) . We were particularly interested in a cluster in the right caudate nucleus and a pair of bilateral clusters in the ventromedial thalamus that are consistent with the location of the thalamic subnucleus that is reciprocally connected to motor cortex ( Nambu , 2011 ) . The nature of the contrast used in the whole-brain analysis does not necessarily confirm that the pattern of responses in the right caudate and bilateral thalamus are consistent with the drift modulation model , as the effect of go trial probability could cause a greater decrease in activation during no-go trials ( e . g . , the opposite direction of the drift modulation predictions ) . Therefore , we extracted out the condition-specific responses in each region of interest ( ROI ) , for all subjects , using a fivefold cross-validation with anatomical priors so as to avoid performing circular inference . Consistent with our predictions , activity in the right caudate nucleus and the left and right thalamic ROIs show a main effect of decision type on evoked activity ( Figure 8; caudate: F ( 1 , 32 ) = 2 . 69 , p=0 . 03; left thalamus: F ( 1 , 32 ) = 4 . 01 , p=0 . 003; right thalamus: F ( 1 , 32 ) = 4 . 50 , p=0 . 001 ) . In the thalamus ROI responses selectively scaled with go trial probability , even in the trials where no key press was detected ( Figure 8; left: F ( 2 , 32 ) = 4 . 13 , p=0 . 025; right: F ( 3 , 32 ) = 3 . 67 , p=0 . 037 ) , while this effect was not significant in the caudate ROI ( F2 , 32] = 1 . 77 , p=0 . 19 ) . More importantly the interaction between go trial probability and decision type was significant for both thalamus ROIs ( left: F[2 , 32] = 4 . 30 , p=0 . 022; right: F[2 , 32] = 4 . 13 , p=0 . 025 ) , but not the caudate ( F[2 , 32] = 1 . 22 , p=0 . 31 ) . It should be noted that these results do not significantly change when only the subjects with reliable EMG signals are included in the analysis , confirming that spurious movements do not drive this pattern . 10 . 7554/eLife . 08723 . 015Figure 8 . Observed interaction between trial outcome and response expectation on BOLD activation . Contrast maps for the comparison of no-go responses , modulated by go trial probability , against modulated go responses in the proactive task ( center panel ) . Warm colors show areas where the modulation was more positive during no-go trials than go trials . This is the direction predicted by the drift modulation model ( Figure 7A ) . All voxels are corrected to a false discovery rate of 0 . 05 ( i . e . , q <0 . 05 ) . Region of interest ( ROI ) clusters were thresholded to a minimum of 40 continuously connected voxels ( i . e . , k >40 ) . The side and bottom panels show individual responses for the go trial probability condition in seven ROIs . Red bars illustrate BOLD responses during trials where no key press was registered ( no-go trials ) and green bars show BOLD fits to trials where a response was registered ( go ) . BOLD: blood oxygenation level dependent; AMA: supplementary motor area; preSMA: pre-supplementary motor area; IFG: inferior frontal gyrus; IPL: inferior parietal lobule; PCC: posterior cingulate cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 01510 . 7554/eLife . 08723 . 016Figure 8—figure supplement 1 . In-scanner EMG protocol . ( A ) Example of modified EMG set-up using the in-house Siemens physiological monitoring unit for monitoring cardiac signals . The recording sensors were placed to estimate the electrical vector from the first dorsal interosseous ( FDI ) muscle of the responding right hand . ( B ) Trial-wise EMG responses detected for the trials where a response was made . Trials are time-locked to the start of the preparatory cue . White line indicates time in which the bar intersected the target . Recording for responses continued well after the end of the trial to identify late responses . ( C ) An example raw EMG signal from a single subject with peaks coinciding with finger movements . DOI: http://dx . doi . org/10 . 7554/eLife . 08723 . 016 Two cortical regions , the inferior frontal gyrus ( IFG ) and supplementary motor area ( SMA ) near the preSMA , were also of particular interest because they have been previously implicated in inhibitory control ( Figure 8 ) . In the IFG , we observed a lateralized interaction between go trial probability and response type in the right hemisphere ( F[2 , 38] = 4 . 05 , p=0 . 027 ) , but not in the left ( F[2 , 38] = 1 . 45 , p=0 . 25 ) . This rightward lateralization is consistent with previous reports of the right IFG being implicated in control of the hyperdirect pathway ( Aron et al . , 2014; van Belle et al . , 2014; Zandbelt and Vink , 2010 ) . The SMA also exhibited an interaction between go trial probability and response type in both the right ( F[2 , 38] = 4 . 86 , p=0 . 014 ) and left ( F[2 , 38] = 4 . 52 , p=0 . 019 ) hemispheres . These results are consistent with previous studies implicating SMA and preSMA in uncertainty-related and conflict-related modulation of boundary height in the traditional drift-diffusion model , achieved through connections with the striatum and subthalamic nucleus ( STN ) ( Forstmann et al . , 2010; Frank et al . , 2015; Jahfari et al . , 2012 ) . This finding supports the notion that modulating the execution drift-rate in our model reflects similar computations as adapting the threshold in the traditional drift-diffusion models without a dynamic pressure on the drift rate . Thus , activation patterns in distributed corticobasal ganglia circuits are largely consistent with the predicted responses from our computational model of competing decision dynamics on execution process drift rates .
The current results provide two novel insights into the dynamics of inhibitory control . First , a DPM , where the initial state of the action cancellation signal depends on the current state of the execution decision , provides a much better fit to reactive stopping performance than models where these two processes are independent . As we showed , this dependency is similar to the interactive race model , proposed by Boucher et al . ( 2007 ) to describe inhibitory control of saccadic eye movements , in which an accumulating go signal is directly suppressed by a much stronger stop signal , thereby deterministically canceling subthreshold actions . In this model the critical parameter for determining stopping performance is the delay between a stop cue and when the stop signal is actually registered . In contrast , the critical parameters in our model are the race between the execution and braking processes and the state of the execution process when the stop signal occurs . This dependency is largely consistent with the circuit-level architecture of corticobasal ganglia pathways ( Figure 1A; Utter and Basso , 2008 ) where the direct ( facilitation ) , indirect ( suppression ) , and hyperdirect ( brake ) pathways converge at the primary output nucleus of the basal ganglia ( i . e . , the GPi in primates ) . Here we show , for the first time , that this dependency better fits behavioral data in the reactive task than the interactive race model , suggesting that there may not be equivalent control dynamics between oculomotor and manual response decisions . Schmidt et al . ( 2013 ) found that , in rodents , an external stop cue reliably produces a burst of activity in the STN , but only results in a successful stop when this signal reaches the substantia nigra pars reticulata ( similar to the GPi in primates ) . Consistent with classic race models of action cancellation , early activity of movement encoding striatal neurons was sufficient to predict whether hyperdirect stop signaling would reach the substantia nigra pars reticulata in time to prevent a response , rising rapidly on fast go and failed stop trials and more gradually on slow go and successful stop trials . Parts of this action cancellation signal from the STN may also be relegated via arkypallidal pathways ( orange pathways in Figure 1A ) that relay inhibitory signals back up to the striatal neurons , offering a route for hyperdirect signals to terminate action plans within the striatum itself ( Dodson et al . , 2015 ) . Both temporal competition between hyperdirect and descending striatal signals at the GPi and cancellation via arkypallidal pathways are largely consistent with the DPM presented here . Our modeling and imaging results also provide novel insights into the nature of proactive No-go decisions . Our framing of the execution process as a dynamic competition between direct and indirect pathways is motivated by the architecture of corticobasal ganglia pathways ( Figure 1A ) , as well as recent electrophysiological and modeling evidence showing that cortical inputs to the striatum dynamically control the balance of direct and indirect signals throughout the basal ganglia ( Wei et al . , 2015 ) . There are many possible routes for a competition between direct–indirect pathways with the basal ganglia . First , given that STN and globus pallidus projections terminate on common outputs , it is possible that a simple summation of inhibitory and excitatory signals at the GPi itself results in a separate race for disinhibiting pallidal projections to the thalamus . It is also possible that branching collaterals from direct pathway neurons that terminate on indirect pathway neurons in the globus pallidus ( Wei et al . , 2015; Wu et al . , 2000 ) would allow for crosstalk between the two major striatal efferent pathways . Finally , as mentioned previously , indirect pathway output , at the level of the STN , may be relayed back up to striatal direct pathway cells via arkypallidal connections ( Dodson et al . , 2015 ) . Disambiguating which of these mechanisms may proactively modulate the accumulation of evidence toward an execution decision requires careful physiological analysis within each pathway itself and is left to future studies . By reconceptualizing the dynamics of action selection as an ongoing competition between direct ( facilitation ) and indirect ( suppression ) pathways , our model may explain some of the discrepancies in previous studies of inhibitory control . For example , proactive inhibition has been associated with hemodynamic responses in the dorsal striatum ( Majid et al . , 2013 ) , interpreted as an increased recruitment of the indirect pathway to suppress an action . Yet , electrophysiological recordings in rodents find concurrent activation of both the direct and indirect pathways during action initiation and cancellation ( Cui et al . , 2013 ) . Both of these findings are consistent within the context of our model , in which the relative concurrent activation of direct and indirect pathways determines rate of integration toward the execution decision threshold . Previous studies have largely found that proactive control is implemented as a change in the distance to threshold ( Cavanagh et al . , 2011b; Forstmann et al . , 2010; Frank et al . , 2015; Jahfari et al . , 2012; Wei et al . , 2015 ) ; whereas in our model , these dynamics are captured by the rate of the execution process itself . While drift rate and threshold modulations have qualitatively dissociable effects in a standard two-choice drift-diffusion model—increasing the decision threshold leads to a speed-accuracy trade-off , whereas increasing drift rate has a positive effect on both outcome measures—this distinction does not hold true for our DPM . Rather , slowing the rate of the execution process leads to greater stopping accuracy at the expense of response speed . This is consistent with speed-accuracy trade-off produced by raising the response threshold in traditional drift-diffusion models . Also , Ratcliff and Frank ( 2012 ) found that synthetic behavioral data generated by a basal ganglia neural network were best described by a drift-diffusion model with a decision threshold that collapsed at a rate that depended on the conflict between choices . Here we included a similar mechanism , but as a dynamic gain signal applied to the execution process , and compared models in which response uncertainty modulated its drift-rate , onset time , or the combination of these parameters . Our findings suggest that in the presence of these temporal dynamics , contextual factors like control demands impose a stronger influence on the rate of the growing urgency to respond than on the onset of the decision process . This finding is consistent with the context-dependent rate decay in the decision threshold found by Ratcliff and Frank ( 2012 ) . However , by capturing these dynamics in the accumulating process itself , the DPM allows for incorporating the effects of context in reactive ( i . e . , penalty and reward structure ) and proactive ( i . e . , probabilistic expectations ) control in a single unifying parameter ( i . e . , execution drift rate ) , while at the same time distinguishing between action cancellation and no-go decision processes . Compared with stationary diffusion models , which predict positively skewed RT distributions , models that assume a temporal decay in the threshold or a nonconstant rate of accumulation predict response times that are more normally distributed ( Hawkins et al . , 2015 ) . Indeed , both tasks produced RT distributions that appeared more normal in shape and lacked a positive skew , adding justification to the assumption of a dynamic bias in the execution process . Compared with the reactive task ( Figure 4—figure supplement 1 ) , however , proactive RT distributions ( Figure 6—figure supplement 1 ) were much narrower , spiking near the end of the trial with a slight negative skew . Correspondingly , best-fit parameter estimates show a muc stronger influence of the dynamic bias signal on ‘go’ decisions in the proactive than reactive task , suggesting that gain adaptation is differentially recruited during these two forms of inhibitory control . Future studies will be necessary to investigate the contextual constraints and neural plausibility of nonstationary decision mechanisms , particularly with respect to their implementation in corticobasal ganglia networks . Other models of behavioral and electrophysiological data in countermanding tasks , such as the anti-saccade task , have suggested that proactive constraints shift the onset of the decision process . This is particularly true in cortical neurons that plan the action ( Pouget et al . , 2011; Purcell et al . , 2010 ) . However , here we show that the onset modulation model did not fit the behavior or imaging results as well as the drift modulation model . Our model was specifically designed to capture constraints in corticobasal ganglia systems , which are upstream of motor execution neurons , such as those recorded in previous countermanding tasks . Thus , it is possible that the dynamics described here are the mechanism for the delayed onset of activity in downstream cortical motor areas , that is , the onset of action selection neurons happens when the decision processes from the basal ganglia reach their boundary , which becomes delayed as the rate of the execution process is decelerated by the indirect pathway . It is also well established by now that subjects do not rely on a single mechanism for incorporating context into the decision process ( Heitz and Schall , 2012; Standage et al . , 2011 ) . While the pattern of observed BOLD responses was generally consistent with the drift modulation predictions , it did share some characteristics with the predictions of the onset modulation model ( i . e . , increasing activity with go trial probability on ‘no-go’ trials and greater overall activation on ‘go’ trials ) , allowing for the possibility that some subjects may alternate between strategies of modulating the drift and onset across trials or use both simultaneously . Only careful analysis of coincident electrophysiological recordings in key basal ganglia and cortical regions will fully confirm the compatibility of the two models . Taken together , our results provide crucial evidence that reactive stopping ability depends on the competitive dynamics of ongoing proactive decision processes , suggesting these two forms of control arise from convergent , but separable control signals in the basal ganglia . This novel reconceptualization of how proactive no-go decisions are distinguished from reactive stopping is consistent with the growing body of evidence to suggest that inhibitory control is a multifaceted process comprised of separable , but not independent , control mechanisms .
Neurologically healthy adults ( behavioral study: N=60 , mean age=22 years , 28 male; imaging study: N=20 , mean age=25 years , 9 male ) were recruited from the local university population . All procedures were approved by the local institutional review board at Carnegie Mellon University . The general trial procedure for go trials was the same for all experiments . On each trial , participants would see a bar fill upwards toward a white target line on the screen . The bar would intersect the target line 500 ms after trial onset . Participants could stop the bar at any point in the trial by pressing a button with their right hand ( space key in behavioral study , index button on a response pad in the fMRI experiment ) and were instructed to stop the bar as close to the target line as possible . The trial was terminated when the subject made a key press or when the bar passed the upper limit of the screen ( reactive task , 650 ms; proactive task , 555 ms ) . On each trial a monetary bonus was given based on how close the top of the bar was to the target line . On baseline trials , this was determined as the inverse of the distance from the target . On caution trials , this was determined as the inverse of the squared distance . At the end of each trial , participants received visual feedback of their accuracy and monetary score . In the reactive experiment , on a subset of trials ( stop trials , n = 100 ) the bar would stop and turn red at one of five predefined points in its trajectory ( the SSD ) . SSDs ranged from 200 to 400 ms ( n=20 trials per SSD ) . On these trials participants were instructed not to produce a key press . Subjects were told that the color change was a cue that the bar had stopped . If subjects correctly withheld a response , then they would receive a bonus score . If subjects erroneously made a response , they received a penalty of equal value to the bonus during successful trials ( 100 points in baseline trials; 200 points in caution trials ) . The first block of trials ( n=22 ) were all go trials so as to familiarize subjects with the timing of the response . On the remaining nine blocks ( n=20 trials per block ) , the stop and go trials were randomly interleaved with a 50% probability . This ratio was selected so as to match the stopping probabilities with the proactive experiment . In the proactive experiment , only one SSD was used ( 450 ms ) for the entire experiment . On each trial the color of the bar indicated the probability that the bar would intersect with the target . Blue bars indicated 100% probability , red bars indicated 0% probability ( i . e . , always a stop trial ) , and shades of purple represented mixture probabilities . On each block ( n=24 trials/block , 10 blocks total ) , each go trial probability was uniformly sampled four times . In order to detect late key presses on no-go trials , the program would continue monitoring for key presses for an extra 100 ms after the bar left the screen . The feedback structures for the baseline and caution conditions were the same as in the reactive experiment . All experiments were run using Psychophysics Toolbox ( version 3 . 0 ) in Matlab ( R2012b ) on a Linux platform ( Ubuntu 12 . 04 ) . A repeated-measures ANOVA was performed to test for a statistically significant interaction between the effects of cued go trial probability ( 0% , 20% , 40% , 60% , 80% , 100% ) and penalty structure ( baseline , caution ) in the mean probability of a no-go decision . The same test was performed for mean go RTs but collapsing across 0% , 20% and 40% go trial probability conditions to increase statistical power , as fewer responses were recorded in these conditions . To account for hardware limitations in precision of RT measurements , mean go RTs were calculated for all responses recorded prior to the earliest recorded ‘no-go’ outcome ( <∼555 ms ) . Both dependent measures violated the sphericity assumption , thus all reported values were corrected accordingly using the Greenhouse-Geisser method . The PSE on the stop and no-go probability curves was estimated using logistic regression . Specifically the probability of stopping or not-going , y , was modeled as a function of cue conditions , x . ( 1 ) y = λx λ0 For the reactive task , x was the stop-signal delay . For the proactive task , x was the probability of a go trial . The PSE was then calculated using the constant , λ0 , and slope , λ , parameters from this regression model . ( 2 ) PSE = -1λ0λ For all simulations , we designed an extension of the standard drift-diffusion model which assumes evidence for competing choices is stochastically sampled and accumulated until terminating at one of the choice thresholds , determining the decision outcome and response time . Our modified model assumes that competing facilitatory ( i . e . , direct ) and suppressive ( i . e . , indirect ) signals are integrated into a single execution process ( θe ) . The linear drift and diffusion ( φe ) of the execution process is shown by the stochastic differential equation in Equation 3 , accumulating with a mean rate of ve ( i . e . , drift rate ) and a standard deviation described by the Wiener diffusion process ( e . g . , white noise ) with diffusion constant σ . The execution process is fully described by Equation 4 in which the linear accumulation described by Equation 3 is scaled by a dynamic bias signal γ , modeled as a hyperbolic function of time with gain xb . ( 3 ) dφe = υedt σdW ( 4 ) θet = φet · γt The execution process begins accumulating after an onset delay of tr . It is worth noting that it is possible for the onset delay parameter to capture both pre- and postdecision delays ( i . e . , sensory encoding and motor execution delays ) . A response and RT is recorded if θe reaches the execution boundary ( a ) before the end of the trial window ( b ) and before the braking process reaches the lower ( 0 ) boundary ( see below ) . In the event of a stop cue , the braking process ( θb ) is initiated at the current state of θe with a negative drift rate ( −vb ) . If θb reaches the 0 boundary before θe reaches the execution boundary no response or RT is recorded from the model . For the reactive stopping task we fit three different models where the dynamics of θe and θb . ( 5 ) dφb = υbdt σdWθb ( SSD ) = θe ( SSD ) ( 6 ) dθb = υbdt σdW ( 7 ) θb ( SSD ) = θe ( tr ) = 0 We also compared different models of contextual modulation in which penalty and proactive cuing either modulated the drift rate , onset time , or boundary height of the execution process . All contextual modulation models in the reactive task were fit using the DPM framework ( Equations 3–5 ) with two free parameters in the drift ( vb , vc ) , onset ( trb , trc ) , and boundary ( ab , ac ) modulation models and four free parameters in the combined drift and onset modulation model ( vb , vc , trb , trc ) . The corresponding models fit to the proactive task contained six free parameters ( one each go trial probability cue ) in the drift ( v0−v100 ) , boundary ( a0−a100 ) , and onset ( tr0−tr100 ) modulation models and 12 free parameters in the combined drift and onset model ( v0−v100; tr0−tr100 ) . We report two complexity-penalized goodness-of-fit statistics to account for differences in the number of parameters of alternative models , AIC and BIC . All models were fit to the behavioral data by minimizing a cost function equal to the sum of the squared and weighted errors between vectors of observed and simulated RT quantiles ( 0 . 1 , 0 . 3 , 0 . 5 , 0 . 7 , 0 . 9 ) and response probabilities , that is , stop probability curve in the reactive task and the no-go probability curve in the proactive task . Weights applied to the RT quantiles were calculated by estimating the standard error for each of the five quantiles ( Maritz and Jarrett , 1978 ) and then dividing the interpolated median standard error by that of each quantile . For the response probabilities , weights were calculated by taking the standard deviation for each condition across subjects and then dividing the mean of those standard deviations by that of each condition . This approach represents the variability of each value in the vector as a ratio ( Ratcliff and Tuerlinckx , 2002 ) , where values closer to the mean are assigned a weight close to 1 , and values associated with higher variability a weight <1 , lower variability a weight >1 . The cost function for all models fit to the reactive task data is shown in Equation 8 below . This included the probability of a response on go trials ( Pg ) , the stop accuracy at each of the five SSD conditions ( Pssd ) , the five RT quantiles for correct responses ( Qc ) and five RT quantiles for error responses ( Qe ) for both the baseline and caution conditions ( j ) separately . Error RTs ( i . e . , responses on trials when the stop cue was presented ) were collapsed across all SSDs due to there being few trials in which the subject made a response in the early SSD conditions . The weights of the correct ( wc ) and error ( we ) RT quantiles were multiplied by the respective probability of a response on correct ( Pcorr ) and error ( Perr ) trials . All variables with the hat accent are model-predicted values and are subtracted from the corresponding empirical subject-averaged statistic . ( 8 ) CostReactive=∑j=12[wg· ( Pgj - P^gj ) 2 ∑wd·d=15 ( Pdj - P⏜dj ) 2 Pcorrj·∑i=15wijc· ( Qijc - Q^ijc ) 2 Perrj·∑i=15wije· ( Qije - Q^ije ) 2] Proactive models were fit using a similar cost function that included the probability of a response in each of the six probabilistic cue conditions ( Pprob , ) , five RT quantiles calculated from trials in which the cue indicated a go probability higher that 50% ( Qhi ) and five RT quantiles calculated from trials in which the cue indicated a go probability of lower than 50% ( Qlo ) . The weights applied to high ( wh ) and low ( wl ) RT quantiles were multiplied by the respective probability of a response in high ( Phigh ) and low ( Plow ) go probability conditions . ( 9 ) costProactive = ∑j=16wpPj - P^j2 Phigh∑h=15WhQh - Q^h2 Plow∑i=15wiQi - Q^i2 The response times were split up in this manner to account for the low trial counts in the lower go trial probability conditions . During the model fits , this was performed by allowing free parameters in the model to vary across all six cue conditions , in order to calculate the predicted response probabilities . All trials in which the execution process reached threshold before the time limit were then collapsed into two vectors , one containing simulations from the low go trial probability conditions and the other from high go trial probability conditions . RT quantiles were then calculated for each aggregate vector of RTs and appended to the vector of response probabilities before being submitted to the cost function . All variables with the hat accent are model-predicted values and are subtracted from the corresponding empirical subject-averaged statistic . To ensure that parameter estimation did not terminate prematurely , models were fit to the data using a combination of global and local minima optimization . First , an initial set of parameters was obtained for each model type by leaving all parameters free and using the basin hopping algorithm provided by SciPy ( Blanco-Silva , 2013 ) to identify a global error basin . The best fitting parameters are then optimized using a Nelder–Mead simplex algorithm ( Nelder and Mead , 1965 ) which explores the basin reached in the previous step for its local minimum . Basin hopping seeks to identify a set of parameters at the global minimum of a cost function by stochastically searching the parameter space for error basins , which is especially desirable when there is a high degree of the variability in the model error . However , stochastic search methods are not well suited for fine-grained error minimization which is best achieved by gradient based methods like the simplex algorithm . For reactive models , this average cost function is similar to that shown in Equation 8 but without the summation over j conditions and with all empirical statistics representing the average value collapsing baseline and caution conditions . For proactive models , this cost function becomes a single response probability , taken as the average response probability across cues , and five RT quantiles calculated by averaging the quantiles for the high and low go trial probability conditions . At this stage the basin hopping algorithm is iterated a maximum of 100 times or until it returns 40 failed attempts to find a new global minimum Following parameter optimization to the average data , alternative models were tested by allowing only one or a few of these parameters to vary across conditions while holding all others constant . First , the average parameter estimates are submitted to another global optimization procedure in which all parameters are held constant except for the parameter of interest that is optimized individually to each condition of interest using the same basin hopping parameters as above . This is performed prior to the final simplex optimization to avoid initiating the parameter ( s ) at the same value for each condition being fit , as we found this to increase the chances that the Nelder–Mead simplex terminates very close to its initial state and overall generally worse fits . The final simplex minimization was run holding all parameters constant except for those of interest that were left free to vary across conditions . For all models , this procedure was run from start to finish at least three times and the model with the lowest AIC was selected . This includes all compared models of contextual modulation in the reactive and proactive tasks , meaning that constant parameters are held constant across conditions , but not models . By running every model through the same global and local optimization routine multiple times , this avoids the issue of biasing model selection to favor those that are simply better able to explain the data within the constraints of a single set of constant parameters . Thus , parameter estimates listed in Tables 1 and 2 for parameters held constant across conditions reflect the best-fit parameters to the average data for the best-of-three model . Example data as well as model code , simulations , cost function weights , and animations can be found here: https://www . github . com/coaxlab/radd In addition to fitting different models of proactive control to behavioral data , we evaluated these models on their ability to make general predictions regarding the change in the BOLD response across probabilistic cues on go and no-go trials . Based on the assumption that the neural activity in regions involved in a competitive execution decision will reflect information about the time and distance of the accumulation process , we generated BOLD predictions from each of the models using the following procedure . For each model , the best fitting parameter estimates from the proactive task were used to simulate 100 trials for each go trial probability condition . Because we were interested in comparing fundamental differences in how each model predicted BOLD activity across conditions rather than trial-wise or subject-wise predictions , simulations were performed using the best fit parameters from the behavioral experiment rather than performing separate fits to behavior from the scanning session . This was done to take advantage of the fact that the behavioral effects of the task were similar inside and outside the scanner , but with significantly more trials/observations in the behavioral-only session . The simulated trials were then separated by trial outcome ( ‘go’ trials in which the execution process reached threshold and ‘no-go’ in which the execution process failed to reach threshold . To approximate the same experimental conditions tested in the imaging analysis , ‘go’ trials in the lower go trial probability conditions ( ≤60% ) were binned together and all ‘no-go’ trials in the higher go trial probability conditions ( ≥40% ) were binned together . For each condition the predicted BOLD magnitude was calculated by taking the cumulative sum of the full execution process over the length of each trial , then averaging across trials . By taking the cumulative sum of the execution process this effectively represents the distance-to-threshold and time-to-threshold of each trial as a single area-under-the-curve estimate . Taking the average area-under-the-curve across trials can then be compared with the observed magnitude of the BOLD response in the corresponding condition . All imaging data were collected in a 3T Siemens Verio MRI at the Scientific Imaging and Brain Research Center on the Carnegie Mellon University campus . All functional images were collected with echo planar imaging sequence ( TE = 20 ms; TR = 1500 ms ) . Thirty contiguous slices ( 3 . 2 mm x× 3 . 2 mm × 4 mm ) were collected in an ascending and sequential fashion parallel to the anterior and posterior commissures . One subject was removed from analysis due to an error in saving the output files for the scan run . One subject failed to complete the reactive experiment due to time constraints . All experiments utilized a rapid event related design with a jittered intertrial interval sampled from an exponential distribution ( range 4–18 s , mean = 8 s ) with 334 volumes per run . In the proactive experiment , a total of four scan runs were collected with 60 trials collected for each go trial probability condition . All trials were preceded by a 300 ms onset cue During scanning electrical activity of the FDI muscle of the right hand was recorded using an MR compatible EMG sensor ( 400 Hz sampling rate ) . Sensors were put over the lateral and medial portions of the FDI muscle , with the ground electrode placed on the bone of the wrist ( Figure 8—figure supplement 1 ) . The onset and offset of recording were time-locked to the onset of the functional scans . In a subset of subjects ( N = 6 ) the electrode placement slipped or a recording buffer error truncated data collection . Thus EMG data for these subjects was not included in the final analysis . The muscle activity for all trials was smoothed using a Gaussian smoothing kernel ( 4 s FWHM ) and time-locked to the onset of each trial . Significant muscle activity was determined within each trial when the root mean squared EMG signal exceeded 3 standard deviations of the pre-trial signal . All fMRI analysis was performed with SPM8 ( http://www . fil . ion . ucl . ac . uk/spm/software/spm8/ ) . Prior to analysis , the echo planar images for each participant were realigned to the first image in the series and corrected for differences in the slice acquisition time . All images were then coregistered to MNI-space ( Montreal Neurological Institute ) using a nonlinear spatial normalization approach ( Internation Consortium for Brain Mapping-152 space template regularization , 16 nonlinear iterations ) and smoothed using a 4 mm isotropic Gaussian kernel . Estimates of task-related responses at each voxel were determined using a reweighted least squares GLM approach ( Diedrichsen and Shadmehr , 2005 ) that minimizes the influence of movement related noise in the signal . Statistically significant effects were determined using a false discovery rate threshold across all voxels of 0 . 05 ( q < 0 . 05 ) . Clusters of 40 or more continuously active voxels were extracted from the parametric ROIs and their condition-wise BOLD responses are reported in Supplementary file 1 . Seven areas were identified in the whole brain parametric GLM analysis as being of particular interest and selected for follow-up ROI analysis . To avoid circular inference , we used the Harvard–Oxford cortical and subcortical atlases ( Desikan et al . , 2006; Frazier et al . , 2005 ) to anatomically identify the right caudate nucleus and bilateral inferior frontal gyrus ( operculum ) , SMA , and thalamus . Using a fourfold cross validation , voxels with a significant ( p <0 . 005 , uncorrected ) positive no-go–go parametric modulation were identified within each anatomical ROI using three-fourths of the sample . Then the regression coefficients for each condition were extracted from this subset of voxels within the ROI in the remaining subjects . | Imagine you are playing baseball . You can decide not to swing the bat at the incoming ball if you see that it is a wild pitch that will be way outside the strike zone; this is known as reactive control . Alternatively , you may decide not to move because you were coached never to swing at the first pitch ( proactive control ) . It is thought that the brain processes these signals separately with reactive control being a quick way to put the brakes on a planned movement and proactive control being a more specific suppression of unwanted actions . However , some researchers have argued that real-life “inhibitory” control decisions are more likely to be made using a combination of both reactive and proactive signals . In primates , reactive and proactive signals are both processed by a region of the brain called the basal ganglia . However , it is not clear whether these signals pass through the same set of nerve cells , or whether they use separate sets of cells that run in parallel . Dunovan et al . studied how these signals are processed in the human basal ganglia using a combination of experiments and computational models . The model assumes that reactive and proactive signals are carried by two pathways that are initially separate but eventually converge in the basal ganglia . If these pathways converge , then proactive control signals should complement reactive decisions to “apply the brakes” . For example , if reactive signals suggest that the ball may not come over the home plate , the fact that it is also the first pitch would make it easier for the brain to decide not to swing the bat . Dunovan et al . tested this model by asking human volunteers to complete two tasks where the decision to respond to a stimulus is made proactively using prior knowledge , or reactively using an explicit stop cue . The experiments also used a technique called functional magnetic resonance imaging ( fMRI ) to measure the activity in the basal ganglia of each volunteer . Simulations from the model were able to predict the observed patterns of behavior and brain activity in several regions that are key to inhibitory control , including the output of the basal ganglia . These findings provide a possible mechanism for how reactive and proactive control may interact in the brain . Because of the limitations in imaging the human brain , the next step will be to test whether the model is able to predict the behavior of individual nerve cells in the brains of other animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2015 | Competing basal ganglia pathways determine the difference between stopping and deciding not to go |
Pheromones play an important role in the behavior , ecology , and evolution of many organisms . The structure of many insect pheromones typically consists of a hydrocarbon backbone , occasionally modified with various functional oxygen groups . Here we show that sex-specific triacylclyerides ( TAGs ) are broadly conserved across the subgenus Drosophila in 11 species and represent a novel class of pheromones that has been largely overlooked . In desert-adapted drosophilids , 13 different TAGs are secreted exclusively by males from the ejaculatory bulb , transferred to females during mating , and function synergistically to inhibit courtship from other males . Sex-specific TAGs are comprised of at least one short branched tiglic acid and a long linear fatty acyl component , an unusual structural motif that has not been reported before in other natural products . The diversification of chemical cues used by desert-adapted Drosophila as pheromones may be related to their specialized diet of fermenting cacti .
Chemical communication significantly influences many complex social behaviors , including aggression , kin recognition , and courtship ( Wyatt , 2003 ) . The chemical structures and functions of insect pheromones have been intensely studied because of the fascinating diversity of behavioral properties and relevance to questions of speciation , reproductive isolation , and applications to pest control ( Witzgall et al . , 2010 ) . Since the discovery of Bombykol in 1959 ( Butenandt et al . , 1959 ) , hundreds of insect pheromones have been identified , including straight chain and branched alkanes and alkenes , oxygen-containing derivatives such as wax esters , fatty alcohols , and aldehydes , sterols , and isoprene-based compounds ( Tillman et al . , 1999; El-Sayed , 2012 ) . In Drosophila , pheromones are produced by oenocytes ( specialized epithelial cells in both males and females ) and the male ejaculatory bulb and subsequently secreted onto the cuticular surface and anogenital region , respectively ( Billeter et al . , 2009; Yew et al . , 2009 ) . Previous studies of Sophophora and Drosophila flies identified alkanes , alkenes , and oxygen-modified hydrocarbons as the major lipids used as pheromones ( Jallon and David , 1987; Greenspan and Ferveur , 2000 ) . Recently , triacylglycerides ( TAGs ) , which are normally found in the fat bodies and used for energy storage , were observed on the cuticles of flies from the Drosophila repleta and Drosophila quinaria groups ( Yew et al . , 2011; Curtis et al . , 2013 ) . However , almost nothing is known about the structure , chemical diversity , conserved expression , and functional roles of these exogenously secreted TAGs . To explore the role of TAGs as pheromones and the ubiquity of their expression in Drosophila , we used ultraviolet laser desorption/ionization mass spectrometry ( UV-LDI MS ) to analyze the cuticular profiles of flies from seven major Sophophora and Drosophila groups . We also investigated the chemical structures of sex-specific TAGs and their role as sex pheromones in species from the D . repleta group . Our studies indicate that TAGs are a broadly conserved , structurally atypical class of Drosophila pheromones that has been overlooked .
We used UV-LDI MS to perform a broad survey of cuticular lipid profiles of flies from the Drosophila and Sophophora subgenera . UV-LDI MS provides spatially resolved chemical profiling from single , intact insects by probing the cuticular surface with a 200 μm laser ( Yew et al . , 2009 , 2011 ) . Chemical signatures consistent with TAG structures were found to be largely conserved across 3 different Drosophila groups: the repleta radiation ( including Drosophila hydei , Drosophila buzzatii , Drosophila navojoa , Drosophila wheeleri , and Drosophila aldrichi ) , the virilis group ( Drosophila americana , Drosophila virilis , and Drosophila montana ) , and within the robusta group ( Drosophila robusta ) ( Figure 1; Figure 1—figure supplement 1 ) . The TAGs were expressed only in the ejaculatory bulb of males . In contrast , sex-specific TAGs were not detected in any of the species tested from the Sophophora subgenus . Many of the TAG-producing species are capable of feeding and reproducing on cacti , fungi ( mushroom ) , and tree sap or slime fluxes , substrates that contain high levels of toxins , plant defensive compounds , or bacteria , a characteristic that may be related to their ability to produce sex-specific TAGs . 10 . 7554/eLife . 01751 . 003Figure 1 . Male-specific TAG expression is broadly conserved across the Drosophila subgenus and not found in species from Sophophora . The primary diets of each species are indicated , based on the previous studies . Branches for TAG-producing species are colored in red . Branch lengths are proportional to evolutionary time . *Evidence for TAG-expression is based on Curtis et al . , 2013 . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 00310 . 7554/eLife . 01751 . 004Figure 1—figure supplement 1 . Representative UV-LDI spectra from distantly related drosophilids in the Drosophila subgenus . Signals corresponding to sex-specific TAGs ( in red ) are identified based on the exact mass measurements , predicted elemental composition , and number of double bonds . Labeled signals correspond to potassiated molecules [M + K]+ . Unlabeled peaks correspond to sodiated molecules [M + Na]+ . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 004 To characterize the structures and functions of sex-specific TAGs , we focused on desert-adapted drosophilids from the D . repleta group , Drosophila arizonae and Drosophila mojavensis , two well-characterized models for speciation , reproductive isolation , and ecological studies ( Ruiz et al . , 1990; Markow , 1996; Etges and Jackson , 2001 ) . Analysis by UV-LDI MS detected 13 TAGs and several long-chain acetyldienyl acetates , 30 or 32 carbons in length ( referred to as long OAcs ) exclusively in the anogenital region of D . arizonae and D . mojavensis males and not on virgin females of either species ( Figure 2A , B; Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 01751 . 005Figure 2 . Pheromone profiles and age-related increase in sex-specific TAGs . ( A and B ) UV-LDI MS allows spatially resolved detection of high molecular weight lipids directly from intact insects , with minimal damage to the cuticle . Representative mass spectra from the anogenital region ( inset ) of D . arizonae and D . mojavensis males show signals corresponding to triglycerides ( TAGs , red ) and long chain alkadienyl acetates ( long OAcs , blue ) . The hydrocarbon C35:2 ( number of carbons: number of double bonds ) is found on cuticles of males and females . Labeled signals correspond to potassiated molecules [M + K]+ . Scale bar: 1 mm . ( C and D ) Relative intensity of TAGs and long OAcs on male D . arizonae and D . mojavensis , respectively . TAGs and long OAcs increase with age , with trace quantities first appearing at 4 day old . The signal intensity for all detected TAGs or long OAcs was normalized to the signal intensity of C35:2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 00510 . 7554/eLife . 01751 . 006Figure 2—figure supplement 1 . UV-LDI MS profiles from the anogenital region of virgin females and dissected ejaculatory bulb ( eb ) and accessory glands . *: Background peak from fly wing matrix . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 006 We next tested whether sex-specific TAG expression is correlated with male sexual maturity . Chemical profiling of D . arizonae from 0 to 15 days old indicated that the expression of the TAGs and long OAcs increased in abundance as males get matured , with little or no expression in the first 4 days after eclosion and higher expression towards the age of maturity at approximately 8 day old , the age when males exhibit full courtship behaviors ( Markow , 1981 ) ( Figure 2C ) . D . mojavensis followed a similar maturation profile ( Figure 2D ) . Analysis of dissected male reproductive organs by UV-LDI MS revealed qualitatively similar chemical profiles exclusively in the ejaculatory bulbs ( Figure 2—figure supplement 1 ) . In addition , no predicted precursors of these compounds such as diacylglycerol or glycerol-3-phosphate were detected from the accessory glands or other reproductive organs . These results indicate that the TAGs and long OAcs are synthesized in the ejaculatory bulb . From D . arizonae and D . mojavensis , we isolated both TAG and long OAc lipid classes using thin layer chromatography ( TLC; Figure 3—figure supplement 1 ) . Chemical derivatization of the long OAc fraction confirmed the presence of an acetyl group ( Figure 3—figure supplement 2 ) . Gas chromatography MS ( GCMS ) analysis of transesterified TLC fractions indicated tiglic acid , a 5-carbon branched unsaturated acid , as one of the fatty acyl moieties ( Figure 3—figure supplement 3 ) . No other 5-carbon fatty acid methyl esters were detected . Tandem MS with low energy collision-induced dissociation ( CID ) analysis provided the chain length and degree of unsaturation of each of the acyl chains present in each TAG . A similar motif was revealed among each of the molecules: a single long-chain fatty acyl component , 16–18 carbons in length , together with 2 short-chain fatty acyl side chains , each 2–5 carbons in length ( Figure 3A; Figure 3—figure supplement 4 ) . For several of the more abundant TAG molecules , the position of the acyl chains on the glycerol backbone could be deduced based on the relative abundance of the product ions in the CID spectra . As described by Hsu and Turk ( 1999 and 2010 ) , fragments reflecting the loss of substituent at the sn-2 carbon ( middle of the glycerol backbone ) are less abundant than ions reflecting losses at either the sn-1 or sn-3 carbons . Based on this observation , long-chain fatty acids are predominantly located at either sn-1 or sn-3 for the major TAGs at [M + Na]+ 543 and 541 ( Figure 3A; Figure 3—figure supplement 4 ) and [M + Li]+ 499 , 501 , and 527 ( Figure 3—figure supplement 4 ) . It was not possible to distinguish between sn-1 and sn-3 positions . The position assignments on the backbone are supported by analysis of synthetic standards in which a long-chain fatty acid is placed at sn-1 . The relative abundances of fragment ions are similar to those observed from crude extract ( Figure 3—figure supplement 6–8 ) . Notably , CID analysis of a fourth major TAG at [M + Li]+ 487 resulted in low-intensity signals corresponding to loss of the C18:1 fatty acyl substituent , suggesting that this component is likely to reside at the sn-2 position ( Figure 3—figure supplement 4 ) . Low-abundance signals for isobaric TAGs containing C18:2 and C18:1 fatty acids were also observed at [M + Li]+ 501 , 499 , and 487 and are likely to represent minor components . In these cases , it was not possible to assign substituent positions . 10 . 7554/eLife . 01751 . 007Figure 3 . Structural elucidation of sex-specific TAGs . ( A ) The low energy collision-induced dissociation ( CID ) mass spectrum of a TAG-related signal from crude D . arizonae extract ( [M + Na]+ 543 ) shows fragments corresponding to losses of a 5 carbon fatty acid with a single double bond ( C5:1 ) and an 18 carbon fatty acid with a single double bond ( C18:1 ) . Both sodiated ( major peak ) and protonated chain side losses are observed . The schematic rationalizes the product ions formed during CID of mass-selected [M + Na]+ of unsaturated lipids . ( B ) Ozone-induced dissociation ( OzID ) of a TAG-related signal ( shown in A ) indicates isomers with variant double bond positions . The fragments at m/z 461 and m/z 433 are aldehyde products consistent with double bonds ( db ) at positions n-7 and n-9 , respectively . The fragment at m/z 531 confirms the n-2 double bond position found in the tiglic acyl component . The corresponding Criegee product ions ( m/z 477 and m/z 449 , respectively ) are also observed . The schematic rationalizes the product ions formed during OzID of mass-selected [M + Na]+ of unsaturated lipids . Product ions are assigned as outlined by Thomas et al . , 2008 and Brown et al . , 2011 . ( C ) CID and OzID MS analyses of the most abundant sex-specific TAGs reveal significant combinatorial complexity . A generic TAG molecule consisting of a glycerol backbone and 3 fatty acyl ( FA ) side chains , R1 , R2 , and R3 , is shown . Each TAG species is comprised of 2 short chain and 1 long chain FA component . Shaded boxes indicate the composite side chains of each TAG species . The glycerol backbone positions for several TAGs are assigned based on the comparison with synthetic standards and ion product abundance patterns ( dark gray boxes ) . Ambiguous backbone positions are in light gray . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 00710 . 7554/eLife . 01751 . 008Figure 3—figure supplement 1 . Thin layer chromatography ( TLC ) separation of D . arizonae male cuticular lipid extract . Direct analysis of contents using fly-assisted laser desorption/ionization ( FALDI ) MS reveal signals corresponding to male-specific triacylglycerides ( TAGs ) , long OAcs , conventional TAGs ( from internal fat stores ) , and cuticular hydrocarbons . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 00810 . 7554/eLife . 01751 . 009Figure 3—figure supplement 2 . Direct analysis in real time ( DART ) MS spectrum of the TLC fraction containing long OAcs after derivatization by base hydrolysis confirms the presence of an acetyl functional group . Based on the high mass accuracy predictions , the molecules [M + H]+ 449 and 477 ( respectively , [M + K]+ 487 and 515 in Figure 2 UV-LDI spectra ) are predicted to contain three double bonds . Derivatization of the putative acetyl functional group with base hydrolysis resulted in a mass shift ( loss of m/z 42 ) consistent with the replacement of an acetyl with a hydroxyl group . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 00910 . 7554/eLife . 01751 . 010Figure 3—figure supplement 3 . Structural elucidation of sex-specific TAGs using gas chromatography MS . GCMS analysis of the male-specific TAG fraction following trans-esterification confirms that tiglic acid is one of the fatty acyl components . The retention time ( at 2 . 98 min ) and the electron ionization ( EI ) spectrum of synthetic methyl tiglate are identical to the analysis of a TAG fraction purified from crude extract . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01010 . 7554/eLife . 01751 . 011Figure 3—figure supplement 4 . Structural elucidation of TAGs by CID MS reveals fatty acid ( FA ) components with 2 , 3 , 5 , 16 , or 18 carbons in length and 0–2 double bonds . For more abundant TAG molecules , the FA substituent positions on the glycerol backbone are assigned based on the relative abundances of the product ions in the CID spectra , as described by Hsu and Turk ( 1999 and 2010 ) . For minor TAG components , only the composite FA side chains are indicated . Substituent positions are ambiguous due to low-signal intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01110 . 7554/eLife . 01751 . 012Figure 3—figure supplement 5 . Analysis of TAGs by OzID reveals double bond positions of acyl side chains . Fragments resulting from OzID analysis are aldehyde products with m/z consistent with double bonds at positions n-2 , n-7 , and n-9 for fatty acyl monoene constituents and at both n-6 and n-9 for a fatty acyl diene . n indicates the position of the terminal methyl group . The corresponding Criegee product ions ( difference of m/z 16 ) were also found in each spectrum . Although the CID spectrum for the TAG species at [M + Li]+ 499 indicated a C18:2 FA component , the intensity of this signal was too low to allow for double bond placement by OzID . Red lines indicate putative cleavage points . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01210 . 7554/eLife . 01751 . 013Figure 3—figure supplement 6 . Spectra obtained from CID MS and OzID analyses of a synthetic TAG comprised of an oleic acid ( cis-9-Octadecenoic acid ) and tiglic acid side chains are consistent with the analysis of a TAG molecule with identical m/z found from crude extract . The relative abundance of fragment ions resulting from fatty acyl side chain loss is consistent between natural and synthetic products and supports the backbone substituent positions . Following OzID analysis , the overall and relative abundance of aldehyde and Criegee ions to each other are similar between natural and synthetic products , indicating cis-double bond geometry . Difference in parent ion intensity between synthetic and natural products is likely due to isobaric interference from other minor components found in the crude extract . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01310 . 7554/eLife . 01751 . 014Figure 3—figure supplement 7 . The spectrum obtained from CID MS analysis of a synthetic TAG ( 16:1/5:1/5:1 ) is consistent with the spectrum from a TAG molecule with identical m/z found from crude extract . The relative abundance of fragment ions resulting from fatty acyl side chain loss is consistent between natural and synthetic products , supporting the backbone substituent positions . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01410 . 7554/eLife . 01751 . 015Figure 3—figure supplement 8 . Spectra obtained from CID MS and OzID analyses of a synthetic TAG consisting of linoleic acid ( cis , cis-9 , 12-Octadecadienoic acid ) and tiglic acid side chains are consistent with spectra from analysis of a TAG molecule with identical m/z found from crude extract . The relative abundance of fragment ions resulting from fatty acyl side chain loss is consistent between natural and synthetic products and supports the backbone substituent positions . Following OzID analysis , the overall and relative abundance of aldehyde and Criegee ions to each other are similar between natural and synthetic products , indicating cis and cis-double bond geometry . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 015 To determine the double bond positions within each fatty acid , we used ozone-induced dissociation ( OzID ) mass spectrometry ( Thomas et al . , 2008; Brown et al . , 2011 ) . Individual TAG species were mass-selected within an ion-trap mass spectrometer where they were exposed to ozone vapor . The resulting gas-phase ion–molecule reaction facilitates targeted oxidative dissociation of carbon–carbon double bonds present in the acyl chains . Fragmentation of the ozonide leads to formation of characteristic aldehyde and Criegee ions with a mass indicative of the positions of each double bond . OzID analysis of the TAG fraction revealed numerous positional isomers , with double bond positions between C9-C10 ( n-9 ) and C11-C12 ( n-7 ) , indicating oleic and palmitoleic acid side chains , and between C2-C3 ( n-2 ) , consistent with tiglic acid ( Figure 3B , C; Figure 3—figure supplement 5 ) . Spectra from OzID analysis of TAG standards synthesized with oleic acid ( C18:1 , n-9 ) or linoleic acid ( C18:2 , n-6 , 9 ) support the double bond position assignments ( Figure 3—figure supplement 6 and 8 ) . Double-bond geometry could also be deduced for two of the more abundant TAGs . cis- and trans-alkenes exhibit differential reactivity to ozone , resulting in differences in the overall abundances of the fragment ions and the relative abundance of the Criegee and aldehyde product ions ( Poad et al . , 2010 ) . The relative abundance of the aldehyde and Criegee ions for the molecules at [M + Na]+ 543 and 541 are consistent with those of synthetic TAG standards synthesized with oleic acid and linoleic acid , both of which contain cis- double bonds ( Figure 3—figure supplement 6 and 8 ) . In summary , MS analysis revealed considerable variation in the carbon chain length , degree of unsaturation , positions of fatty acyl chains , and double bond positions of both the short chain and long chain fatty acyl components ( Figure 3C ) . All of the analyzed TAGs contained tiglic acid . The unusual combination of short odd-branched chain fatty acids with a single linear long-chain component has not been reported before in natural products . To determine the contribution of diet to TAG production , we compared TAG levels between males raised on standard fly media for 2 generations vs media supplemented with cactus powder and banana . Thin layer chromatography of the lipid contents of ejaculatory bulbs indicated that males raised on standard media produced a significantly lower amount of some of the sex-specific TAGs , including [M + K]+ 559 , one of the most abundant molecules ( Figure 4 ) . The results show that although a specialized diet is not essential for sex-specific TAG production , precursors derived from food can influence the quantity of several of the TAGs . 10 . 7554/eLife . 01751 . 016Figure 4 . Diet changes the quantity but not composition of sex-specific TAGs . ( A ) TAGs from individual ejaculatory bulbs of males raised on standard fly food ( n = 10 ) or cactus-banana supplemented food ( n = 9 ) were quantified using direct tissue thin layer chromatography . Each lane contains a single bulb . c: control band ( point of origin ) used for normalization; f2 and f3: fractions containing sex-specific TAGs . ( B ) The amount of TAGs in f3 from ejaculatory bulbs of males raised on standard food is significantly lower than supplemented food conditions ( Student’s t-test , two-tailed , p=0 . 0016 ) . TAGs found in f2 were not significantly different ( p=0 . 062 ) . Error bars indicate SEM . **: p<0 . 005; ns: not significant; a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 016 In some species of insects , males anoint the females with anti-aphrodisiacs during mating to suppress subsequent courtship from other males ( Zawistowski and Richmond , 1986; Bownes and Partridge , 1987; Wigby et al . , 2009; Yew et al . , 2009 ) . We hypothesized that sex-specific TAGs may play a similar role based on the sexually dimorphic pattern of expression and localization to a male sex organ . To test this prediction , a mate choice assay was used in which a naïve male was given a choice to court either a virgin female or a recently mated female ( Figure 5A ) . Male Drosophila courtship behavior consists of a sequence of stereotyped , quantifiable features , including wing vibration ( ‘singing’ ) , foreleg tapping , proboscis extension , and copulation ( Spieth , 1974 ) . Courtship initiation and copulation preferences were measured since both indicate male choice while the latter is also influenced by female rejection behavior . Males from D . arizonae and D . mojavensis were significantly more attracted to virgin females than recently mated females ( Figure 5B ) . Notably , significant levels of both TAGs and long OAcs were found on the anogenital regions of D . arizonae females shortly after mating but decreased by approximately 80% at 2–4 hr post-mating and were almost negligible at 8 hr post-mating ( Figure 5C; Figure 5—figure supplement 1 ) . Mated D . arizonae females became increasingly attractive over time , correlating with a decrease of the levels of TAGs and long OAcs on the cuticle ( Figure 5D ) . From 4 hr onwards , males showed no significant preference between mated and virgin females . Female remating was observed starting only at 8 hr post-mating ( Figure 5—figure supplement 2 ) . Taken together , these results show that the presence of male-transferred lipids on female cuticles is accompanied by a concomitant decrease in female attractiveness . 10 . 7554/eLife . 01751 . 017Figure 5 . Sex-specific lipids suppress male mating behavior . ( A ) To measure male courtship behavior , one male fly is placed with 2 females , one mated ( M ) , and one virgin ( V ) . ( B ) D . arizonae ( Dari ) and D . mojavensis ( Dmoj ) prefer to court virgin females over recently mated females ( n = 20 , Fisher’s exact test , p=0 . 00123; n = 31 , p=0 . 0105 ) . **: p<0 . 01; ns: not significant . A preference score of 1 indicates all males initiate courtship first with the virgin female; −1 indicates all males initiated courtship first with the mated female . ( C ) Levels of male-transferred TAGs and long OAcs on the female cuticle after first mating decreases by 2 hr post-mating . ( D ) Females are significantly less attractive for up to 2 hr after mating . By 4 hr , males do not exhibit significant preference for courting mated vs virgin females . **: Fisher’s exact test , p=0 . 00123; *: p=0 . 0256 . ( E ) D . arizonae males are more reluctant to initiate courtship with females perfumed with the contents of [1 . 25] ejaculatory bulb ( eb ) ( n = 27 , Fisher’s exact test , p=0 . 000624 ) or [0 . 5] eb ( n = 28 , p<0 . 0001 ) but not [0 . 25] eb ( n = 28 , p=0 . 176 ) . Extracts from immature male ebs were ineffective at inhibiting male courtship ( n = 28 , p=0 . 000389 ) . D . mojavensis and D . navojoa ( Dnav ) males also avoided virgin females perfumed with eb contents ( Dmoj: n = 21 , p=0 . 00160; Dnav: n = 21 , p<0 . 0002 ) . **: p<0 . 002; ***: p<0 . 0001 . C: solvent control; P: perfumed . ( F ) Suppression of D . arizonae courtship initiation is elicited only when TAGs and long OAcs are combined ( n = 28 , Fisher’s exact test , p<0 . 0001 ) . TAGs alone are ineffective ( n = 28 , p=0 . 593 ) . Long OAcs on their own could be attractive to males ( n = 28 , p=0 . 006 ) . *: p<0 . 05; ***: p<0 . 0001 . ( G ) D . arizonae copulation is suppressed in the presence of TAGs alone ( n = 28; Fisher’s exact probability test , p=0 . 0287 ) or TAGs combined with long OAcs ( n = 28; p<0 . 0001 ) , but not long OAcs alone ( n = 28; p=0 . 0543 ) . *: p<0 . 05; ***: p<0 . 0001 . A copulation choice score of 1 indicates all males copulated with solvent-perfumed females; −1 indicates all males copulated with TAG-perfumed females . ( H ) Perfuming with synthetic TAGs recapitulates copulation suppression . Oleic acid ( C18:1 ) -containing TAGs produced significant effects at high and low doses ( 750 ng: n = 21 , Fisher’s exact test , p<0 . 0001; 75 ng: n = 21 , p=0 . 00167 ) . Only the ( R ) -18:1 stereoisomer was bioactive ( 75 ng: n = 21 , p=0 . 00480 ) ; the ( S ) -18:1 stereoisomer did not elicit a significant behavioral response ( 75 ng: n = 21 , p=1 ) . *: p<0 . 01; **: p<0 . 002; ***: p<0 . 0001 . ( I ) Two combinations of TAGs produced synergistic effects on copulation suppression: oleic acid-TAG paired with stearic acid-TAG ( n = 21 , p=0 . 000139 ) and stearic acid-TAG paired with linoleic acid-TAG ( n = 19 , p=0 . 022 ) . The oleic acid-containing TAG is not bioactive at a dose of 37 . 5 ng/fly ( n = 20 , Fisher’s exact test , p=0 . 751 ) . *: p<0 . 05; ***: p<0 . 0005 . ( J ) UV-LDI MS spectra of females perfumed with TLC fractions or synthesized TAGs . Signals for TAGs and long OAcs are indicated in red and blue , respectively . *: C35:2 Pentatriacontadiene reference peak , m/z 527 . 5 [M + K]+ . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01710 . 7554/eLife . 01751 . 018Figure 5—figure supplement 1 . UV-LDI spectra of mated D . arizonae female anogenital regions reveal signals corresponding to male-specific TAGs ( red ) and long OAcs ( blue ) for up to 12 hr post-mating . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01810 . 7554/eLife . 01751 . 019Figure 5—figure supplement 2 . D . arizonae mated females are receptive to copulation starting at 8 hr after the first mating . Copulation choice indicates the percentage of females that copulated with males from all trials within 30 min . In some cases , males copulated with both virgin and mated female targets within the same trial . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 01910 . 7554/eLife . 01751 . 020Figure 5—figure supplement 3 . D . arizonae males avoided copulating with females perfumed with the contents of male ejaculatory bulbs ( eb ) at a concentration of [1 . 25] eb ( n = 27 , Fisher’s exact test , p=0 . 0003 ) or [0 . 5] eb ( n = 28 , Fisher’s exact test , p<0 . 0001 ) but not [0 . 25] eb ( n = 28 , p=0 . 176 ) . Eb extract from immature males was significantly less effective than extract from mature males ( n = 28; *: Fisher's exact test , p<0 . 0001 ) . D . mojavensis exhibited a tendency to avoid eb-perfumed females ( n = 21 , Fisher's exact test , p=0 . 0578 ) . D . navojoa males courted but seldom copulated within the 1 hr observation time . Only 4 out of 21 males copulated , preferring the non-perfumed females . **: p<0 . 001; ***: p<0 . 0001 . C: solvent control; P: perfumed . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 02010 . 7554/eLife . 01751 . 021Figure 5—figure supplement 4 . D . arizonae courtship and copulation preferences towards partly olfactory and gustatory perception-deficient virgin females that are perfumed with control or ejaculatory bulb extract . Leaving gustatory perception intact and removing the antennae and maxiliary palps , there is no effect on male courtship or copulation preference towards control females . Males still showed a preference for control females ( no ANT , MP; n = 20 , p<0 . 0001; and p<0 . 01 ) . Virgins without antennae , maxiliary palps , and having T3-5 segments of forelegs painted over with nail polish ( T3-5 ) , also did not affect male courtship choice ( no ANT , MP , T3-5; Fisher's exact test , n = 21 , p<0 . 01 ) , suggesting that female behavior did not affect courtship initiation . However , these females are less likely to allow copulation ( only 5 out of 21 copulated ) ( no ANT , MP , T3-5; n = 21 , p=0 . 343 ) . *: p<0 . 01; ***: p<0 . 0001 . C: solvent control; P: perfumed . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 02110 . 7554/eLife . 01751 . 022Figure 5—figure supplement 5 . D . arizonae courtship preferences towards virgin females perfumed with synthetic TAGs together with extract-purified long OAcs . Perfuming at the higher dose with the oleic acid ( C18:1 ) or linoleic acid ( C18:2 ) -containing . TAGs suppressed courtship initiation ( n = 21 , Fisher's exact test , p=0 . 00167; 750 ng: n = 21 , p= 0 . 00158 ) . Oleic acid-TAG was also effective at the 75 ng dose ( n = 21 , p=0 . 0126 ) paired with long OAcs , but not by itself . Only the ( R ) -18:1 stereoisomer was bioactive ( 75 ng: n = 21 , p=0 . 000139 ) ; the ( S ) -18:1 stereoisomer did not elicit a significant behavioral response ( 75 ng: n = 21 , p=1 ) . *: p<0 . 02; **: p<0 . 002; ***: p<0 . 0002 . C: solvent control; P: perfumed . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 02210 . 7554/eLife . 01751 . 023Figure 5—figure supplement 6 . D . arizonae courtship preferences towards virgin females perfumed with different combinations of synthetic TAG together with extract-purified long OAcs . Three combinations worked synergistically to inhibit courtship imitation: TAGs containing oleic acid ( C18:1 ) paired with those containing stearic acid ( C18:0; Fisher's exact test , n = 21 , p<0 . 0001 ) or palmitoleic acid ( C16:1; n = 21 , p=0 . 00292 ) and TAGs containing stearic acid ( C18:0 ) paired with those containing linoleic acid ( C18:2; n = 19 , p=0 . 00023 ) . **: p<0 . 005; ***: p<0 . 0005 . C: solvent control; P: perfumed . DOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 023 The reluctance of males to court mated females could be the result of female rejection behavior or the presence of additional transferred compounds . To determine whether male-transferred lipids account solely for the loss of female attractiveness , we tested the preference of males from D . arizonae , D . mojavensis , and D . navojoa when presented with a choice between virgin females perfumed with evaporated solvent or extracts from ejaculatory bulbs . In these species , fewer males chose to initiate courtship with extract-perfumed females ( Figure 5E ) . D . arizonae was also significantly less likely to copulate with bulb-perfumed females , whereas D . mojavensis showed a tendency to avoid perfumed females ( Figure 5—figure supplement 3 ) . D . navojoa only rarely copulated under these experimental conditions , likely because males prefer to court in the presence of more flies . However , the few males that copulated preferred to do so with control females . These results suggest that the use of these sex-specific TAGs as anti-aphrodisiacs is conserved across several species . Further analysis of D . arizonae indicated that the higher the concentration of the extract , the greater the aversion exhibited by males . A minimum dose of extract from approximately 0 . 5 ejaculatory bulbs was needed to achieve a significant behavioral effect ( Figure 5E ) . Copulation choice was similarly affected ( Figure 5—figure supplement 3 ) . Ejaculatory bulb extract from immature males ( and therefore , containing negligible amounts of TAGs and long OAcs ) did not suppress male courtship ( Figure 5E; Figure 5—figure supplement 3 ) . Thus , courtship inhibition is not a generalized aversion to other compounds extracted from ejaculatory bulb tissue . Since virgin females were used , female rejection behavior was not a major contributing factor to copulation choice . Furthermore , it is unlikely that female rejection behavior was triggered by females’ sensory feedback from perfuming . D . arizonae males continued to avoid mating with perfumed females from which the major olfactory and gustatory organs have been removed ( Figure 5—figure supplement 4 ) . Taken together , the presence of male-specific long OAcs and TAGs on females was sufficient to fully recapitulate the loss of attractiveness observed in recently mated females . To determine whether TAGs and long OAcs function synergistically with each other , we examined male response to females perfumed with these two classes of compounds separately and together ( Figure 3—figure supplement 1 shows TLC separation ) . Perfuming with both long OAcs and TAGs strongly suppressed courtship initiation and copulation ( Figure 5F , G ) . The presence of TAGs alone significantly reduced the likelihood of copulation ( Figure 5G ) . Interestingly , long OAcs had an attractive effect when used alone ( Figure 5F ) . These results suggest that both TAGs and long OAcs are needed to suppress courtship initiation , whereas TAGs are important for discouraging later stages of courtship . We next tested whether individual TAG species plays a role in suppressing courtship . We synthesized four of the postulated TAGs along with a TAG containing stearic acid as racemic mixtures ( Table 1 ) . Additionally , individual ( R ) - and ( S ) -enantiomers were synthesized for a TAG species containing oleic acid , the most abundant of the sex-specific TAGs ( Mori , 2012 ) . Males were given a choice of solvent-perfumed females or females perfumed with a single TAG species together with long OAcs . Under these conditions , TAGs containing oleic acid or linoleic acid significantly suppressed courtship initiation ( Figure 5—figure supplement 5 ) . However , only the former was capable of suppressing copulation as well and at a dose of 75 ng per female ( Figure 5H ) . Moreover , only the ( R ) configuration of this TAG was bioactive ( Figure 5H; Figure 5—figure supplement 5 ) . Perfuming females with the ( S ) -enantiomer showed no significant effect on male choice , signifying that courtship aversion is specific to the stereochemistry of the TAG and not due to general avoidance of a foreign molecule or masking of female aphrodisiacs . 10 . 7554/eLife . 01751 . 024Table 1 . Synthetic TAGs used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 01751 . 024Calculated [M + K]+Fatty acyl components* , †Long chain fatty acid531 . 31C16:1 ( n-7 ) -C5:1-C5:1Palmitoleic acid533 . 32C16:0-C5:1-C5:1Palmitic acid557 . 32C18:2 ( n-6 ) -C5:1-C5:1Linoleic acid559 . 34C18:1 ( n-9 ) -C5:1-C5:1Oleic acid561 . 36C18:0-C5:1-C5:1Stearic acid559 . 34C18:1 ( n-9 ) -C5:1-C5:1 ( R isomer ) Oleic acid559 . 34C18:1 ( n-9 ) -C5:1-C5:1 ( S isomer ) Oleic acid*Synthesized as racemic mixtures unless otherwise noted . †Notation indicates number of carbons followed by number of double bonds for each fatty acyl component; double bond position indicated in brackets . To examine the possibility that TAGs function synergistically with each other , we paired several combinations of synthetic TAGs together with long OAcs . At a dose of 37 . 5 ng per fly , none of the TAGs were effective by themselves . However , two combinations of TAGs reduced copulation: 1-steroyl-2 , 3-ditigloyl glycerol together with either 1-oleoyl-2 , 3-ditigloyl glycerol or 1-linoleyl-2 , 3-ditiglyoyl glycerol ( Figure 5I ) . Both combinations also inhibited courtship initiation ( Figure 5—figure supplement 6 ) . Additionally , courtship initiation was affected by the combination of long OAcs with 1-palmitoleyl-2 , 3-ditiglyol and 1-oleyl-2 , 3-ditiglyol ( Figure 5—figure supplement 6 ) . Thus , low amounts of several TAG species , inactive on their own , can act in synergy with each other to suppress male courtship and copulation . It is notable that none of the TAG combinations tested were effective without long OAcs despite our finding that a mixture of TAGs purified from extract was by itself sufficient to deter male attraction . It may be the case that a combination of several different TAG species is needed for courtship inhibition without the presence of long OAcs .
Synergistic interactions between multiple sensory cues in chemical communication have been described in several forms , in many cases involving a combination of enantiomers ( Borden et al . , 1976 ) or a blend of molecules from the same chemical class ( Lecomte et al . , 1998; Srinivasan et al . , 2008 ) . Famously , the honey bee queen uses a blend of least nine different fatty acids and alcohols secreted by multiple glands ( Keeling et al . , 2003 ) The parasitic wasp Lariophagus distinguendus was recently reported to use TAGs together with a branched alkane to promote mating behavior ( Kühbandner et al . , 2012 ) . Food odors are also known to synergize with aggregation pheromones in beetles ( Lin et al . , 1992 ) and with sex pheromones in Drosophila ( Grosjean et al . , 2011 ) . Pheromonal synergism between completely different classes of molecules is rare and may be a mechanism to increase combinatorial complexity . Interestingly , only some of the D . arizonae TAGs appear to play a role as an anti-aphrodisiac . The quiescent stereoisomers could be used as potential future chemical cues ( Niehuis et al . , 2013 ) . Sex-specific TAGs found on desert-adapted drosophilids are the first examples of natural products bearing combinations of branched and linear fatty acyl side chains . Conventional naturally occurring TAGs found in plant oils and animal fat typically consist of linear fatty acyl moieties that have 16 , 18 , or 20 carbons ( Nelson et al . , 2008 ) . Although medium and short chain fatty acyls have been found in TAGs from , respectively , whale blubber ( Litchfield et al . , 1971 ) and bovine milk fat ( Breckenridge and Kuksis , 1968 ) , they are only in combination with linear and even-numbered carbon acyls . In contrast , sex-specific TAGs from flies exhibit an unusual combination of short and long chain acyl components with odd and even numbers of carbons . To what extent can these unconventional structures be attributed to diet ? Previous and current work has shown that altering the diet of desert-adapted flies results in a quantitative change in hydrocarbons and sex-specific triglycerides ( Figure 4 ) ( Toolson et al . , 1990; Etges et al . , 2006; Etges et al . , 2009; Yew et al . , 2011 ) . Notably , many of the substrates on which the drosophilids subsist and oviposit contain compounds that can be toxic for other animals . For example , desert-adapted D . arizonae and D . mojavensis feed on fermenting cacti , which have high levels of triterpene glycosides , medium chain fatty acids , and sterol diols , compounds which can serve as toxic plant defense chemicals ( Fogleman and Danielson , 2001 ) . Similarly , Drosophila subquinaria and Drosophila recens are found exclusively on mushrooms that are rich in secondary metabolites such as isoprenoids and fatty acids in a variety of lengths , from 4 to 26 carbons ( Wandati et al . , 2013; Pedneault et al . , 2008 ) but also contain high levels of toxic compounds like alpha-amanitin ( Jaenike et al . , 1983; Courtney et al . , 1990 ) . Slime fluxes , on which D . virilis and D . robusta can be found , have large bacterial communities that can be inhospitable to other drosophilids ( Carson and Stalker , 1951; Powell , 1997 ) . Bacterial wetwood infections have been shown to produce acetate , butyrate , valerate , hexanoate , and propionate ( Ward and Zeikus , 1980 ) , which could be directly incorporated into the TAGs or serve as precursors for branched or linear fatty acids , as has been observed for nitidulid beetles ( Carpophilus spp ) ( Bartelt and Weisleder , 1996 ) . The ability of drosophilids that produce sex-specific TAGs to thrive on these specialized substrates alludes to the possibility that enzymes used for detoxification may have been adapted for TAG synthesis . Notably , cytochrome P450 monooxygenases have been identified in other insects as playing a crucial role in detoxification and cuticular lipid synthesis ( Li et al . , 2004; Qiu et al . , 2012 ) . In bark beetles , it has been suggested that a key cytochrome P450 enzyme used in pheromone synthesis was previously used for detoxification ( Blomquist et al . , 2010 ) . Alternatively , conservation of TAG expression may be more related to phylogenetic effects than functional adaptation ( Oliveira et al . , 2011 ) . A second pathway for TAG synthesis relies on de novo production of precursors . Each of the desert-adapted drosophilids tested in this study are still able to produce TAGs despite being raised on standard laboratory fly media . Thus , desert-adapted drosophilids are capable of using precursors from the environment and synthesizing the components de novo though significant quantitative differences are found for some TAG molecules . Based on these observations , we expect that lab-raised flies are likely to have quantitative differences in lipid profiles compared to natural populations because of the differences between the natural diet and a highly simplified lab diet . Both pathways for precursor synthesis are used by numerous Coleoptera beetle species for the production of aggregation pheromones ( Tillman et al . , 1999 ) . The ability to utilize different production pathways may enable insects to switch host plants while preserving conspecific signaling . Elucidation of the biochemical pathways underlying TAG synthesis is needed to better understand whether dietary sequestration or de novo production is preferentially used by desert-adapted drosophlids and to determine the ancestral state of TAG production . In this study , we have identified an unusual chemical class of pheromones in the form of triacylglycerides and described their function as anti-aphrodisiacs . Specialized TAGs were prevalent across other Drosophila species and may also be found in other insect orders and have other functions . For example , triolein in the fire ant , Solenopsis invicta , acts as a brood pheromone and has application as bait when combined with toxicants ( Bigley and Vinson , 1975 ) . The chiral 1 , 2-dioleyl-3-palmitolyl glycerol is also a brood pheromone in the honey bee , Apis mellifera ( Koeniger and Veith , 1984 ) . Taken together , triacylglycerides represent a broadly conserved and largely overlooked class of pheromones . Ultimately , to understand the evolutionary origin of these unusual molecules , it will be important to determine the behavioral function of sex-specific TAGs in other species and the underlying biosynthetic pathways . Using a broad range of analytical methods for chemical profiling will expand the detectable range of chemical classes used in communication . Correlating gene expression in the ejaculatory bulb across multiple TAG-expressing species will enable us to identify candidate biosynthetic enzymes and to provide molecular markers that will allow the evolution and function of this surprising chemical phenotype to be traced .
D . arizonae , D . aldrichi , and D . navojoa in this study were wild caught from Las Bocas , Sonora , Mexico , in March 2009 . D . mojavensis were caught from Santa Catalina Island , California and D . wheeleri from Punta Onah , Sonora , Mexico , in November 2007 by sweep netting over fermenting bananas . All stocks are available from WJE at the University of Arkansas . Drosophila melanica and D . robusta were obtained from UC San Diego Drosophila Stock Center . D . virilis , D . hydei , D . americana , D . buzzatii , and D . montana were obtained from Ehime-Fly Drosophila Stocks of Ehime University . Flies were reared on autoclaved yeast-sucrose-cornmeal-agar food or food supplemented with added banana ( ca . 110 g/20 half-pint bottles ) and cactus powder ( ca . 2 . 3 g/20 half-pint bottles; Nopal cactus powder , Oro Verde , Mexico ) in a 23 . 3°C room on a 12:12 hr light:dark cycle at 60% humidity level . Adult flies were transferred to fresh food contained in half pint bottles every 3–5 days for female oviposition . After pupal eclosion , all emerging adults were sexed under CO2 every 5 hr during the day . Virgin females were grouped in groups of 20–30 individuals in a new food vial , whereas males were isolated individually to keep them socially naive . The flies were allowed to reach sexual maturity ( 8–10 day old ) at 23 . 3°C before behavioral analysis . Ejaculatory bulbs from ca . 500 mature males were dissected and soaked in hexane in a 1 . 8 ml glass vial with a Teflon-lined cap ( Wheaton , Millville , New Jersey , USA ) for 20 min . The extract was placed in a clean glass vial , evaporated with N2 , and stored at −20°C until analysis . To obtain individual fractions , extract was overlaid onto a 10 × 10 cm thin layer chromatography silica plate ( Merck , Darmstadt , Germany ) and developed with a solution of hexane/diethyl ether/acetic acid ( 90:9:1; per vol ) . Silica from fractions containing the male-specific TAGs and acetates were scraped into a disposable borosilicate glass Pasteur pipette ( 15 cm length ) stuffed with glass wool fiber ( Pall Corporation , Ann Arbor , Michigan , USA ) and eluted with hexane . The contents were divided into 3 aliquots , evaporated under a gentle stream of N2 , and kept at −20°C until analysis . Individual ejaculatory bulbs were dissected from 9-day-old D . arizonae males raised on either standard yeast-sucrose-cornmeal-agar food for two generations or cactus and banana-supplemented fly food . Each bulb was placed in its own lane on a 10 × 10 cm thin layer chromatography silica plate ( Merck , Darmstadt , Germany ) . To release the contents and ensure complete elution , each bulb was first gently punctured then overlaid 10 times with 0 . 5 µl of hexane , allowing the solvent to fully evaporate between each solvent application . The plate was run in a solution of hexane/diethyl ether/acetic acid ( 90:9:1; per vol ) , developed with primuline ( 0 . 1% in 20% acetone ) , and imaged with the Gel Doc XR system ( Bio-Rad Laboratories , Inc . , USA ) using Quantity One software ( v 4 . 5 . 2 , Bio-Rad Laboratories , Inc . , USA ) . Intensities of the bands in the image were processed and analyzed using ImageJ ( v 1 . 43 , NIH , USA ) to produce a plot of peaks according to brightness of the bands . Intensity values of the fractions were normalized to the intensity of the control band at the origin . 200 µl of methanolic HCl ( Supelco Analytical , Sigma–Aldrich Co . , St . Louis , MO ) was added to dried , crude whole fly extract from about 500 flies and incubated for 1 . 5 to 2 hr at 60°C with occasional vortexing . After the acid-based catalysis , the reaction was cooled on ice , followed by the addition of 50 µl of water and 50 µl of hexane , and brief vortexing . The hexane layer ( which contains the fatty acid methyl esters ) was removed for GCMS analysis . Concurrently , synthetic standards containing 5 carbons ( tiglic acid , trans-2-pentenoic acid , trans-3-pentenoic acid , and 3-methyl crotonic acid [TCI Chemicals Co . , Tokyo , Japan] ) were treated with the same reactions . Methyl angelate ( TCI Chemicals Co . ) was not treated . The long OAc fraction obtained from TLC was derivatized with 200 µl of 0 . 2M KOH in 80% isopropanol for 2 hr at 60°C . After incubation , 50 µl of 1M HCl was added and evaporated under a gentle stream of N2 . 200 µl of hexane was added prior to analysis by DART MS and GCMS . Synthesized TAGs used in this study are shown in Table 1 . Synthesis procedures were previously described ( Mori , 2012 ) . UV-LDI MS analysis and the procedures for preparing the flies were described in detail previously ( Yew et al . , 2009 , 2011 ) . Measurements were performed on a QStar Elite ( ABSciex ) orthogonal time-of-flight mass spectrometer equipped with an intermediate pressure oMALDI2 source and a N2 laser ( λ = 337 nm , 40 Hz repetition rate , 200 µm beam diameter , pulse duration 3 ns ) . Ions are generated in a buffer gas environment using 2 mbar of N2 . Individual flies were attached to a cover slip with adhesive tape and mounted onto a custom-built sample plate . During data acquisition , the anogenital region was irradiated for 30 s , corresponding to 1200 laser shots . Mass accuracy for the mass spectrometer was approximately 20 ppm . Elemental composition and number of double bonds are predicted from exact mass measurements . CID spectra were acquired on a linear ion trap mass spectrometer ( Thermo Fisher Scientific LTQ , San Jose , CA ) that has been modified to allow ozone-induced dissociation ( OzID ) experiments ( Thomas et al . , 2008; Brown et al . , 2011 ) . Methanolic solutions of lipid samples ( ca . 10 µM ) in the presence of either sodium or lithium acetate ( ca . 10 mM ) were infused into the electrospray ionization source of instrument with a flow rate of 5 µl/min; a spray voltage of 5 kV; a capillary voltage of 21 V; a tube lens voltage 125 V; and the temperature of the heated transfer-capillary was set to 275°C . To acquire CID spectra of triacylglycerol alkali metal adduct ions , the [M + Na]+ or [M + Li]+ were isolated with an isolation width of 1–2 Da and a normalized collision energy of 32–35% was applied . Individual TAG species generated by electrospray ionization of the extract was mass-selected within an ion trap mass spectrometer where they were exposed to ozone vapor . The resulting gas-phase ion–molecule reaction facilitates targeted oxidative dissociation of carbon–carbon double bonds present in the acyl chains . Fragmentation of the ozonide leads to formation of characteristic aldehyhde and Criegee ions with a mass indicative of the positions of each double bond . To determine the carbon–carbon double bond positions within TAGs , alkali metal adduct ions were mass-selected within the modified linear ion trap mass spectrometer ( see above ) and allowed to react with ozone seeded in the helium buffer gas ( Thomas et al . , 2008 ) . To acquire OzID spectra , ions were isolated in the absence of collision energy and the reaction time ( set by adjusting the activation time parameter within the XCalibr instrument control software ) was typically 5–10 s per scan . OzID spectra reported here correspond to the average of at least 50 scans . Reaction of ozone with carbon–carbon double bonds in the TAG acyl chains produces fragment ions that identify their position within the chain . Location of each double bond is indicated by the traditional nomenclature ‘n-x’ where ‘n’ refers to the number of carbon atoms in the chain and subtracting ‘x’ provides the location of the double bond ( i . e . , x represents the position of the double bond with respect to the methyl terminus ) . Prepared extracts were re-dissolved in 60 µl of hexane and transferred into GCMS vials ( Supelco ) . Analysis was run in a 5% phenyl-methylpolysiloxane ( DB-5 , 30 m length , 0 . 32 i . d . , 0 . 25-μm film thickness , Agilent ) column and GCMS QP2010 system ( Shimadzu ) with an initial column temperature of 50°C for 2 min and increment to 300°C at a rate of 15°C/min in splitless mode . Fly wings were detached and washed in a solution of chloroform/methanol ( 2:1 , vol/vol ) to ensure existing cuticular hydrocarbons were completely removed . The wings were then attached onto a MS customized sample plate with double-sided tape . The fraction-containing vials were re-dissolved with 20 µl of hexane each and 2 µl from each vial was pipetted onto different wings . The sample plate was placed into the UV-LDI MS instrument and the fractions analyzed using the same parameters for UV-LDI MS analyses of intact flies . For more details see Yew et al . , 2011 . Analysis was performed using the following ion source settings: the gas heater was set to 200°C; the glow discharge needle was set at 3 . 5 kV . Electrode 1 was set to +150 V and electrode 2 was set to +250 V . Helium gas flow was set to 2 . 5 L/min . Under these conditions , TAGs were detected as [M + NH4]+ molecules and long OAcs were detected as [M + H]+ molecules . Clean borosilicate glass capillaries ( World Precision Instruments ) were used for sampling the solution . The capillary was placed in the DART stream for 5 s . Polyethylene glycol ( Sigma–Aldrich ) was used as calibrant . Analysis was done with MassCenter ( version 1 . 3 . 0 . 1 ) ( JEOL ) program . To generate mated female targets in assays where males choose virgin or recently mated females , naive males were first paired with virgin females in 35 × 10 mm tissue culture dishes ( Nunclon , Denmark ) and observed for copulation . Immediately after copulation , the dish containing the copulated pair was placed on ice to anaesthetize the flies temporarily . In clean tissue culture dishes , a mated female and a virgin female are placed into each dish and allowed to recover for 45 min to an hour before performing the courtship assay . New socially naive males are aspirated into each dish and assayed for first courtship event lasting more than 5 s , including wing vibration , foreleg tapping , proboscis extension , and copulation . At least 20 trials of each set of experiments were performed . For perfuming assays , virgin females were perfumed by lightly vortexing eight individuals in a 1 . 8-ml glass vial with Teflon-lined caps containing extract or evaporated solvent ( control ) for 3 bouts of 20 s each with 5–10 s rest between each bout . A fly from each vial was checked for perfuming efficiency using UV-LDI MS , whereas the other 7 were used for the courtship assay . Approximately , 20% of contents from the extract were transferred to all eight flies during perfuming ( Billeter et al . , 2009 ) . Therefore , each fly was perfumed with the equivalent of 2 . 5% of the total concentration from the vial . In assays using synthetic TAGs ( Table 1 ) , an amount of 750 ng or 75 ng per fly was used . Where TAGs are paired , each female was perfumed with 37 . 5 ng of each TAG for a total 75 ng per fly . Figure 5J shows the spectra of perfumed females , indicating that the amounts perfumed are moderate compared to mated females . The phylogram was generated using Mesquite 2 . 75 ( Maddison and Maddison , 2011; http://mesquiteproject . org ) . Distances and primary diets were based on the previous studies ( Throckmorton , 1975; Lemeunier et al . , 1986; Spicer , 1991; Jeffs et al . , 1994; Russo et al . , 1995; Powell , 1997; Shoemaker et al . , 1999; Remsen and O’Grady , 2002; Gao et al . , 2007; Reed et al . , 2007; Routtu , 2007; Flores et al . , 2008; Mcdermott and Kliman , 2008; Oliveira et al . , 2012; Curtis et al . , 2013 ) . | For animals , the ultimate purpose of life is to have sex , as nothing is more important than passing down your genes to future generations . A wide range of strategies are therefore employed throughout nature to maximize the chances of sexual success , from ostentatious courtship rituals to the subtle subliminal signals sent out using chemicals called pheromones . Plants and animals release pheromones to influence the behavior of other plants and animals , often without the recipient being aware of it . Hundreds of different insect pheromones have been discovered . Fruit flies release a number of different pheromones , all with similar chemical structures . Now , Chin et al . have discovered that male flies belonging to several species of fruit fly that live in the desert release chemicals called triacylglycerides ( TAGs ) , which are commonly used for energy storage by many organisms as pheromones . During sex , the male fly rubs the TAGs onto the body of the female , which makes her less attractive to other male flies for several hours , thus increasing his chances of parenthood and passing his genes to future generations . TAGs are also found in other insect species , but have been largely overlooked as pheromones . Moreover , the TAGs discovered by Chin et al . have an unusual structure , not previously seen in nature , which may result from the diet of fermenting cacti the desert-dwelling fruit flies enjoy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"biochemistry",
"and",
"chemical",
"biology"
] | 2014 | Sex-specific triacylglycerides are widely conserved in Drosophila and mediate mating behavior |
Animals and humans need to move deftly and flexibly to adapt to environmental demands . Despite a large body of work on the neural control of walking in invertebrates and vertebrates alike , the mechanisms underlying the motor flexibility that is needed to adjust the motor behavior remain largely unknown . Here , we investigated optomotor-induced turning and the neuronal mechanisms underlying the differences between the leg movements of the two body sides in the stick insect Carausius morosus . We present data to show that the generation of turning kinematics in an insect are the combined result of descending unilateral commands that change the leg motor output via task-specific modifications in the processing of local sensory feedback as well as modification of the activity of local central pattern generating networks in a body-side-specific way . To our knowledge , this is the first study to demonstrate the specificity of such modifications in a defined motor task .
An animal's motor system faces complex demands in order to ensure proper function and survival . Such demands arise either from the intrinsic need to find food , or a mating partner , or the extrinsic needs to avoid predators or simply overcome obstacles . The investigation of the neural mechanisms underlying the generation of locomotion has unraveled considerable detail on how those neural networks that produce the basic motor outputs for swimming , flying and walking of invertebrates and vertebrates are organized and operate ( Grillner et al . , 2008; Goulding , 2009; Lehmann , 2004 ) . At the level of the locomotor organs , a basic motor output is the result of the activity of local central pattern generating networks ( CPGs ) , and its adjustment through sensory feedback from the locomotor organs ( Büschges and Gruhn , 2008; Pearson , 2008 ) . Particularly important for the control of the step cycle in this context is load ( Schmitz and Stein , 2000; Duysens et al . , 2000 ) . The activity of the CPGs is initiated and modified depending upon the behavioral tasks the animal faces . Evidence suggests that descending signals from brain centers such as reticulospinal neurons in the vertebrate hind brain/basal ganglia ( Stephenson-Jones et al . , 2011 ) or the head ganglia of arthropods with the putative arthropod basal ganglia homolog ( Strausfeld and Hirth , 2013 ) , the central complex in the cerebral ganglion ( Strauss and Heisenberg , 1993; Guo and Ritzmann , 2013; Bidaye et al . , 2014; Martin et al . , 2015 ) contribute to both the basic drive and its modulation ( Brocard et al . , 2010; Shik et al . , 1969; Martin et al . , 2015 ) . How these descending signals influence the motor activity at the level of the local neural networks that control the locomotor organs , however , is only beginning to be understood , in particular for walking , the most ubiquitous form of locomotion in terrestrial animals ( Ridgel et al . , 2007; Dyson et al . , 2014; Martin et al . , 2015 ) . When changing course , all walking animals need to change the kinematics of each leg substantially in order to separate the activity between the leg and body segments and produce the appropriate locomotor movements on the two body sides ( Jander , 1985; Gruhn et al . , 2009; Rivera et al . , 2006; Dürr and Ebeling , 2005; Mu and Ritzmann , 2005; Jindrich and Full , 1999; Strauss and Heisenberg , 1990; Musienko et al . , 2012 ) . In a six-legged curve walking insect , for example , outside legs push the animal , while inside legs pull the animal into the direction of the curve ( Jindrich and Full , 1999; Dürr and Ebeling , 2005; Gruhn et al . , 2009 ) , sometimes even reversing the stepping direction in single inside legs ( Gruhn et al . , 2009; Cruse et al . , 2009 ) . Currently , little insight exists , into how the nervous system generates such motor flexibility ( Guo and Ritzmann , 2013; Huang et al . , 2013; Hellekes et al . , 2012; Ridgel et al . , 2007; Fagerstedt et al . , 2001; McClellan , 1984; Martin et al . , 2015 ) . It seems clear that sensory input from vestibular/antennal or optical areas is processed in the brains of vertebrates and insects alike , and transformed into descending neuronal signals that induce changes in the activity of the local networks which produce the ongoing motor patterns ( Guo and Ritzmann , 2013; Mu and Ritzmann , 2008a; Ridgel et al . , 2007; Strauss and Heisenberg , 1993; Huang et al . , 2013; Musienko et al . , 2012; McClellan , 1984; Martin et al . , 2015 ) . The nature of these changes that generate body-side-specific motor output , however , is still largely unknown . In the present study we have therefore investigated the neural mechanisms underlying the generation of the motor output during curve walking in an insect . We first analyzed stepping activity as well as the activity and timing of leg muscles in tethered , intact animals , walking freely on a slippery surface during optomotor-induced curved walking . In a next step , we investigated the underlying mechanisms by studying the local processing of load signals on the respective inside and outside of the curve . Finally , we investigated the centrally generated local motor output in a deafferented preparation and tested if the observed behavioral changes involved also changes in the activity of local central pattern generating networks . Our results show that the leg motor output in optomotor induced curve walking insects results from body-side-specific modulation of the local , segmental processing of sensory feedback and the modulation of the activity of CPGs that control the individual leg joints .
During curved walking in insects , outside legs usually perform slightly modified forward steps , while the inside leg pulls the animal into the direction of the curve , and may produce forward , sideward and backward steps independent of the conditions under which walking is studied ( Mu and Ritzmann , 2005; Jindrich and Full , 1999; Szczecinski et al . , 2014; Strauss and Heisenberg , 1990; Dürr and Ebeling , 2005; Gruhn et al . , 2009; 2011 ) . To understand the neuronal basis for these differences , we first studied turning of intact stick insects on a slippery surface ( Gruhn et al . , 2006 ) . Despite some differences to walking on solid ground that will be discussed below , this arrangement excludes inter-leg influences from mechanical coupling through the ground , and a passive entrainment of leg activity through that of neighboring legs . Quantitative analysis of 14 inside and outside curve walking sequences from 10 intact animals showed that the period of front leg ( FL ) inside steps ( Pin=0 . 83s; SD=0 . 21 ) between animals was significantly shorter than that of FL outside steps ( Pout=1 . 21s; SD=0 . 37 ) in 11 out of 14 sequences ( p<0 . 05 ) . Likewise , the average period of middle leg ( ML ) inside steps ( Pin=0 . 76s; SD=0 . 15 ) was significantly shorter than that of ML outside steps ( Pout=1 . 13s; SD=0 . 28 ) ( p<0 . 01 , Figure Figure 1a ) . Neither the stepping periods of the inside , nor those of the outside FLs and MLs were significantly different from one another ( pin =0 . 43s , pout= 0 . 6s , Figure 1a ) . We also looked for changes in motor output at the level of the muscle activity between the inside and outside legs . The only clear difference in the average EMG activity was observed in the most proximal leg joint , the thorax-coxa-joint , through which movements of a leg along the body axis are controlled . During inside stepping , the muscles retractor coxae and its antagonist , the protractor coxae , both served as either stance or swing muscles . Figures 1bi and bii show examples of ML pro- and retractor coxae EMG activity during inside and outside walking , and the averages of the rectified and smoothed EMG activity from 182 inside ( Figure 1ci ) and 204 outside steps ( Figure 1cii ) from five animals . Interestingly , the protractor activation in this case often also overlapped with that of the retractor , which suggests co-contraction and matches sideward steps observed during inside stepping . In addition , depressor activity on the outside was slightly prolonged ( data not shown ) . Altogether , these findings corroborate the relative independence of the CPGs for each joint , and between hemiganglia that has been long known ( Büschges et al . , 1995; Dürr and Ebeling , 2005 ) . 10 . 7554/eLife . 13799 . 003Figure 1 . Differences between inside and outside middle leg ( ML ) activity in the intact tethered stick insect that is turning on the slippery surface . ( a ) Significant difference between the periods of inside ( orange ) and outside ( yellow ) turning of the front ( striped ) and middle ( solid ) leg , and no significant differences between the periods of front and middle leg stepping on either the inside or the outside; asterisks mark significance levels: * p<0 . 05 , ** p>0 . 01 . ( bi–ii ) Original and rectified EMG traces of protractor ( Pro ) and retractor coxae ( Ret ) with stance and swing ( grey shaded bars ) monitored electrically . The example in bi ( orange ) shows examples of backward steps during inside stepping with the retractor being active mostly in swing . Asterisks ( * ) mark very weak crosstalk between the two channels; bii ( yellow ) , original and rectified EMG traces of the same muscles during outside steps . Here the retractor is always active in stance . ci and cii . Protractor and Retractor latency of the first EMG spike during inside ( ci , orange ) and outside ( cii , yellow ) steps from N=5 animals . Note the bimodal distribution during inside steps due to the occurrence of forward and backward steps . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 003 How do these drastic differences in leg kinematics , and the activity in the thorax-coxa leg joint , come about ? We examined whether sensory feedback encoding load , which is known to be crucial for the control of stance motor activity during forward and backward walking , is controlling leg kinematics during turns ( Akay et al . , 2007 ) . During outside stepping sequences , load stimuli either initiated or enhanced retractor activity ( Figure 2; 359 stimuli during outside walking in 13 animals ) . The timing of the load stimuli with respect to the front leg steps was distributed uniformly around FL stepping activity ( Figure 2c and Figure 2—figure supplement 1 ) . Processing of loading signals on the outside of curve walking animals is thus similar to straight stepping and ensures leg retraction throughout stance ( Akay et al . , 2007; Borgmann et al . , 2009 ) . On the inside , however , the response to a loading stimulus was markedly different . Figure 3a shows an example of mesothoracic protractor and retractor MN activity during application of load stimuli while the animal was standing , and during inside steps of the FL ipsilateral to the recording site . 10 . 7554/eLife . 13799 . 004Figure 2 . Influence of local load stimuli on mesothoracic motor activity during outside turning . ( a ) Mesothoracic pro- and retractor activity ( 2nd and 3rd row , resp . ) before , during and after a front leg outside walking sequence ( FL stance begin marked in top row event channel ) and simultaneous application of load stimuli to the mesothoracic leg stump ( bottom trace ) . ( bi ) Peristimulus time histogram ( PSTH ) showing the increase in mesothoracic retractor MN activity in response to local load stimuli during FL outside stepping; note the increased background activity . ( bii ) PSTH of retractor activity timed to the begin of the CS stimulus ramp in the quiescent animal ( control ) . ( c ) Increased retractor MN activity upon CS stimuli did not depend on the phase of the CS stimulus within the FL step cycle during outside steps for the animal shown in a and b; 0 on the y axis marks the normalized baseline response in the 100ms before the CS stimulus . Figure 2—figure supplement 1 shows the same data for all 13 animals . FL: front leg , ML: middle leg , pro: protractor , ret: retractor . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 00410 . 7554/eLife . 13799 . 005Figure 2—figure supplement 1 . Distribution of response amplitudes of retractor MNs to CS stimuli throughout the FL step cycle during outside steps from all 13 animals , normalized to the activity in the 100ms preceding the CS stimulus . Retractor activity mostly increased over already ongoing retractor activity without any phase preference . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 00510 . 7554/eLife . 13799 . 006Figure 3 . Influence of local load stimuli on mesothoracic motor activity during inside turning . ( a ) Mesothoracic pro- and retractor activity ( 2nd and 3rd row , resp . ) before and during a front leg inside walking sequence ( front leg [FL] stance begin marked in top row event channel ) , and simultaneous application of load stimuli to the mesothoracic leg stump ( bottom trace ) . Four example responses ( grey shaded areas ) with expanded time resolution are shown below , the first one during quiescence , the other three during inside steps of the ipsilateral FL . ( bi–iii ) PSTHs showing the forms of mesothoracic retractor MN response to local load stimuli during FL inside stepping: bi , peristimulus time histogram ( PSTH ) of retractor activity timed to the begin of the CS stimulus ramp in the quiescent animal . bii , retractor activation upon load stimulus; biii , termination of retractor activity ( activation of protractor activity is not shown as PSTH ) . ( c ) Distribution of response strength of retractor MNs to CS stimuli throughout the FL step cycle during inside steps with retractor activation from the animal shown in a; note that there is no phase preference for either an increase or a decrease in retractor activation compared to controls , this was also found for all other animals ( Figure 3—figure supplement 1 ) . ( d ) Example for no response to a loading stimulus . ( e ) Phase plot showing the distribution of CS stimuli in the FL step cycle that lead to protractor activation from N=7 animals; no significant phase preference was detectable . ( f ) Phase plot showing the distribution of CS stimuli in the FL step cycle that did cause neither re- nor protractor activation from N=10 animals; a significant phase preference was detectable at 0 . 89 . FL: front leg , ML: middle leg , pro: protractor , ret: retractor . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 00610 . 7554/eLife . 13799 . 007Figure 3—figure supplement 1 . Distribution of response amplitudes of retractor MNs to CS stimuli throughout the FL step cycle during inside steps with retractor activation from all 13 animals ( note that 9a and 9b constitute data from two files from one animal ) . Maximum retractor activation in a time window of 150 ms after CS stimulation was normalized to the average maximum response during control stimulations . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 007 Upon start of FL inside stepping , the mesothoracic protractor MNs showed an increase in activity . In contrast to the standing animal , however , no stereotypical activation of retractor activity was observed upon load stimuli ( n=617 ) . Instead , load stimuli either started or terminated retractor , and vice versa , protractor activity . Activation of the retractor MNs in variable strength compared to control upon load stimuli was observed in all 13 animals ( 16–100% of the stimuli/animal ) , termination of retractor , and activation of protractor MNs was observed in 54% of the cases ( 7 out of 13 animals , 2–25% of the stimuli ) . In addition , in 10 out of 13 animals , and on average upon 18 . 8% of the stimuli ( SD=15 . 2 ) , load stimuli did neither elicit retractor nor protractor activation . This is depicted for one animal in Figure 3 , including PSTH analysis of the responses to CS stimuli during standing and during inside turning sequences ( Figure 3bi–iii ) . Timing of the load stimuli with respect to the front leg steps was distributed uniformly throughout the FL step cycle . We also tested whether the strength of the retractor activation was dependent on the timing of the CS stimulus during the FL step cycle . This was not the case , as shown in the plot in Figure 3c for the example animal shown in Figure 3a , nor for all other animals ( Figure 3—figure supplement 1 ) . We equally tested whether either activation of the protractor motor activity or the lack of a response occurred during a preferred phase of the step cycle . No phase preference was found for the activation of protractor activity during FL step cycle ( Figure 3d ) . However , we found a significantly increased likelihood for failure to elicit a response to a CS stimulus at 0 . 89 of the FL step cycle ( p=0 . 028 ) although this phenomenon could occur during all phases of the FL step cycle ( Figure 3e ) . In summary , these results clearly show side-specificity in the processing of local load signals during inside curve walking , modified from that during forward and outward stepping , such that the stereotypical influence of load feedback is no longer present . In light of the significant role sensory feedback plays in the generation of the motor output for stepping in the stick insect , the results presented above can offer some explanation for observed differences in stance kinematics in the curve stepping animal ( Gruhn et al . , 2009; Dürr and Ebeling , 2005 ) . Nevertheless , the question as to the mechanism behind the differences in cycle periods of outside and inside legs upon removal of mechanical coupling remains . We therefore checked for potential differences in the activity of those local central networks that generate alternating activity in the coxal protractor and retractor motoneurons ( Büschges et al . , 1995 ) . To do so , we analyzed their motor output in the deafferented mesothoracic ganglion during curve walking . Figure 4 shows mesothoracic protractor and retractor MN activity during inside ( a–c ) , and outside turning ( d–f ) of the FL ipsilateral to the recording site . When the FL performed inside steps , protractor activity in the mesothorax was increased over retractor activity in all animals ( N=17 ) , and alternating activity between the two MN pools was observed ( Figure 4a ) . The rhythmic activity of the mesothoracic coxal motoneurons was tightly correlated and largely in phase with the steps of the ipsilateral FL , as previously described for straight walking of the FL on a treadmill ( Borgmann et al . , 2007 ) ( Figure 4bi–ii ) . Protractor activity was strongest at about 270° of the step cycle in the FL in about 60% of the experiments ( N=17 ) , retractor activity at about 90° in 86% of the experiments ( N=14 ) ( Figure 4ci–ii ) . During very few inside turning sequences protractor activity became almost tonic , while retractor activity was almost absent . 10 . 7554/eLife . 13799 . 008Figure 4 . Motor output of the protractor and retractor coxae motor neuron pools in the deafferented mesothoracic ganglion during front leg inside and outside stepping . ( a ) Flexor EMG recording of the front leg ( FL , top trace , black ) together with simultaneous recording of ipsilateral mesothoracic protractor and retractor nerve activity ( 2nd and 3rd row , resp . , orange ) during an inside stepping sequence . ( bi ) Phase histogram of protractor and retractor activity with respect to the step cycle of the FL . bii . Cross correlation function of protractor and retractor activity . ( c ) Polar plots of directionality vectors with respect to the step cycle of the ipsilateral FL for protractor ( ci ) and retractor ( cii ) from N=17 and N=14 animals , resp . . ( d ) Flexor EMG recording of the FL ( top trace , black ) together with simultaneous recording of ipsilateral mesothoracic protractor and retractor nerve activity ( 2nd and 3rd row , resp . , yellow ) during an outside stepping sequence from the same animal as in a . ( ei ) Phase histogram of protractor and retractor activity with respect to the step cycle of the FL . ( eii ) Cross correlation function of protractor and retractor activity . ( f ) Polar plots of directionality vectors with respect to the step cycle of the ipsilateral FL for protractor ( fi ) and retractor ( fii ) from N=17 and N=15 animals , resp . . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 008 In contrast , during outside turns of the ipsilateral FL , protractor activity was strongly reduced compared to retractor activity , and the activity of both motor neuron pools was no longer alternating , but generally tonic ( Npro=17 , Nret=15; Figure 4d ) . Consequently , no significant phase coupling to the FL steps was present in the majority of cases , that is in about 76% of the animals for the protractor , and in 80% of the animals for the retractor ( Figure 4e–f ) . In summary , coxal motoneuron activity on the inside of the curve walking animal was rhythmic and alternating , coupled to front leg stepping , while motoneuron activity on the outside was tonic and mostly limited to the retractor motoneurons . The above results gave rise to the question whether the observed side-specific changes in coxal motoneuron activity were the result of influences on the activity of the local hemiganglionic CPGs that drive both MN pools ( Büschges et al . , 1995 ) . We therefore isolated the mesothoracic ganglion in a split-bath preparation , and , after recording mesothoracic pro- and retractor activity in control conditions ( Figure 5ai , 6ai ) , activated the local CPGs in the mesothoracic ganglion pharmacologically through superfusion with the muscarinic agonist pilocarpine ( Büschges , 1995;Borgmann et al . , 2009 ) . This elicited slow , alternating motor activity in the standing animal ( Figure 5aii , 6aii , aiii ) . During subsequent inside steps of the ipsilateral FL , protractor activity increased , and the pilocarpine-induced rhythm sped up significantly in all experiments from an average 0 . 27Hz ( SD=0 . 1 ) to 1 . 16 Hz ( SD=0 . 36; N=11 , p<0 . 001; Figure 5b ) . In about two thirds of the cases ( 62 . 5%; N=8 ) the frequency of the rhythmic mesothoracic MN activity was not significantly different from that of the stepping FL . Likewise , phase-coupling of mesothoracic MN activity to FL stepping was present , with protractor activity peaking around 270° , and retractor activity around 90° of the step cycle ( Figure 5c , d ) . 10 . 7554/eLife . 13799 . 009Figure 5 . Motor output of the protractor and retractor coxae motor neuron pools in the deafferented control , and while the mesothoracic ganglion was superfused with a pilocarpine solution during front leg ( FL ) inside stepping in split-bath configuration . ( a ) Traces show the time with and without FL activity ( top trace ) together with simultaneous mesothoracic protractor and retractor nerve recordings ( 2nd and 3th row , resp . ) ipsilateral to the FL , in control conditions ( ai ) and with pilocarpine on the mesothoracic ganglion ( aii ) of the same animal . ( b ) frequency of the mesothoracic pilocarpine rhythm in the quiescent animal ( clear bar ) , and during inside steps of the FL ( horizontally shaded bar ) from N=11 animals . In addition , diagonally shaded bar shows the frequency of the stepping FL from the eight animals where the FL EMG was recorded . In all 11 cases the frequency of the pilocarpine rhythm during stepping sequences was significantly ( p<0 . 001 ) increased over that in the quiescent animal . ( c ) Phase histogram of protractor and retractor activity from the same animal with respect to the step cycle of the FL during pilocarpine superfusion . ( d ) Polar plots with directionality vectors of protractor ( di ) and retractor ( dii ) with respect to the step cycle of the ipsilateral FL when the mesothoracic ganglion was superfused with pilocarpine ( N=8 animals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 00910 . 7554/eLife . 13799 . 010Figure 5—figure supplement 1 . Effect of unilateral transsection of the connective between the pro- and mesothoracic ganglion during inside stepping of the ipsilateral front leg ( FL ) . Left panel: Activity in the mesothoracic ganglion during FL inside steps in the intact animal . Right panel: Activity in the mesothoracic ganglion during FL inside steps in the same animal after lesion of the connective . The residual activity after lesion is due to the activity of breathing MNs . Events mark stance begin in the ipsilateral FL . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 010 In contrast , during outside turns of the FL , similar to the control condition without pilocarpine , retractor activity in the mesothoracic segment was strongly increased ( Figure 6ai ) . No increase in mean frequency of the protractor/retractor MN activity was observed ( N=8; Fctrl=0 . 27 , SD= 0 . 12 , Fout=0 . 33 , SD= 0 . 16 ) , and the frequency was always significantly lower than that of the stepping FL ( N=5; Figure 6b ) . In individual walking sequences , the frequency of coxal motoneuron activity even appeared to be locked in retractor phase ( Figure 6aii and aiii ) . No systematic phase coupling to the outside stepping FL was present , again similar to controls ( Figure 6c , d ) . In summary , our results thus show that side-specific changes in mesothoracic motor activity occur during turning that result from local changes in CPG activity . These changes in local motor activity appear to be influenced by FL stepping on the inside but not the outside . 10 . 7554/eLife . 13799 . 011Figure 6 . Motor output of the protractor and retractor coxae motor neuron pools in the deafferented control and when the mesothoracic ganglion was superfused with a pilocarpine solution during front leg ( FL ) outside stepping in split-bath configuration . ( a ) Traces show the time with and without FL activity ( top trace ) together with simultaneous mesothoracic protractor and retractor nerve recordings ( 2nd and 3th row , resp . ) ipsilateral to the FL , in control conditions ( ai ) and with pilocarpine on the mesothoracic ganglion of the same animal ( aii ) , and in a 2nd animal ( aiii ) . ( b ) Frequency of the mesothoracic pilocarpine rhythm in the quiescent animal ( clear bar ) and during inside steps of the FL ( horizontally shaded bar ) from N=8 animals . In addition , the diagonally shaded bar shows the mean FL stepping frequency from the five animals where the EMG was recorded . Note that except for 2 animals the frequency of the pilocarpine rhythm during stepping sequences was not significantly different from control in the quiescent animal . ( c ) Phase histogram of protractor and retractor activity with respect to the step cycle of the front leg during pilocarpine superfusion from the animal in aii . ( d ) Polar plots with directionality vectors of protractor ( di ) and retractor ( dii ) with respect to the step cycle of the ipsilateral FL when the mesothoracic ganglion was superfused with pilocarpine ( N=5 animals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13799 . 011 To test whether these side-specific modifications of the mesothoracic motor output were based either on descending unilateral drive , through the corresponding connective , or bilateral drive through both connectives , we lesioned the connective between the pro- and mesothoracic ganglia ipsilateral to the recording site . This lesion abolished all of the above described modifications of MN activity during curve stepping ( Figure 5—figure supplement 1 , N=4 ) , incl . the influence on the activity of the segmental CPGs in the split-bath configuration ( N=2 ) , while the contralateral activity on the intact side remained unaffected . These findings demonstrate that the modifications in mesothoracic motor activity observed during FL curve stepping are the result of unilateral information flow through the ipsilateral connectives .
Previous studies on curve walking reported varying degrees of decoupling and asymmetry between the limbs on the inside and outside of a curve ( cat: ( Musienko et al . , 2012 ) ; turtle: ( Rivera et al . , 2006 ) , stick insect: ( Jander , 1985; Dürr and Ebeling , 2005; Gruhn et al . , 2009 ) ; cockroach: ( Jindrich and Full , 1999; Mu and Ritzmann , 2005 ) ; crayfish: ( Cruse and Saavedra , 1996 ) ; Drosophila: ( Strauss and Heisenberg , 1990 ) ) . The success of the turning motor behavior can either be deduced from changes in the stride length ( Strauss and Heisenberg , 1990 ) or from the sideward or backward steps or strokes of the inside leg ( Gruhn et al . , 2009; Strauss and Heisenberg , 1990; Mu and Ritzmann , 2005; Rivera et al . , 2006; Szczecinski et al . , 2014 ) . In the stick insect , the ML movement on the inside is characterized by a flexion around the femur-tibia joint , reduced angle of movement in the thorax-coxa joint , and the occurrence of forward , sideward and backward steps in immediate succession . In contrast , outside MLs only show an increased front to back angle of movement , and reduced flexion in the femur tibia joint both under conditions of turning on normal substrate and during turns on the slippery surface ( Dürr and Ebeling , 2005; Gruhn et al . , 2009- , 2011 ) . Most studies on turning of arthropods have been performed on air cushioned balls , and also reported a decrease in stepping frequency of the inside compared to the outside legs ( Zolotov et al . , 1975; Jander , 1985 ) , sometimes no change in frequencies ( Cruse and Saavedra , 1996 ) . These reports differ from our observation that the average cycle period of inside steps was significantly shorter than that of outside steps . However , all of the studies cited above investigated curve walking while the individual legs of the animals were mechanically coupled through the ground , and could not express their inherent cycle period . A plausible explanation for the difference to our findings and similar ones in the cockroach ( Mu and Ritzmann , 2005 ) is that we studied curve walking on a slippery surface . During walking above greased plates , the mechanical coupling between the legs , and passive influences caused by ground contact are reduced or even abolished . Therefore , turning on the slippery surface must be considered a special case , in that the traction for the legs and the resistance the legs experience is reduced . In addition , stepping is not affected by passive displacement through the movements of the other legs on the ground via the mechanical coupling . Hence , the resulting sensory feedback is altered from walking on normal ground . Interestingly , turning kinematics between slippery surface and air-cushioned ball are still remarkably similar for both cockroach and stick insect ( Tryba and Ritzmann , 2000; Gruhn et al . , 2009; Dürr and Ebeling , 2005; Szczecinski et al . , 2014 ) . Because of these differences , walking on the slippery surface allows one to draw conclusions about the centrally generated motor output of the nervous system . The significantly higher stepping frequency observed on the inside compared to the outside under these conditions may therefore reflect exactly this changed output during turning which , under more naturalistic conditions , is masked and altered by sensory feedback . How are the differences in stepping frequency and kinematics generated ? Experimental data point towards two alternative ways for the neuronal control of turning , that are not necessarily mutually exclusive . In lamprey , curve swimming is achieved by increasing the descending excitatory drive unilaterally which ultimately leads to steering into the direction of the body side with the increased drive ( Fagerstedt et al . , 2001; Grillner et al . , 2008 ) . During terrestrial locomotion , however , the rigid coupling through ground contact exerts much stronger influences than mechanical coupling in fluids and locomotion involves more than trunk movement . This might explain why changes in sensory processing have been reported to form the basis of context-dependent changes in motor output in insects ( Akay et al . , 2007; Hellekes et al . , 2012; Mu and Ritzmann , 2008b; Martin et al . , 2015 ) . We have shown that local load feedback signals influence coxal MN activity depending on whether the load stimulus is applied to the ML stump ipsilateral to an outside or inside front leg ( FL ) . Similar context-dependent processing of sensory feedback was reported before in vertebrates and insects ( Pearson and Collins , 1993; Akay et al . , 2007; Hellekes et al . , 2012; Mu and Ritzmann , 2008b; Martin et al . , 2015; Mu and Ritzmann , 2008b ) . Feedback about flexion of the stick insect tibia , signaled by the femoral chordotonal organ ( fCO ) , leads to a reinforcement of flexion in inside legs , while no such effect is observed in outside legs ( Hellekes et al . , 2012 ) . Similarly , feedback from ML campaniform sensilla ( CS ) of stick insects , elicited by a load-mimicking stimulus in the horizontal plane , like that applied in our experiments , was shown to trigger a switch from ML protractor to retractor activity during forward , and in the opposite direction during backward stepping of the FL on a treadwheel ( Akay et al . , 2007 ) . All these changes in sensory processing could be depending on descending commands from the brain , as lesioning the connective anterior to the prothoracic ganglion ( Mu and Ritzmann , 2008b ) and stimulation of turn-inducing central complex ( CX ) neurons have been shown to alter similar reflexes elicited by stimulation of the chordotonal organ in cockroach ( Martin et al . , 2015 ) . Our results suggest that processing of a loading stimulus on the outside is similar to that during forward stepping , leading to retractor activation . Processing of the same stimulus on the inside was not found to be so stereotypic , ranging from activation of retractor to different degrees and termination of protractor activity , through activation of neither motor neuron pool , to protractor activation and termination of retractor activity ( Figure 7 ) . Except for a very weak phase dependence in the failure to respond to a CS stimulus to the final fourth of the FL step cycle , all other types of response to loading stimuli were not dependent of the FL step cycle . On the outside , the response matches the functional stance movement of the outside leg in the intact animal and , similar to straight walking , load in this context supports retraction of the leg ( Akay et al . , 2007 ) . On the inside , the variability in the influence is in good agreement with the observations from turning intact animals , where inside ML steps show the full spectrum of movement directions from forward to backward steps both on slippery ground and under more natural conditions ( Dürr and Ebeling , 2005; Gruhn et al . , 2009; Cruse et al . , 2009 ) . A context-dependent influence of load feedback as part of a unilateral mechanism to generate behavioral adaptivity had been shown ( Akay et al . , 2007 ) . From our results , it seems clear that graded input from yet unknown origin anterior to the mesothoracic ganglion during optomotor induced curve walking of the front legs locally modifies strength and sign of this load stimulus-dependent influence . At least two possible mechanisms for the processing of load feedback are conceivable . One is presynaptic inhibition ( Stein and Schmitz , 1999 ) that could modify the gain of load feedback onto the neural networks controlling the thorax-coxa joint CPG . The existence of presynaptic modulation of sensory synapses in the CNS has previously been described in a number of invertebrate and vertebrate locomotor systems ( Clarac and Cattaert , 1996; Büschges and Wolf , 1999; Sirois et al . , 2013 ) . Alternatively , the weighing of parallel pathways from the load sensors , the CS , to the premotor interneurons could be shifted and thereby either promote retractor or protractor activation , as previously shown for gain control of reflexes ( Driesang and Büschges , 1996 ) and recently implemented in a simulation model for cockroach turning ( Szczecinski et al . , 2014 ) . Taken together , the turning related modifications of the sensory processing on the two body sides always support stance . The difference between the two sides is that on the inside , flexion of the tibia and the inward directed movement of the leg are prioritized through strong assisting input from the fCO ( Hellekes et al . , 2012 ) . In this context , loading feedback is of lesser importance for determining forward or backward movement . In contrast , on the outside , processing of sensory feedback produces the opposite effect in that loading supports retraction to promote the front-aft movement of the leg . In addition to altered processing of local load feedback , we showed that in the intact animal phasing of the subcoxal protractor and retractor muscles with respect to stance and swing is markedly different between inside and outside steps . In addition , the motor output in the deafferented mesothoracic motor control system is also modified in a side-specific manner during curve walking of the FLs: first , there is pronounced alternating activity between pro- and retractor MNs on the inside , and tonic activity on the outside . Second , there is a strong bias towards protractor activation on the inside , and towards retractor activation on the outside . Our results thus indicate that the activity of the local subcoxal CPGs is under the influence of descending input from more anterior ganglia to support the generation of curve walking kinematics ( Figure 7 ) . How can such an influence be mediated ? Up to now , there are only two animal neural networks for which there exists a well founded idea on how turning is realized on the local level . The first are the networks that generate swimming in lamprey ( Grillner et al . , 2008 ) , and the other are the networks that generate flight in locust ( Rowell , 1993 ) . In lamprey , curve swimming is presently viewed to be mediated by a unilateral increase in descending excitatory drive from the brainstem onto the segmental CPGs which leads to a bias in the contraction strength of the segmental trunk muscles , and consequently to swimming towards the side with the increased drive ( Fagerstedt et al . , 2001; Grillner et al . , 2008 ) . In the locust , asymmetries in wing beat amplitude , phase shifts of wing beat between the sides , bending the abdomen and ruddering with the hind legs , all form part of steering maneuvers ( Rowell , 1993 ) . Especially , if the pronation angle in both wings on one side changes , turns are induced as a result of differences in lift and thrust between the two body sides ( Wolf , 1990 ) . These asymmetries are created by activation of thoracic interneurons that receive input from descending interneurons about changes in body orientation . This appears to directly affect MN output on one side of the body ( Reichert , 1989; Rowell , 1993 ) . However , in contrast to lamprey and the stick insect , in the locust , local CPG activity is unaffected by this descending input ( Rowell , 1993 ) . In the curve walking stick insect , it is quite conceivable that a unilateral mechanism not completely unlike that in curve swimming lamprey is at work , as severing one of the two connectives between pro- and mesothoracic ganglion only prevented turning-related changes in motor activity on the lesioned side and not on the other . The existence of a descending drive from anterior to the mesothoracic ganglion to the pattern generating networks of the mesothoracic joints was shown previously in physiological experiments ( Westmark et al . , 2009; Borgmann et al . , 2007 ) , and successfully implemented in models for the stick insect ( Knops et al . , 2013; Toth et al . , 2012 ) and also for cockroach walking and turning ( Szczecinski et al . 2014 ) . In contrast to lamprey steering , however , this drive appears to affect the segmental CPGs on each side of the stick insect . Moreover , the drive to the mesothorax during turning produces particularly strong activity in the respective opposite coxal MN pools in the homologous CPGs of the two sides . The above mentioned model hypothesized , and our results suggest an independent descending drive to the two body sides , which can independently strengthen or weaken the motor outputs to single leg muscles on each side of the animal ( Knops et al . , 2013; Toth et al . , 2012; Szczecinski et al . , 2014 ) . Currently , little information is available on where the drive onto the mesothoracic segment that induces the modification of local network activity in the stick insect comes from . The coupling of mesothoracic inside motor activity , clearly demonstrates a phasic interganglionic thoracic influence through the FL activity . However , this influence is not present on the outside . This suggests that higher order centers in the subesophageal and supraesophageal ganglia may play the decisive role in shaping thoracic motor activity and sensory processing . Evidence for higher order centers in the insect brain as a source of the descending drive come from studies on the central complex ( CX ) of the cerebral ganglion . Lesion and stimulation experiments in the cockroach have shown that this neuronal structure is necessary for proper turning ( Ridgel et al . , 2007; Martin et al . , 2015 ) . Fruit flies with mutations in the CX , show severe deficits in turning and start/stop maneuvers ( Strauss and Heisenberg , 1993 ) . By lesioning the connectives anterior to the prothoracic ganglion , in cockroach and locust , or stimulation of the CX , descending signals from the brain were also shown to be responsible for altering reflex sign or gain ( Mu and Ritzmann , 2008b; Knop et al . , 2001; Martin et al . , 2015 ) as part of the generation of turning kinematics . The nature of the descending signals and the cellular pathways are still unknown . Assuming a similar source in the stick insect , alteration of motor activity and sensory processing in the turning stick insect could be the result of increased excitation following disinhibition on one side ( Roeder , 1937 ) or , through crossed inhibition , on the contralateral side ( Gal and Libersat , 2006 ) , and direct inhibition from the cerebral ganglion ( Gal and Libersat , 2006 ) . Our data allow both mechanisms to play a role in turning behavior , as both , inside and outside mesothoracic pro-/retractor CPGs show increased activity , only in opposite MN pools . Recently , a neuron class was described in Drosophila that seems both necessary and sufficient to drive backward walking ( Bidaye et al . , 2014 ) . Sustained activation of the so-called MDN neurons ( two per cerebral ganglion hemisphere ) allow the flies to reverse stepping direction . Their axons arborize in both hemispheres of the cerebral and subesophageal ganglia , and cross the midline to project to the contralateral thoracic ganglia . From the report , it appears that this crossing takes place in the ventral protocerebrum ( Bidaye et al . , 2014 ) . Unilateral activation of similar neurons in the stick insect could provide means to promote turning . In summary , our results indicate that motor output of single legs in the turning stick insect is under individual hemi-segmental influence by descending signals . We have shown that these signals to the mesothoracic ganglion act together to produce turning kinematics by at least three mechanisms: first , task-dependent processing of local load signals , modified on the inside , and unchanged on the outside; second , a shift in the weighting of motor output to the protractor on the inside , and the retractor on the outside , via modification of mesothoracic CPG activity; and third , task-dependent influence from the ipsilateral front leg which is strong on the inside and weak on the outside . Together or separately , in full strength or gradually , these three mechanisms could provide the animal with the means to perform optomotor induced turns in all their different curvatures . Previous work suggests that the origin for the modulatory influence on mesothoracic CPG activity resides largely anterior to the prothoracic ganglion . Our results provide the first comprehensive evidence for the combined local action of different central neuronal mechanisms acting together at the level of the thoracic ganglia to drive this specific kind of behavior in a legged animal .
All experiments were performed in a darkened Faraday cage at room temperature ( 22°–24°C ) on adult female stick insects ( Carausius morosus; Brunner ) , of 7 . 5 cm in length . Animals were raised on blackberry leaves fed ad libitum and kept at a 12 hr:12 hr light dark cycle . In all experiments , animals walked on a 13 . 5 cm x 13 . 5 cm polished nickel-coated brass plate covered with a lubricant ( 95% glycerin , 5% saturated NaCl ) as described earlier ( Gruhn et al . , 2006 ) . This produced slipperiness and allowed recording of tarsal contact when desired ( Gruhn et al . , 2006 ) . Animals were glued ventral side down on an 80 mm long and 3 mm wide balsa rod using dental cement ( ProTempII , ESPE , Seefeld Germany ) so that legs and head protruded from the rod , and all joints were unrestrained . Animal height above the substrate was adjustable , but was typically 10 mm . Walking was elicited as an optomotor response by projecting a striped pattern ( pattern wave length 21° ) onto two 13 . 5 cm diameter round glass screens placed at right angles to each other and at a 45°angle to the walking surface , approximately 6–7 cm away from the eyes of the animal ( Gruhn et al . , 2006 ) . If the animal did not begin locomotion spontaneously , walking was elicited by light brush strokes to the abdomen . For experiments with two-legged animals , we induced autotomy of the meso- and metathoracic legs with a pair of forceps or cut the legs at the level of the coxa after recording from the intact animal ( Borgmann et al . , 2007 ) . After that we allowed a minimum of 30 min for recovery . Walking sequences were recorded from above with a high speed video camera ( Marlin F-033C , Allied Vision Technologies , Stadtroda , Germany ) at 100fps as described ( Gruhn et al . , 2009; 2011 ) . Legs were marked at the distal end of the tibia , using yellow fluorescent pigments as markers ( catalogue #56150 , Dr . Georg Kremer Farbmühle , Aichstetten , Germany ) which were dissolved in two-component glue ( ProTempII , ESPE , see above ) . During the recording , the animal was illuminated with blue LED arrays ( 12 V AC/DC , Conrad Electronic , Germany ) . A yellow filter in front of the camera lens was used to suppress the short wavelength of the activation light . Video files were analyzed using motion tracking software ( WINanalyze , Vers . 1 . 9 , Mikromak service , Berlin , Germany ) . Stepping frequencies between inside and outside front and middle legs were determined using the times of touchdown ( at the anterior extreme position , AEP ) and lift-off ( at the posterior extreme position , PEP ) , as well as the period between consecutive AEPs . These were verified visually by an additional mirror placed at an angle of 45° behind the animal . All inside and outside steps of complete 20–30 s stepping sequences were averaged in their respective groups and evaluated . The sample size for the inside and outside walks was N=10 animals , and 1–2 stepping sequences per animal ( total of 14 ) , the walks used for the analysis of front and middle leg stepping frequency were the same . Latencies of muscle activation were determined using the electronic tarsal touchdown signal of the ML as described ( Gruhn et al . , 2006 ) . In brief , a square wave signal of 2–4 mV amplitude was generated with a pulse generator ( Model MS501 , electronics workshop , Zoological Institute , Cologne ) , and applied to the slippery surface and a lock-in amplifier ( electronics workshop , Zoological Institute , Cologne ) as a reference signal . A copper wire ( 49 µm outer diameter ) with its insulation removed at the tip was tied around the tibia of the monitored leg , and connected to the lock-in amplifier with an alligator clip . The electrical resistance between the cuticle and copper wire was reduced with a drop of electrode cream ( Marquette Hellige , Freiburg , Germany ) placed at the area of contact , allowing a 2–4 nA current to pass through tarsus and tibia . During stance , current flowed from the plate through tarsus and tibia into the copper wire . The digitized using an AD converter ( Micro 1401k II , CED , Cambridge , UK ) and Spike2 software ( Vers . 5 . 05 , CED , Cambridge , UK ) . Due to the low-pass filter properties of the lock-in amplifier and the gradual lift-off/touchdown of the tarsus , the signal was not exactly square . We used thresholds close to the transition point to define the timing of tarsal contact and manually checked each event . Touchdowns could be determined at a resolution of less than 1 ms . Lift-off transitions were less steep and more delayed because of delayed tearing of the lubricant from the tarsus due to capillary action and occasional upward movements of the leg during stance without complete lift-off . To have comparable lift-off times in all experiments , we always defined lift-off as the time point with the steepest ascending slope . Muscle activity ( electromyogram , EMG ) was recorded as described ( Rosenbaum et al . , 2010 ) , using two twisted , coated copper wires ( 40 µm outer diameter ) placed in each muscle approximately 1 mm apart and held in place with dental cement ( ProTempII , ESPE ) or tissue adhesive ( 3 M Vetbond , St . Paul , MN ) . All recordings were differentially amplified . The EMG signal was pre-amplified 100 fold ( electronics workshop , Zoological Institute , Cologne ) , band-pass filtered ( 100 Hz-2000 Hz ) , when necessary further amplified 10–1000 fold , and imported into Spike2 ( Vers . 5 . 05 , CED ) through an AD converter ( Micro 1401k II , CED ) . A reference electrode was placed in the abdomen of the animal . For the analysis of muscle latency in the inside or outside ML , in most experiments , two antagonistic joint muscles were recorded simultaneously . Protractor coxae and retractor coxae EMG`s were recorded in the thorax , depressor trochanteris and levator trochanteris in the coxa , and extensor tibiae and flexor tibiae in the femur . In 2 experiments three muscles , in 3 experiments four , and in one all six muscles were recorded from , simultaneously . These experiments gave the same results as the others . For the experiments in the two-leg preparations , EMGs in the flexor tibiae muscle of each leg served as a reference for the step cycle of the front leg . For Figure 1 , the EMG activity was placed in the reference frame of the electrically determined swing and stance phases . Extracellular nerve recordings were performed as follows: all except the two front legs were amputated at mid-coxa ( Borgmann et al . , 2007 ) . Animals were fixed with dental cement ( Protemp II , ESPE ) dorsal side up on a foam platform with the same width as the balsa stick for the intact animals , but for a small stretch along the thorax behind the front legs where the platform was 8mm wide to accommodate the gut . The dorsal side of the thorax was opened , the gut moved aside , and connective tissue carefully removed to expose the connectives , the meso- and metathoracic ganglia and respective leg nerves . The coxae of the amputated middle and hind legs were fixed with dental cement , and afferent feedback was prevented by severing all lateral nerves of the meso- and metathoracic ganglia , except nl2 and nl5 ( Marquardt , 1940 ) of the mesothorax , which contain the axons of the mesothoracic coxal protractor and retractor motor neurons ( MNs ) , respectively . The body cavity was filled with Carausius saline ( Weidler and Diecke , 1969 ) , except for in split-bath experiments ( see below ) . Mesothoracic motor activity was recorded extracellularly from leg nerves nl2 , and nl5 ( Borgmann et al . , 2007 ) on both sides of the animal , using monopolar hook electrodes ( modified after ( Schmitz et al . , 1991 ) . For the lesion experiments , the pro-to-mesothoracic connective under investigation was cut with a pair of micro-surgery scissors . Split-bath experiments with the muscarinic agonist Pilocarpine were conducted as described ( Borgmann et al . , 2007; 2009 ) . A small , 2 mm stretch of cuticle between the pro- and meso- , and the meso- and metathoracic ganglia was removed except for the connectives , and the gap filled with vaseline ( Bad Apotheke , Bad Rothenfelde , Germany ) . The compartments were first filled with normal Carausius saline for control recordings , and the saline in the mesothoracic compartment was subsequently replaced by 3 mM pilocarpine solution in saline . ML campaniform sensilla ( CS ) were stimulated according to ( Akay et al . , 2007 ) . One-half to two-thirds of the ML femur were left intact . We mimicked the effect of CS-activation through ML loading during walking by rhythmically bending the femur with a piezoelectric device driven by a ramp generator ( both from electronics workshop , University of Cologne ) while recording from mesothoracic pro- and retractor MNs in the otherwise deafferented mesothoracic ganglion ( Akay et al . , 2007 ) . N denotes the number of animals used for a given condition , n the number of steps evaluated . For comparisons of EMG activity , EMG traces were rectified and smoothed ( τ = 50 ms , Spike 2 , CED , UK ) , and each single data point of each step exported to Excel ( Microsoft Corp . , Seattle , USA ) for averaging . For each step , the minimum muscle activity was set to zero and the maximum to 1 . In some cases , weak crosstalk from the antagonist muscle was removed by the following procedure: the activity of the EMG in the antagonist and in the agonist were triggered simultaneously ( i . e . at lift-off or touchdown of the tarsal contact trace ) , and exported in the same way as above . Then the antagonists minimum activity was set to 0 , but its maximum to an arbitrary value of 0 . 5 , due to the smaller size of the antagonist signal in the agonist EMG . The normalized activity of the antagonistic muscle was then subtracted from the corresponding value of the muscle under investigation . Latencies of the first spike with respect to lift-off or touchdown were calculated relative to the tarsal contact signal ( see above ) . The absolute latency was normalized with respect to the corresponding step cycle and averaged . Average swing/stance phase duration was calculated from each evaluated step from lift-off to touchdown for swing , and from touchdown to lift-off for stance . For stepping period analysis , we used 14 walks from N=10 animals , the muscle activation profiles during inside and outside were determined from N=5 animals . For the reduced preparations of two-legged animals , we used head movement in addition to leg kinematics to determine the direction of the turn . By using head posture alone and not having four more legs to judge turning behavior from , we may have introduced a bias towards more easily identifiable narrow turns in our sample of the two-leg preparations . However , this has no consequence for the conclusions drawn . For the analysis of the processing of load feedback we analyzed walking sequences from N=13 animals . Data for the analysis of descending influence of two-legged walking on the motor activity of ML pro-/retractor was taken from N=14 to 17 animals depending on the muscle and the direction of the turn ( numbers given in the text ) . For the split-bath experiments , data from N=5 animals for outside , and N=8 animals for inside turns were analyzed . The lesion experiments were performed on N=4 animals ( lesion of the pro-mesothoracic connective ) , and CS stimulation in N=13 animals . Data processing and figure preparation were performed in Spike2 ( Vers . 5 . 05 , CED ) and MATLAB 7 . 0 ( The MathWorks , Inc . , Natick ) . In some analyses , extracellular recordings were rectified and smoothed . For EMG analysis , smoothing was performed with the Spike2 smoothing function by calculating the average value of the input data points from time t - T to t + T seconds for each sample point . T was 0 . 05 s in our analyses . To analyze the activity of the retractor in response to CS stimuli we first rectified and then smoothed the retractor activity with a running Gaussian average ( width of 100 ms ) . For responses during inside steps we then determined the maximum retractor activation in a time window of 150 ms after CS stimulation and normalized these to the average maximum response during control stimulations . The resulting activation strength was then plotted against the instantaneous phase of the FL step cycle . To characterize retractor responses during outside steps we first determined the average retractor activity in a time window of 100 ms before CS stimulation and 100 ms after CS stimulation . We then subtracted the average activity after CS stimulation from the activity before CS stimulation and divided the difference by the sum of the two activities; the resulting value ( also known as Michelson contrast ) varied between -1 and 1 . Values larger than 0 thereby indicated an increase in retractor activation after CS stimulation , values below 0 indicated a decrease . These values were plotted against the phase of the FL step cycle , defined by the start of one stance phase to the start of the next , using circular statistics ( Berens , 2009 ) . In addition , we determined the instantaneous phase of FL steps during inside steps at the times when either no retractor activation occurred in response to CS stimulation or when protractor activation was observed . For the analysis of the mesothoracic motor activity during FL stepping , data were first analyzed with respect to walking leg step cycle , again defined by the start of one stance phase to the start of the next , using circular statistics ( Berens , 2009 ) . Phase histograms were used to compare motor neuron activity in steps with different cycle periods . Polar plots for mesothoracic motor activity show mean vectors of activity in the step cycle for each experiment . Vectors that had significant lengths are marked with an asterisk ( Rayleigh test with a value of 0 . 005 [Batschelet , 1981] ) . The vector length from most experiments was highly significant due to the large number of spikes . For the overall mean vector of all experiments no test of significance was done due to the varying number of spikes and steps in the experiments . A cross-correlation between pro- and retractor MN activities was done for the complete recording time including the time between stepping sequences . Between stepping sequences , pro- and retractor MNs are tonically active ( Büschges and Schmitz , 1991 ) . The cross-correlation function mirrors the episodic occurrence of the stepping sequences and , if it exists , a periodic coupling between pro- and retractor MN activities . Significance levels marked with asterisks are as follows: * p<0 . 05 , ** p<0 . 01 , *** p<0 . 001 where applicable . Figures were prepared with Origin ( Vers . 8 . 5 , Origin Lab Corp . , Northampton , MA ) and Photoshop software ( Vers . 12 . 0 , Adobe Systems Inc . , San Jose , CA ) . | Walking along a curve or turning is a complex manoeuvre for the nervous system , as it must coordinate different leg movements on each side of the body . Rhythmic processes such as walking are controlled by networks of neurons called central pattern generators . The resulting movements can be adjusted by feedback from sense organs in response to environmental conditions . For example , sensory feedback that provides information about the load placed on each leg , allows the animal to control the duration of a stance . How the nerve cells , or neurons , involved in these processes work together to produce complex , flexible movements such as turning is largely unknown . Previous work on how the brain negotiates turning movements has been carried out mostly in animals that swim or fly . To understand what happens during walking , Gruhn et al . monitored stick insects that walked in a curve on a slippery surface , and recorded the electrical activity within the animals' nervous system as they turned . By comparing the activity of the nervous system on each side of the body while the insects walked a curve , Gruhn et al . found that the nervous system uses at least three different mechanisms to produce the different movements on the inside and outside . Firstly , the sensory feedback signals that communicate the load on the leg are processed in the legs on the outside of the curve to support forward steps , while they are processed on the inside legs to support forward , sideward , and backward steps . Secondly , the motor activity produced by the central pattern generator is modulated to be stronger for the muscle that moves the leg backward on the outside of the curve . At the same time , this activity is stronger for the muscle that moves the leg forward on the inside of the curve . Thirdly , signals from a front leg influence the movement of the other legs on the same side of the body . This influence is strong on the inside and weak on the outside of the curve . Together or separately , these three mechanisms could provide the animal with the means to perform turns in all their different curvatures . Future work will need to work out exactly which local neurons process the signals sent from the brain to control movement . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Body side-specific control of motor activity during turning in a walking animal |
The heterotetrameric AP and F-COPI complexes help to define the cellular map of modern eukaryotes . To search for related machinery , we developed a structure-based bioinformatics tool , and identified the core subunits of TSET , a 'missing link' between the APs and COPI . Studies in Dictyostelium indicate that TSET is a heterohexamer , with two associated scaffolding proteins . TSET is non-essential in Dictyostelium , but may act in plasma membrane turnover , and is essentially identical to the recently described TPLATE complex , TPC . However , whereas TPC was reported to be plant-specific , we can identify a full or partial complex in every eukaryotic supergroup . An evolutionary path can be deduced from the earliest origins of the heterotetramer/scaffold coat to its multiple manifestations in modern organisms , including the mammalian muniscins , descendants of the TSET medium subunits . Thus , we have uncovered the machinery for an ancient and widespread pathway , which provides new insights into early eukaryotic evolution .
The evolution of eukaryotes some 2 billion years ago radically changed the biosphere , giving rise to nearly all visible life on Earth . Key to this transition was the ability to generate intracellular membrane compartments and the trafficking pathways that interconnect them , mediated in part by the heterotetrameric adaptor complexes , APs 1–5 and COPI ( Dacks et al . , 2008; Field and Dacks , 2009; Hirst et al . , 2011; Koumandou et al . , 2013 ) . In mammals , APs 1 and 2 and COPI are essential for viability , while mutations in the other APs cause severe genetic disorders ( Boehm and Bonifacino , 2002; Hirst et al . , 2013 ) . The AP and COPI complexes share a similar architecture , due to common ancestry predating the last eukaryotic common ancestor ( LECA ) . All six complexes consist of two large subunits of ∼100 kD , a medium subunit of ∼50 kD , and a small subunit of ∼20 kD ( Figure 1A ) . Their function is to select cargo for packaging into transport vesicles , and together with membrane-deforming scaffolding proteins such as clathrin and the COPI B-subcomplex , they facilitate the trafficking of proteins and lipids between membrane compartments in the secretory and endocytic pathways . The recent discovery of the evolutionarily ancient AP-5 complex , found on late endosomes and lysosomes , added a new dimension to models of the endomembrane system , and raised the possibility that other undetected membrane-trafficking complexes might exist ( Hirst et al . , 2011 ) . Therefore , we set out in search of additional members of the AP/COPI subunit families . 10 . 7554/eLife . 02866 . 003Figure 1 . Diagrams of APs and F-COPI . ( A ) Structures of the assembled complexes . All six complexes are heterotetramers; the individual subunits are called adaptins in the APs ( e . g . , γ-adaptin ) and COPs in COPI ( e . g . , γ-COP ) . The two large subunits in each complex are structurally similar to each other . They are arranged with their N-terminal domains in the core of the complex , and these domains are usually ( but not always ) followed by a flexible linker and an appendage domain . The medium subunits consist of an N-terminal longin-related domain followed by a C-terminal μ homology domain ( MHD ) . The small subunits consist of a longin-related domain only . ( B ) Jpred secondary structure predictions of some of the known subunits ( all from Homo sapiens ) , together with new family members from Dictyostelium discoideum ( Dd ) and Arabidopsis thaliana ( At ) . See also Figure 1—figure supplements 1–4 , Figure 1—source data 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00310 . 7554/eLife . 02866 . 004Figure 1—source data 1 . Large subunit homologues found by reverse HHpred in different organisms . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00410 . 7554/eLife . 02866 . 005Figure 1—source data 2 . Medium and small subunit homologues found by reverse HHpred in different organisms . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00510 . 7554/eLife . 02866 . 006Figure 1—figure supplement 1 . PDB entries used to search for adaptor-related proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00610 . 7554/eLife . 02866 . 007Figure 1—figure supplement 2 . Summary table of all subunits identified using reverse HHpred . The lighter shading indicates where an orthologue was found either below the arbitrary cut-off , by using NCBI BLAST ( see Figure 1—figure supplement 3 ) , or by searching a genomic database ( e . g . , AP-1 μ1 |Naegr1|35900| , JGI ) . The new complex is called ‘TSET’ . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00710 . 7554/eLife . 02866 . 008Figure 1—figure supplement 3 . Subunits that failed to be identified using reverse HHpred , but were identified by homology searching using NCBI BLAST . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00810 . 7554/eLife . 02866 . 009Figure 1—figure supplement 4 . TSET orthologues in different species . The orthologues were identified by reverse HHpred , except for those in italics , which were found by BLAST searching ( NCBI ) using closely related organisms . TTRAY1 and TTRAY2 were initially identified by proteomics in a complex with TSET , but could also have been predicted by reverse HHpred as closely related to β′-COP using the PDB structure , 3mkq_A . In all other organisms TTRAY1 and TTRAY2 were identified by NCBI BLAST ( italics ) . Note that orthologues of TSAUCER in P . patens , and TTRAY 2 in M . pusilla were identified in Phytozome , which is a genomic database hosted by Joint Genome Institute ( JGI ) . Note orthologues of TCUP in D . purpureum and TSPOON in D . discoideum were identified by searching genomic sequences using closely related sequences , and have been manually appended in DictyBase . In these cases corresponding sequences are not at present found at NCBI . Whilst S . moellendorffii and V . vinifera were included in the reverse HHpred database , they were not included in the Coulson plot . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 00910 . 7554/eLife . 02866 . 010Figure 1—figure supplement 5 . Identification of ENTH/ANTH domain proteins and the AP complexes with which they associate , using reverse HHpred . Reverse HHpred searches were initiated using the key words ‘epsin’ or ‘ENTH’ . The PDB structures used were: 1eyh_A ( Chain A , Crystal Structure Of The Epsin N-Terminal Homology ( Enth ) Domain At 1 . 56 Angstrom Resolution ) ; 1inz_A ( Chain A , Solution Structure Of The Epsin N-Terminal Homology ( Enth ) Domain Of Human Epsin ) ; 1xgw_A ( Chain A , The Crystal Structure Of Human Enthoprotin N-Terminal Domain ) ; 3onk_A ( Chain A , Yeast Ent3_enth Domain ) , and the output was assimilated in Excel as described for the adaptors . The identity of the hits was determined using NCBI BLAST searching . Note that all of the organisms that have lost AP-4 have also lost its binding partner , tepsin . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 010
Because we were unable to find any promising candidates for new AP/COPI-related machinery using sequence-based searches , we developed a more sensitive tool , designed to search for structural similarity rather than sequence similarity . Using HHpred to analyse every protein in the RefSeq database from 15 organisms , covering a broad span of eukaryotic diversity , we built a ‘reverse HHpred’ database . This database contains potential homologues for >300 , 000 different proteins ( http://reversehhpred . cimr . cam . ac . uk ) , and can be searched with structures from the Protein Data Bank ( PDB ) . As proof of principle , we used this database to identify all four subunits of the AP-5 complex ( Figure 1—figure supplements 1 and 2; Figure 1—source data 1 , 2 ) , even though in our previous study only the medium subunit was initially detectable by bioinformatics-based searching ( Hirst et al . , 2011 ) . In addition to known proteins , our reverse HHpred database revealed novel candidates for each of the four subunit families , with orthologues present in diverse eukaryotes including plants and Dictyostelium ( Figure 1—figure supplements 2–4 , Figure 1—source data 1 , 2 ) . Secondary structure predictions confirmed that the new family members have similar folds to their counterparts in the AP complexes and COPI ( Figure 1B ) . Only one of these proteins had been characterised functionally: TPLATE ( NP_186827 . 2 ) , an Arabidopsis protein related to the AP β subunits and β-COP , found in a microscopy-based screen for proteins involved in mitosis and localised to the cell plate ( Van Damme et al . , 2006; Van Damme et al . , 2011 ) . There is some variability between orthologous subunits in different organisms: for instance , Arabidopsis has added an SH3 domain to the C-terminal end of its ‘γαδεζ’ large subunit , while Dictyostelium has lost the μ homology domain ( MHD ) at the end of its medium subunit; and in general there seems to be much less selective pressure on these genes than on those encoding other AP/COPI family members ( e . g . , the AP-1 β1 subunits are 58 . 01% identical in Dictyostelium and Arabidopsis , while the new β family members are only 14 . 63% identical ) . To determine whether the four new candidate subunits identified in our searches actually form a complex , we transformed D . discoideum with a GFP-tagged version of its small ( σ-like ) subunit ( Figure 2A ) , and then used anti-GFP to immunoprecipitate the construct and any associated proteins from cell extracts ( Figure 2B ) . Precipitates were analysed by mass spectrometry , yielding ten proteins considered to be specifically immunoprecipitated ( Figure 2—figure supplement 1a ) . Two of these were the small subunit itself and its GFP tag . Three others were the remaining candidate subunits: XP_639969 . 1 ( the β-like subunit ) , XP_640471 . 1 ( the γαδεζ-like subunit ) , and XP_629998 . 1 ( the μ-like subunit ) , confirming their presence in a complex . Quantification by iBAQ indicated that these three proteins were present in the immunoprecipitate at approximately equimolar levels ( Figure 2C , Figure 2—figure supplement 1A ) , while the small subunit and GFP tag were in ∼15-fold molar excess , probably due to overexpression . 10 . 7554/eLife . 02866 . 011Figure 2 . Characterisation of the TSET complex in Dictyostelium . ( A ) Western blots of axenic D . discoideum expressing either GFP-tagged small subunit ( σ-like ) or free GFP , under the control of the Actin15 promoter , labelled with anti-GFP . The Ax2 parental cell strain was included as a control , and an antibody against the AP-2α subunit was used to demonstrate that equivalent amounts of protein were loaded . ( B ) Coomassie blue-stained gel of GFP-tagged small subunit and associated proteins immunoprecipitated with anti-GFP . The GFP-tagged protein is indicated with a red asterix . ( C ) iBAQ ratios ( an estimate of molar ratios ) for the proteins that consistently coprecipitated with the GFP-tagged small subunit . All appear to be equimolar with each other , and the higher ratios for the small ( σ-like/TSPOON ) subunit and GFP are likely to be a consequence of their overexpression , which we also saw in a repeat experiment in which we used the small subunit's own promoter ( Figure 2—figure supplement 1 ) . ( D ) Predicted structure of the N-terminal portion of D . discoideum TTRAY1 , shown as a ribbon diagram . ( E ) Stills from live cell imaging of cells expressing either TSPOON-GFP or free GFP , using TIRF microscopy . The punctate labelling in the TSPOON-GFP-expressing cells indicates that some of the construct is associated with the plasma membrane . See Videos 1 and 2 . ( F ) Western blots of extracts from cells expressing either TSPOON-GFP or free GFP . The post-nuclear supernatants ( PNS ) were centrifuged at high speed to generate supernatant ( cytosol ) and pellet fractions . Equal protein loadings were probed with anti-GFP . Whereas the GFP was exclusively cytosolic , a substantial proportion of TSPOON-GFP fractionated into the membrane-containing pellet . ( G ) Mean generation time ( MGT ) for control ( Ax2 ) and TSPOON knockout cells . The knockout cells grew slightly faster than the control . ( H ) Differentiation of the Ax2 control strain and two TSPOON knockout strains ( 1725 and 1727 ) . All three strains produced fruiting bodies upon starvation . ( I ) Assay for fluid phase endocytosis . The control and knockout strains took up FITC-dextran at similar rates . ( J ) Assay for endocytosis of membrane , labelled with FM1-43 , showing the time taken to internalise the entire surface area . The knockout strains took significantly longer than the control ( *p<0 . 05; **p<0 . 01 ) . See also Figure 2—figure supplements 1 and 2 , Figure 2; Videos 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 01110 . 7554/eLife . 02866 . 012Figure 2—figure supplement 1 . Further characterisation of Dictyostelium TSET . ( A ) iBAQ ratios for the proteins that coprecipitated with TSPOON-GFP , normalized to the median abundance of all proteins across five experiments . ND = not detected . ( B ) Fluorescence and phase contrast micrographs of cells expressing GFP-tagged TSPOON under the control of its own promoter ( Prom-TSPOON-GFP ) . The construct appears mainly cytosolic . ( C ) Homology modeling of TTRAYs from A . thaliana , D . discoideum , and N . gruberi , revealing two β-propeller domains followed by an α-solenoid . ( D ) Disruption of the TSPOON gene . PCR was used to amplify either the wild-type TSPOON gene ( in Ax2 ) or the disrupted TSPOON gene . The resulting products were either left uncut ( U ) or digested with SmaI ( S ) , which should not cut the wild-type gene , but should cleave the disrupted gene into three bands . Several clones are shown , including HM1725 ( 200/1 A1 ) . ( E ) Spore viability after detergent treatment was used to test for integrity of the cellulosic spore and the ability to hatch in a timely manner . The control ( Ax2 ) strain and the knockout ( HM1725 ) strain both showed good viability . ( F ) Expansion rate of plaques on bacterial lawns . The rates for control ( Ax2 ) and knockout ( HM1725 , 1727 , and 1728 ) strains were similar initially , but by 2 days the control plaques were larger . ( G ) Micrographs of plaques from control and knockout strains . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 01210 . 7554/eLife . 02866 . 013Figure 2—figure supplement 2 . Distribution of secG . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 01310 . 7554/eLife . 02866 . 014Figure 2—figure supplement 3 . Distribution of vacuolins . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 014 Interestingly , two of the other proteins in the immunoprecipitate , also approximately equimolar to the three coprecipitating subunits , were XP_642289 . 1 and XP_637150 . 1 . Both proteins are predicted to consist of two N-terminal β-propeller domains followed by an α-solenoid ( Figure 2D , Figure 2—figure supplement 1C ) . This type of architecture is found in several coat components , including clathrin heavy chain , SPG11 ( associated with AP-5 ) , the α-COP and β′-COP subunits of the COPI coat ( B-COPI ) , and the Sec31 subunit of the COPII coat ( Devos et al . , 2004 ) . HHpred analyses show that the closest matches for both XP_642289 . 1 and XP_637150 . 1 are β'-COP , followed by α-COP . Probable orthologues of XP_642289 . 1 and XP_637150 . 1 can be found in other organisms that have the four core subunits ( Figure 1—figure supplement 4 ) . Because proteins with this architecture often act as a coat for transport vesicles , we hypothesize that these proteins may provide a scaffold for the newly identified heterotetramer . The other three proteins in the immunoprecipitate , secG and vacuolins A and B , appear to be less widespread taxonomically ( Figure 2—figure supplement 2 and 3 ) , but are nonetheless suggestive of function . SecG is related to the plasma membrane- and endosome-associated ARNO/cytohesin family of Arf GEFs in animal cells ( Shina et al . , 2010 ) , and also appears to be equimolar with the core complex . Vacuolins are members of the SPFH ( stomatin-prohibitin-flotillin-HflC/K ) superfamily . They have been shown to associate with the late vacuole just before exocytosis and also with the plasma membrane ( Rauchenberger et al . , 1997; Gotthardt et al . , 2002 ) , and to contribute to vacuole function ( Jenne et al . , 1998 ) . However , the amounts of coprecipitating vacuolins were more variable , suggesting that they are less tightly associated with the complex ( Figure 2—figure supplement 1A ) . Thus , like TPLATE , both SecG and the vacuolins have been implicated in membrane traffic , acting at the plasma membrane and/or endosomal compartments . As the β-like subunit is already named TPLATE , we propose similar nomenclature for the other three subunits of the heterotetramer , relating to their relative sizes: TSAUCER , TCUP , and TSPOON . For the two associated β-propeller/α-solenoid proteins , we propose TTRAY1 and TTRAY2 , and for the conserved heterohexamer , we propose the name TSET . One of the key properties of coat proteins is their ability to cycle on and off membranes . Although by widefield fluorescence microscopy TSPOON-GFP looked diffuse and cytosolic ( Figure 2—figure supplement 1B ) , TIRF imaging showed a punctate pattern , especially in the cells with lower expression , indicating that some of the construct is associated with the plasma membrane ( Figure 2E , Figure 2; Video 1 ) . In contrast , free GFP appeared to be entirely cytosolic ( Figure 2E , Figure 2; Video 2 ) . In addition , high speed centrifugation of a post-nuclear supernatant showed a substantial amount of TSPOON-GFP coming down in the membrane-containing pellet , in contrast to free GFP , which was exclusively in the supernatant ( Figure 2F ) . These findings indicate that like other coat proteins , the complex is transiently recruited onto a membrane ( specifically , the plasma membrane ) from a cytosolic pool . 10 . 7554/eLife . 02866 . 015Video 1 . Related to Figure 2 . TIRF microscopy of D . discoideum expressing TSPOON-GFP , expressed off its own promoter in TSPOON knockout cells . One frame was collected every second . Dynamic puncta can be seen , indicating that the construct forms patches at the plasma membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 01510 . 7554/eLife . 02866 . 016Video 2 . Related to Figure 2 . TIRF microscopy of D . discoideum expressing free GFP , driven by the Actin15 promoter in TSPOON knockout cells . One frame was collected every second . The signal is diffuse and cytosolic . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 016 Silencing TPLATE in Arabidopsis produces a very severe phenotype , with impaired growth and differentiation , thought to be caused by defects in clathrin-mediated endocytosis ( Van Damme et al . , 2006; Van Damme et al . , 2011 ) . To investigate the function of TSET in Dictyostelium , we disrupted the TSPOON gene by replacing most of the coding sequence with a selectable marker ( Figure 2—figure supplement 1D ) . Surprisingly , the resulting knockout cells grew at least as fast a control axenic strain ( Figure 2G shows the mean generation time ) ; and differentiation also appeared normal , with fruiting bodies forming under appropriate stimuli ( Figure 2H ) . Uptake of FITC-dextran , an assay for fluid phase endocytosis , was unimpaired in the TSPOON knockout cells ( Figure 2I ) ; however , uptake of FM1-43 , a membrane marker , was slower than in the control ( Figure 2J shows the time taken to internalise the entire surface area ) , indicating that TSET plays a role in plasma membrane turnover , consistent with studies on Arabidopsis . Nevertheless , it is clear that in contrast to Arabidopsis , Dictyostelium can thrive without a functional TSET complex . Very recently , the discoverers of TPLATE used tandem affinity purification to identify TPLATE binding partners , and found the Arabidopsis orthologues of the TSET components that we identified independently in the present study ( Gadeyne et al . , 2014 ) . The Arabidopsis pulldowns did not contain any proteins resembling secG or the vacuolins , supporting our hypothesis that these proteins are add-ons to the core heterohexamer . However , Arabidopsis TSET is associated with two additional proteins containing EH domains , which we did not find in our Dictyostelium pulldowns . Some of the Arabidopsis pulldowns also brought down components of the machinery for clathrin-mediated endocytosis , including clathrin itself . Although we also found clathrin and associated proteins in our Dictyostelium immunoprecipitates , these proteins were equally abundant in control immunoprecipitates from non-GFP-expressing cells , indicating that they were contaminants . The differences in proteins that coprecipitate with TSET in the two organisms are probably a reflection of functional differences: TSET knockouts in Arabidopsis are lethal and knockdowns profoundly affect clathrin-mediated endocytosis , while TSET knockdowns in Dictyostelium produce a very mild phenotype . When TPLATE was discovered in Arabidopsis , it was reported to be unique to plant species ( Van Damme et al . , 2006; Van Damme et al . , 2011 ) . Similarly , in the more recent Arabidopsis study , the authors concluded that the complex was plant-specific ( Gadeyne et al . , 2014 ) . However , these conclusions were based on analyses of plants , yeast , and humans only . Our identification and characterization of homologues of all six subunits in Dictyostelium discoideum , as well as their presence in the excavate Naegleria gruberi , suggested that the evolutionary distribution was much more extensive . In depth homology searching identified orthologues in genomes from across the broad diversity of eukaryotes ( Figure 3 , Figure 3—source data 1 , Figure 3—figure supplement 1 ) , strongly suggesting that the complex was present prior to the LECA . 10 . 7554/eLife . 02866 . 017Figure 3 . Distribution of TSET subunits . ( A ) Coulson plot showing the distribution of TSET in a diverse set of representative eukaryotes . Presence of the entire complex in at least four supergroups suggests its presence in the last eukaryotic common ancestor ( LECA ) with frequent secondary loss . Solid sectors indicate sequences identified and classified using BLAST and HMMer . Empty sectors indicate taxa in which no significant orthologues were identified . Filled sectors in the Holozoa and Fungi represent F-BAR domain-containing FCHo and Syp1 , respectively . Taxon name abbreviations are inset . Names in bold indicate taxa with all six components . ( B ) Deduced evolutionary history of TSET as present in the LECA but independently lost multiple times , either partially or completely . See also Figure 3—source data 1 , Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 01710 . 7554/eLife . 02866 . 018Figure 3—source data 1 . Sequences used for phylogenetic analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 01810 . 7554/eLife . 02866 . 019Figure 3—figure supplement 1 . Models used for phylogenetic analyses . WC = with COPI; WOC = without COPI . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 019 Although TSET is clearly ancient , its relationship to the other heterotetrameric complexes was unclear from homology searching alone . Consequently , after analyses of the individual subunits ( Figure 4—figure supplements 1–7 ) , we performed a phylogenetic analysis on the concatenated set of the four core subunits for direct comparison of TSET with the other AP and COPI complexes ( Figure 4A , Figure 4—figure supplement 8 ) . This provided moderate support for TSET as a clade , but strong resolution excluding it from the APs and COPI , as well as backbone resolution between the heterotetramer clades . Thus , TSET is clearly an ancient component of the eukaryotic membrane-trafficking system , distinct from the known heterotetramers . 10 . 7554/eLife . 02866 . 020Figure 4 . Evolution of TSET . ( A ) Simplified diagram of the concatenated tree for TSET , APs , and COPI , based on Figure 4—figure supplement 8 . Numbers indicate posterior probabilities for MrBayes and PhyloBayes and maxium-likelihood bootstrap values for PhyML and RAxML , in that order . ( B ) Schematic diagram of TSET . ( C ) Possible evolution of the three families of heterotetramers: TSET , APs , and COPI . We propose that the earliest ancestral complex was a likely a heterotrimer or a heterohexamer formed from two identical heterotrimers , containing large ( red ) , small ( yellow ) , and scaffolding ( blue ) subunits . All three of these proteins were composed of known ancient building blocks of the membrane-trafficking system ( Vedovato et al . , 2009 ) : α-solenoid domains in both the large and scaffolding subunits; two β-propellers in the scaffolding subunit; and a longin domain forming the small subunit . The gene encoding the large subunit then duplicated and mutated to generate the two distinct types of large subunits ( red and magenta ) , and the gene encoding the small subunit also duplicated and mutated ( yellow and orange ) , with one of the two proteins ( orange ) acquiring a μ homology domain ( MHD ) to form the ancestral heterotetramer , as proposed by Boehm and Bonifacino ( 12 ) . However , the scaffolding subunit remained a homodimer . Upon diversification into three separate families , the scaffolding subunit duplicated independently in TSET and COPI , giving rise to TTRAY1 and TTRAY2 in TSET , and to α- and β′-COP in COPI . COPI also acquired a new subunit , ε-COP ( purple ) . The scaffolding subunit may have been lost in the ancestral AP complex , as indicated in the diagram; however , AP-5 is tightly associated with two other proteins , SPG11 and SPG15 , and the relationship of SPG11 and SPG15 to TTRAY/B-COPI remains unresolved , so it is possible that SPG11 and SPG15 are highly divergent descendants of the original scaffolding subunits . The other AP complexes are free heterotetramers when in the cytosol , but membrane-associated AP-1 and AP-2 interact with another scaffold , clathrin; and AP-3 has also been proposed to interact transiently with a protein with similar architecture , Vps41 ( Rehling et al . , 1999; Cabrera et al . , 2010; Asensio et al . , 2013 ) . So far no scaffold has been proposed for AP-4 . Although the order of emergence of TSET and COP relative to adaptins is unresolved , our most recent analyses indicate that , contrary to previous reports ( Hirst et al . , 2011 ) , AP-5 diverged basally within the adaptin clade , followed by AP-3 , AP-4 , and APs 1 and 2 , all prior to the LECA . This still suggests a primordial bridging of the secretory and phagocytic systems prior to emergence of a trans-Golgi network . The muniscins arose much later , in ancestral opisthokonts , from a translocation of the TSET MHD-encoding sequence to a position immediately downstream from an F-BAR domain-encoding sequence . Another translocation occurred in plants , where an SH3 domain-coding sequence was inserted at the 3′ end of the TSAUCER-coding sequence . See also Figure 4—figure supplements 1–10 . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02010 . 7554/eLife . 02866 . 021Figure 4—figure supplement 1 . Phylogenetic analysis of TPLATE , β-COP , and β-adaptin , with TPLATE robustly excluded from the β-COP clade . In this and all other figure supplements to Figure 4 , AP subunits are boxed in blue , F-COPI subunits are boxed in red , and subunits of TSET are boxed in yellow . Node support for critical nodes is shown . Numbers indicate Bayesian posterior probabilities ( MrBayes ) and bootstrap support from Maximum-likelihood analysis ( RAxML ) . Support values for other nodes are denoted by symbols ( see inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02110 . 7554/eLife . 02866 . 022Figure 4—figure supplement 2 . Phylogenetic analysis of TPLATE and β-adaptin subunits ( β-COP removed ) showing , with weak support , that TPLATE is excluded from the adaptin clade . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02210 . 7554/eLife . 02866 . 023Figure 4—figure supplement 3 . Phylogenetic analysis of TSAUCER , γ-COP , and γαδεζ-adaptin subunits , with TCUP robustly excluded from the γ-COP clade , and weakly excluded from the adaptin clade . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02310 . 7554/eLife . 02866 . 024Figure 4—figure supplement 4 . Phylogenetic analysis of TSAUCER and γαδεζ-adaptin subunits ( γ-COP removed ) , showing weak support for the exclusion of TSAUCER from the adaptin clade . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02410 . 7554/eLife . 02866 . 025Figure 4—figure supplement 5 . Phylogenetic analysis of TCUP , δ-COP , and μ-adaptin subunits , with TSAUCER robustly excluded from the δ-COP clade and weakly excluded from the adaptin clade . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02510 . 7554/eLife . 02866 . 026Figure 4—figure supplement 6 . Phylogenetic analysis of TCUP and μ-adaptin subunits ( δ-COP removed ) , showing weak support for the exclusion of TCUP from the adaptin clade . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02610 . 7554/eLife . 02866 . 027Figure 4—figure supplement 7 . Phylogenetic analysis of TSPOON with ζ-COP and σ–adaptin subunits with moderate support for the exclusion of TSPOON from both the COPI and adaptin clades , in addition to moderate support for the monophyly of the TSPOON clade . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02710 . 7554/eLife . 02866 . 028Figure 4—figure supplement 8 . TSET is a phylogenetically distinct lineage from F-COPI and the AP complexes . Phylogenetic analysis of the heterotetrameric complexes: F-COPI ( orange ) , TSET ( purple ) , and AP ( magenta , blue , red , green , and yellow for 5 , 3 , 1 , 2 , and 4 , respectively ) , shows strong , weak , and moderate support for clades of each complex , respectively . Node support for critical nodes is shown . Numbers indicate Bayesian posterior probabilities ( MrBayes and PhyloBayes ) and bootstrap support from Maximum-likelihood analysis ( PhyML and RAXML ) . Support values for other nodes are denoted by symbols ( see inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02810 . 7554/eLife . 02866 . 029Figure 4—figure supplement 9 . Phylogenetic analysis of TTRAY1 , TTRAY2 , α-COP , and β′-COP . TTRAYs 1 and 2 , and COPI α and β′ , arose from separate gene duplications , indicating that the ancestral complex had only one such protein , although possibly present as two identical copies . Phylogenetic analysis of α- and β′-COPI ( red ) , and TTRAYs 1 and 2 ( yellow ) , shows a well supported COPI clade excluding all of the TTRAY1 and 2 sequences , suggesting that the duplications giving rise to these proteins occurred independently , and the utilization of two different outer coat members occurred through convergent evolution . Node support for critical nodes is shown . Numbers indicate Bayesian posterior probabilities ( MrBayes ) and bootstrap support from Maximum-likelihood analysis ( RAxML ) . Support values for other nodes are denoted by symbols ( see inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 02910 . 7554/eLife . 02866 . 030Figure 4—figure supplement 10 . Muniscin family members identified by reverse HHpred , using the following PDB structures . 2V0O_A ( Chain A , Fcho2 F-Bar Domain ) ; 3 G9H_A ( Chain A , Crystal Structure Of The C-Terminal Mu Homology Domain Of Syp1 ) ; 3G9G_A ( Chain A , Crystal Structure Of The N-Terminal EfcF-Bar Domain Of Syp1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02866 . 030 Phylogenetic analysis of the TTRAYs and their closest relatives , β′-COP and α-COP ( Figure 4—figure supplement 9 ) , showed that the paralogues are due to ancient duplications in the TSET and COPI families respectively , which occurred prior to the divergence of the LECA . Together , these findings imply that the ancestor of the TSET , COPI , and AP complexes was a heterohexamer rather than a heterotetramer , consisting of five different proteins , with the two scaffolding proteins present as two identical copies ( Figure 4B , C ) . These scaffolding subunits then duplicated independently in COPI and TSET . The ancestral AP complex may have lost its original scaffolding subunits , although AP-5 , the first AP to branch away , is closely associated with SPG11 , a β-propeller + α-solenoid protein whose relationship to the TTRAYs and B-COPI is as yet unclear . None of the other APs has any closely associated proteins with this architecture , but AP-1 and AP-2 transiently interact with clathrin , and there may also be a transient association between AP-3 and another β-propeller + α-solenoid protein , Vps41 ( Rehling et al . , 1999; Cabrera et al . , 2010; Asensio et al . , 2013 ) . Although TSET is deduced to have been present in LECA , the complex appears to have been entirely or partially lost in various lineages ( Figure 3B ) . None of the subunits has a full orthologue in opisthokonts ( animals and fungi ) , indicating secondary loss in the line leading to humans . However , the C-terminal domain of TCUP is homologous to the C-terminal domains of the muniscins , opisthokont-specific proteins ( Gadeyne et al . , 2014 ) ( Figure 4—figure supplement 10 ) . This suggests that in opisthokonts , the TCUP gene retained its 3′ end , which then combined with a new 5′ end encoding an F-BAR domain to generate the muniscin family ( Figure 4C ) . These include the vertebrate proteins FCHo1/2 and the yeast protein Syp1 , important players in the endocytic pathway ( Reider et al . , 2009; Henne et al . , 2010; Cocucci et al . , 2012; Umasankar et al . , 2012; Mayers et al . , 2013 ) . The muniscins constitute one of eight families of MHD proteins in humans , and the only family whose evolutionary origin was unexplained until now . The present study indicates not only that the muniscins are homologous to TCUP , but also that they are the sole surviving remnants of the full TSET complex that existed in our pre-opisthokont ancestors . TSET is the latest addition to a growing set of trafficking proteins that have ancient distributions , but are frequently lost ( Schlacht et al . , 2014 ) , or in the case of TSET reduced perhaps with neofunctionalization ( Figure 3 ) . This is consistent with the uneven distribution of the individual components ( in contrast to the all-or-nothing distribution of AP-5 ) , the additional apparently lineage-specific binding partners in Dictyostelium , and the acquisition of extra domains ( e . g . , F-BAR in opisthokonts and SH3 in plants ) adding lineage-specific function . Studies on the muniscins may help to explain the different phenotypes of TSET knockouts in Dictyostelium and Arabidopsis . Like Arabidopsis TSET , the muniscins interact with EH domain-containing proteins and participate in clathrin-mediated endocytosis ( Reider et al . , 2009; Henne et al . , 2010; Cocucci et al . , 2012; Umasankar et al . , 2012; Mayers et al . , 2013 ) . Dictyostelium has lost its TCUP MHD , and it seems likely that concomitant with this loss , it also lost some of TSET's binding partners and functions . Nevertheless , we suspect that TSET may predate clathrin-mediated endocytosis , for two reasons . First , AP-1 and AP-2 , the two AP complexes that function together with clathrin , are the most recent additions to the AP family ( Figure 4A ) ; and second , TSET already has its own β-propeller + α-solenoid scaffold , so it is not clear why it would need clathrin as well . Thus , the interaction between TSET and the clathrin pathway may have evolved considerably later than TSET itself , although still pre-LECA . It is tempting to speculate that TSET was part of the original endocytic machinery , which then became redundant in some organisms as the clathrin pathway took over . Thus , our bioinformatics tool , reverse HHpred , is able to find novel homologues of known proteins , and could potentially be used to identify new players both in membrane traffic and in other pathways ( Figure 1—figure supplement 5 ) . Using this tool , we were able to find the four core subunits of an ancient complex belonging to the same family as the APs and COPI . This ancient complex , TSET , is therefore both the answer to the question of the origin of the last set of MHD proteins in humans , and a major new piece of the puzzle to be incorporated alongside the other membrane-trafficking machinery , as we delve into the history of the eukaryotic cell .
The proteomes of various organisms ( detailed in Figure 1—figure supplement 2 ) were downloaded from the National Center for Biotechnology Information archives at ftp://ftp . ncbi . nih . gov/refseq/release/ . The * . protein . faa . gz files obtained were then split into separate files , each containing one protein sequence . These were stored such that each directory contained information from only one species ( the total number of protein ‘faa’ files searched for each organism were: Arabidopsis thaliana , 35270; Caenorhabditis elegans , 23903; Dictyostelium discoideum , 13262; Dictyostelium purpureum , 12399; Drosophila melanogaster , 22256; Giardia lamblia , 6502; Homo sapiens , 32977; Micromonas pusilla , 10269; Mus musculus , 29897; Naegleria gruberi , 15756; Physcomitrella patens , 35893; Saccharomyces cerevisiae , 5882; Schizosaccharomyces pombe , 5004; Selaginella moellendorffii , 31312; Vitis vinifera , 23492; Volvox carteri , 14429 ) . The latest protein data bank ( pdb70 ) , which contains all publicly available 3D structures of proteins , was downloaded from the Gene Center Munich , Ludwig-Maximilians-Universität ( LMU ) Munich via their web site at: ftp://toolkit . lmb . uni-muenchen . de/pub/HHsearch/databases/hhsearch_dbs/ . The linux rpm version 2 . 0 . 11 of the hhsuite software was downloaded from the same website at ftp://toolkit . lmb . uni-muenchen . de/pub/HH-suite/releases/ . Each of the faa files was then compared to the pdb70 databank using the hhsearch program from the above suite . The files were tested using the default parameters . Once each protein sequence was tested , the output file was parsed and the hits were extracted and then inserted into a mysql database . The database is searchable by keywords in PDB entries , and therefore is limited to searches where the structure of a given domain structure has been solved . The database is accessible using the link http://reversehhpred . cimr . cam . ac . uk , and searches can be initiated using keywords . Should the link become unavailable , or if you are interested in hosting this yourself please email jpn25@cam . ac . uk for more information . A conceptually similar database , ‘BackPhyre’ , has independently been generated , using Phyre ( Kelley and Sternberg , 2009 ) rather than HHpred as a starting point to identify homologues of known proteins based on predicted structural similarities . Like reverse HHpred , BackPhyre is able to find three of the four TSET subunits in Arabidopsis; however , the only eukaryotes represented in BackPhyre are A . thaliana , D . melanogaster , H . sapiens , M . musculus , P . falciparum , and S . cerevisiae; and without additional organisms , such as D . discoideum and N . gruberi , we would not have been able to find the entire TSET complex . The large adaptor subunits share sequence and structure homology , as do the medium and small subunits . Therefore , we were able to combine searches for novel large subunits , or for medium/small subunits . Using the key words ‘clathrin’ , ‘adaptor’ , ‘adapter’ , ‘adaptin’ , ‘AP1’ , ‘AP2’ , ‘AP3’ , ‘AP4’ , we searched in PDB for solved structures of any large or medium/small subunit in a given organism ( 11 solved structures for the large subunits and six solved structures for the medium/small subunits were used to initiate searches [Figure 1—figure supplement 1] ) . These structures span different domains found within the subunits . For each search , a list was output of any proteins found to contain structural homology . Included in this information are the precise amino acids encompassing the region of similarity , the probability score , and most importantly the ‘result number’ . A protein with a ‘result number’ of ‘1’ means that there was no other structure in the PDB database that it is more like . Since multiple structures for the various subunits were used , we could also factor in the number of times a particular protein was identified in a search ( ‘repeats’ ) . These parameters were used as key pieces of evidence to determine how likely a hit in these searches would be . Once the primary data were outputted , all other manipulations were performed in Excel . For the large subunits there were 11 data sets ( the 11 structures used to search for homologues ) , and for the medium/small subunits there were six data sets . The data manipulation was standardised at this point , and the following steps performed to assimilate the data . The data sets were sorted by result number to preclude anything with a result number of >50 ( this means that there are 49 other structures in the PDB database that this protein is more similar to ) . Duplicates , where a protein was identified in multiple searches , were removed with the highest ranking ( in ‘result’ terms ) kept , and the number of times it was identified recorded in a new column ( ‘repeats’ ) . The results were the ordered with the lowest ‘Result number’ and the highest ‘Probability’ to give a final list of proteins ( Figure 1—source data 1 , 2 ) . Generally only proteins with a ‘Result number’ <10 , ‘Probability’ >50% , at least 100 amino acids of homology ( ‘thstt’ to ‘thend’ ) , and ‘Repeats’ at least two times were considered to be real hits . For ease of visualisation , only proteins with Result number <10 or ‘Repeats’ >2 are shown , and other proteins of interest ( e . g . , FCHo1 , Syp1 ) with Result number <10 that did not fit the criteria listed above are greyed out . The ‘IDs’ have been deduced using NCBI BLAST searches , and have not been experimentally verified . Where the identity is ambiguous ( such as the identity of a β-adaptin ) , a shared homology is suggested . While searching for genes encoding potential components of the complex in four dictyostelid genomes , we could find complete sets in Polysphondylium pallidum and Dictyostelium fasciculatum , but one component each was missing in the databases of predicted proteins of D . discoideum ( σ-like subunit ) and D . purpureum ( μ-like subunit ) . We identified these genes by tblastn ( Camacho et al . , 2009 ) , using the most closely related orthologous sequence as query and the chromosomal sequences as target . Gene models were created and refined using the Artemis tool ( Carver et al . , 2012 ) . These two genes have been given the DictyBase IDs DDB_G0350235 ( D . discoideum TSPOON ) and DPU0040472 ( D . purpureum TCUP ) ( www . dictybase . org ) . The σ-like ( TSPOON ) coding sequence ( CDS ) was synthesised ( GeneCust ) with a BglII restriction site inserted at its 5′ end , its stop codon removed , and a SpeI site inserted at its 3' end , then cloned into pBluescript KSII and sequenced . The CDS was then transferred into a derivative of pDM1005 ( Veltman et al . , 2009 ) as a BglII/SpeI fragment , placing GFP at the C terminus , with expression driven from the constitutive actin15 promoter , to generate plasmid pJH101 . In addition , the TSPOON promoter and the first 105 bases of the CDS were amplified from Ax2 gemonic DNA by PCR , using primers ( 5′TATCTCGAGCGTCTTCATCTTCACTATCATTTAATG-3′ ) and ( 5′-TAAAAGCTTTTCATATTCACTCTGTTTCTCGTC-3′ ) . The product was cut with XhoI/HindIII , and the 536-bp fragment cloned into the pBluescript KSII plasmid already containing the TSPOON CDS , via the XhoI site in the vector and the silent HindIII site introduced at nucleotide +97 of the TSPOON CDS during its synthesis . The resulting promoter-driven TSPOON CDS was removed by digestion with XhoI/SpeI and inserted into the corresponding sites of pDM323 and pDM450 , resulting in expression constructs containing the TSPOON CDS with GFP fused at its C terminus and driven by its own promoter ( pDT61 and pDT58 respectively ) . All of the methods used for cell biological studies on Dictyostelium are described in detail at Bio-protocol ( Hirst et al . , 2015 ) . D . discoideum Ax2-derived strains were grown and maintained in HL5 medium ( Formedium ) containing 200 µg/ml dihydrostreptomycin on tissue culture treated plastic dishes , or shaken at 180 rpm , at 22°C ( Kay , 1987 ) . Cells were transformed with expression constructs ( 30 µg/4 × 106 cells ) by electroporation using previously described methods ( Knecht and Pang , 1995 ) . Transformants were selected and maintained in axenic medium supplemented with 60 µg/ml hygromycin ( pDT58 and pJH101 ) and 20 µg/ml G418 ( pDT61 and Actin15_GTP; Traynor and Kay , 2007 ) . For the TSPOON knockout , 17 . 5 µg of the blasticidin disruption cassette , freed from pDT70 by digestion with ApaI and SacII , was added to 4 × 106 Ax2 cells before electroporation . Transformants were selected and maintained in HL5 medium containing 10 µg/ml blasicidin . Cells were transformed with GFP driven by the actin 15 promoter ( A15_GFP; Traynor and Kay , 2007 ) , or with TSPOON-GFP driven by either the actin 15 promoter ( A15_TSPOON -GFP ) or its own promoter ( promoter_TSPOON-GFP ) . For microscopy , the cells were washed in KK2 ( 16 . 5 mM KH2PO4 , 3 . 8 mM K2HPO4 , 2 mM MgSO4 ) at 2 × 107/ml and then transferred into glass bottom dishes ( MatTek , Ashland , MA ) at 1 × 106/cm2 . They were either imaged immediately ( vegetative ) or allowed to starve for a further 6–8 hr ( developed ) before imaging live on a Zeiss Axiovert 200 inverted microscope ( Carl Zeiss , Jena , Germany ) using a Zeiss Plan Achromat 63 × oil immersion objective ( numerical aperture 1 . 4 ) , an OCRA-ER2 camera ( Hamamatsu , Hamamatsu , Japan ) , and Improvision Openlab software ( PerkinElmer , Waltham , MA ) . Various treatments including with or without starvation , fixation , pre-fixation saponin treatment did not reveal obvious membrane-associated labelling in cells expressing either promoter_TSPOON-GFP and A15_TSPOON expressing cells . For TIRF microscopy , TSPOON-GFP was expressed in the TSPOON null cell lines HM1725 and HM1727 ( see below ) , using the promoter_TSPOON-GFP plasmids pDT58 or pDT61 . Transformants were selected and maintained in 30 µg/ml hygromycin ( pDT58 ) or 10 µg/ml G418 ( pDT61 ) . As a control , free GFP was expressed in the null cells using the plasmid A15_GFP . Cells were harvested from tissue culture dishes when they formed a semi-confluent monoloyer and washed in KK2C ( KK2 containing 0 . 1 mM CaCl2 ) . Approximately 3 × 104 cells were added to 35-mm glass bottom ( No . 1 . 5 coverglass ) microwell dishes ( MatTek ) containing 2 . 5 ml of KK2C . They were incubated at 22°C for 2 hr to allow residual fluorescence associated with ingested axenic medium to dissipate , and 20 min before imaging , the KK2C was with fresh KK2C , containing 50 µg/ml L-ascorbic acid as an antioxidant to reduce the effects of phototoxicity . Cells were visualised using a Nikon N-STORM microscope operating in the TIRF mode with a 100x lens ( NA 1 . 49 ) and a zoom of 1 . 5x . For fractionation , cells expressing A15_GFP or promoter_TSPOON-GFP were grown until they reached a density of 2–4 × 106/ml in selective media , and by microscopy >50% of cells were expressing GFP . Starting with a maximum of 8 × 108 cells , the cells were washed in KK2 buffer and then pelleted at 600 × g for 3 min . The cells were resuspended in PBS with a protease inhibitor cocktail ( Roche ) , lysed by 8 strokes of a motorized Potter–Elvehjem homogenizer followed by 5 strokes through a 21-g needle , and centrifuged at 4100 × g for 32 min to get rid of nuclei and unbroken cells . The postnuclear supernatant was then centrifuged at 50 , 000 rpm ( 135 , 700 × g RCFmax ) for 30 min in a TLA-110 rotor ( Beckman Coulter ) to recover the membrane pellet . The cytosolic supernatant and pellet were run on pre-cast NUPAGE 4–12% BisTris Gels ( Novex ) at equal protein loadings , and Western blots were probed with an antibody against GFP ( Seaman et al . , 2009 ) . Pulldowns were performed using Dictyostelium discoideum stably expressing TSPOON-GFP under a constitutive ( A15_ TSPOON-GFP ) and its own promoter ( prom_TSPOON-GFP ) . Similar results were found with both cell lines regardless of the promoter . Non-transformed cells were used as a control . Cells were grown until they reached a density of 2–4 × 106/ml in selective media , and by microscopy >50% of cells were expressing GFP . Starting with a maximum of 8 × 108 cells , they were pelleted by centrifugation ( 600×g for 2 min ) and washed twice in KK2 buffer before being resuspended at 2 × 107 cells/ml in KK2 buffer and starved for 4–6 hr at 22°C by shaking at 180 rpm . The cells were then pelleted at 600×g for 3 min and then lysed in 4 ml PBS 1% TX100 plus protease inhibitor cocktail tablet ( Roche ) for 10 min on ice , and then spun 20 , 000×g 15 min to get rid of debris and insoluble material . By protein assay the resulting lysate contained 10–15 mg total protein . The lysates were pre-cleared using PA-sepharose 30 min , and then immunoprecipitated using anti-GFP overnight with rotation at 4°C . PA-sepharose was added for 60 min and then the antibody complexes washed with PBS 1%TX100 followed by PBS before elution from beads with 100 mM Tris , 2% SDS 60°C for 10 min . The eluted proteins were precipitated with acetone overnight at −20°C , recovered by spinning 15 , 000×g 5 min and then resuspending in sample buffer . The samples were run on pre-cast NUPAGE 4–12% BisTris Gels ( Novex ) , stained with SimplyBlue Safe Stain ( Invitrogen ) and then cut into 8 gel slices . Each gel slice was processed by filter-aided sample preparation solution digest , and the sample was analyzed by liquid chromatography–tandem mass spectrometry in an Orbitrap mass spectrometer ( Thermo Scientific; Waltham , MA ) ( Antrobus and Borner , 2011 ) . Proteins that came down in the non-transformed control were eliminated , as were any proteins with less than 5 identified peptides , proteins that did not consistently coimmunoprecipitate in three independent experiments , or proteins of very low abundance compared with the bait ( i . e . , molar ratios of <0 . 002 ) . The remaining ten proteins were considered to be specifically immunoprecipitated . Normalized peptide intensities were used to estimate the relative abundance of the specific interactors ( iBAQ method; Schwanhäusser et al . , 2011 ) . For each protein , the values from all five repeats were plotted , including the bait protein and GFP which are clearly overrepresented by overexpression . The relative abundances of proteins were normalized to the median abundance of all proteins across each experiment ( i . e . , median set to 1 . 0 ) and values were then log-transformed and plotted . The TSPOON disruption plasmid was constructed by inserting regions amplified by PCR from upstream and downstream of the TSPOON gene into both side of the blasticidin-resistance cassette in pLPBLP ( Faix et al . , 2004 ) . The primer pair used to amplify the 5′ region was TCP1 ( 5′-ACTGGGCCCTGATGTTTACCTCTCTTTGGGTCATCCCATTCTATAC-3′ ) with σ-TCP2 ( 5′-AAAAAGCTTTATTACCATTGTTATTGGTAATTAACAAACTATTGATC-3′ ) and for the 3′ homology TCP3 ( 5′-A CCGCGGCCGCATAATTCAAAGAGGTCATTTAGATCAAGTTCAATTAG-3′ ) with TCP4 ( 5′-CCTCCGCGGCTTCAGGCATTGGTTCAACTTCTTGATTATTCTCAAC -3' ) . The PCR products were inserted as ApaI/HindIII and NotI/SacII fragments into the corresponding sites in pLPBLP , yielding pDT70 . Growth of control vs mutant strains was assayed in HL5 medium , by calculating the mean generation time , and on Klebsiella aerogenes bacterial lawns , by monitoring the expansion of a spot of 104 cells . Spore viability was also assayed , both with and without detergent treatment , by clonally diluting spores on bacterial lawns and counting the resultant plaques ( Kay , 1982 ) . Membrane uptake was measured in real time at 22°C with 2 × 106 cells in 1 ml of KK2C containing 10 µM FM1-43 ( Life Technologies ) . Briefly , a 2-ml fluorimeter cuvette containing 0 . 9 ml of KK2C plus 11 µM FM1-43 was placed in the fluorimeter ( PerkinElmer LS50B ) with stirring set on high . The uptake was initiated by the addition of 100 µl cells at 2 × 107/ml in KK2C and data collected every 1 . 2 s at an excitation of 470 nm ( slit width 5 nm ) and emission of 570 nm ( slit width 10 nm ) for up to 360 s . The uptake curves were biphasic and the data were normalized against the initial rise in fluorescence , when the cells were first added to the FM1-43 , as this essentially corresponds to the dye incorporation into the plasma membrane only ( Aguado-Velasco and Bretscher , 1999 ) . The uptake rate was calculated from linear regression of the initial linear phase of the uptake using GraphPad Prism software . The surface area uptake time is 1/slope of the initial phase . Fluid phase uptake was measured at 22°C using FITC-dextran 70 kDa ( Sigma FD-70 ) by adding 2 mg/ml ( final ) to cells ( 1 × 107/ml ) in filtered HL5 medium that was shaken at 180 rpm . Duplicate 0 . 5 ml samples were taken at each time point and diluted in 1 ml of ice-cold HL5 in a microcentrifuge tube held on iced water . Cells were pelleted , the supernatant aspirated , and the pellet washed twice by centrifugation in 1 . 5 ml ice-cold wash buffer ( KK2C plus 0 . 5%wt/vol BSA ) before being lysed in 1 ml of buffer [100 mM Tris–HCl , 0 . 2% ( vol/vol ) Triton X-100 , pH 8 . 6] and fluorescence then determined ( excitation 490 nm , slit width 2 . 5 nm; emission 520 nm , slit width 10 nm ) . Data were normalized to protein content ( Traynor and Kay , 2007 ) . Sequences from Arabidopsis thaliana , Dictyostelium discoideum , and Naegleria gruberi were obtained with our new reverse HHpred tool . These sequences were used to build HMMs for each subunit using HMMer v3 . 1b1 ( http://hmmer . org ) . HMMs were used to search the protein databases for the organisms in Figure 3A ( see Figure 3—source data 1 for the location of each genomic database ) . Sequences identified as potential homologues were verified through reciprocal BLAST into the genomes of each of the original three sequences . Sequences were considered homologues if they retrieved the correct orthologue as the reciprocal best hit in at least one of the reference genomes , with an e-value at least two orders of magnitude better than the next best hit . New sequences were incorporated into the HMM prior to searching a new genome in order to increase the sensitivity and specificity of the HMM . Genomic protein databases were also searched by BLAST using the closest related organism with an identified sequence as the reference genome . Nucleotide databases ( scaffolds or contigs ) were also searched using tblastn to ensure that no sequences were missed resulting from incomplete protein databases . The distribution of TSET components is displayed in Coulson plot format using the Coulson plot generator v1 . 5 ( Field et al . , 2013 ) . Identified sequences were combined with the adaptin and COPI sequences from Hirst et al . ( 2011 ) into subunit-specific data sets with the intention of concatenation . Data sets were aligned using MUSCLE v3 . 6 ( Edgar , 2004 ) and masked and trimmed using Mesquite v2 . 75 . Phylogenetic analysis was carried out using MrBayes v . 3 . 2 . 2 ( Ronquist and Huelsenbeck , 2003 ) and RAxML v7 . 6 . 3 ( Stamatakis , 2006 ) , hosted on the CIPRES web portal ( Miller et al . , 2010 ) . MrBayes was run using a mixed model with the gamma parameter until convergence ( splits frequencey of 0 . 1 ) . RAxML was run under the LG + F + CAT model ( Lartillot et al . , 2009 ) and bootstrapped with 100 pseudoreplicates . The resulting trees were visualized using FigTree v1 . 4 . Initial data sets were run and long branches were removed . Data sets were then re-aligned and re-run as above . Opisthokont adaptin and COPI sequences were also removed from all data sets except from the TCUP alignment . Data sets were realigned and new phylogenetic analyses were carried out . Remaining sequences were used for concatenation . Sequences were aligned and trimmed , as above , and concatenated using Geneious v7 . 0 . 6 . Subsequent phylogenetic analysis was carried using PhyloBayes v3 . 3 ( Lartillot et al . , 2009 ) under the LG + CAT model until a splits frequency of 0 . 1 and 100 sampling points was achieved , and PhyML v3 . 0 , with model testing carried out using ProtTest v3 . 3 . MrBayes and RAxML were used as above . Raw phylogenetic trees were converted into figures using Adobe Illustrator CS4 . The models of amino acid sequence evolution are provided in Figure 3—figure supplement 1 . The database identifiers of all sequences and their abbreviations and figure annotations are provided in Figure 3—source data 1 . All alignments are available in Supplementary file 1 . The Phyre v2 . 0 web server ( Kelley and Sternberg , 2009 ) was used to predict the 3D structures of each TTRAY from A . thaliana , D . discoideum , and N . gruberi . Default settings were used for structural predictions , and structures were visualized using MacPyMOL ( www . pymol . org ) . | Eukaryotes make up almost all of the life on Earth that we can see around us , and include organisms as diverse as animals , fungi , plants , slime moulds , and seaweeds . The defining feature of eukaryotes is that , unlike nearly all bacteria , they have membrane-bound compartments—such as the nucleus—within their cells . Moving molecules , such as proteins , between these compartments is essential for living eukaryotic cells , and these molecules are usually trafficked inside membrane-bound packages called vesicles . Two similar sets of protein complexes—each containing four different subunits—ensure that the molecules are packaged inside the correct vesicles . However , it is not clear how these two protein complexes ( called the AP complexes and the COPI complex ) are related to each other , and when and where they originated in the history of life . Now , Hirst , Schlacht et al . have discovered a new—but very ancient–protein complex that they refer to as the ‘missing link’ between the AP and COPI complexes . The four subunits inside this new complex were found by searching for proteins with shapes that were similar to those of the AP and COPI proteins , rather than just searching for proteins with similar sequences of amino acids . This approach identified related protein subunits in groups as diverse as plants and slime moulds , which suggests that this protein complex evolved in the earliest of the eukaryotes . The four subunits identified in a slime mould were confirmed to interact , and also shown to bind to the plasma membrane of living cells . One of the subunits had already been named TPLATE , so Hirst , Schlacht et al . decided to call the complex TSET; the other three subunits were named TSAUCER , TCUP and TSPOON , and two other proteins that interacted with the complex were both called TTRAY . While most of the TSET complex itself has been lost from humans and other animals , one of subunit appears to have evolved into a family of proteins that help molecules get into cells . The discovery of TSET reveals another major player in vesicle-trafficking that is not only important for our understanding of how modern eukaryotes work , but also how ancient eukaryotes evolved . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"cell",
"biology"
] | 2014 | Characterization of TSET, an ancient and widespread membrane trafficking complex |
Mutations in the catalytic subunit of phosphoinositide 3-kinase ( PIK3CA ) and other PI3K-AKT pathway components have been associated with cancer and a wide spectrum of brain and body overgrowth . In the brain , the phenotypic spectrum of PIK3CA-related segmental overgrowth includes bilateral dysplastic megalencephaly , hemimegalencephaly and focal cortical dysplasia , the most common cause of intractable pediatric epilepsy . We generated mouse models expressing the most common activating Pik3ca mutations ( H1047R and E545K ) in developing neural progenitors . These accurately recapitulate all the key human pathological features including brain enlargement , cortical malformation , hydrocephalus and epilepsy , with phenotypic severity dependent on the mutant allele and its time of activation . Underlying mechanisms include increased proliferation , cell size and altered white matter . Notably , we demonstrate that acute 1 hr-suppression of PI3K signaling despite the ongoing presence of dysplasia has dramatic anti-epileptic benefit . Thus PI3K inhibitors offer a promising new avenue for effective anti-epileptic therapy for intractable pediatric epilepsy patients .
The phosphoinositide-3 kinase ( PI3K ) -AKT pathway is a central player of intracellular signaling , conserved from yeast to mammals . Activating mutations in genes of PI3K-AKT signaling pathway , especially PIK3CA , encoding the catalytic p110α isoform of the PI3K complex , have long been linked to cancer ( Cheung and Testa , 2013; Engelman , 2009; Hennessy et al . , 2005; Iwabuchi et al . , 1995; Samuels and Waldman , 2011; Gymnopoulos et al . , 2007 ) . Germline and mosaic mutations of PIK3CA and other pathway genes also cause a wide range of brain and body overgrowth disorders; all anomalies caused by somatic PIK3CA mutations are now collectively termed PIK3CA-Related Overgrowth Spectrum ( PROS ) ( Keppler-Noreuil et al . , 2014 ) . The broad spectrum of brain overgrowth disorders caused by PIK3CA activating mutations is impressive . Three strongly activating PIK3CA mutations found most commonly in cancer ( hotspot mutations ) result in severe segmental cortical dysplasia ( SEGCD ) , which includes bilateral dysplastic megalencephaly ( MEG ) , hemimegalencephaly ( HMEG ) and focal cortical dysplasia ( FCD ) types 2a/2b ( Lee et al . , 2012; D'Gama et al . , 2015; Conway et al . , 2007; Jansen et al . , 2015 ) . Other mutations , resulting in intermediate or weak PIK3CA activation , cause MEG or MEG with polymicrogyria ( MEG-PMG ) as part of the MEG-capillary malformation syndrome ( MCAP ) ( Conway et al . , 2007; Mirzaa et al . , 2012; Rivière et al . , 2012 ) . Developmental features of these brain disorders include cortical malformations , hydrocephalus , Chiari malformation , intellectual disability , autism and epilepsy ( Keppler-Noreuil et al . , 2014; Mirzaa et al . , 2012 ) . FCD represents one of the most common causes of intractable epilepsy ( Bast et al . , 2006; Fauser et al . , 2015; Fauser , 2006 ) . Conditional mouse alleles for the H1047R and E545K Pik3ca hotspot mutations have been generated to study tumor formation and assess anti-cancer activities of pathway inhibitors ( Kinross et al . , 2012; Liu et al . , 2011; Meyer et al . , 2011; Robinson et al . , 2012; Yuan et al . , 2013 ) . To understand the cellular mechanisms behind PIK3CA-related brain overgrowth disorders , we used a series of cre-drivers to activate expression of H1047R and E545K alleles in subsets of neural progenitors . Dramatic phenotypes resulted , faithfully modeling the entire spectrum of PIK3CA-associated human brain disorders including enlarged brain size , hydrocephalus , cortical dysplasia and epilepsy . The severity of these brain phenotypes critically depended on the Pik3ca allele and its time of activation . Notably , Pik3ca-associated epilepsy in mice was independent of brain overgrowth and cortical dysplasia . Further the seizures of adult megalencephalic mice were suppressed by acute 1 hr-administration of pan-PI3K inhibitor BKM120 ( Maira et al . , 2012 ) . We conclude that epilepsy in these models represents an active Pik3ca-driven process that can be restricted by dynamic modulation of PI3K pathway activity in dysmorphic brains . These results raise the exciting prospect of new molecular based epilepsy therapies in patients whose seizures have been intractable to current anti-seizure therapies .
Two conditional Pik3ca activating alleles ( H1047R and E545K ) were crossed with cre-drivers to overactivate p110α in progressively restricted sets of neural progenitors and their progeny . The broadest distribution was achieved with Nestin-cre , being expressed in nearly all neural progenitors from early embryonic stages . A subset of late embryonic progenitors was targeted by hGFAP-cre; while tamoxifen-inducible Nestin-creER line drove cre-expression in a small subset of neural progenitors around birth . Expression of H1047R was dependent upon a tri-allelic system with tet-inducible mutant human cDNA activated by cre-dependent expression of the tet-activator protein ( Liu et al . , 2011 ) ( Figure 1—figure supplement 1 ) . The E545K mutation was knocked into the endogenous Pik3ca locus and a lox-stop-lox cassette introduced upstream of the initiation-coding exon , rendering the mutant allele cre-dependent ( Robinson et al . , 2012 ) . The activity of all cre drivers was confirmed using reporter lines ( Figure 1—figure supplement 2 ) . The most severe phenotype was achieved in hGFAP-cre;H1047R mutants , when doxycycline was administered from embryonic day ( E ) 0 . 5 . All mutants exhibited progressive hydrocephalus and died prior to weaning . Hydrocephalus was evident as a domed forehead at postnatal day ( P ) 21 ( Figure 1b ) . Hematoxylin-eosin stained P3 sections showed ventriculomegaly in the megalencephalic H1047R mutant brains . Strikingly the hippocampus was not evident in these mutants . Instead , the medial tissue was highly dysplastic with multiple infoldings along its entire length ( Figure 1c , d ) . In contrast , when pups were treated with doxycycline from P1 , no morphological differences were observed between the control and the hGFAP-cre;H1047R mutant ( Figure 1—figure supplement 3 ) . Thus the effect of H1047R mutation on brain size was dependent on time of activation . 10 . 7554/eLife . 12703 . 003Figure 1 . Embryonic Pik3ca overactivation in mice causes MEG . ( a , b ) Compared to control , P21 hGFAP-cre;H1047R mutants had domed foreheads . ( c , d ) Coronal section of H&E-stained P3 H1047R mutant showed bigger brain and enlarged lateral ventricles compared to control . Mutant neocortex ( nctx ) was dysplastic and medial tissue highly infolded ( arrowhead; d ) . ( e–g ) P35 hGFAP-cre;E545K and Nestin-cre;E545K brains were noticeably larger than controls , while Nestin-creER;E545K mutants had normal-sized brains compared to controls . Red color of Nestin-creER;E545K brain is due to presence of a lox-stop-lox-Tomato reporter allele , and shows successful induction of cre activity . Controls for e , f and g are of genotypes Pik3ca E545K , hGFAP-cre , Nestin-cre and Nestin-creER . ( h ) MRI volumetric analyses of mutant and corresponding control brains . *p<0 . 0001; ns , not significant . Each data point in the graph represents 1 mouse . ( i–l ) Nissl-stained coronal sections of representative control and mutant brains . Scale bars: 1 mm ( c , d ) ; 2 mm ( i-l ) . See also Figure 1—figure supplements 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 00310 . 7554/eLife . 12703 . 004Figure 1—figure supplement 1 . Genetic strategy for Pik3ca mouse models . ( a ) Schematic of PIK3CA functional domains , highlighting positions of E545K and H1047R activating mutations . ( b ) Genetic strategy for tet-activated H1047R transgenic mice ( Liu et al . , 2011 ) : the human H1047R mutation was activated in the combined presence of cre recombinase and doxycycline ( dox ) . rtTA , reverse tetracycline-controlled transactivator . ( c ) Genetic strategy for E545K conditional knock-in mice ( Robinson et al . , 2012 ) : exon 9 of PIK3CA gene was replaced by an exon containing E545K mutation; and a STOP cassette flanked by loxP recombination sites is introduced in the intron immediately upstream of the exon encoding the transcription initiation site . Cre recombination resulted in removal of STOP cassette , allowing the transcription of the mutant E545K allele . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 00410 . 7554/eLife . 12703 . 005Figure 1—figure supplement 2 . Expression of cre lines . Table of cre expression for ( a ) Nestin-cre , ( b ) hGFAP-cre and ( c ) Nestin-creER induced by tamoxifen at P0 and P1 , using Ai14 and Rosa26-LacZ reporter lines . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 00510 . 7554/eLife . 12703 . 006Figure 1—figure supplement 3 . Neonatal activation of H1047R mutation show no effect on brain morphologyhGFAP-cre;H1047R mutant display normal brain morphology , when doxycycline was administered postnatally from P1 . Scale bar: 2 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 006 E545K mice with the same cre-driver ( hGFAP-cre;E545K ) had a milder phenotype , surviving as adults without hydrocephalus , though their brain size was significantly larger compared to control littermates ( Figure 1e , h , j ) . This provides evidence that with identical time of activation by the same cre driver , the brain phenotypes are dependent on specific Pik3ca allele . Earlier activation of E545K mutation with Nestin-cre led to an even more striking 54 . 4% volumetric increase , with mild ventriculomegaly and no hydrocephalus ( Figure 1f , h , k ) . Interestingly , neonatal activation of E545K using Nestin-creER had no apparent impact on brain size ( Figure 1g , h , l ) . Enlarged head size in both Nestin-cre;E545K and hGFAP-cre;E545K mutants was evident at birth and brain size of all the three adult E545K mutants was relatively stable ( data not shown ) . Unlike H1047R mutants , gross morphology was normal for all E545K mutants . We conclude that brain phenotypes caused by Pik3ca-overactivation are both allele and time dependent . Further , we conclude that to cause brain overgrowth , overactivation of Pik3ca function is required during embryogenesis . To assess the mechanisms causing Pik3ca-dependent embryonic brain enlargement , we focused our analysis on hGFAP-cre;H1047R ( doxycycline from E0 . 5 ) and Nestin-cre;E545K mutants , since these allelic combinations had the most extreme megalencephalic phenotypes . The inner cortical length of hGFAP-cre;H1047R mutants was longer than controls at both E14 . 5 ( p<0 . 01 ) and E16 . 5 ( p<0 . 001; Figure 2b–g ) . This was accompanied by enlarged nuclear and cell size at both ages and decreased cell density at E16 . 5 , but not increased proliferation or cell cycle exit ( Figure 3c–l ) . Total cell number per cortical column length was not significantly different between control and H1047R mutant both at E14 . 5 and E16 . 5 . Also , TUNEL+ cell number was significantly lower in E16 . 5 mutant cortex than in control ( p<0 . 01 ) , indicating reduced apoptosis ( Figure 3—figure supplement 1a , c ) ; however the overall TUNEL+ cell numbers for both control and mutant were small . Together these results indicate that cortical expansion in hGFAP-cre;H1047R mutant was not primarily driven by increased proliferation or reduced apoptosis; rather reduced cell density and increased cell size during embryogenesis were major contributing factors . 10 . 7554/eLife . 12703 . 007Figure 2 . Pik3ca activating mutations lead to increased embryonic cortical length . ( a ) Schematic shows how cortical length and thickness were measured . F , fimbria/cortical hem . Nissl-stained coronal sections of control ( b , e , h , k ) and mutant ( c , f , I , l ) brains . ( b–g ) Cortical length of hGFAP-cre;H1047R mutant at E14 . 5 and E16 . 5 was longer than control; cortical thickness was not different . ( h–m ) Cortical length of Nestin-cre;E545K mutant was longer than control at E16 . 5 but not at E14 . 5; thickness was not different at E14 . 5 but was smaller than control at E16 . 5 . Data are represented as mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001 . Scale bars: 300 μm ( b , c , e , f , h , i , k , l ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 007 By contrast , in Nestin-cre;E545K mutants , the inner cortical length was comparable to controls at E14 . 5 but elongated at E16 . 5 ( p<0 . 01 ) ; cortical thickness was slightly reduced in E16 . 5 ( p<0 . 05 ) as compared to controls ( Figure 2h–m ) . The labeling index was similar to the control at E14 . 5 but increased in E16 . 5 mutants ( p<0 . 05 ) , indicating more proliferation ( Figure 3m , o , r ) . Cell density in E545K mutant neocortex was similar to the control at E14 . 5 but was reduced at E16 . 5 ( p<0 . 05 ) . Total cells per cortical column length did not change in the E545K mutant; but the TUNEL+ cell number was lower in E16 . 5 mutant cortex than in control ( p<0 . 001 ) . Intriguingly , the nuclear size of these mutant cells was similar to controls at both E14 . 5 and E16 . 5 but cell somas were significantly larger ( p<0 . 05 ) at E16 . 5 ( Figure 3n , p , q , s , t ) . The quit fraction indicative of rate of cells exiting cell cycle was significantly higher ( p<0 . 05 ) in the E545K mutant between E15 . 5 and E16 . 5 ( Figure 3u , v ) . At P35 , the neocortical cells of Nestin-cre;E545K mutant were still larger ( p<0 . 05 ) compared to controls ( Figure 3—figure supplement 2a , b ) . Notably , in the adult P35 Nestin-creER;E545K mutant mice , where activation was initiated in neonates , and brain size was not different from controls , E545K-activated ( YFP+ ) cells have the same size as controls ( Figure 3—figure supplement 2c , d ) . We conclude that increased cell size due to E545K overactivation also has a critical embryonic period . Further , changes in multiple developmental parameters including proliferation , cell cycle exit , cell size and density contribute to MEG of Nestin-cre;E545K Pik3ca embryonic overactivation . 10 . 7554/eLife . 12703 . 008Figure 3 . H1047R and E545K mutations differentially affect proliferation , cell density and size in neocortex . ( a ) Schematic shows area of interest ( red box ) in E16 . 5 mouse coronal section , as depicted in c , d , k , m , n , u . ( b ) Experimental outline of the proliferation and cell cycle exit assays . For labeling index , E14 . 5 and E16 . 5 control and mutant brains , harvested after a 1hr BrdU pulse , were processed for BrdU and DAPI staining ( c , m ) . For quit fraction analysis , E16 . 5 control and mutant brains , pulsed with BrdU at E15 . 5 , were processed for BrdU and Ki67 ( k , u ) . Magnified view of DAPI-stained cortical nuclei shows differences in size and density between controls and mutants ( d , n ) . ( e-j , l ) E14 . 5 and E16 . 5 H1047R mutants had similar labeling indices ( BrdU+ cells/total DAPI+ cells ) ; E16 . 5 H1047R mutant neocortex displayed reduced cell density ( x105 DAPI+ cells/mm3 volume ) , larger nuclear and cell size ( μm2 ) and similar quit fraction ( BrdU+Ki67- cells/total BrdU+ cells ) . ( o-q ) E14 . 5 E545K mutant neocortex was similar to control in labeling index , cell density and nuclear size . ( r-t , v ) E16 . 5 E545K mutant showed significantly higher labeling index and quit fraction , reduced cell density , and enlarged cell and nuclear size , compared to controls . Data are represented as mean ± SEM ( e , f , h , i , I , o , p , r , s , v ) or as median-centered box-and-whisker plots ( g , j , q , t ) ; *p<0 . 05; **p<0 . 001; ***p<0 . 0001 . Scale bars: 50 μm ( c , d , m , n ) ; 100 μm ( k , u ) . See also Figure 3—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 00810 . 7554/eLife . 12703 . 009Figure 3—figure supplement 1 . Effect of PIK3CA mutations on total cell numbers and apoptosis . ( a , b ) No significant differences in the total cell numbers per cortical column length were observed in Nestin-cre;E545K and hGFAP-cre;H1047R mutants when compared with their respective control littermates , both at E14 . 5 and E16 . 5 . ( c , d ) TUNEL-positive cell number at E16 . 5 is significantly lower in Nestin-cre;E545K and hGFAP-cre;H1047R mutants than the respective controls . *p<0 . 01; **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 00910 . 7554/eLife . 12703 . 010Figure 3—figure supplement 2 . E545K mutation affects cell size when activated embryonically but not postnatally . ( a ) Cells of P35 control and Nestin-cre;E545K mutant neocortex are marked by pS6 . ( b ) Cell size of P35 E545K mutant was significantly larger than that of control littermate . ( c ) YFP-positive cells of P51 control ( Nestin-creER;YFP ) and mutant ( Nestin-creER; E545K;YFP ) , induced by tamoxifen at P0 and P1 are the cre-recombined cells , DAPI stains the nuclei . ( d ) Size of these cells was not significantly different between the control and Nestin-creER;E545K mutants . Data are represented as median-centered box-and-whisker plot ( b , d ) . white open arrows , control cells; white arrowheads , mutant cells . Scale bars: 100 μm ( a , c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 010 Since disordered lamination is a key feature of human SEGCD ( Arai et al . , 2012; Rossini et al . , 2014 ) , we assessed neocortical organization and development in both hGFAP-cre;H1047R and Nestin-cre;E545K mutants . First , we studied the effect of Pik3ca overactivation on the Nestin-positive radial glial fibers , the scaffold for glial-guided neuronal migration , at multiple developmental stages . In H1047R mutants , the radial glial scaffold was slightly fasciculated and irregular at E14 . 5 and E16 . 5 . Irregularities were very prominent at P0 when a disrupted pial surface was associated with irregular clusters of enlarged radial glial end-feet ( Figure 4—figure supplement 1 ) . The radial glial phenotype was much more subtle in the E545K mutant at E14 . 5 and E16 . 5; however at P0 , we observed thinning of radial glial fibers and irregular clusters of end-feet at the intact pial surface ( Figure 5—figure supplement 1 ) . Cajal-Retzius cells expressing Reelin , a major regulator of radial migration , were normally present in an ordered array in the marginal zone ( layer I ) of controls and Nestin-cre;E545K mice ( Figure 4b; Figure 5b , c ) . However , these cells were dysplastic in hGFAP-cre;H1047R mice at E16 . 5 ( Figure 4c , c’ ) . We did not observe ectopic Reelin-positive cells within the cortical column in either mutant . As expected , within the developing wildtype neocortex , Ctip2 and Tbr1 were expressed predominantly in the early-born , deep layers ( layers V-VI ) , while Cux1 was expressed in late-born upper layers ( layers II-IV ) . hGFAP-cre;H1047R mutants displayed a marked disorganization of all layers . Ctip2/Tbr1-positive as well as Cux1-positive cells in the E16 . 5 H1047R mutant were dispersed throughout the cortical plate , with both early- and late-born neurons severely mislocalized ( Figure 4d–g ) . Laminar disorganization was less severe in E16 . 5 Nestin-cre;E545K brains , but deep Ctip2/Tbr1-positive neurons and upper Cux1-positive neurons were dispersed throughout the cortical plate ( Figure 5f , g ) . 10 . 7554/eLife . 12703 . 011Figure 4 . H1047R mutant mice display abnormal neocortical layering . ( a ) Schematics of mouse brain and section; section corresponds to the marked coronal plane; red box in the section marks the area of neocortex ( nctx ) depicted in the images below . ( b–g ) and ( h–m ) correspond to ages E16 . 5 and P3 respectively . In control cortex , Reelin is in layer I ( b ) , Ctip2 and Tbr1 in deep layers VI and V ( d , j ) , Cux1 in upper layers II-IV ( f , l ) and NeuN in all matured neurons ( h ) . H1047R mutants displayed abnormal distribution of cells for all neocortical layers , observed at E16 . 5 and P3 ( c , e , g , i , k , m ) . ( b’ , c’ ) Magnified view of Reelin-positive cells in control and H1047R mutant . P3 H1047R mutant showed enlarged area between ventricular zone ( vz ) and cortical plate and absence of clear subplate ( sp ) boundary ( h-k ) . A , anterior; P , posterior; yellow dashed lines , lateral ventricular lining; white dotted lines , pial surface; I-VI , neocortical layers; arrowheads , mispositioned mutant cells . Scale bars: 25 μm ( b’ , c’ ) , 50 μm ( b-i ) , 150 μm ( j-o ) . See also Figure 4—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01110 . 7554/eLife . 12703 . 012Figure 4—figure supplement 1 . Nestin expression in hGFAP-cre; H1047R mutant . Nestin-positive radial glial fibres appeared slightly irregular and hyperfasciculated in E14 . 5 H1047R mutant ( b ) but show progressively dysplastic morphology at E16 . 5 ( d ) and P0 ( f ) , compared to respective controls ( a , c , e ) . Yellow boxes ( e , f ) show broken pia and disrupted radial glial end-feet in the P0 H1047R mutant ( f ) . Arrows indicate Nestin-fibers crossed the broken pial surface in the mutant . Scale bars: 50 μm ( a , b ) , 150 μm ( c-f ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01210 . 7554/eLife . 12703 . 013Figure 4—figure supplement 2 . hGFAP-cre;H1047R mutant displays distinct white matter dysplasia . Nissl-stained coronal sections of P3 control ( a , c , e ) and mutant ( b , d , f ) brains; areas with faint or absence of Nissl stain consist of axon fibre tracts . ( a , b ) and ( c , d ) show two comparable antero-posterior planes of section between control and H1047R ( hGFAP-cre;H1047R ) mutant . ( a , c ) P3 Control sections showed presence of anterior commissure ( black arrow ) , corpus callosum ( cc ) and hippocampal commissure ( hc ) . H1047R mutants lacked corpus callosum ( asterisk , b ) while other commissures were present ( d ) . Magnified view of neocortex showed expansion of white matter ( wm ) areas in H1047R mutant compared to control ( e , f ) . ( ei ) and ( fi ) correspond to the dotted boxes in e and f respectively , illustrating presence of increased number of Olig2-positive cells in the expanded white matter area of H1047R mutant . The mutant also had an unclear subplate boundary ( white dotted line; f , fi ) , which is normally seen in the control ( sp; e , ei ) . CP , cortical plate; vz , ventricular zone . Scale bars: 50 μm ( ei , fi ) ; 300 μm ( e , f ) ; 1 mm ( a-d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01310 . 7554/eLife . 12703 . 014Figure 5 . E545K mutant mice display abnormal neocortical upper layers . ( a ) Schematics of mouse brain and section; section corresponds to the marked coronal plane; red box marks the area of neocortex ( nctx ) depicted in the images below . ( b-g ) and ( h-m ) correspond to ages E16 . 5 and P35 respectively . ( b-g ) Compared to control , in E16 . 5 E545K mutant , layer I appeared normal; deep layers lacked normal arrangement while Cux1-positive cells were dispersed throughout the cortical plate . Extent of dispersion was reduced postnatally ( h-m ) . vz , ventricular zone; yellow dashed lines , lateral ventricular lining; white dotted lines , pial surface; I-VI , neocortical layers; arrowheads , mispositioned mutant cells . Scale bars: 50 μm ( b-g ) , 150 μm ( h-m ) . See also Figure 5—figure supplements 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01410 . 7554/eLife . 12703 . 015Figure 5—figure supplement 1 . : Nestin expression in Nestin-cre;E545K mutant . Nestin-positive radial glial fibres appeared normal in E14 . 5 and E16 . 5 Nestin-cre;E545K mutant ( b , d ) but subtle abnormalities in the glial end-feet were observed at P0 ( asterisks , f ) , compared to respective controls ( a , c , e ) . Scale bars: 50 μm ( a , b ) , 150 μm ( c-f ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01510 . 7554/eLife . 12703 . 016Figure 5—figure supplement 2 . Nestin-cre;E545K mutant displays distinct white matter dysplasia . ( a , b ) P3 Nestin-cre;E545K mutants had all three major commissures as in control littermates . ( c , d ) The mutant corpus callosum as well as the corona radiata were thickened compared to controls . ci , cii , di and dii correspond to the dotted boxes in ( c ) and ( d ) respectively . E545K mutants had increased numbers of Olig2-positive cells in an expanded corpus callosum ( di ) and lateral fiber tract ( dii ) with respect to the respective controls ( ci , cii ) . The subplate in the E545K mutant though defined was less packed than the control . CP , cortical plate; cc , corpus callosum; hc , hippocampal commissure; sp , subplate . Scale bars: 50 μm ( ci , cii , di , dii ) ; 1 mm ( a-d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01610 . 7554/eLife . 12703 . 017Figure 5—figure supplement 3 . Astrocytes show no gross dysmorphology in adult Nestin-cre;E545K and Nestin-creER;E545K mutants . S100 is expressed in astrocytes ( a-d ) . No gross change in morphology or number was observed in the Nestin-cre;E545K and Nestin-creER;E545K;YFP mutants compared to controls . Insets ( a , b ) show magnified cells . Scale bars: 100 μm ( a-d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 017 Laminar patterns in postnatal animals remained disrupted in both mutants , with hGFAP-cre;H1047R cortex more affected Nestin-cre;E545K mutant cortex ( Figure 5h–m ) . Thus a simple developmental delay was not the cause of dysplasia ( Figure 4h–m; Figure 5h–m ) . In P3 hGFAP-cre;H1047R mutants , NeuN-positive mature cortical neurons were found within the normally cell-sparse marginal zone as well as in the cortical white matter and residual ventricular zone , a feature reported in SEGCD patients ( Arai et al . , 2012 ) . Further , the cortical subplate was not readily discernible in these mutants ( Figure 4h–k; Figure 4—figure supplement 2e , f ) , blurring the boundary between grey and white matter – a feature often observed in SEGCD patients ( Rossini et al . , 2014 ) . To determine whether the mislocalization of neocortical cells was due to defects in cell fate specification and/or migration , we labeled cells at either E12 . 5 or at E16 . 5 with pulse of BrdU and assessed cortical neuronal location and fate ( layer V; Ctip2+ and layers II/III; Cux1+ ) ( Figure 6a ) . The total numbers of BrdU+ P0 cells , born at E12 . 5 and E16 . 5 , were not significantly different between controls and either hGFAP-cre;H1047R or Nestin-cre;E545K mutants ( Figure 6b , j ) . The distribution of BrdU+ cells showed significant differences between controls and H1047R mutants labelled during both early and late embryonic stages . At P0 , more BrdU+ cells were localized in the lower cortical plate ( CP ) and white matter ( Figure 6c ) . For E545K mutants the BrdU+ cell numbers were not different at either age . The distribution was subtly , yet significantly different only for the early born neurons ( Figure 6k ) . For both the H1047R and E545K Pik3ca activating alleles , total layer V ( Ctip2+ ) cell numbers at P0 were not significantly different between controls and mutants ( Figure 6d , l ) . Also , the numbers of Ctip2/BrdU double-labeled cells were the same in controls and mutants , indicating that cell fate specification for these deep layer neurons was unaffected by either the H1047R or E545K Pik3ca allele ( Figure 6e , m ) . However , similar to the overall BrdU+ cell distribution , the specific distribution of layer V neurons was abnormal in H1047R mutants , with ectopic Ctip2/BrdU double-labeled cells in the upper and lower CP and white matter , instead of mid CP ( Figure 6f ) . In E545K mutants , fewer Ctip2+ cells were positioned in the mid CP ( Figure 6n ) , although the phenotype was much less severe . 10 . 7554/eLife . 12703 . 018Figure 6 . Birthdating assays demonstrate defects in laminar distribution . ( a ) Experimental outline of birthdating assays: BrdU was injected at E12 . 5 and E16 . 5 and analyzed at P0 ( B12 . 5;P0 and B16 . 5;P0 ) . Total number of BrdU+ cells at P0 generated at E12 . 5 and E16 . 5 ( b , j ) , and total number of Ctip2+ cells ( layer V neurons; d , l ) were not significantly different between respective controls and mutants , for both H1047R and E545K lines . ( c ) Distribution of BrdU+ cells in the neocortex was significantly different between control and hGFAP-cre;H1047R mutant for both early and late assays , with more cells residing in the lower cortical plate and white matter instead of mid and upper zones of the cortical plate . ( e , m ) Total number of layer V neurons in both H1047R and E545K mutants , born at E12 . 5 and at E16 . 5 , did not significantly differ from the respective controls; but showed significant difference in their zonal distribution with Ctip2+BrdU+ cells predominating the lower cortical plate in both the mutants ( f , n ) . Total number of Cux1+ neurons ( layers II/III neurons; g , o ) was significantly higher in both the mutants compared with the respective controls . The colocalization of Cux1 and BrdU was not significantly different in the H1047R mutant and control for both ages ( h ) ; but number of Cux1+ cells born at E16 . 5 was significantly higher in E545K mutant than in the control ( p ) . ( i , q ) Zonal distribution of Cux1+ cells was significantly different between controls and mutants , with more Cux1+ cells residing at the lower portion of the P0 cortical plate . The H1047R mutant phenotype is more extreme than the E545K mutant . Data are represented as mean ± SEM . *p<0 . 05; **p<0 . 001; ***p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 018 Upper layer ( Cux1+ ) neuronal numbers and distributions were significantly different in both H1047R and E545K mutants , compared to their respective controls ( Figure 6g , o ) . In E545K mutants , the increase in total Cux1+ cell numbers in the E545K mutant corresponded to increased Cux1/BrdU double-labeled cells , born at E16 . 5 ( Figure 6p ) . However , no such correlation was observed in E12 . 5 or E16 . 5-born Cux1+ cells in the H1047R mutant ( Figure 6h ) . These extra cells were therefore likely born between E16 . 5 and P0 . The distribution of Cux1+ cells was disrupted in both mutants , with the H1047R mutant displaying the more severe phenotype ( Figure 6i , q ) . Together , these data indicate that cell fates are largely correctly specified in both Pik3ca mutants and that cortical dysplasia is more likely caused by aberrant neuronal migration . In P3 hGFAP-cre;H1047R mutants , although the cortical plate itself was not dramatically thicker than controls , the underlying cortical white matter was much thicker ( Figure 4j , k; Figure 4—figure suplement 2e , f ) . This was less pronounced but readily discernible in P3 Nestin-cre;E545K mutants ( Figure 5—figure supplement 2c , d ) . In P3 H1047R mutants , there was complete absence of corpus callosum , although hippocampal and anterior commissures were present ( Figure 4—figure supplement 1a-d ) . In contrast , all major tracts were present in P3 E545K mutants ( Figure 5—figure supplement 2 ) . These data are consistent with the wide spectrum of white matter dysplasia reported in MEG and SEGCD patients ( Conway et al . , 2007; Adamsbaum et al . , 1998; De Rosa et al . , 1992; Jansen et al . , 2015 ) . Moreover , increased number of Olig2-positive cells was observed in the white matter area of both H1047R and E545K mutants ( Figure 4—figure supplement 2 , Figure 5—figure supplement 2 ) . Although astrocytosis is observed when mTOR signaling is activated by TSC mutations in humans and mice ( Sosunov et al . , 2008; Zeng et al . , 2008 ) , it is not a feature of PIK3CA-pathology in our mouse models ( Figure 5—figure supplement 3a-d ) . Epilepsy is one of the most important clinical features of SEGCD ( Bast et al . , 2006; Fauser et al . , 2015; Fauser , 2006; Arai et al . , 2012 ) . Since most of the H1047R mutants were not viable post-weaning , we assessed Nestin-cre;E545K ( megalencephalic ) and Nestin-creER;E545K ( normocephalic ) adults for epilepsy phenotypes . Baseline sleep EEG recordings in both animal models revealed epileptiform activity including sets of spikes/polyspikes , and regional and generalized spike and wave discharges during non-rapid eye movement ( NREM ) sleep ( Figure 7b , c ) . We also conducted additional 2 hr of continuous EEG recording immediately after 5 hr of total sleep deprivation of the Nestin-creER;E545K mice . Sleep deprivation is commonly implemented during epilepsy diagnostic studies in mice and humans and increases the sensitivity and specificity of EEG diagnosis for epilepsy ( De Rosa et al . , 1992; Giorgi et al . , 2013; Binnie and Prior , 1994; Kalume et al . , 2015 ) . The frequency of epileptiform interictal activity was increased in post sleep deprivation EEG recordings , and clinically relevant spontaneous seizures including myoclonic ( MC ) seizures , frequent isolated spikes , and train of spikes , were observed in the Nestin-creER;E545K mice ( Figure 7d , e ) . 10 . 7554/eLife . 12703 . 019Figure 7 . PI3K activity acutely modulates epileptic seizures . ( a ) Schematic shows electrode placement for EEG recordings . LF=Left Frontal , LP= Left Posterior , RF=Right Frontal , RP= Right Posterior . Only 2 electrodes were placed in P35 Nestin-cre;E545K . ( b ) EEG-EMG tracings of Nestin-cre;E545K mutant showed bilateral spikes/polyspikes , myoclonic ( MC ) seizures , fast and slow wave discharges , not associated with movement on video or EMG activity . ( c ) Generalized ( G ) and regional ( R ) spike and wave discharges were observed in Nestin-creER;E545K mice . Scale: 1s , 1mV . ( d , e ) Sleep deprivation ( SD ) enhances epileptiform EEG activity in Nestin-creER;E545K mutant . EEG tracings of a Nestin-creER;E545K mutant mouse after 5 hr of normal sleep ( Pre SD ) and after 5 hr of total sleep deprivation ( Post SD ) in the same mouse ( d ) , the mutant showing myoclonic ( MC ) seizures and isolated regional spikes ( R ) . Power spectrum analysis , representing the frequency distribution for EEG activity over time , also displayed increased activity of the mutant post SD ( e ) . ( f ) Bar chart showing average number of seizures ( SZ ) in PTZ-induced P35 Nestin-cre;E545K and control over time . ( g ) Experimental outline for BKM120-PTZ test . ( h ) Total number of seizures was significantly higher in P35 mutants than controls . Acute administration of BKM120 reduced number of seizures in mutants . ( i ) Duration of sustained generalized tonic-clonic seizure state ( Racine 5 ) , normalized to the total time of test , was significantly longer in P35 Nestin-cre;E545K mutants than controls . BKM120 significantly reduced the duration . Data are represented as mean ± SEM . *p<0 . 05; **p<0 . 0001 . See also Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 01910 . 7554/eLife . 12703 . 020Figure 7—figure supplement 1 . Seizure activity of E545K mutants at old age . ( a ) Experimental outline for ~P180 constitutive ( Nestin-cre;E545K ) mice: Seizures were induced in mice by administering PTZ subcutaneously and then video recorded for 30 min . ( b , c ) Total number of seizures was not significantly different in P180 Nestin-cre;E545K while duration spent by these mutants in severe seizure attacks ( Racine 5 ) , measured as a percentage of the total time of recording , was significantly longer compared to respective controls . ( d ) Table showing percentage of generalized tonic clonic seizures ( GTC SZ ) and mean latency ( in mins ) across different genotypes and age groups . ( e , f ) In ~P35 Nestin-creER;E545K mutants , total seizure number was higher while duration spent in Racine 5 , measured as a percentage of the total time of recording , was the same as respective controls . ( g , h ) In ~P180 Nestin-creER;E545K mutants , total seizure number was comparable to the controls but the duration in Racine 5 was different compared to respective controls . *p<0 . 01; **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 020 When challenged with the chemoconvulsant pentylenetetrazol ( PTZ ) , a GABA-A receptor antagonist ( Macdonald and Barker , 1978 ) , both the megalencephalic and normocephalic E545K mouse models exhibited lower seizure thresholds compared to controls at both P35 and P180 ( Figure 7f–i; Figure 7—figure supplement 1d ) . In the 30 min post PTZ injection , both models showed shorter latencies to first generalized tonic clonic ( GTC ) seizures , more myoclonic seizures , and a prolonged seizure load . We conclude that Pik3ca overactivation is sufficient to cause epilepsy . Further our data indicate that Pik3ca-related epilepsy is dissociable from brain overgrowth and cortical dysplasia . BKM120 , a 2 , 6-dimorpholino pyrimidine derivative , is an orally available pan-Class I PI3K inhibitor currently in clinical trials for solid tumors ( Maira et al . , 2012; Bendell et al . , 2012; Brachmann et al . , 2012 ) and may represent a novel therapeutic agent for PIK3CA-related epilepsy . Preclinical studies show that BKM120 maximally inhibits downstream phosphorylation of Akt , 1hr post-administration ( Maira et al . , 2012 ) . To test its anti-seizure effects in our adult Pik3caE545K gain-of-function megalencephalic and normocephalic models , we administered 50 mg/kg BKM120 ( Maira et al . , 2012 ) by oral gavage 1hr prior to PTZ-challenge at ~P35 . BKM120 increased the seizure threshold of control animals . More importantly , despite the presence of megalencephaly and considerable cortical dysplasia in P35 Nestin-cre;E545K megencephalic animals , BKM120 dramatically decreased the seizure number and duration to untreated control levels and marginally increased seizure latency in the mutant mice ( Figure 7h , i; Figure 7—figure supplement 1d ) . These data powerfully demonstrate that dynamic Pik3ca-dependent processes , independent of cortical and cellular dysplasia , cause Pik3ca-related epilepsy and they are highly amenable to therapeutic intervention . To begin to dissect the cell signaling mechanisms underlying Pik3ca-driven epilepsy , we conducted reverse phase protein array ( RPPA ) analysis to measure protein levels of a comprehensive panel of cell signaling molecules ( Tibes et al . , 2006 ) . We assessed subdissected cortical and hippocampal tissue from untreated ( - ) and PTZ , BKM120 and BKM120 PTZ treated adult control and Nestin-cre;E545K mutants ( Figure 8 , Figure 8—figure supplement 1 ) . As expected , untreated Nestin-cre;E545K mutants exhibited significant elevations of phospho ( p ) S473-Akt and pT346-NDRG1 , consistent with PI3K pathway activation . Notably , baseline pS473-Akt levels in the Nestin-cre;E545K hippocampus were prominently higher than the cortical levels . In both E545K mutant and control brain tissues , PTZ treatment alone increased the levels of pAkt , especially pS473-Akt , of pS6 ( pS235/S236 , pS240/244 ) , pT346-NDRG1 and pS2448-mTOR . As expected , acute BKM120 treatment alone reduced phosphorylation of multiple PI3K pathway members , including AKT , S6 , NDRG1 , GSK3 and 4EBP1 . Most remarkably , BKM120 also inhibited the increased phosphorylation levels induced by PTZ , notably returning mutant hippocampal pS473-AKT levels to baseline untreated control levels . 10 . 7554/eLife . 12703 . 021Figure 8 . BKM120 acutely alter PI3K pathway protein profile . ( a–g ) Graphs show differential protein levels in P35 Nestin-cre;E545K mutant and control brains due to different treatments: untreated ( - ) ; BKM120; PTZ; BKM120+PTZ . Data are represented as mean ± SEM . *p<0 . 05 . Inset shows simplified PI3K pathway; BKM120 significantly regulated the highlighted molecules . See also Figure 8—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 02110 . 7554/eLife . 12703 . 022Figure 8—figure supplement 1 . RPPA analysis graphs . Protein profile of cortical and hippocampal samples from untreated and treated P35 control and Nestin-cre;E545K mutant . Graphs show differential protein levels due to PTZ and BKM120 treatments on control and Nestin-cre;E545K mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 022
Activating PIK3CA mutations have been associated with many human overgrowth disorders categorized based on severity and distribution of the mutation . Involvement of multiple tissues results in CLOVES or Klippel-Trenaunay syndrome with highly mosaic mutation levels ( 0 . 8–32% ) in affected tissues ( Luks et al . , 2015; Kurek et al . , 2012 ) . Involvement of single tissue or body segment results in epidermal nevi , lymphatic malformations or other localized phenotypes with usually no mutations detected in unaffected tissues ( Keppler-Noreuil et al . , 2014; Luks et al . , 2015; Kurek et al . , 2012; Osborn et al . , 2015; Groesser et al . , 2012; Lindhurst et al . , 2012; Hafner et al . , 2007; Rios et al . , 2013; Cohen et al . , 2014 ) . Too few patients and insufficient quantitative data have been reported to observe allele-specific differences . In the brain , mosaic hotspot mutations result in SEGCD , classified as dysplastic MEG , HMEG or FCD2a based on extent of lesion ( Jansen et al . , 2015 ) . PIK3CA mutations were detected in 9/73 patients with HMEG and 1/33 with FCD2 ( Lee et al . , 2012; D'Gama et al . , 2015; Jansen et al . , 2015 ) . These overlapping SEGCD are associated with severe and usually intractable epilepsy ( Bast et al . , 2006; Fauser et al . , 2015; Fauser , 2006; Pasquier et al . , 2002; Russo et al . , 2003; Tassi , 2002; Devlin , 2003; Blümcke et al . , 2011 ) . ~20 other PIK3CA mutant alleles have been seen in MCAP , characterized by MEG or MEG-PMG , hydrocephalus and less severe epilepsy ( Gymnopoulos et al . , 2007; Mirzaa et al . , 2012; Rivière et al . , 2012; Tatton-Brown and Weksberg , 2013 ) . We activated the two most common hotspot mutations , E545K and H1047R , in mouse brain at different developmental timepoints to generate the first models of human PIK3CA-related SEGCD . Our mouse models faithfully recapitulated the most important PIK3CA-related phenotypes of MEG , hydrocephalus , cortical and white matter dysplasia , and epilepsy ( Table 1 ) . 10 . 7554/eLife . 12703 . 023Table 1 . Table displays comparison of the key features across different Pik3ca genetic models used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 12703 . 023Mutant allele H1047R E545K Cre driver hGFAP-cre ( dox from E0 . 5 ) hGFAP-cre ( dox from P1 ) Nestin-cre hGFAP-cre Nestin-creERT2 ( tamoxifen @P0/P1 ) GoF expression onset Late embryonicNeonatalEarly embryonicLate embryonicNeonatalViability Lethal by weaning ageViableViableViableViableMegalencephaly ✓X✓✓ ( intermediate ) XHydrocephalus ✓XXXXIncreased cell size ✓Not tested✓XXCortical dysplasia ✓X✓XXWhite matter dysplasia ✓X✓✓ ( data not shown ) XEpilepsy Not testedNot tested✓Not tested✓ Data from cancer biology suggests that H1047R mutation is more severe than E545K mutation . For example , E545K mutation accounts for 1932/7548 ( 26% ) and H1047R for 2898/7548 ( 38% ) of PIK3CA-coding mutations detected in the COSMIC database of cancer mutations ( http://cancer . sanger . ac . uk/cancergenome/projects/cosmic ) . We found that H1047R and E545K mutations caused distinct phenotypes in mice , H1047R being more severe than E545K . hGFAP-cre;H1047R mutants had severe hydrocephalus and died pre-weaning . In contrast all mice with the E545K allele survived through adulthood without hydrocephalus . Developmental analyses of hGFAP-cre;H1047R and Nestin-cre;E545K embryos revealed common mechanisms , such as larger neurons and lower cell densities , contributing to enlarged brain size , with differences more significant in H1047R mutants . E545K mutation also elevated cortical proliferation and cell cycle exit during late neurogenesis . We do not believe that overexpression of the transgenic H1047R allele versus the knock-in design of the E545K allele underlies the phenotypic differences . The PI3K enzyme is made of p110 ( encoded by Pik3ca ) and p85 subunits . p110 stability is entirely dependent on levels of p85 ( Geering et al . , 2007; Fruman et al . , 2000; Yu et al . , 1998 ) and we have not altered p85 . Rather , the phenotypic differences more likely reflect distinct allele-specific overactivation of PI3K signaling . The H1047R mutation increases the level and duration of response to extracellular ligand , while E545K alters the helical domain resulting in constitutive low level signaling with a blunted response to extracellular ligands ( Miled et al . , 2007; Zhao and Vogt , 2008 ) . These differences likely reflect distinct mechanisms differentially altering PI3K signaling . Whereas H1047R mutation increases the level and duration of response to extracellular ligand , E545K alters the helical domain resulting in constitutive low level signaling with a blunted response to extracellular ligands ( Miled et al . , 2007; Zhao and Vogt , 2008 ) . By activating the E545K mutation in progressively limited progenitor pools , we decreased the size of brain and cells in a graduated fashion . Postnatal E545K activation had no impact on cell/brain size . We conclude that the PIK3CA-related brain overgrowth must arise from mosaic mutations in embryonic neural progenitors . Although neuronal size was enlarged in both hGFAP-cre;H1047R and Nestin-cre;E545K mutants , it was less than that observed in Pten null mice or in cultured hippocampal neurons constitutively overexpressing Akt ( Kumar , 2005; Kwon et al . , 2001 ) . Multiple models of Pten deletion cause progressive increases in postnatal neuronal size and increased brain size without continued proliferation ( Kwon et al . , 2001; Fraser , 2004; Fraser et al . , 2008; Backman et al . , 2001 ) . Our Nestin-cre;E545K mutants had enlarged brain size evident at birth , without progressive increases in postnatal cell size . This is congruent with the analysis of resected human brain tissue from SEGCD patients . Mild cellular enlargement was observed with PIK3CA mutations in contrast to marked enlargement with PTEN or AKT3 mutations ( Jansen et al . , 2015 ) . Brains of SEGCD patients show mild to moderate migration defects in early-born cortical neurons and more severe defects in late-born neurons ( Arai et al . , 2012; Rossini et al . , 2014 ) . Similarly , embryonic activation of H1047R and E545K in mice caused abnormal neocortical lamination , with late-migrating Cux1-positive neurons severely affected in both the mutants . Birthdating studies support the conclusion that Pik3ca activation does not alter cell fate and that cortical dysplasia is predominantly a result of aberrant migration . The severity of dyslamination in H1047R mutants likely reflects the dysplastic Reelin-positive Cajal-Retzius cells . However the Reelin-positive layer remained well defined in both mutants . This is in contrast to the ectopic Reelin expression in neurons expressing high levels of overactive pAKT introduced by electroporation into embryonic mouse cortex ( Baek et al . , 2015 ) . In human SEGCD , late migrating neurons often fail to migrate to the upper layers ( Arai et al . , 2012 ) , a phenotype more severe than seen in any of our mouse models . However , NeuN immunohistochemistry in H1047R mutants confirmed the presence of ectopic neurons in the subcortical white matter , as seen in human SEGCD brain ( Arai et al . , 2012; Salamon , 2006 ) . Human MEG is associated with a wide range of white matter dysplasia ranging from agenesis of corpus callosum to thickening of subcortical axon bundles ( Conway et al . , 2007; Adamsbaum et al . , 1998; De Rosa et al . , 1992; Jansen et al . , 2015; Salamon , 2006 ) . These features were also faithfully recapitulated in our mouse models . Both adult megalencephaic Nestin-cre;E545K and normocephalic Nestin-creER;E545K mice exhibited spontaneous seizures as well as lowered seizure thresholds upon PTZ-seizure induction . Although cortical dysplasia resulted from embryonic activation of Pik3ca in Nestin-cre;E545K mice , postnatal activation of Nestin-creER;E545K did not cause increased cell size or megalencephaly or altered cortical lamination . Thus Pik3ca-dependent epilepsy is independent of dysmorphology . Further , inhibitory interneurons were not grossly perturbed in Nestin-creER;E545K mice ( data not shown ) . This is congruent with the fact that these interneurons are born at embryonic stages and their migration is almost complete before birth ( Batista‐Brito and Fishell , 2009 ) . Therefore altered interneuron development in Nestin-cre;E545K may contribute to epilepsy , aberrant interneuron development cannot represent a common mechanism for epilepsy in both models . The observation that acute BKM120 treatment is sufficient to inhibit PTZ-induced seizures even in adult megalencephalic mice supports the argument that the epileptic seizures are independent of dysplasia since the latter is not reversed over the short course of treatment . This is an important finding since a large portion of FCD patients who do not show detectable dysplasia suffer from intractable epilepsy ( Bernasconi et al . , 2011 ) . Proteomic analyses of cell signaling networks in megalencephalic cortical and hippocampal tissue at baseline and treated with PTZ and/or BKM120 provide insight into the mechanism of Pik3ca-dependent epilepsy . Nestin-cre;E545K mutants had elevated PI3K signaling with a more robust upregulation of mTOR-dependent pS473-Akt than the direct PDK1-dependent pT308-Akt , similar to the recent findings in human brain samples with PIK3CA mutations ( Jansen et al . , 2015 ) . The modest changes in signaling compared to controls is congruent with previous studies which demonstrated only modest changes in the steady-state levels of PI3K signaling in breast cancer cells with PIK3CA mutations ( Stemke-Hale et al . , 2008 ) . Higher signaling levels in hippocampus versus cortex suggest a more prominent role of hippocampus in the seizure phenotype . PTZ administration alone in both controls and megalencephalic Nestin-cre;E545K mutants caused upregulation of many core components of PI3K-AKT pathway , including pAkt , pS6 and pNDRG1 . This is congruent with a report showing PTZ-induced seizures in rats upregulated PI3K-AKT-mTOR pathway ( Zhang and Wong , 2012 ) and suggests that elevated baseline PI3K signaling levels are epileptogenic . Indeed , there is extensive human and mouse evidence that elevated mTOR signaling is epileptogenic although the mechanisms for the epilepsy are incompletely understood . A number of mechanisms including altered development , cell size , growth , proliferation and circuitry have been reported ( Wong and Crino , 2012 ) . Most remarkably however , our acute BKM120 administration data clearly demonstrates that histopathological mechanisms are not the primary epilepsy drivers . Acute 1 hr of BKM120 administration was sufficient to completely inhibit the increased phosphorylation levels induced by PTZ , notably returning mutant hippocampal mTOR-dependent pS473-AKT levels to baseline untreated control levels . This was sufficient to normalize the PTZ-seizure induction threshold , despite continued dysplasia in Nestin-cre;E545K mutants . We conclude that elevated PI3K signaling is itself actively epileptogenic , independent of underlying developmental pathology . The discovery that Pik3ca-related epilepsy is independent of dysplasia and susceptible to acute modulation is a major and paradigm shifting finding . Since PIK3CA resides at the top of the PI3K-AKT pathway , our mouse models represent surrogates for the entire group of patients with segmental brain overgrowth , including patients with somatic mosaic mTOR and AKT3 mutations ( Keppler-Noreuil et al . , 2014; Lee et al . , 2012; D'Gama et al . , 2015; Conway et al . , 2007; Jansen et al . , 2015; Mirzaa et al . , 2012 ) . SEGCD is associated with early onset , severe and frequently intractable epilepsy that responds poorly to standard seizure medications ( Bast et al . , 2006; Fauser et al . , 2015; Fauser , 2006; Pasquier et al . , 2002; Russo et al . , 2003; Tassi , 2002 ) . Epilepsy surgery has been comparatively more successful ( 73% ) in combating seizures in the same children ( Fauser et al . , 2015 ) . A drug-based therapy however , would clearly be preferable . mTOR inhibition with rapamycin has shown therapeutic promise in FCD patients and animals models ( Baek et al . , 2015; Curatolo and Moavero , 2013; Lim et al . , 2015; Moon et al . , 2015 ) ; however , rapamycin treatments are not acute . Our data demonstrates that acute small molecule-based modulation of PI3K signaling , despite the presence of dysplasia , has dramatic therapeutic benefit . This suggests that PI3K inhibitors offer a promising new avenue for effective antiepileptic therapy for large cohorts intractable pediatric epilepsy patients .
The following mouse lines were used: Nestin-cre ( Jackson Labs , Bar Harbor , Maine , USA; Stock #003771 ) , Nestin-creERT2 lines ( Jackson Laboratory , Bar Harbor , Maine , USA , MGI:3641212 and line generated in SJB’s lab , Zhu et al . , 2012 ) , human glial fibrillary acidic protein ( hGFAP ) -cre ( Jackson Labs , Stock #004600 ) , Pik3caH1047R transgenic ( human H1047R transgene expression is under the control of a tetracycline-inducible promoter ( TetO ) ) ( Liu et al . , 2011 ) , Rosa26-rtTA line ( Jackson Labs , Stock #005670 ) , Pik3caE545K knock-in ( Robinson et al . , 2012 ) , Ai-9 ( Jackson Labs , Stock #007905 ) , Ai-14 ( Jackson Labs , Stock #007914 ) , R26-LSL-EYFP ( Jackson Labs , Stock #006148 ) , Rosa26-LacZ ( Jackson Labs , Stock #003474 ) . We have designated the Pik3caH1047R and Pik3caE545K conditional mutant mice as H1047R and E545K mutants/lines throughout the manuscript . All lines were maintained on a mixed genetic background , comprising of FVB , C57Bl6 , 129 and CD1 strains . Noon of the day of vaginal plug was designated as embryonic day 0 . 5 ( E0 . 5 ) . The day of birth was designated as postnatal day 0 ( P0 ) . The H1047R and Rosa26-rtTA lines were intercrossed and female mice positive for both these alleles were crossed with hGFAP-cre;RosartTA;Pik3caH1047R males . To ensure that cre and Pik3caH1047R mutant transgene expression was correlated plugged females were treated with doxycycline ( Sigma; 2 mg/ml ) from E0 . 5 available ad libitum in drinking water . For the neonatal induction experiment , the pups were treated with doxycycline from P1 . The E545K line was crossed to reporter lines to obtain E545K floxed allele and the reporter in the same mouse line . Tamoxifen ( Sigma T5648 ) was dissolved at 37°C in corn oil ( Sigma ) at 5 mg/ml and was administered intraperitoneally to pups of the cross Nestin-creER X Pik3caE545K mice at a dose of 75 μg/g body weight , once a day at P0 and P1 , to activate the E545K mutation postnatally . hGFAP–cre , Nestin–cre and Nestin–creERT2 mice were genotyped by PCR using primers for the cre coding region , as previously described ( Chizhikov , 2006 ) . Genotyping of other alleles were done according to the following references: H1047R and Rosa+/- ( Liu et al . , 2011 ) , E545Kfloxed/+ ( Robinson et al . , 2012 ) , EYFP/+ and Ai9/+ ( Zhu et al . , 2012 ) . All mouse procedures were approved by the Institutional Animal Care and Use Committees . Embryos and postnatal pups were harvested in phosphate buffer saline ( PBS ) ; brains fixed in 4% paraformaldehyde ( PFA ) for 4 hr , equilibrated in 30% ( wt/vol ) sucrose made in PBS , and sectioned at 25 μm on a freezing microtome . Adult mice were perfused with 4% PFA , brains collected and fixed in 4% PFA overnight , sunk in 30% sucrose in PBS , embedded in optimum cutting temperature ( OCT ) compound and sectioned at 12 μm on a cryostat . Sections were then processed for Nissl , hematoxylin and eosin ( H&E ) or immunohistochemical staining . Immunohistochemistry: Sections were washed thrice in PBS , boiled in 10 mM Sodium citrate solution for antigen retrieval , blocked in 5% serum in PBS with 0 . 1%Triton X-100 and then incubated overnight at 4°C with primary antibodies . The next day , sections were washed thrice in PBS , incubated with appropriate species-specific secondary antibodies conjugated with Alexa 488 , 568 , 594 or 647 fluorophores ( Invitrogen ) for 2 h at room temperature and then counterstained with DAPI to visualize nuclei . Sections were coverslipped using Fluorogel ( EMS #17985 ) mounting medium . Immunostained sections were imaged in Zeiss LSM 710 Imager Z2 laser scanning confocal microscope using Zen 2009 software and later processed in ImageJ software ( NIH , Bethesda , Maryland , USA ) . Primary antibodies used are: rat anti-BrdU ( Abcam ) , mouse anti-BrdU ( Roche ) , rabbit anti-Tbr1 ( EMD Millipore ) , mouse anti-Tbr2 ( EMD Millipore ) , rat anti-Ctip2 ( Abcam ) , rabbit anti-Cux1/CDP ( Santa Cruz Biotechnology ) , rabbit anti-pS6 ( Cell Signaling ) , mouse anti-NeuN ( EMD Millipore ) , mouse anti-Reelin ( EMD Millipore ) , rabbit anti-Laminin ( Sigma ) , rabbit anti-Olig2 ( EMD Millipore ) , mouse anti-Nestin ( EMD Millipore ) , chicken anti-YFP ( Abcam ) , rabbit anti-Ki67 ( Vector Lab ) , mouse anti-S100 ( Abcam ) . Nissl and H&E staining: Sections were stained in 0 . 1% cresyl violet solution for 10 min , rinsed quickly in distilled water , dehydrated in 95% ethanol , and left in xylene before being coverslipped with Permount ( Fischer Scientific ) . H&E staining was performed by passing the sections through Harris modified Hematoxylin solution ( Fisher Scientific ) and EosinY ( Sigma ) and then dehydrating them in increasing grades of ethanol before dipping in xylene and coverslipping . Brightfield images were taken in Leica MZFLIII microscope using Leica DFC425 camera and LAS V3 . 8 software . Bromodeoxyuridine ( BrdU; Life Technologies ) was administered intraperitoneally ( 100 μg/g of body weight ) to pregnant mice at E14 . 5/16 . 5 for 1 hr , at E15 . 5 for 1 day and at E12 . 5/E16 . 5 for proliferation assays , cell cycle exit and birthdating experiments respectively . S-phase labeling index ( LI ) was calculated by dividing total BrdU+ cells by total number of DAPI+ cells . Quit fraction was calculated by dividing BrdU+Ki67− cells by total number of BrdU+ cells . Brain sections were briefly fixed , washed in wash buffer at room temperature and then stained overnight at 37°C in the staining solution comprising of the X-gal substrate . The sections were then washed in wash buffer at room temperature and stored at 4°C . TUNEL staining was processed on E16 . 5 control and mutant sections using Roche In situ Cell Death Detection Kit , Fluorescein . At least 5 mice of each genotype ( age P40–60 ) were used for volumetric analyses . MRI study was performed using a 7 T Bruker ClinScan system ( Bruker BioSpin MRI GmbH , Germany ) equipped with 12S gradient coil . A 2-channel surface coil was used for MR imaging . Animals were anesthetized and maintained with 1 . 5% isoflurane during MRI sessions . Transverse T2-weighted turbo spin echo images were acquired for volume measurements ( TR/TE = 3660/50 ms , FOV = 25 × 25 mm , matrix = 320 × 320 , NEX = 1 , thickness = 0 . 4 mm , scan time = 6 . 5 min ) . Total brain volumes were obtained by manually segmenting brain regions from olfactory bulbs to cerebellum , and computing volumes using OsiriX ( Pixmeo , Switzerland ) . Each data point in the graph represents 1 mouse . Mice obtained from the following crosses were used for experiments at ~P35 ( young age ) and ~P180 ( old age ) : Nestin-cre/+ X Pik3caE545Kfloxed/+ and Nestin-creER/+ X Pik3caE545Kfloxed/+ . At least 5 animals of each genotype were used per treatment experiment . Pentylenetetrazole ( PTZ ) seizure test . Mice were subcutaneously injected with PTZ ( Sigma ) , a GABA ( A ) receptor-antagonist , at 40 mg/kg body weight and digital videos of the mice were recorded for 30 min post-PTZ injection . Principal behavior in each 10 second-bin of the recorded video was scored as 4 or 5 using the Racine scale of seizure severity ( 4 , rearing with forelimb clonus; and 5 , rearing and falling with forelimb clonus ) ( Kalume , 2013; Racine , 1972 ) . Treatment trials . Pan-PI3K inhibitor BKM-120 ( Novartis; 50 mg/kg body weight , dissolved in 0 . 5% Tween-80 , 0 . 5% methylcellulose ) or saline was administered by oral gavage to the mice 1hr before PTZ seizure test . Sleep deprivation ( SD ) . To permit control of circadian variations of sleep in these experiments , baseline ( control ) sleep data ( Pre SD ) were recorded from mice one day before they were submitted to total sleep deprivation . Mice were allowed to sleep normally for 5 continuous hours beginning at 8:00 AM , and then baseline sleep video-EEG recordings were obtained continuously in the 1 subsequent hour . On the following day , beginning at 8:00 AM , the same mice were kept awake for 5 consecutive hours by random gentle touches with a rotating light curtain attached to a motor mounted on the lid of the sleep deprivation chamber . The motor was in turn , connected to a computer via Power Lab ( ADInstruments , Colorado Spring , CO ) . The random direction and speed of the motor rotation were custom-programmed in the stimulator panel dialog box of LabChart 8 Software ( ADInstruments , Colorado Spring , CO ) . The specific parameters used are tabulated as Supplementary file 1 . Post sleep deprivation , mice were not disturbed and post SD recordings were obtained for 2hr . Video-electroencephalagraphy-electromyography ( Video-EEG-EMG ) recording . These experiments were performed as previously described ( Kalume , 2013 ) . Briefly , mice underwent survival surgery to implant fine ( diameter: 130 µm bare; 180 µm coated ) silver wire EEG and EMG electrodes under isoflurane anesthesia . Four EEG electrodes were placed bilaterally through the small cranial burr holes over the posterior and frontal cortices and were fixed in place with cyanoacrylate glue and dental cement ( Lang Dental Manufacturing Co . , Inc . , Wheeling , IL ) . Similarly , one reference electrode was placed above the cerebellum . A ground electrode was inserted subcutaneously over the back . EMG electrodes were placed in back muscles . Only 2 electrodes were implanted in the young Nestin-cre;E545K mutant and control mice . Mice were allowed to recover from surgery for 2-3 days . Simultaneous video-EEG-EMG recordings were collected from conscious mice on a PowerLab 8/35 data acquisition unit using LabChart 7 . 3 . 3 software ( AD Instruments , Colorado Spring , Co ) . All bioelectrical signals were acquired at 1KHz sampling rate . The EEG signals were processed off-line with a 1-70 Hz bandpass filter and the ECG signals with a 3-Hz highpass filter . Interictal spikes were identified as transient , clearly distinguished from background activity , with pointed peak and short duration . Myoclonic seizures were identified as shock-like jerks of the muscles on video associated with a spike or polyspike-wave complex on EEG . Cortex and hippocampus were dissected out of P35 control and E545K mutant mice , following different treatments ( vehicle only ( - ) , +BKM120 , +PTZ , BKM120+PTZ ) , and flash-frozen in liquid nitrogen then sent to the RPPA Core Facility at MD Anderson Cancer Center , University of Texas . Three independent biological replicates per sample were analyzed . Analysis was performed as previously described ( Tibes et al . , 2006 ) . The mouse brain tissue samples were lysed and underwent protein extraction . Cellular protein was denatured by SDS sample buffer and serial dilution was made for each sample . Cell lysates were then probed with different validated antibodies . Signals were detected by DAB colorimetric reaction and intensity was quantified using ArrayPro software . Protein concentration was determined by super curve fitting . All the data points were normalized for protein loading and transformed to linear value . These linear values were used to make bar graphs for comparative analysis . See http://www . mdanderson . org/education-and-research/resources-for-professionals/scientific-resources/core-facilities-and-services/functional-proteomics-rppa-core/index . html ) for a detailed antibody list and protocols . For quantitative analysis of embryos , data was collected from comparable sections of a minimum of 3 embryos of each genotype ( from 2 or more independent litters ) at each developmental stage . Cortical length was measured in the lateral ventricular lining from the tip of the fimbria/cortical hem to the pallial-subpallial boundary . Cortical length and thickness were measured using ImageJ software ( NIH , Bethesda , Maryland , USA ) ; the data was normalized to the control value . Cell counts from E14 . 5 and E16 . 5 brains were obtained from 25% of the neocortex . Area of interest was derived by dividing the whole length of neocortex into quarters and then taking images of the total area , from pia to ventricle , in the third quartile from dorsal midline . Confocal stacks of immunostained sections of each developmental stage were generated by scanning at intervals of 0 . 99 μm using filters of appropriate wavelengths at 20X and 40X magnifications . Confocal images of DAPI-stained brain sections and NeuN/pS6-immunostained sections were used to measure nuclear and cell size respectively . Measurements for labeling index , quit fraction , birthdating studies , cell density and size were calculated using ImageJ . For zonal quantification of cells , the cortical column was divided into 5 different parts – the ventricular-subventricular zone ( vz/svz ) , white matter , and 3 equally divided zones of the neocortical plate ( lower , mid , upper ) . Statistical significance was assessed using 2-tailed unpaired t-tests ( for cortical length and thickness , cell density , nuclear size , TUNEL assay , labeling index , quit fraction , total cell counts and seizure data ) and ANOVA followed by Bonferroni ( for cell size , BKM treatment data , birthdating experiments ) and Tukey ( for RPPA graphs ) post-tests . These analyses were performed in GraphPad Prism v5 . 01 ( GraphPad Software Inc . , San Diego , USA ) or in Igor Pro v6 . 3 . 6 . 4 , Igor Pro Software , Lake Oswego , USA . Differences were considered significant at P< 0 . 05 . | An enzyme called PI3K is involved in a major signaling pathway that controls cell growth . Mutations in this pathway have devastating consequences . When such mutations happen in adults , they can lead to cancer . Mutations that occur in embryos can cause major developmental birth defects , including abnormally large brains . After birth , these developmental problems can cause intellectual disabilities , autism and epilepsy . Children with this kind of epilepsy often do not respond to currently available seizure medications . There are several outstanding questions that if answered could help efforts to develop treatments for children with brain growth disorders . Firstly , how do the developmental abnormalities happen ? Do the abnormalities themselves cause epilepsy ? And can drugs that target this pathway , and are already in clinical trials for cancer , control seizures ? Now , Roy et al . have made mouse models of these human developmental brain disorders and used them to answer these questions . The mice were genetically engineered to have various mutations in the gene that encodes the catalytic subunit of the PI3K enzyme . The mutations were the same as those found in people with brain overgrowth disorders , and were activated only in the developing brain of the mice . These mutations caused enlarged brain size , fluid accumulation in the brain , brain malformations and epilepsy in developing mice – thus mimicking the human birth defects . The severity of these symptoms depended on the specific mutation and when the mutant genes were turned on during development . Next , Roy et al . studied these mice to see if the seizures could be treated using a drug , that has already been developed for brain cancer . This drug specifically targets and reduces the activity of PI3K . Adult mutant mice with brain malformations were treated for just one hour; this dramatically reduced their seizures . These experiments prove that seizures associated with this kind of brain overgrowth disorder are driven by ongoing abnormal PI3K activity and can be treated even when underlying brain abnormalities persist . Roy et al . suggest that drugs targeting PI3K might help treat seizures in children with these brain overgrowth disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Mouse models of human PIK3CA-related brain overgrowth have acutely treatable epilepsy |
How dietary selection affects genome evolution to define the optimal range of nutrient intake is a poorly understood question with medical relevance . We have addressed this question by analyzing Drosophila simulans and sechellia , recently diverged species with differential diet choice . D . sechellia larvae , specialized to a nutrient scarce diet , did not survive on sugar-rich conditions , while the generalist species D . simulans was sugar tolerant . Sugar tolerance in D . simulans was a tradeoff for performance on low-energy diet and was associated with global reprogramming of metabolic gene expression . Hybridization and phenotype-based introgression revealed the genomic regions of D . simulans that were sufficient for sugar tolerance . These regions included genes that are involved in mitochondrial ribosome biogenesis and intracellular signaling , such as PPP1R15/Gadd34 and SERCA , which contributed to sugar tolerance . In conclusion , genomic variation affecting genes involved in global metabolic control defines the optimal range for dietary macronutrient composition .
Animals require macronutrients to sustain growth , reproduction and repair over their lifetimes and the balance between nutrients has been shown to have significant effects on development , reproduction and longevity ( Raubenheimer and Simpson , 1997; Lee et al . , 2008; Arganda et al . , 2017 ) . Most animals consume a variety of different foods to meet their nutritional needs ( Raubenheimer and Simpson , 1997; Lee et al . , 2008 ) and even closely related species make highly distinct diet choices ( Tinker et al . , 2008; Goldman-Huertas et al . , 2015; Salinas-Ramos et al . , 2015; Costello et al . , 2016; Han et al . , 2016 ) . Therefore , it is conceivable that the impact of nutrient composition on various life history traits depends on the genetic makeup of the animal . Some closely related species are distinguished by variation in morphological structures that are specialized for obtaining nutrients from unique resources ( trophic morphology ) ( Malinsky et al . , 2015; Parsons et al . , 2016; Santana and Cheung , 2016 ) . Darwin’s finches are the classic example of species that are differentiated in part by interspecific competition and specialization on an under-used food type ( De León et al . , 2014 ) . Darwin’s finches are considered ‘imperfect dietary generalists’ ( De León et al . , 2014 ) having similar preferred diets that overlap among species but specialize on a unique food when the preferred diet is limiting . Such difference in diet flexibility suggests that in addition to morphological differences , animals might also display differential metabolic flexibility , that is the capacity to adapt nutrient use to nutrient availability . While much recent attention has been paid to the genetics that underlie plasticity of trophic morphology in animals ( Malinsky et al . , 2015; Ledogar et al . , 2016; McGirr and Martin , 2017; Burress et al . , 2017; Zelditch et al . , 2017 ) , less focus has been placed on metabolic regulators with regard to diet choice ( Turner and Thompson , 2013 ) . Metabolic phenotypes and diet tolerance is observed to vary with ecological diversification within and between species ( Matzkin et al . , 2009; Reed et al . , 2010; Matzkin et al . , 2011 ) , and these phenotypic changes correlate with changes in gene expression ( Nazario-Yepiz et al . , 2017 ) . However , it remains poorly understood , which genetic changes are causally important during evolution of diet choice and what kind of metabolic tradeoffs might emerge from adaptation to a new macronutrient composition . Flexibility in the usage of metabolic pathways allows animals to accommodate changes in food nutrient content and availability . At the level of the organism , systemic nutrient levels are actively monitored by the so-called nutrient-sensing pathways composed of intra- and intercellular signaling pathways and gene regulatory networks , which ultimately control the activity of metabolic pathways ( Mattila and Hietakangas , 2017 ) . There are specific nutrient-sensing mechanisms for each type of macronutrient . For example , protein kinases mTOR complex 1 ( mTORC1 ) and GCN2 respond to changes in amino acid availability ( Efeyan et al . , 2015 ) , while the transcription factor complex Mondo/ChREBP-Mlx is activated in response to sugars ( Havula and Hietakangas , 2012 ) . In Drosophila , genetic mutations that impair nutrient-sensing pathways have revealed diet-specific phenotypes . Adult mutants of 4EBP , a target of mTORC1 , are indistinguishable from controls when fed a protein-rich diet , but are sensitive to amino acid starvation ( Teleman et al . , 2005 ) . On the other hand , mlx mutant larvae have impaired function of the sugar sensor Mondo/ChREBP-Mlx and grow normally when fed a low sugar diet ( LSD ) , but are intolerant of high-sugar diet ( HSD ) ( Havula et al . , 2013 ) . The mlx null mutants exhibit impaired growth , increased larval development time and reduced larvae to pupae survival when the high dietary sugar concentration is within the range available from natural food sources ( Havula et al . , 2013 ) . Thus , nutrient-sensing pathways define the tolerated lower and upper limits of nutrient intake . These limits for each nutrient will further depend on the availability of other nutrients . We call the inclusive matrix of tolerated macronutrient contents the ‘macronutrient space’ . Here , we aimed to explore the natural variation of macronutrient space in closely related species . We hypothesized that the natural variation of diet choice as well as diet flexibility ( specialist vs . generalist ) is affected by genetically encoded differences that define the macronutrient space . To test this hypothesis , we studied two closely related Drosophila species that differ in diet choice , namely the generalist D . simulans and its specialist relative , D . sechellia . In nature , D . simulans larvae consume a range of decaying fruits that may contain high levels of sugars , whereas D . sechellia larvae grow on the unripe fruits of Morinda citrifolia , which has a low-sugar content ( Singh et al . , 2012 ) . The two species occur together on islands of the Seychelles archipelago; however , D . sechellia adults and larvae are found infrequently on fruits other than that of Morinda ( R'Kha et al . , 1991; Matute and Ayroles , 2014 ) . D . simulans and D . sechellia show strong dietary differentiation yet they are closely related and can form fertile female F1 hybrids ( Lachaise et al . , 1986 ) . This makes the two species and their hybrids a tractable system for studying the genetics associated with determination of macronutrient space .
Because the natural larval diet of the generalist species D . simulans may have a highly variable sugar content compared to that of the specialist species D . sechellia , we predicted that egg to pupa development time of these species would be dissociated along the sugar axis in a yeast × sugar macronutrient space . To test this prediction , we characterized larval development time and survival to pupa for both species across a macronutrient space consisting of a 5 × 5 grid of diets that varied in sucrose and yeast content . The species showed different phenotypes across the grid of diets ( Figure 1A , Table 1 ) . D . simulans larvae displayed rapid development and high larval survival on diets up to and including 20% sugar . In contrast , D . sechellia larvae displayed a more restricted space , with slowed development and reduced survival on high-protein diets containing >10% sugar and complete lethality on diet composed of 20% sucrose/20% yeast ( Figure 1A , Table 1 ) . For D . sechellia larvae , dietary sucrose concentration showed significant ( p < 0 . 001 ) positive correlation with lengthened development time and significant ( p < 0 . 001 ) negative correlation with survival , but such correlations were not observed for D . simulans larvae ( Table 1 ) . For both species , larval development time was negatively correlated and survival was positively correlated with dietary yeast concentration; however , the correlation was weaker for D . sechellia than for D . simulans ( Table 1 ) . To test the possibility that D . simulans and D . sechellia larvae differed by a behavioral feeding response to sugar , we assayed mouth hook extension rate for both species in the presence and absence of 20% sucrose ( Shen , 2012; Scheiner et al . , 2014 ) . ANOVA showed no significant effect of species ( F ( 1 , 12 ) = 0 . 06 , p = 0 . 81 ) , sugar concentration ( F ( 1 , 12 ) = 0 . 42 , p = 0 . 53 ) , or their interaction ( F ( 1 , 12 ) = 4 . 47 , p = 0 . 06 ) on feeding behavior ( Figure 1—figure supplement 1 ) . Since nutrition affected both larval development time and survival similarly , we combined the data and calculated a so-called ‘pupariation index’ ( Pupind ) that takes both parameters into account . A high Pupind score is achieved with shorter larval development time and higher survival to pupal stage . Analysis of the Pupind of D . simulans and sechellia confirmed the poor performance of sechellia on high-sugar diets ( Figure 1A , Table 1 ) . We further analyzed our data by using a full generalized linear model ( glm ) , which showed significant effects of genotype , percent sugar , percent yeast , and all interactions of the main effects on larval development time , larval survival , and Pupind ( Table 2 ) . However , while the effect of yeast on Pupind was stronger for D . simulans than for D . sechellia ( ω2 = 0 . 94 and 0 . 33 , respectively ) , the effect of sugar on Pupind was substantially stronger for D . sechellia ( ω2 = 0 . 01 and 0 . 35 for D . simulans and D . sechellia , respectively ) ( Table 3 ) . This further supports the conclusion that D . sechellia is sugar intolerant . To confirm a genetic basis for the observed differential sugar tolerance between species , we assayed the larval development time and survival to pupa of D . simulans × D . sechellia F1 hybrid larvae across the 5 × 5 grid of diets . D . simulans and D . sechellia are closely related , having diverged from a common ancestor roughly 0 . 4 million years ago ( Kliman et al . , 2000 ) , and hybrid females from the cross , D . sechellia male × D . simulans female , are fertile ( Lachaise et al . , 1986 ) . We observed a clear rescue of sugar tolerant larval development and survival in the F1 hybrid ( Figure 1B ) . This implied that the sugar intolerance phenotype of D . sechellia may be caused by altered function of genes underlying sugar tolerance To generate flies having the minimal D . simulans genomic regions essential for sugar tolerance in a mostly D . sechellia genomic background , we sought to introgress the sugar tolerance phenotype from D . simulans into a mostly D . sechellia genetic background . To do this , we used the phenotype-based introgression approach of Earley and Jones ( Earley and Jones , 2011 ) ( Figure 2A ) . Dietary sugar content of 20% provided a strong selection , since no survivors of the D . sechellia parental line were observed in these conditions . After 10 generations of backcrossing with selection , we observed that tolerance of an HSD ( 20% yeast/20% sugar ) in the backcross larvae was equal to that of D . simulans ( Figure 2B ) . A control fly line that was backcrossed in the same manner , but not selected on a high-sugar diet , showed only minimal tolerance for HSD ( Figure 2B ) . Since the introgression was performed by repeated backcrossing of D . sechellia males with the hybrid line ( maternally D . simulans ) , the mitochondrial genomes of the selected and control lines are the same . Therefore , the observed phenotypic differences are due to the nuclear genome . Morphologically , the introgressed lines resemble D . sechellia , including genital arch morphology ( data not shown ) . Metabolic analysis of the parental and introgressed lines revealed that the sugar intolerant D . sechellia and no-selection control lines were less efficient in clearing glucose from circulation after challenge with high-sugar diet ( Figure 2C ) . This suggests that the pathways controlling energy metabolism or their response toward high-sugar diet are affected by the genomic regions underlying sugar tolerance . In order to achieve a genome-wide view of the gene expression profiles in the sugar tolerant and intolerant lines , we used RNAseq analysis and assayed 3rd instar larvae fed continuously on LSD ( 20% yeast ) as well as following acute exposure to HSD ( 20% yeast/20% sugar for 8 hr ) ( Figure 3A ) . Global comparison of the gene expression by sample clustering revealed striking association between expression profiles and sugar tolerance . The gene expression profile of the sugar-selected hybrid had high similarity with that of D . simulans , while the sugar intolerant control hybrid clustered close to D . sechellia ( Figure 3B ) . This implies , surprisingly , that the genetic factors underlying the differences in sugar tolerance explain the majority of the differential gene expression between the parental species . We further identified the genes that were differentially expressed in the tolerant vs . intolerant genotypes , focusing on genes that differed significantly ( p < 0 . 05 ) when both sugar tolerant genotypes were compared to both sugar intolerant genotypes on HSD . Genes with reduced expression in both sugar intolerant lines displayed significant ( p < 0 . 05 ) enrichment in functional categories related to mitochondrial ribosome , detoxification ( e . g . cytochrome P450 and glutathione metabolism ) , growth control ( ribosome biogenesis ) , carbohydrate metabolism ( starch and sucrose metabolism ) as well as several categories related to amino acid metabolism ( Figure 3C ) . On the other hand , genes with high expression in sugar intolerant lines displayed overrepresentation of proteolysis and lysosome ( Figure 3C ) . We have earlier observed that the mlx1 null mutant larvae , lacking functional sugar sensing by Mondo-Mlx , display strong sugar intolerance , similar to D . sechellia ( Havula et al . , 2013 ) . To test if the sugar intolerant D . sechellia lines show similarities to mlx1 mutants in gene regulation , we compared the simulans/sechellia RNAseq dataset with that of mlx1 null mutant , published earlier ( Mattila et al . , 2015 ) . There was significant similarity between the gene expression profiles of the sugar intolerant genotypes ( Figure 4A ) . For example , 30% ( 174/587; p = 1 . 2×10−69 ) of the genes downregulated in mlx1 mutants displayed reduced expression in D . sechellia and the control hybrid line ( Figure 4A ) . Global comparison of all genes with similar gene expression differences in tolerant and intolerant lines revealed a high degree of similarity in the gene expression patterns ( Figure 4B ) . These overlapping gene sets included several genes that were upregulated upon sugar feeding in sugar tolerant genotypes , but the activation was either absent or reduced in the sugar intolerant genotypes ( Figure 4B ) . These include , for example the transcription factor sugarbabe and Phosphoserine phosphatase astray ( Figure 4C , D ) , which we have earlier shown to be essential for sugar tolerance ( Mattila et al . , 2015 ) . To determine the introgressed genomic regions and genes associated with the sugar tolerant phenotype , we sequenced the whole genomes of the two parental , sugar-selected and non-selected control fly lines and identified species-specific SNPs across the genome . SNP analysis showed three small and one large region of D . simulans SNPs on chromosome arm 2R , while all other regions of the genome showed an almost completely D . sechellia SNP signature ( Figure 5A and B ) . Locations of the introgressions relative to nucleotide positions on the D . melanogaster chromosome arm 2R were from approximately 5 , 758 , 067 to 6 , 085 , 625 ( spanning 25 annotated D . melanogaster genes ) ; from 6 , 600 , 044 to 6 , 810 , 530 ( spanning 35 annotated genes ) ; and 21 , 774 , 876 to 24 , 092 , 079 ( spanning 312 annotated genes ) ( Supplementary file 1 ) . Majority of the introgressed genes showed no significant changes in gene expression ( Figure 5C ) . In total , 40 introgressed genes were significantly ( p < 0 . 05 ) downregulated in the intolerant lines , while 24 displayed elevated expression associated with sugar intolerance ( Figure 5C; Supplementary file 1 ) . This suggests that the global gene expression differences between sugar tolerant and intolerant lines are likely due to a small number of loci , which control the expression of a large number of downstream genes . To identify genes from the introgressed chromosome 2R regions that were potentially responsible for the sugar tolerance phenotype , we utilized the genetic toolkit of D . melanogaster , a dietary generalist and close relative of D . simulans and sechellia with sugar tolerance similar to that of D . simulans ( Figure 6—figure supplement 1 ) . From the total number of 372 introgressed genes , we selected 102 genes based on their annotation , with a putative metabolic or regulatory function to be screened for survival on low ( 20% yeast ) and high ( 20% yeast/20% sugar ) sugar diets ( Supplementary file 1 ) . The screen identified several genes with a sugar intolerant phenotype . Interestingly , three of the identified sugar tolerance genes , mRpL43 , CG4882 and bonsai encode components of the mitochondrial ribosome ( Figure 6A–C ) . Furthermore , all of them displayed reduced expression in sugar intolerant genotypes ( Figure 6D–F ) , implying that reduced capacity mitochondrial protein biosynthesis contributes to the sugar intolerance phenotypes . In addition to mitochondrial genes , our screen identified several sugar tolerance genes with a role in signaling . RNAi knockdown of Sarco/endoplasmic reticulum Ca2+-ATPase ( SERCA ) , Protein phosphatase 1 regulatory subunit 15 ( PPP1R15 , Gadd34 ) , or Phosphotidylinositol 3 kinase 59F ( Pi3K59F ) led to strongly impaired larval growth on high-sugar diet , with only a few larvae surviving to pupae ( Figure 7A–C ) . Furthermore , the expression of SERCA was downregulated in D . sechellia and in the sugar intolerant control line ( Figure 7D–F ) . All these three genes have been linked to metabolic processes . SERCA pumps Ca2+ into endoplasmic reticulum ( ER ) and is involved in control of lipid homeostasis ( Bi et al . , 2014 ) , while Pi3K59F is a known regulator of autophagy ( Juhász et al . , 2008 ) . PPP1R15/Gadd34 is best known for its function as a negative regulator of the integrated stress response pathway , including amino acid sensing kinase GCN2 ( Malzer et al . , 2013 ) . Furthermore , we found three additional genes ( Taldo , Dpit47 , GlcT-1 ) displaying milder phenotypes , namely reduced eclosion on high-sugar diet ( Figure 7—figure supplement 1 ) . Next we wanted to assess the level of genomic variation in the candidate genes identified , first focusing on the four genes ( mRpL43 , CG4882 , bonsai , and SERCA ) that displayed reduced expression in sugar intolerant lines . We mapped the density of nucleotide differences between D . simulans and sechellia using a sliding window of 100 bases within these specific gene regions . While all genomic regions displayed areas of high SNP density , the promoter region of the SERCA gene was found to be particularly variable ( Figure 8A; Figure 8—figure supplement 1 ) . To validate the functional impact of this variation , we cloned 1 . 2 kB fragments with putative promoter regions of D . simulans and sechellia SERCA gene in front of a lacZ reporter and generated in vivo reporter lines in D . melanogaster ( Figure 8A ) . Indeed , the D . sechellia -derived promoter displayed significantly ( p < 0 . 01 ) lower activity than the respective region of D . simulans , confirming the functional importance of the SNPs in the SERCA promoter ( Figure 8B ) . We also looked for potential coding region changes in the candidate genes . For all of the hits , the D . sechellia DNA sequence contained nucleotide substitutions that cause amino acid differences in the encoded protein as compared to D . simulans ( Table 4 ) . Most genes in the set of hits had substantially higher number of silent than amino acid altering nucleotide differences , which implies purifying selection . In contrast , there were 10 amino acid changing and only five silent nucleotide differences in the PPP1R15 sequence of D . sechellia compared to that of D . simulans . The rate of amino acid changing to silent mutations ( Ka/Ks ) in PPP1R15 was 0 . 58 indicating reduced purifying selection along the D . sechellia lineage ( Table 4 ) . Plausible alternative explanations for the higher Ka/Ks include the introduction of a new selective pressure on the founder population of D . sechellia . Low sugar tolerance in D . sechellia could be due to genetic drift in a low sugar dietary environment lacking selection or may be caused by a trade-off for an altered function that provides D . sechellia with selective advantage . To test if sugar tolerance is associated with Morinda toxin tolerance , we selected hybrid larvae on Morinda toxin and performed three generations of selection . Selection for tolerance of the Morinda toxins had no significant impact on sugar tolerance ( Figure 9—figure supplement 1 ) . Moreover , the sugar tolerant introgression lines were not sensitive to Morinda toxins ( Figure 9—figure supplement 1 ) , further implying that toxin tolerance is genetically independent of sugar tolerance . Since no association between sugar tolerance and toxin tolerance was found , we hypothesized that poor sugar tolerance is associated with improved fitness in low-sugar nutrient space . Therefore , we determined regions of the diet space where D . sechellia larvae would have an advantage compared to those of D . simulans . We subtracted the pupariation rate of D . simulans from that of D . sechellia and plotted the difference across the diet space . The subtracted diet space surface shows that D . sechellia and D . simulans larvae have clearly separated peaks where they hold an advantage ( Figure 9A ) . The strongest advantage for D . sechellia was observed when the yeast content was 5% or lower and sugar levels were close to zero . In order to test , whether the high fitness of D . sechellia on low energy diet is a tradeoff for sugar tolerance , we raised the sugar selected and control introgression lines on 2 . 5% yeast diet . Strikingly , the sugar tolerant introgression line performed like parental D . simulans , while the sugar intolerant control line displayed higher fitness on 2 . 5% yeast , similar to the parental D . sechellia larvae ( Figure 9B ) . This implies that the genetic loci of D . simulans , which provide high sugar tolerance cause a disadvantage in conditions of low-energy diet . Given the observed tradeoff in sugar tolerance and starvation tolerance , we next tested the Drosophila melanogaster RNAi lines with sugar-intolerant phenotype for their performance on low-energy diet . Interestingly , PPP1R15 knockdown animals showed elevated pupariation on low-energy diet when compared to corresponding control animals ( Figure 9C ) . This implies that genetic changes affecting individual regulatory genes can contribute to the optimal macronutrient space of the animal .
In this study , we show that the macronutrient space of two closely related species that have different dietary choices is dissociated in concordance with their natural diets . Larvae of the dietary specialist D . sechellia that feed on a low-sugar diet in nature exhibited intolerance of high-sugar diet . In contrast , the dietary generalist D . simulans broadly tolerated dietary sugar , but performed poorly on low-energy-content diets . Sugar intolerance was rescued in F1 hybrids suggesting complementation of D . sechellia alleles with those of D . simulans . To identify the genomic regions associated with sugar tolerance , we introgressed a sugar tolerant phenotype into a mostly D . sechellia genomic background through multiple rounds of backcrossing and selection on a high-sugar diet . The sugar selected fly lines exhibited sugar tolerance equal to that of the D . simulans parent while the introgression control lines that were not selected on high-sugar diet exhibited very poor sugar tolerance that was only slightly better than the D . sechellia parent . It remains a possibility that the dietary composition affects the growth of commensal microbes , which may differentially affect the growth of Drosophila species . However , we observed impaired clearance of circulating glucose in the sugar intolerant lines as well as differential gene expression response to sugar feeding . These data , together with our loss-of-function phenotypes on metabolic regulators , strongly suggests that differences in the regulation of energy metabolism were the primary causes for the observed differences in sugar tolerance . It should also be noted that our study relied on the use of single representative lines for D . simulans and D . sechellia and future studies with a larger number of lines are needed to determine the degree of natural variation of sugar tolerance within the species . Our study represents evidence for rapid ( ~0 . 4 MY ) evolution of macronutrient space in a multicellular animal . Evolution of metabolism is known to occur through multiple mechanisms , such as nonsynonymous coding region mutation , copy number variation or mutation of regulatory regions of a gene encoding a metabolic enzyme ( Wagner , 2012 ) . Examples of recent evolution of animal metabolism by altered function of a single enzyme include the lactase persistence in human populations ( Gerbault et al . , 2011 ) as well as increase in copy number of amylase-encoding gene upon dog domestication ( Axelsson et al . , 2013 ) . In contrast to the aforementioned examples , D . simulans and D . sechellia display deviation of the macronutrient spaces along the carbohydrate/protein axis , likely requiring much more widespread reprogramming of core metabolic pathways . In line with this prediction , our RNAseq analysis revealed global changes in carbohydrate and amino acid metabolism , mitochondrial function , ribosome biogenesis and stress response pathways associated with sugar tolerance . In conclusion , our data demonstrates that global changes in macronutrient space caused by global rewiring of core metabolic pathways can occur in animals in relatively short evolutionary timeframe . In order to reprogram large metabolic networks through mutations of genes encoding metabolic enzymes , a number of independent mutations would need to occur simultaneously , which is unlikely to occur . A plausible model for obtaining such rapid global changes in metabolic pathways is through genetic changes in metabolic ‘hub’ genes , including mitochondrial genes and signaling pathway components , whose activity is reflected to multiple metabolic pathways simultaneously . Several genes involved in mitochondrial ribosome were included into the introgression regions associated with sugar tolerance . The importance of mitochondrial ribosome biogenesis in survival on carbohydrate-rich food has been observed earlier ( Kemppainen et al . , 2016 ) . Furthermore , reduced mitochondrial ribosome biogenesis is widely reflected to central carbon metabolism and redox balance of the animal ( Kemppainen et al . , 2016 ) , which is consistent with the global gene expression differences observed in our sugar tolerant vs . intolerant lines . One of the identified genetic determinants of sugar tolerance was SERCA , an ATP-dependent Ca2+ pump in the ER membrane . We observed that SERCA displayed significantly reduced gene expression in sugar intolerant lines and that RNAi-mediated knockdown of SERCA caused significant sugar intolerance D . melanogaster . Sequence comparison of genomic regions of D . simulans and sechellia led to identification of a high level of sequence variation at the promoter , which was sufficient to explain the lower expression of SERCA in D . sechellia , based on the in vivo reporter experiment . Previous evidence shows that SERCA has a critical role in metabolic control . In Drosophila , SERCA mutant fat body cells contain reduced number and size of lipid droplets compared to wild-type ( Bi et al . , 2014 ) . SERCA2b expression is strongly downregulated in livers of obese mice and restoring its expression is sufficient to improve glucose tolerance ( Park et al . , 2010 ) . Similar beneficial effects have been obtained by pharmacological activation of SERCA in ob/ob mice ( Kang et al . , 2016 ) . Furthermore , mice mutant for sarcolipin , a muscle-specific regulator of SERCA , are obese and have poor glucose tolerance . Thus , regulation of SERCA expression and activity has a significant and conserved role in the control of glucose metabolism . While it remains unclear how intracellular calcium homeostasis mechanistically regulates energy metabolism , it has been proposed that Ca2+ transport from ER to mitochondria plays a key role ( Kaufman and Malhotra , 2014 ) . Interestingly , regulation of SERCA activity appears to be involved in human evolution as well . SNPs in the gene THADA , which encodes a regulator of SERCA activity , are among the most strongly positively selected SNPs during the evolution of modern humans , based on comparative analyses with the Neanderthal genome ( Green et al . , 2010 ) . THADA interacts with SERCA and acts as a SERCA uncoupling protein , controlling lipid homeostasis and feeding as well as cold resistance in Drosophila ( Moraru et al . , 2017 ) . In human , there is further evidence of THADA selection upon adaptation to cold climate ( Cardona et al . , 2014 ) . Future studies should explore further the role of SERCA and its regulators in other evolutionary processes associated with global changes in energy metabolism . Moreover , considering the role of SERCA in thermogenesis , it will be interesting to test , whether the lower SERCA expression in D . sechellia affects its cold tolerance . Another interesting candidate gene identified in our study was PPP1R15 , which had an impact on both sugar tolerance as well as survival on low-energy diet in D . melanogaster . Moreover , our data show that the PPP1R15 coding region contains several amino acid changing nucleotide changes , displaying a significantly higher number than expected when compared with the other candidate genes identified ( χ2 = 17 . 03 , p < 0 . 01 , d . f . = 5 ) . This indicates that there has been reduced pressure of purifying selection on the gene in D . sechellia compared to D . simulans , possibly reflecting a new habitat with reduced need to maintain sugar tolerance . However , since PPP1R15 affects both sugar tolerance and starvation resistance and these two traits form a trade-off , it is also possible that the pressure for development on low-energy diet has favored an alternative form of PPP1R15 , explaining the high degree of amino acid changing mutations . Functionally PPP1R15 is an excellent candidate for a gene that enables rapid evolution of macronutrient space , since it controls major metabolic and energy consuming processes . PPP1R15 is a regulatory subunit of protein phosphatase one and it controls translation by dephosphorylating Ser51 of eIF2alpha ( Novoa et al . , 2001 ) . eIF2alpha Ser51 is the target of the so-called integrated stress response pathway , including for example GCN2 , the sensor of amino acid deprivation as well as PERK , a sensor of ER stress ( Harding et al . , 1999; Harding et al . , 2000 ) . It should be noted that SERCA-dependent ER Ca2+ homeostasis has a critical role in counteracting ER stress ( Park et al . , 2010; Lai et al . , 2017 ) , providing a possible functional link between PPP1R15A and SERCA . Downstream of eIF2alpha ( and thus PPP1R15 ) is transcription factor Atf4 , which controls carbohydrate metabolism in Drosophila melanogaster ( Seo et al . , 2009; Lee et al . , 2015 ) . Moreover , PPP1R15A-deficient mice develop obesity , nonalcoholic liver disease , insulin resistance and impaired glucose tolerance ( Nishio and Isobe , 2015 ) . In conclusion , our study provides evidence for natural variation of organismal sugar tolerance and its association with diet choice . Furthermore , our data demonstrate that global changes in metabolic gene expression , substantially affecting macronutrient space can occur in relatively short evolutionary timeframe . Our findings indicate that adaptation to a new metabolic environment , such as one with low-level nutrition , may be broadly reflected to the macronutrient space , for example by lowering the tolerance to sugar overload . This may be conceptually relevant to human health , considering that human populations with distinct histories of diet choice may bear differential vulnerabilities to nutrient overload posed by modern lifestyles .
Drosophila simulans line C167 . 4 ( 14021-0251-199 ) and D . sechellia line SynA ( 14021-0248-28 ) were obtained from the Drosophila Species Stock Center , University of California , San Diego ( now located at the College of Agriculture and Life Science , Cornell University ) . D . melanogaster strain Oregon R was a gift from Tapio Heino , University of Helsinki . The following VDRC RNAi lines were used: mRpL43: 104466 , CG4882: 106629 , bonsai: 104412 , SERCA: 107446 , PPP1R15 ( Gadd34 ) : 107545 , Pi3K59F: 100296 , Taldo: 106308 , Dpit47: 110401 , GlcT-1: 108064 ( see VDRC web site for specific information ) . Ubi-GAL4 , tub-GAL4 and cg-GAL4 were obtained from the Bloomington Drosophila Stock Center . Fb-GAL4 ( FBti0013267 ( Grönke et al . , 2003 ) was also used . All fly stocks and parents of the experimental flies were maintained on a common laboratory diet containing 2 . 4% ( v/v ) nipagin , 0 . 7% ( v/v ) propionic acid . All experiments took place at 25°C , 50% RH with 12 hr light , 12 hr dark daily cycle and at controlled density ( 30 larvae per vial ) . Driver lines crossed with w1118 containing landing site VIE-260B ( VDRC ID: 60100 ) were used as controls for the RNAi experiments . Different driver lines were used in the knockdown , depending on the strength of the phenotypes . For selection on Morinda toxins flies were reared on 0 . 5% agar 20% yeast diet supplemented with 0 . 5% hexanoic acid and 0 . 01% octanoic acid ( Earley and Jones , 2011 ) . We determined the optimal nutrient space for larval growth for each species . Parents of the experimental flies were released into egg-laying chambers provided with apple-juice-agar plates supplemented with yeast and allowed to lay eggs for 2 hr intervals . Yeast was removed from the egg-laying plates and they were incubated 24 hr at 25°C , 50% RH . Thirty 1st instar larvae were placed into five replicate vials of 0 . 5% agar-based media in a 5 × 5 grid of baker’s yeast ( 1 . 25 , 2 . 5 , 5 , 10% and 20% ) and sucrose ( 0 , 5 , 10 , 15 , and 20% ) . Estimated caloric contents of the diets are presented in Table 5 . Vials were scored for the number of larvae pupariated at 24 hr intervals for 408 hr total . Because nutrition affects both developmental timing and larval survival , we calculated the pupariation index ( Pupind ) that is the maximum rate of pupariation . Specifically , Pupind = max ( { ( pt/t ) : t = 24 , 48 , … , 408} ) ; where pt = number of larvae that have pupariated at observation time t hours after egg-laying ( hAEL ) . The maximum Pupind was analyzed using a full general linearized model with main effects of genotype , yeast , sucrose and their interactions using JMP software ( SAS Institute ) . The strengths of the main effects were determined by calculating ω2 ( Yigit and Mendes , 2018 ) . Feeding behavior was assayed by quantifying the rate of larval mouth hook extensions using the method described by Shen ( 2012 ) . Parents of the experimental larvae were allowed to lay eggs for 2 hr on apple juice agar plates spread with yeast paste . Plates were incubated at 25°C overnight and then first instar larvae were transferred to vials of 20% yeast in groups of 30 and raised to pre-wandering third larval instar . On the morning of the assay , larvae were transferred to Petri plates containing 20% yeast colored with blue food dye and were allowed to feed for 1 hr . Actively feeding larvae with visible blue food in the gut were selected for assay . Media for the mouth hook extension assay was prepared by mixing 12 g dry agar with a solution of 1 × PBS buffer or with 1 × PBS buffer containing 20% sucrose to final volume of 100 ml . The mixtures were incubated overnight at 4°C to hydrate the agar completely . The thickened assay media was poured into small Petri plates and allowed to equilibrate to RT for 3 hr . To begin the assay , 30 larvae were transferred from the blue-dyed LSD plate onto a plate of assay media . Larvae were viewed on the plate using a dissecting microscope and the number of mouth hook extensions in 1 min was counted for 10 individual larvae . Larvae float on the assay media and are unable to move from where they are placed . Four no sucrose and four 20% sucrose plates were observed . Counts were recorded using a handheld cell counter and each observed larva was removed from the plate before proceeding to the next . Mean mouth hook extension rate was calculated per plate from the 10 larvae and compared by ANOVA with species and sucrose percent as main effects . We introgressed the phenotype of sugar tolerance from D . simulans into a mostly D . sechellia genomic background using the crossing scheme outlined in Figure 2A . To produce F1 hybrids , groups of 100 unmated D . simulans females were collected and crossed to 100 D . sechellia males in 1-litre plastic population cages . Females were allowed to lay eggs on apple-juice agar plates for 24 hr , then eggs were collected and seeded at approximately 200 eggs per 240 ml bottle onto 20% yeast media ( Clancy and Kennington , 2001 ) . For the first backcross generation ≥100 unmated F1 hybrid females were collected from the bottles and crossed to 100 D . sechellia males in 1-liter plastic population cages . Females from the cross were allowed to lay eggs on apple-juice agar plates for 24 hr then eggs were collected and seeded at approximately 200 eggs per 240 ml bottle onto 20% yeast/20% sucrose media to select the sugar tolerant phenotype or onto 20% yeast for the control backcross . Bottles of eggs that were collected from a single crossing cage were kept together , separate from those collected from replicate crosses . For generation 2 through 10 , virgin females were collected from the bottles , crossed to D . sechellia males , eggs were collected and seeded into fresh 20% yeast/20% sucrose ( selected ) or 20% yeast ( control ) media . Crossing populations consisted of ≥100 backcross females from the previous generation and 100 D . sechellia males . After 10 generations of backcrossing the lines were sibling mated for three generations before beginning experiments . Backcross lines were continuously maintained on 20% yeast/20% sucrose ( selected ) or 20% yeast ( control ) diet throughout sibling mating . Genomic DNA was extracted from 30 female flies for two no-selection control lines , three sugar-selected lines , and from each parental D . sechellia and D . simulans line using a PureGene DNA extraction kit ( Qiagen ) . Resequencing was performed at the University of North Carolina DNA sequencing facility . Regions of introgression were mapped using the PSI-seq method of Earley and Jones ( 2011 ) . Original datasets have been placed into a public repository ( NCBI ) : https://www . ncbi . nlm . nih . gov/bioproject/PRJNA486014/ We determined the hemolymph glucose concentration of the parental and introgression lines after feeding on a high sugar diet . Larvae were collected onto 20% yeast media and raised to pre-wandering third larval instar . Five replicates of 50 pre-wandering third instar larvae were transferred to media containing 20% sucrose for 2 hr and then to 0% sucrose media for 2 hr . Hemolymph was collected from 10 larvae at time points 0 hr - 0% , 2 hr - 20% sucrose and 2 hr - clearance ( 0% sucrose ) . Hemolymph glucose was assayed using a Glucose Oxidase/Peroxidase assay kit ( Sigma ) ( Havula et al . , 2013 ) and data were analyzed by comparing to D . simulans using Dunnett’s test implemented in JMP software ( SAS institute ) . Dunnett’s test queries whether the difference between the mean of the control group and an experimental group differs by greater than a critical value . We extracted and sequenced total RNA from third instar D . simulans ( C167 . 4 ) , D . sechellia ( SynA ) , backcross control , and backcross selected larvae that were fed LSD ( 20% yeast ) or HSD ( 20% yeast/20% sucrose ) for 8 hr using a Nucleospin RNA II kit ( Macherey-Nagel ) . Larvae were prepared from three replicates of each fly line . Parents of the experimental larvae were released into egg collection cages and allowed to lay eggs for 2 hr on apple juice agar plates supplemented with yeast paste . Plates of eggs were placed at 25°C for 24 hr . Following incubation , 100 first instar larvae were transferred to plates containing LSD and placed at 25°C for 48 hr after which pre-wandering third instar larvae were transferred to plates of LSD or HSD that contained 2% blue food dye . Larvae were allowed to feed for 2 hr at 25°C then larvae that did not have blue dye in their gut were removed from the plates . Feeding larvae that had blue dye in their gut were kept 6 hr at 25°C and then collected for RNA extraction . RNA sequencing libraries we prepared using the TruSeq Stranded mRNA Library Prep Kit ( Illumina ) and sequenced ( single-end 76 bp reads ) using Illumina NextSeq 500 technology . The quality of the reads was assessed with FASTQC ( v . 0 . 11 . 2 ) ( Babraham Bioinformatics , Cambridge , UK ) . The reads were trimmed with Trimmomatic ( v . 0 . 33 ) ( Bolger et al . , 2014 ) . Reads were scanned with sliding window of 20 and if the average quality per base dropped below 20 , the read was discarded . Additionally , base length of 40 was required for the reads and for both leading and trailing ends quality score of 30 was required . For the mlx mutant RNAseq dataset ( Mattila et al . , 2015 ) , trimming was performed with sliding window of 4 bases with average per base quality requirement of 15 . Required base length was set to 36 and required strand quality in both at the end and start was set to 36 . Tophat ( v . 2 . 1 . 0 ) ( Trapnell et al . , 2009 ) was used for aligning the reads to D . melanogaster reference genome ( Flybase R6 . 10 ) . HTseq ( v . 2 . 7 . 6 ) ( Anders et al . , 2015 ) was used for strand-specific quantification of exons . Reads with quality score below 10 were discarded . The quality of the samples was assessed with multi-dimensional scaling and variety of sample clustering methods ( pearson correlation , euclidean ward , euclidean complete ) using R/Bioconductor's package pvclust ( Suzuki and Shimodaira , 2006 ) . Based on these results , one sample ( sim . 0 . 1 ) was defined as outlier and thus removed from the analysis . The differential expression analysis was performed with R/Bioconductor package limma ( Ritchie et al . , 2015; Law et al . , 2014 ) . The samples were required to have >1 CPM ( counts per million ) in all replicates in either tolerant or intolerant group . For mlx mutant datasets , no outliers were discovered , and >1 CPM was required in majority of samples in at least one of the conditions . The Benjamini-Hochberg correction was used for adjusting p values ( Benjamini and Hochberg , 1995 ) . Sample clustering was performed using R/Bioconductor package pvclust ( Suzuki and Shimodaira , 2006 ) . Correlation was used as distance matrix . The gene set enrichment was performed with hypergeometric test for the manually downloaded pathway sets from KEGG and GO . The pathway was defined as enriched if the adjusted p value < 0 . 05 . The heatmaps were generated using scaled log2CPM values for means of each sample group . The scaling was performed separately for the two datasets . The row-wise clustering was performed using correlation distance . Original datasets have been placed into a public repository ( NCBI ) : https://www . ncbi . nlm . nih . gov/bioproject/PRJNA486014/ . Species-specific nucleotide sequences for SERCA , CG4882 , RpL49 , and bonsai genomic regions were compiled by mapping Illumina sequence reads from D . simulans ( C167 . 4 ) and D . sechellia ( SynA ) to D . melanogaster sequences using BWA-MEM in the Burrows-Wheeler alignment software package ( Li and Durbin , 2009 ) . Sequence read alignments were edited by hand using Geneious 11 . 1 . 5 software ( Biomatters Ltd . , Aukland , NZ ) to produce simple majority consensus sequences . For each gene , the consensus sequences were aligned and nucleotide differences between species called using Geneious 11 . 1 . 5 software . The average frequency of nucleotide differences was calculated in 100 base windows slid forward in steps of 25 bases . The genomic average frequency of nucleotide differences between D . simulans and D . sechellia was calculated for three randomly chosen 20 kb regions on chromosomes 2R , 2L , and X , and was subtracted from each window to correct for background noise . Frequencies were charted using JMP Pro 14 software ( SAS Institute Cary , NC ) . The likely SERCA promoter region was identified based on the nine chromatin stage model ( Kharchenko et al . , 2011 ) , the selected fragment corresponds to the ‘red’ chromatin type ( Active promoter/transcription start site region ) . The selected 1 . 2 kB promoter regions of SERCA from D . simulans ( C167 . 4 ) and D . sechellia ( SynA ) were cloned into the placZ-2 . attB vector using restriction enzyme sites NotI and XhoI and ligase-dependent cloning ( Bischof et al . , 2013 ) . Successful generation of plasmids was verified with Sanger sequencing . Injection was performed by GenetiVision ( Houston , TX ) into a D . melanogaster w1118 line with landing site attP2 ( 3L ) 68A4 . Cloning was performed using the following primers: SERCA F: 5’-TAAGCGGCCGCTCTTCGTTCAGTGGCCTGTG-3’ SERCA R: 5’-TAACTCGAGTCGTGATAAGGATTTCAGTTCG-3’ Eight early 3rd instar larvae were collected for each sample and RNA was extracted using the Nucleospin RNA kit ( Macherey-Nagel ) according to the manufacturer’s protocol . RNA was reverse-transcribed using the SensiFAST cDNA Synthesis kit ( Bioline ) according to the manufacturer’s protocol , and qPCR was performed using SensiFAST SYBR No-ROX kit ( Bioline ) with Light cycler 480 Real-Time PCR System ( Roche ) with three technical replicates per sample . Actin 42A was used as a reference gene . Following primers were used: LacZ F: 5’-CGAATCTCTATCGTGCGGTG-3’ LacZ R: 5’-CCGTTCAGCAGCAGCAGAC-3’ Act42A F: 5’-CCGTACCACAGGTATCGTGTTG-3’ Act42A R: 5’-GTCGGTTAAATCGCGACCG-3’ | Animals meet their nutritional needs in a variety of ways . Some animals are specialists feeding only on one type of food; others are generalists that can choose many different kinds of food depending on the situation . Despite these differences in diet , animals have similar needs for basic cellular metabolism . This suggests that generalist and specialist species likely process the foods they eat in different ways in order to meet their basic needs . For example , the metabolism of generalist species may be more flexible to adapt to changing food sources . To learn more about how metabolism evolves to respond to diet , scientists can study closely related species that eat different foods . For example , a species of fruit fly called Drosophila simulans is a generalist and its larvae can grow and develop by feeding on different kinds of decaying fruits and vegetables . Larvae of a closely related fruit fly called Drosophila sechellia are specialized to eat only the nutrient-poor Morinda fruit . Looking at how genetic differences between these species affect metabolism may provide scientists with clues about how these feeding strategies evolved . Melvin et al . grew larvae of D . sechellia and D . simulans in different conditions . D . sechellia larvae thrived in low nutrient conditions , but died when exposed to high sugar foods . By contrast , D . simulans larvae tolerated high sugar levels , but did poorly in low-nutrient conditions . Melvin et al . then bred the two species with each other , selecting flies that are genetically similar to D . sechellia but have the genes necessary for larvae to tolerate sugar . Analyzing the selected hybrid flies revealed genetic changes that explain the different survival abilities of each species . These changes suggest that D . sechellia rapidly evolved to thrive in low nutrient conditions , but the trade-off was losing their ability to tolerate high sugar levels . Overall , the results presented by Melvin et al . suggest that genetic adaptions to food sources can occur quickly and drastically change metabolism . Further research will be needed to confirm if similar metabolic trade-offs developed as part of human evolution . If so , human populations that survived with limited nutrition for many generations may have a harder time adapting to high-sugar modern diets . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2018 | Natural variation in sugar tolerance associates with changes in signaling and mitochondrial ribosome biogenesis |
Ascorbic acid ( vitamin C ) is an enzyme co-factor in eukaryotes that also plays a critical role in protecting photosynthetic eukaryotes against damaging reactive oxygen species derived from the chloroplast . Many animal lineages , including primates , have become ascorbate auxotrophs due to the loss of the terminal enzyme in their biosynthetic pathway , l-gulonolactone oxidase ( GULO ) . The alternative pathways found in land plants and Euglena use a different terminal enzyme , l-galactonolactone dehydrogenase ( GLDH ) . The evolutionary processes leading to these differing pathways and their contribution to the cellular roles of ascorbate remain unclear . Here we present molecular and biochemical evidence demonstrating that GULO was functionally replaced with GLDH in photosynthetic eukaryote lineages following plastid acquisition . GULO has therefore been lost repeatedly throughout eukaryote evolution . The formation of the alternative biosynthetic pathways in photosynthetic eukaryotes uncoupled ascorbate synthesis from hydrogen peroxide production and likely contributed to the rise of ascorbate as a major photoprotective antioxidant .
Ascorbate ( vitamin C ) plays an essential role in eukaryotes as an enzyme co-factor in hydroxylation reactions , contributing to diverse processes such as the synthesis of collagen and the demethylation of histones and nucleic acids ( Mandl et al . , 2009; Blaschke et al . , 2013 ) . Ascorbate also plays an antioxidant role in eukaryotes to help protect against reactive oxygen species ( ROS ) derived from metabolic activity . The majority of hydrogen peroxide ( H2O2 ) generated in some organelles is likely reduced by other antioxidant systems , such as the peroxiredoxins and glutathione peroxidases in the mitochondria , and catalases in the peroxisome ( Mhamdi et al . , 2012; Sies , 2014 ) . However , ascorbate plays an important role in protecting photosynthetic cells against ROS derived from the chloroplast ( Smirnoff , 2011 ) . Ascorbate peroxidase ( APX ) , which is found in both the cytosol and the chloroplast of photosynthetic eukaryotes , is central to this photoprotective role . Thylakoid- and stroma-localised APX removes H2O2 produced by photosystem I through the activity of the ascorbate-glutathione cycle and this process may account for 10% of photosynthetic electron transport flow . Ascorbate also plays a critical role in preventing lipid peroxidation in the thylakoid membranes and acts as a co-factor for violaxanthin de-epoxidase in the xanthophyll cycle . In addition , ascorbate in the thylakoid lumen may prevent photoinhibition in high light by directly donating electrons to the photosynthetic electron transport chain ( Smirnoff , 2011 ) . These roles have been demonstrated in a range of ascorbate deficient plants that display sensitivity to high light and to oxidants ( Smirnoff , 2011 ) . Photosynthetic eukaryotes arose following the endosymbiotic acquisition of a cyanobacterial ancestor by a non-photosynthetic eukaryote in the Archaeplastida ( Plantae ) lineage ( Keeling , 2010 ) . Several other eukaryote lineages , including the diatoms , haptophytes and euglenids , subsequently gained plastids through a secondary endosymbiosis with either a red or green alga . These plastid endosymbioses were accompanied by lateral gene transfer on a massive scale from the symbiont to the host nuclear genome ( known as endosymbiotic gene transfer or EGT ) , giving rise to the complex physiologies of photosynthetic eukaryotes ( Timmis et al . , 2004 ) . The plastids in the major photosynthetic eukaryote lineages are all ultimately derived from the primary endosymbiosis . Whilst acquisition of a photosynthetic endosymbiont may have been beneficial to the host cell in many ways , the plastid is also a major source of potentially damaging ROS ( Dorrell and Howe , 2012 ) . There is evidence for extensive leakage of H2O2 out of plastids via aquaporins , particularly at high light intensities ( Mubarakshina et al . , 2010; Naydov et al . , 2012 ) . Plastid acquisition is therefore associated with a greatly increased requirement for cellular antioxidant systems to prevent photodamage . Cyanobacteria do not possess APX or any of the known enzymes for ascorbate biosynthesis and minimise photo-oxidative stress using alternative mechanisms , such as peroxiredoxins , catalases and glutathione peroxidases ( Zámocký et al . , 2010; Gest et al . , 2013 ) . This suggests that ascorbate was most likely recruited to its photoprotective role after the acquisition of the plastid in ancestral Archaeplastida ( Gest et al . , 2013 ) , although the evolutionary origins of ascorbate biosynthesis are unclear . There is no clear evidence for ascorbate biosynthesis in prokaryotes ( see ‘Materials and methods’ ) , suggesting that the ability to synthesise ascorbate evolved in eukaryotes . The three eukaryote lineages in which ascorbate biosynthesis has been examined extensively ( animals , plants and Euglena ) all exhibit different biosynthetic pathways ( Figure 1 ) . These pathways may have arisen due to convergent evolution , or may represent modifications of an ancestral pathway . An understanding of these evolutionary relationships will provide insight into the cellular roles of ascorbate in eukaryotes , particularly in relation to plastid acquisition in the photosynthetic lineages . 10 . 7554/eLife . 06369 . 003Figure 1 . Major ascorbate biosynthetic pathways in eukaryotes . The scheme depicts the three major ascorbate biosynthetic pathways found in eukaryotes ( Shigeoka et al . , 1979; Wheeler et al . , 1998; Linster and Van Schaftingen , 2007 ) . The plant pathway ( also known as the Smirnoff-Wheeler or D-mannose/l-galactose pathway ) involves no inversion of the carbon chain ( i . e . , C1 of D-glucose becomes C1 of l-ascorbate ) , whereas the euglenid and animal pathways involve inversion of the carbon chain in the conversion from uronic acid to aldonolactone ( i . e . , C1 of D-glucose becomes C6 of l-ascorbate ) . Our analyses focus on enzymes with a dedicated role in ascorbate biosynthesis ( shown in red ) : GULO—l-GulL oxidase; VTC2—GDP-l-galactose phosphorylase; VTC4—l-galactose-1-phosphate phosphatase; l-galDH—l-galactose dehydrogenase; GLDH—l-GalL dehydrogenase . The other enzymes are: PGM—phosphoglucomutase; UGP—UDP-D-glucose pyrophosphorylase; UGDH—UDP-D-glucose dehydrogenase; UGUR—UDP-glucuronidase; GlcUAR—D-glucuronate reductase; SMP30—regucalcin/lactonase; GAE—UDP-D-glucuronate-4-epimerase; GalUAR—D-galacturonate reductase . Enzyme names are not listed for steps where multiple enzymes may be involved or where specific enzymes have not been identified . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 003 Animals synthesise ascorbate via D-glucuronic acid and l-gulonolactone ( l-GulL ) , with l-gulonolactone oxidase ( GULO ) catalysing the oxidation of l-GulL to ascorbate . Many animal lineages have lost the ability to synthesise ascorbate , including haplorhine primates , guinea pigs , teleost fish , some bats and passerine birds . In all of the animal ascorbate auxotrophs that have been examined , the gene encoding GULO is lost ( Nishikimi et al . , 1992 , 1994; Cui et al . , 2011a; Drouin et al . , 2011 ) . GULO uses molecular O2 as its electron acceptor , resulting in H2O2 production , which may have contributed to selective pressure to lose this enzyme in some animals ( Bánhegyi et al . , 1996; Mandl et al . , 2009 ) . The inability of many invertebrates to synthesise ascorbate led to early speculation that ascorbate synthesis may have evolved later in metazoan evolution ( Chatterjee , 1973 ) . However , whilst the loss of GULO in vertebrate lineages has been extensively examined ( Yang , 2013 ) , little is known about its distribution in invertebrates or the non-metazoan members of the holozoa . This information is required to identify the selective pressures underlying the evolution of ascorbate auxotrophy in animals . Two alternative routes to ascorbate biosynthesis have been identified in photosynthetic eukaryotes , which both employ l-galactonolactone dehydrogenase ( GLDH ) as the terminal enzyme instead of GULO . A pathway via D-galacturonic acid and l-galactonolactone ( l-GalL ) was identified in Euglena ( Shigeoka et al . , 1979 ) . This pathway is analogous to the animal pathway and also appears to be functional in some stramenopile algae ( Helsper et al . , 1982; Grün and Loewus , 1984 ) . In contrast , ascorbate biosynthesis in land plants was found to occur via a different route using D-mannose and l-galactose ( Wheeler et al . , 1998 ) . Green algae also use the ‘plant pathway’ ( Running et al . , 2003; Urzica et al . , 2012 ) , but evidence is lacking for the nature of ascorbate biosynthesis in many other evolutionarily important lineages , most notably the rhodophytes ( red algae ) . This paper focuses on the distribution of the three major pathways of ascorbate biosynthesis described above . Alternative routes of ascorbate biosynthesis have been described in trypanosomes and also in the fungi , which synthesise a range of ascorbate analogues ( see ‘Materials and methods’ ) ( Loewus , 1999; Logan et al . , 2007 ) . There is some evidence for the operation of alternative routes to ascorbate in land plants ( Wolucka and Van Montagu , 2003; Lorence et al . , 2004; Badejo et al . , 2012 ) , although molecular genetic evidence from Arabidopsis indicates that the D-mannose/l-galactose pathway is the primary route of ascorbate biosynthesis ( see ‘Materials and methods’ ) ( Conklin et al . , 1999; Dowdle et al . , 2007 ) . The three major pathways of ascorbate biosynthesis therefore all utilise different routes to synthesise an aldonolactone precursor ( l-gulonolactone , l-GulL or l-galactonolactone , l-GalL ) , which is converted to ascorbate by either GULO ( animal pathway ) or GLDH ( plant and euglenid pathways ) ( Shigeoka et al . , 1979; Wheeler et al . , 1998; Loewus , 1999 ) . GULO and GLDH exhibit significant sequence similarity and are both members of the vanillyl alcohol oxidase ( VAO ) family of flavoproteins ( Leferink et al . , 2008 ) . These similar enzymes exhibit important biochemical differences . GULO can oxidise l-GulL and l-GalL , whereas GLDH is highly specific for l-GalL ( Smirnoff , 2001 ) . GULO localises to the lumen of the endoplasmic reticulum ( ER ) , whereas GLDH is associated with complex I in the mitochondrial electron transport chain ( Schertl et al . , 2012 ) . Importantly , GLDH does not generate H2O2 , as it uses cytochrome c rather than O2 as an electron acceptor . Despite the importance of ascorbate in eukaryote physiology , it is not known how the different pathways of ascorbate biosynthesis arose in animals , plants and algae or relate to its differing cellular roles . This manuscript examines the origins of ascorbate biosynthesis in eukaryotes and seeks to address the following important gaps in our current knowledge: ( 1 ) what is the wider distribution of GULO loss and ascorbate auxotrophy in the metazoa ? ( 2 ) do all photosynthetic eukaryotes use an alternative terminal enzyme to animals ? ( 3 ) why do two different pathways using GLDH exist in photosynthetic eukaryotes ? ( 4 ) which pathway is used in the rhodophytes ? Using a combination of molecular and biochemical analyses , we present evidence that GULO is an ancestral gene in eukaryotes that has been functionally replaced by GLDH in the photosynthetic lineages , resulting in the development of their alternative biosynthetic pathways .
To examine the origins of ascorbate biosynthesis in eukaryotes we analysed the distribution of GULO and GLDH in eukaryote genomes . We found that GULO and GLDH have a mutually exclusive distribution ( Figure 2 ) . GULO is absent from many metazoan genomes , including all insects , supporting earlier biochemical evidence that insects are predominately ascorbate auxotrophs ( Supplementary file 1 ) ( Chatterjee , 1973; Dadd , 1973 ) . However , GULO is present in basally derived metazoans , including sponges and cnidarians , and is also present in a filasterean ( Capsaspora owczarzaki ) and in fungi ( Supplementary file 1 ) . This suggests that ascorbate synthesis via GULO is an ancestral trait in the Opisthokonta that has been lost in many lineages . 10 . 7554/eLife . 06369 . 004Figure 2 . Coulson plot indicating the taxonomic distribution of the different ascorbate pathways . 40 eukaryote genomes were analysed for the presence of genes in the ascorbate biosynthetic pathways . The two potential terminal enzymes in the pathway are boxed . GLDH is common to both the ‘plant’ and ‘euglenid’ type pathways . A schematic tree depicts the currently accepted phylogenetic relationships between organisms . The predicted route of ascorbate biosynthesis in each organism is shown . Note that ‘euglenid’ and ‘rhodophyte’ type pathways cannot currently be distinguished from sequence analysis alone and the predictions are based on biochemical evidence . Asterisk denotes a genome assembly was not available for Euglena gracilis and its transcriptome was analysed ( ‘Materials and methods’ ) . Grey circles in VTC4 represent the presence of a highly similar enzyme , myo-inositol-1-phosphate phosphatase that exhibits l-galactose-1-phosphatase activity . GULO in trypanosomes and yeasts acts to oxidise the alternative substrates l-galactonolactone or D-arabinonolactone respectively . VTC3 is not a biosynthetic enzyme , but represents a dual function Ser/Thr protein kinase/protein phosphatase 2C that may play a regulatory role in the plant pathway ( Conklin et al . , 2013 ) . Black cross represents a pseudogene encoding a non-functional enzyme . Organisms with sequenced genomes that were found to lack both of the terminal enzymes in the known pathways ( GULO and GLDH ) are likely to be ascorbate auxotrophs and were not included in the plot . These include Giardia intestinalis , Trichomonas vaginalis , Entamoeba invadens , Plasmodium falciparum and Perkinsus marinus . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 00410 . 7554/eLife . 06369 . 005Figure 2—figure supplement 1 . Distribution of GULO and GLDH in the Archaeplastida . A schematic tree demonstrating the currently accepted phylogenetic positions of the major lineages in the Archaeplastida ( Yoon et al . , 2006; Leliaert et al . , 2012 ) . The presence of either GULO or GLDH in representatives of each lineage is shown . The boxes denote the Viridiplantae ( green ) , Rhodophyta ( red ) and Glaucophyta . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 00510 . 7554/eLife . 06369 . 006Figure 2—figure supplement 2 . Distribution of the different ascorbate pathways . The schematic tree summarises the distribution of the two terminal enzymes in ascorbate biosynthesis , along with VTC2 , the first committed enzyme in plant pathway . Blue lines indicate photosynthetic lineages derived by the primary endosymbiosis ( of a cyanobacterium ) . Red or green lines indicate lineages that have become photosynthetic following a secondary endosymbiosis event with either a red or a green alga respectively . It should be noted that the timing and origin of many secondary endosymbioses remain unclear , particularly within the SAR supergroup where several non-photosynthetic lineages within the stramenopiles , alveolates and even rhizaria may potentially have lost an ancestral plastid . GULO is found in basally derived lineages of the Archaeplastida , Excavata , Opisthokonta , Amoebozoa and the CCTH group . In contrast , GLDH is found predominately in photosynthetic eukaryotes , although it is also found in non-photosynthetic stramenopiles and rhizaria and also in some choanoflagellates . Lineages where there is biochemical evidence determining inversion or non-inversion of the carbon chain in the conversion from D-glucose to ascorbate are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 006 GULO is also present the Apusomonadida ( Thecamonas trahens ) , a sister group to the Opisthokonts , and in members of the Amoebozoa and Excavata . Surprisingly , we also found GULO in basally derived Archaeplastida including the glaucophyte , Cyanophora paradoxa , and the rhodophytes Galdieria sulphuraria and Galdieria phlegrea . The glaucophytes occupy a key position in the evolution of photosynthetic eukaryotes as they diverged from the other Archaeplastida before the split of the red and green algal lineages and have highly unusual chloroplasts ( termed cyanelles ) that retain several features of the cyanobacterial endosymbiont ( Price et al . , 2012 ) . GULO is absent from all other Archaeplastida genomes , although an enzyme family exhibiting weak similarity to GULO has been reported in Arabidopsis ( Maruta et al . , 2010 ) . However , this enzyme forms a distinct phylogenetic clade from all other GULO and GLDH sequences and its role in de novo ascorbate biosynthesis remains unclear . GLDH was found in all Archaeplastida genomes , except for Cyanophora and Galdieria , and in all photosynthetic lineages that have acquired a plastid via secondary endosymbiosis ( including stramenopiles , cryptophytes , haptophytes , chlorarachniophytes and euglenids ) . GLDH was present in several non-photosynthetic organisms including the oomycetes , the foraminifera and in the choanoflagellates , Monosiga brevicollis and Salpingoeca rosetta . The evolutionary history of algal plastids acquired by secondary endosymbiosis remains uncertain and there is some evidence that non-photosynthetic stramenopile ( e . g . , oomycetes ) and rhizarian ( e . g . , foraminifera ) lineages may have once acquired a plastid that was subsequently lost ( Tyler et al . , 2006; Keeling , 2010; Glöckner et al . , 2014 ) . Further identification of GLDH or GULO in the transcriptomes of 165 eukaryotes within the Marine Microbial Eukaryote Trancriptome dataset ( Keeling et al . , 2014 ) confirmed that GLDH was found primarily in photosynthetic organisms ( Supplementary file 2 ) , but also in the non-photosynthetic stramenopiles such as oomycetes , biocosoecids and labryinthulids and in an acanthoecid choanoflagellate . The presence of GULO was restricted to non-photosynthetic organisms , including the heterotrophic flagellate Palpitomonas bilix , which is a non-photosynthetic relative of the Cryptophyte algae ( Yabuki et al . , 2014 ) . The exception was the presence of GULO in the chromerids , Chromera velia and Vitrella brassicaformis , which are photosynthetic relatives of the Apicomplexa ( Janouskovec et al . , 2010 ) . Further searches of Expressed Sequence Tag ( EST ) and Transcriptome Shotgun Assembly ( TSA ) databases identified a GULO sequence in the green alga , Chlorokybus atmophyticus ( JO192417 . 1 ) ( Leliaert et al . , 2012; Timme et al . , 2012 ) . Chlorokybus represents a basal lineage in the charophyte algae , which are a sister group to the land plants ( Figure 2—figure supplement 1 ) . GLDH was identified in the other charophytes Klebsormidium flaccidum , Nitella mirabilis and Nitella hyalina ( JO285109 . 1 , JV744884 . 1 , JO253095 . 1 ) . These searches also revealed the presence of GULO in the craspedid choanoflagellate , Monosiga ovata ( DC478225 . 1 ) . The Craspedida subgroup of the choanoflagellates is divided into two major clades; clade I contains M . brevicollis and S . rosetta ( which both possess GLDH ) and clade II contains M . ovata ( Jeuck et al . , 2014 ) . We conclude that nearly all photosynthetic eukaryotes use GLDH rather than GULO as the terminal enzyme in ascorbate biosynthesis . The exceptions are the basally derived Archaeplastida ( Cyanophora , Galdieria and Chlorokybus ) and the chromerids ( Figure 2—figure supplement 2 ) . No photosynthetic organisms were found to lack both GULO and GLDH , although both genes were absent in many non-photosynthetic organisms , including all insects , Daphnia , Paramecium and Dictyostelium . The genomes of many parasitic groups also appear to lack both enzymes including the diplomonads , parabasalids and apicomplexa ( e . g . , Giardia intestinalis , Trichomonas vaginalis and Plasmodium falciparum ) . Since all documented pathways of ascorbate biosynthesis require either GULO or GLDH , organisms lacking both of these enzymes are likely to be ascorbate auxotrophs ( Chatterjee , 1973 ) . A maximum likelihood tree of GULO and GLDH sequences was produced using the other members of the VAO family as an outgroup to root the tree . GLDH is highly conserved and the phylogenetic analyses strongly support a monophyletic origin for all GLDH sequences ( 100% bootstrap support , posterior probability = 1 ) ( Figure 3 ) . The monophyly of eukaryote GULO sequences is well supported ( 85% bootstrap support , posterior probability = 1 ) . This clade includes trypanosome l-GalL oxidase and ascomycete D-arabinonolactone oxidase ( Huh et al . , 1994; Logan et al . , 2007 ) , indicating that although these enzymes exhibit altered substrate specificity they should be considered within the GULO clade for our evolutionary analyses . There is also moderate support for the monophyly of GULO and GLDH as a clade within the VAO family , supporting hypotheses that these enzymes may have originated from a gene duplication event . GLDH is highly unusual amongst the VAO flavoprotein family in that it does not use O2 as an electron acceptor , but mutation of a single highly conserved alanine residue in GLDH is required to convert it from a dehydrogenase to an oxidase ( Leferink et al . , 2009 ) . The phylogenies within the GULO clade and the GLDH clade are poorly resolved , and so the trees do not provide evidence on the likelihood of lateral gene transfer , such as endosymbiotic gene transfer ( EGT ) or horizontal gene transfer ( HGT ) , of either gene . Further phylogenetic analyses , individually examining each gene using unrooted trees with much greater taxonomic sampling , were unable to provide greater resolution ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 06369 . 007Figure 3 . Phylogenetic analysis of l-gulonolactone oxidase and l-galactonolactone dehydrogenase . A maximum likelihood phylogenetic tree demonstrating the relationships between aldonolactone oxidoreductases involved in ascorbate biosynthesis . A multiple sequence alignment of 263 amino acid residues was used with alditol oxidases from the vanillyl alcohol oxidase ( VAO ) family acting as the outgroup . Photosynthetic organisms are shown in green . There is strong support for a monophyletic origin for GLDH in eukaryotes . Bootstrap values >80% are shown above nodes ( 100 bootstraps ) and Bayesian posterior probabilities >0 . 95 are shown below ( 10000000 generations ) , except for selected key nodes ( circled ) where all values are displayed . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 00710 . 7554/eLife . 06369 . 008Figure 3—figure supplement 1 . Phylogenetic analysis of l-galactonolactone dehydrogenase . An unrooted maximum likelihood phylogenetic tree of GLDH . To improve resolution of GLDH phylogeny , an individual phylogenetic analysis was performed using a larger alignment ( 302 amino acids ) with greater taxonomic sampling ( 151 sequences ) , although relationships between major taxonomic groups remain poorly resolved . Bootstrap values >70% are shown ( 100 bootstraps ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 008 Many of the enzymes preceding GULO or GLDH in the animal and euglenid pathways play other roles within the cell , for example , in uronic acid metabolism or providing pentose intermediates ( Linster and Van Schaftingen , 2007 ) . The presence or absence of these genes is therefore not solely related to ascorbate biosynthesis . However , the plant pathway of ascorbate biosynthesis contains a number of dedicated enzyme steps , allowing a much clearer examination of its distribution . This also enables a distinction to be made between the plant- and euglenid-type pathways , as both utilise GLDH as the terminal enzyme . Plants and green algae use GDP-l-galactose phosphorylase ( VTC2 ) and l-galactose dehydrogenase to generate l-GalL ( Gatzek et al . , 2002; Running et al . , 2003; Dowdle et al . , 2007; Laing et al . , 2007; Linster et al . , 2007 ) , whereas euglenids use D-galacturonate reductase ( Figure 1 ) ( Ishikawa et al . , 2006 ) . We found that l-galactose dehydrogenase is present in all rhodophytes and Viridiplantae , except the prasinophytes Ostreococcus and Micromonas ( see ‘Materials and methods’ ) . Sequences exhibiting similarity to l-galactose dehydrogenase were also found in the diatoms and in some metazoa , but as some of these species are ascorbate auxotrophs , it appears that this enzyme may play alternative metabolic roles . VTC2 is found exclusively in the Viridiplantae ( including Chlorokybus ) , indicating that the definitive ‘plant’ pathway is restricted to this lineage ( Figure 2; Supplementary files 3 , 4 ) . Biochemical evidence from euglenids and stramenopiles ( Shigeoka et al . , 1979; Helsper et al . , 1982; Grun and Loewus , 1984 ) suggests that organisms that lack VTC2 but possess GLDH are likely to operate a ‘euglenid’ pathway , with D-galacturonic acid acting as the precursor of l-GalL . However , it is not clear whether this is also the case in the rhodophytes . All rhodophytes possess a sequence that is highly similar to characterised l-galactose dehydrogenases from plants and bacteria ( Gatzek et al . , 2002; Hobbs et al . , 2014 ) and l-galactose residues are a major constituent of red algal polysaccharides ( Percival , 1979 ) . We therefore examined whether rhodophytes could synthesise ascorbate via a modified ‘plant’ pathway using an alternative route to l-galactose . We used the macroalga , Porphyra umbilicalis , to examine whether rhodophytes can utilise l-galactose in ascorbate synthesis . We detected NAD+-dependent l-galactose dehydrogenase activity in Porphyra thallus extracts ( Figure 4A ) . Feeding 10 mM l-galactose to Porphyra thallus slices increased the concentration of ascorbate ( detected as dehydroascorbate by GC-MS ) ( Figure 4B ) . As red algae synthesise GDP-l-galactose from GDP-D-mannose ( Su and Hassid , 1962 ) , rhodophytes likely use a modified ‘plant pathway’ to synthesise ascorbate , employing an unidentified enzyme activity to generate l-galactose from GDP-l-galactose instead of VTC2 . 10 . 7554/eLife . 06369 . 009Figure 4 . Biochemical evidence for a modified D-mannose/l-galactose pathway in rhodophytes . ( A ) Crude extracts of Porphyra umbilicalis thallus demonstrate l-galactose dehydrogenase activity using 5 mM l-galactose ( l-Gal ) as a substrate . No activity was demonstrated with 5 mM l-fucose ( 6-deoxy-l-galactose ) as a substrate . The result is representative of three different enzyme preparations . ( B ) Feeding 10 mM l-Gal to Porphyra thallus for 24 hr resulted in an accumulation of ascorbate ( detected as dehydroascorbate—DHA ) . D-mannose ( 10 mM ) did not cause an increase in ascorbate in Porphyra , but exogenous D-mannose does not elevate ascorbate in land plants even though it is an intermediate in ascorbate biosynthesis . The bar chart shows mean peak areas of selected fragments ( ±s . d . ) . n = 3 . ( C ) Feeding ascorbate precursors ( 25 mM ) to Galdieria sulphararia from both the plant and animal pathways results in increased cellular ascorbate ( detected as dehydroascorbate using GC-MS ) ( ±s . d . ) . The extent of the increase in cellular ascorbate is influenced by the rate of conversion of the intermediate and the rate of its uptake into the cell . n = 3 . ( D ) Feeding D-[1-13C]-glucose ( 25 mM ) to Galdieria sulphararia results in enrichment of 13C in the 316/317 m/z fragment of dehydroascorbate ( which includes C1 ) , but not in the 245/246 m/z or 157/158 m/z fragment ( which exclude C1 ) ( ±s . d . ) . In contrast , feeding D-[6-13C]-glucose ( 25 mM ) labels all fragments , suggesting that they all include C6 . In combination , this labelling pattern indicates plant-like non-inversion of the carbon chain in the conversion of hexoses to ascorbate . n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 00910 . 7554/eLife . 06369 . 010Figure 4—figure supplement 1 . Positional isotopic labelling of ascorbate biosynthesis . Analysis of ascorbate by GC-MS . Ascorbate is oxidised to dehydroascorbate during the derivatisation process and representative accurate mass spectra are shown . Analysis of a l-[1-13C]-ascorbate standard indicates that the m/z 316 fragment of dehydroascorbate contains C1 whilst other fragments ( m/z 157 and 245 ) do not . Dehydroascorbate from G . sulphuraria extracts exhibits an identical retention time and mass spectra to that of the ascorbate standard . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 010 We then examined ascorbate biosynthesis in Galdieria , which differs from all other rhodophytes ( Supplementary file 4 ) in that it possesses GULO rather than GLDH . l-galactose , l-GalL and l-GulL were all effective precursors of ascorbate in G . sulphuraria ( Figure 4C ) , suggesting that l-galactose is converted to l-GalL , which may then be converted to ascorbate by GULO . A positional isotopic labelling approach indicated that label from D-[1-13C]-glucose was incorporated primarily into carbon 1 ( C1 ) of ascorbate ( Figure 4D , Figure 4—figure supplement 1 ) . This labelling pattern is expected for the plant pathway , while the reduction of a uronic acid intermediate in the animal or euglenid pathways would result in the transfer of label from C1 of glucose into C6 of ascorbate/dehydroascorbate ( Loewus , 1999 ) . G . sulphuraria therefore uses a similar pathway to other rhodophytes , employing GULO instead of GLDH . In combination , these data identify a clear difference between the Archaeplastida and the photosynthetic lineages that have acquired a plastid via secondary endosymbiosis . Whilst both groups use GLDH as the terminal enzyme for ascorbate synthesis , they differ in the route to l-GalL , The Archaeplastida generate l-GalL via l-galactose ( without inversion of the carbon chain of glucose ) , whereas photosynthetic eukaryotes with secondary plastids synthesise l-GalL via D-galacturonate , resulting in inversion of the carbon chain . We have found that nearly all photosynthetic eukaryotes use GLDH to synthesise ascorbate , suggesting that this distribution may be linked to the photoprotective role of ascorbate . We therefore determined the distribution of ascorbate-dependent antioxidant mechanisms in eukaryote genomes . Three main isoforms of APX are found in eukaryotes ( APX , APX-R and APX-CCX , a hybrid enzyme containing both ascorbate and cytochrome c peroxidase domains ) ( Zamocky et al . , 2010; Lazzarotto et al . , 2011; Fawal et al . , 2013 ) . APX and APX-R are found in nearly all photosynthetic eukaryotes and in the choanoflagellates , M . brevicollis and S . rosetta . Euglena and Emiliania do not possess APX or APX-R , although both possess APX-CCX , which is also found in some fungi and in Capsaspora ( Figure 5 ) . The only photosynthetic eukaryote in this analysis that lacks any isoform of APX was C . paradoxa . Cyanophora also lacks all of the remaining enzymes of the plant ascorbate-glutathione cycle: monodehydroascorbate reductase ( MDHAR ) , dehydroascorbate reductase ( DHAR ) and glutathione reductase ( GR ) . The ascorbate-dependent xanthophyll cycle is not present in glaucophytes and so Cyanophora does not require ascorbate for non-photochemical quenching ( Figure 5 ) . Moreover , the cellular concentration of ascorbate in Cyanophora is either very low or absent , as we could not detect ascorbate in Cyanophora extracts using GC-MS ( data not shown ) . It is possible that ascorbate analogues are present that we could not identify . However , in combination with the lack of the plant ascorbate-glutathione cycle and the xanthophyll cycle , we conclude that Cyanophora is unlikely to rely on ascorbate to detoxify peroxides derived from photosynthesis . Cyanophora does however contain several glutathione peroxidases , peroxiredoxins and catalase , as well as a unique peroxidase ( symerythrin ) similar to rubrerythrin of prokaryotes ( Cooley et al . , 2011 ) . These data suggest that glaucophytes rely on alternative mechanisms to detoxify peroxides derived from photosynthesis . As cyanobacteria also do not appear to use ascorbate for photoprotection ( Bernroitner et al . , 2009; Latifi et al . , 2009 ) , the photoprotective role of ascorbate may therefore have emerged in the Archaeplastida after the divergence of the glaucophytes . 10 . 7554/eLife . 06369 . 011Figure 5 . Coulson plot showing the distribution of photoprotective ascorbate-dependent enzymes . Eukaryote genomes were analysed for the presence of enzymes from the plant ascorbate-glutathione cycle , the xanthophyll cycle and other ascorbate-dependent enzymes . We found that eukaryotes possess two distinct isoforms of GSH reductase . PeroxiBase was used to distinguish between the different forms of ascorbate peroxidase ( Fawal et al . , 2013 ) . Boxes highlight the terminal enzymes in the biosynthetic pathway and the ascorbate peroxidase family ( APX , APX-R and APX-CCX ) . MDHAR—monodehydroascorbate reductase; DHAR—dehydroascorbate reductase; GR-I—glutathione reductase isoform I; GR-II—glutathione reductase isoform II; APX—ascorbate peroxidase; APX-R—ascorbate peroxidase-related; APX-CCX—hybrid ascorbate peroxidase/cytochrome c peroxidase; VDE—violaxanthin de-epoxidase; VDE-like—violaxanthin de-epoxidase like; AO—ascorbate oxidase . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 011
Ascorbate ( vitamin C ) is a very familiar metabolite to humans , so it is perhaps surprising that so many aspects of its biosynthesis and metabolism remain uncharacterised . The biosynthetic pathway of ascorbate in plants , which supplies the vast majority of ascorbate in the human diet , remained elusive for many years ( Wheeler et al . , 1998 ) and the major role of ascorbate in DNA demethylation emerged only very recently ( Blaschke et al . , 2013 ) . In order to better understand the cellular roles of ascorbate , we have examined the distribution of the three major pathways of ascorbate biosynthesis in eukaryotes . We identify that the Opisthonkonts ( animals and fungi ) , the Amoebozoa and the non-photosynthetic representatives of the Excavata and CCTH ( Hacrobia ) use GULO for ascorbate biosynthesis . In contrast , the photosynthetic organisms in the Archaeplastida , CCTH ( Hacrobia ) , SAR and the photosynthetic members of the Excavata ( euglenids ) use GLDH . In these photosynthetic organisms , the combination of molecular and biochemical evidence suggests that the non-inversion pathway via L-galactose ( plant pathway ) is restricted to Archaeplastida , whereas the inversion pathway via D-galacturonate ( euglenid pathway ) is used by photosynthetic eukaryotes that acquired plastids via secondary endosymbiosis . The important exceptions to these trends are: firstly , that GLDH is found in several non-photosynthetic organisms , notably in some choanoflagellates ( Opisthokonts ) and stramenopiles and secondly , that GULO is found in several basally derived members of the Archaeplastida . The processes underlying the distribution of the different terminal enzymes are therefore central to our understanding of the evolution of ascorbate biosynthesis . The mutually exclusive distribution of two highly conserved and functionally similar genes in eukaryotes may be explained by either of two evolutionary scenarios: an ancient gene duplication in the last common eukaryote ancestor ( ancient paralogy ) followed by differential loss of either gene , or lateral gene transfer of a novel gene followed by functional replacement of the ancestral gene ( Keeling and Inagaki , 2004 ) . It is likely that one of these evolutionary scenarios underlies the distribution of GULO and GLDH amongst eukaryotes ( Figure 6 ) . 10 . 7554/eLife . 06369 . 012Figure 6 . Evolutionary scenarios for GULO and GLDH . The scheme illustrates two most likely evolutionary scenarios responsible for the distribution of GULO and GLDH in eukaryotes . In the ancient paralogy scenario , an ancient gene duplication in the last common eukaryote ancestor results in the presence of two functionally similar genes , GULO and GLDH , followed by differential loss of either gene in each lineage . In the endosymbiotic gene transfer ( EGT ) scenario , GULO represents the ancestral gene and GLDH represents a novel gene that arose in a specific lineage . EGT of GLDH ( red dashed arrow ) to other photosynthetic lineages ( green ovals ) enables functional replacement of the ancestral gene . Note that GULO represents an ancestral gene in both of these evolutionary scenarios . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 012 The model of ancient paralogy requires that both genes were present in the last common eukaryote ancestor , where they both presumably contributed to ascorbate biosynthesis , and were then differentially lost by every eukaryote lineage . This requires that these two functionally similar enzymes co-existed in multiple lineages throughout eukaryote evolution without significant functional divergence . The distribution of GULO and GLDH in the Archaeplastida and in the choanoflagellates suggests that these enzymes may have coexisted for a time in these lineages ( Figure 2—figure supplement 1 ) . However , there are no clear examples of extant eukaryotes that possess both enzymes , which would be expected if they have co-existed extensively throughout eukaryote evolution . The ancient paralogy model also requires extensive loss of both GULO and GLDH . There is clear evidence for multiple independent losses of GULO in animals and indications that GULO activity may be deleterious under certain conditions , providing a potential selective pressure for gene loss ( Bánhegyi et al . , 1996; Hiller et al . , 2012 ) . Similar evidence for the loss of GLDH in eukaryotes is lacking . In addition , Arabidopsis mutants that lack GLDH cannot correctly assemble mitochondrial complex I ( Pineau et al . , 2008 ) , suggesting that loss of GLDH is likely to have many wider impacts on metabolism . The broad distribution of GULO supports an ancient evolutionary origin for this gene . It is present in all of the eukaryote supergroups , including basally-derived lineages within the CCTH ( Hacrobia ) and Archaeplastida and also in the Apusomonads . Although GLDH is also present in most of the eukaryote supergroups ( except the Amoebozoa ) , its distribution is primarily restricted to lineages that have acquired a plastid or to isolated lineages ( e . g . , choanoflagellates ) . Whilst we cannot discount an ancient origin for GLDH in the last common eukaryote ancestor , its distribution may also be reasonably explained by the lateral gene transfer model . In the lateral gene transfer scenario , either GULO or GLDH could represent a novel gene that arose in a specific lineage and was then acquired by other eukaryotes through horizontal gene transfer ( HGT ) or endosymbiotic gene transfer ( EGT ) . However , the distribution of GULO cannot be reasonably explained by lateral gene transfer , as this requires HGT on a massive scale specifically into non-photosynthetic eukaryotes . In contrast , the distribution of GLDH can be largely explained by EGT during plastid acquisition . GLDH may have arisen specifically in ancestral Archaeplastida after the divergence of the glaucophytes and functionally replaced the ancestral gene ( GULO ) . GLDH could then have been transferred to the other photosynthetic lineages via EGT , resulting in the replacement of GULO in lineages that acquired their plastids via secondary endosymbiosis ( Figure 7 ) . The presence of GLDH in some non-photosynthetic eukaryotes may be explained by the evolutionary acquisition of a plastid that was subsequently lost . For example , there is some evidence to support plastid loss in non-photosynthetic stramenopiles , although the number and timing of plastid acquisition events via secondary endosymbiosis remains a subject of significant debate ( Keeling , 2010 ) . The choanoflagellates have not acquired a plastid at any stage , but there is evidence for large scale horizontal gene transfer ( HGT ) from algae into this lineage , including the HGT of APX ( Nedelcu et al . , 2008 ) . Choanoflagellates may therefore have acquired GLDH via HGT along with these other algal genes . EGT of GLDH is a more parsimonious scenario than ancient paralogy , as it requires fewer independent loss events . However , the poor resolution of the phylogenies within the GLDH clade means that direct evidence for either EGT or HGT between lineages is lacking . 10 . 7554/eLife . 06369 . 013Figure 7 . A proposed evolutionary model of ascorbate biosynthesis . The scheme illustrates the proposed events in the EGT evolutionary model of eukaryote ascorbate biosynthesis . In this scenario , ancestral eukaryotes synthesised ascorbate via GULO . GLDH arose in the Archaeplastida following primary endosymbiosis of a cyanobacterium , after the divergence of the glaucophyte lineage . GLDH functionally replaced GULO in the red and green algal lineages , coinciding with the rise of the photoprotective role of ascorbate . Plastid acquisition via secondary endosymbiosis of either a green or red alga resulted in endosymbiotic gene transfer of GLDH and replacement of GULO . As these organisms became the dominant primary producers in many ecosystems , a series of trophic interactions ( dotted lines ) resulted in the loss of GULO in non-photosynthetic organisms , either by providing a ready supply of dietary ascorbate ( resulting in ascorbate auxotrophy in heterotrophic organisms ) or through putative horizontal gene transfer of GLDH ( e . g . , choanoflagellates ) . For clarity , not all potential trophic interactions are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 013 Both the ancient paralogy and EGT evolutionary scenarios are plausible in the wider context of ascorbate biosynthesis . However , the EGT scenario provides a clear rationale to explain why photosynthetic eukaryotes with primary plastids exhibit a different pathway from those with secondary plastids . Our biochemical evidence suggests that ancestral Archaeplastida developed a non-inversion pathway via l-galactose that employed the broad specificity of GULO to oxidise l-GalL . The development of GLDH in ancestral Archaeplastida would have led to the eventual replacement of GULO in all red and green algal lineages , except Galdieria and Chlorokybus , resulting in the non-inversion plant-type pathway found in extant Archaeplastida . In the photosynthetic eukaryotes with secondary plastids , it is likely that the host initially synthesised ascorbate via an animal-type pathway ( involving inversion of chain and GULO ) and that the red or green algal symbiont used a plant-type pathway ( involving non-inversion of the carbon chain and GLDH ) . However , neither pathway appears to operate in photosynthetic eukaryotes with secondary plastids , which instead use a euglenid-type pathway . We propose that EGT of GLDH from the symbiont could have resulted in functional replacement of GULO in the animal-type pathway of the host , leading to a hybrid biosynthetic pathway that employed D-galacturonate rather than D-glucuronate as an intermediate in order to provide l-GalL as a substrate for GLDH . The hybrid pathway therefore involves inversion of the carbon chain of D-glucose and GLDH . The generation of a hybrid pathway suggests that photosynthetic eukaryotes with secondary plastids only acquired GLDH by EGT rather than the entire plant pathway . In conclusion , the distribution of GULO and GLDH in eukaryotes may be explained by either of two evolutionary models; ancient paralogy followed by differential gene loss or EGT of GLDH followed by GULO loss . We favour the EGT scenario as the most parsimonious and the most consistent with the biochemical evidence , but we cannot rule out either scenario based on the current evidence . Therefore our evolutionary analyses do not allow us to definitively identify the origin of GLDH . However , they do enable clear conclusions to be made on the loss of GULO . Both evolutionary models support GULO as an ancestral gene in the last common eukaryote ancestor , indicating that GULO has been lost in almost all photosynthetic eukaryotes . Therefore , we can conclude that photosynthetic eukaryotes encountered strong selective pressure to replace the function of GULO in ascorbate biosynthesis . The critical role of ascorbate in photoprotection has been demonstrated in a diversity of photosynthetic eukaryotes , including land plants , green algae , diatoms and euglenids ( Lavaud and Kroth , 2006; Ishikawa et al . , 2010; Mellado et al . , 2012; Urzica et al . , 2012 ) . Our analyses indicate that many photosynthetic eukaryotes possessed GULO prior to plastid acquisition , whereas almost all extant photosynthetic lineages use GLDH to synthesise ascorbate . The selective pressure to replace GULO in ascorbate biosynthesis following plastid acquisition could therefore be linked to the photoprotective role of ascorbate . One intriguing possibility is that the production of H2O2 by GULO may have limited the ability of the host cell to protect itself against ROS derived from the chloroplast . Ancestral eukaryotes developed multiple antioxidant mechanisms to protect themselves from ROS derived from organelles such as the peroxisome and the mitochondria . However , the acquisition of a photosynthetic cyanobacterial endosymbiont in the Archaeplastida would have resulted in a greatly increased requirement for cellular antioxidants to protect the host cell from H2O2 secreted by the plastid . Ascorbate , synthesised in the host cell but not in the cyanobacterial endosymbiont , appears to have been recruited to this role after the divergence of the glaucophytes . The recruitment of ascorbate as a major cellular antioxidant in photosynthetic eukaryotes may have led to an increased requirement for ascorbate biosynthesis . However , ascorbate biosynthesis via GULO results in the production of H2O2 in the ER lumen . In mammalian cells , this results in a damaging depletion and oxidation of the glutathione pool when ascorbate synthesis is increased by feeding l-GulL ( Bánhegyi et al . , 1996; Puskás et al . , 1998 ) . Ancestral photosynthetic eukaryotes may have been unable to balance their increasing requirements for ascorbate biosynthesis with maintenance of the redox status within the ER , providing selective pressure to uncouple ascorbate biosynthesis from H2O2 production . This hypothesis is consistent with the presence of GULO rather than GLDH in the glaucophytes . As glaucophytes do not appear to use ascorbate for photoprotection , ascorbate biosynthesis would not have been subjected to the same selective pressures as other photosynthetic eukaryotes . This rationale may also apply to the retention of GULO in Galdieria , which is likely to have both possessed both GULO and GLDH . Galdieria is photosynthetic and expresses a functional APX ( Sano et al . , 2001 ) , but it is very sensitive to even moderate light intensities and grows primarily in an endolithic environment utilising heterotrophic carbon sources ( Oesterhelt et al . , 2007 ) . Thus , photo-oxidative stress in this environment may be minimal , reducing the selective pressure in Galdieria to replace GULO . The evolution of vitamin auxotrophy underpins many important nutritional and ecological interactions between organisms ( Helliwell et al . , 2013 ) . The selective pressures resulting in GULO loss in animals represent a combination of the costs of ascorbate synthesis ( including detoxification of H2O2 derived from GULO ) , the physiological requirements for ascorbate and the ecological factors that determine the supply of dietary ascorbate throughout their life cycle . The development of the photoprotective role of ascorbate in photosynthetic eukaryotes would have significantly altered its availability to many heterotrophic organisms . The leaves of land plants have particularly high cellular concentrations of ascorbate relative to other photosynthetic eukaryotes ( Gest et al . , 2013 ) , which may result from their inability to remove intracellular H2O2 via diffusion to an aquatic medium . Our dataset reveals that in almost all documented cases of ascorbate auxotrophy in animals , the major source of dietary ascorbate derives from GLDH rather than GULO ( Table 1 ) . This is the case even for insectivorous animals , as insects appear to lack GULO and must also obtain ascorbate in their diet , primarily from land plants . Thus , the replacement of GULO with GLDH in photosynthetic organisms may have ultimately been an important contributory factor in the loss of GULO in many animal auxotrophs . 10 . 7554/eLife . 06369 . 014Table 1 . Dietary sources of ascorbate in animal auxotrophsDOI: http://dx . doi . org/10 . 7554/eLife . 06369 . 014Animal auxotrophPrimary dietary source of ascorbateUltimate dietary source of ascorbateEnzyme for ascorbate synthesisReferencesPrimatesLand plantsGLDH ( Milton and Jenness , 1987 ) Guinea pigLand plantsGLDHBatsLand plantsGLDH ( Birney et al . , 1976; Milton and Jenness , 1987; Cui et al . , 2011b ) InsectsLand plantsGLDHFishPhytoplanktonGLDHBloodGULOPasserine birdsLand plantsGLDH ( Drouin et al . , 2011 ) InsectsLand plantsGLDHSmall vertebratesGULOTeleost fishZooplankton ( crustacea ) PhytoplanktonGLDH ( Dabrowski , 1990 ) PhytoplanktonGLDHCrustaceaPhytoplanktonGLDH ( Desjardins et al . , 1985; Hapette and Poulet , 1990 ) Phytophagous insectsLand plantsGLDH ( Pierre , 1962; Dadd , 1973 ) Major sources of dietary ascorbate were identified in known animal auxotrophs . This information allows us to assess which terminal enzyme contributed to the production of dietary ascorbate . In nearly all cases the major source of dietary ascorbate is most likely to have been derived from GLDH . Phylogenetic analyses suggest GULO has been lost on multiple independent occasions throughout the Chiroptera ( bats ) . Although ancestral bats may have been primarily insectivores , various sources of dietary ascorbate may have contributed to GULO loss . The passerine birds that are unable to synthesise ascorbate are primarily herbivores or insectivores . However , some members of the Lanius genus ( shrikes ) feed also on small vertebrates , in addition to insects . Most teleost fish are believed to be ascorbate auxotrophs due to loss of GULO . As zooplankton ( primarily crustacea ) are also ascorbate auxtrophs , phytoplankton are likely to be the ultimate source of dietary ascorbate . Reports suggest the ability of crustacea to synthesise ascorbate is either absent or very weak , although the taxonomic sampling and currently available genomic resources are limited . Most , but not all , phytophagous insects have a dietary requirement for ascorbate , and we did not find GULO in any insect genomes . Note also that some species of insect ( e . g . , cockroaches ) may obtain ascorbate from eukaryote endosymbionts , which may allow them to survive on ascorbate-poor diets . The pseudogenisation of GULO in primates , bats and guinea pigs is one of the best known examples of evolutionary gene loss ( Drouin et al . , 2011 ) . Through a wider analysis of ascorbate biosynthesis , we have identified that GULO has also been lost in photosynthetic eukaryotes . Photosynthetic eukaryotes functionally replaced GULO with an alternative terminal enzyme , GLDH , which uncoupled ascorbate biosynthesis from H2O2 production and potentially aided the important photoprotective role of ascorbate . These developments in photosynthetic eukaryotes may have ultimately contributed to the loss of GULO in many herbivorous animals , by influencing their supply of dietary ascorbate .
A broad range of eukaryote genomes were selected for detailed analyses ( Supplementary file 5 ) . Sequence similarity searches were used to identify candidate genes involved in ascorbate biosynthesis . Genomic searches were initially performed using BLASTP with mouse or Arabidopsis proteins . All instances of protein absence were confirmed using TBLASTN against the genome with additional searches using sequences from closely related organisms . Sequence similarity searches of transcriptomic datasets were used to identify trends in the presence of ascorbate biosynthesis genes , but were not used to infer absence . Proteins recovered from sequence similarity searches were identified using a combination of BLAST score , manual inspection of conserved residues in multiple sequence alignments and their position in phylogenetic trees generated by both neighbour-joining and maximum likelihood method within the MEGA5 software package ( Tamura et al . , 2011 ) . For detailed phylogenetic analysis of GULO and GLDH , multiple sequence alignments were generated using MUSCLE . Poorly aligned regions were removed by manual inspection and the alignments were further refined using GBLOCKS 0 . 91b to remove ambiguously aligned sites ( Talavera and Castresana , 2007 ) , resulting in an alignment of 263 amino acids . ProtTest ( Abascal et al . , 2005 ) was used to determine the best substitution model ( WAG with gamma and invariant sites ) ( Whelan and Goldman , 2001 ) . Maximum likelihood phylogenetic trees were generated using PhyML3 . 0 software with 100 bootstraps . Bayesian posterior probabilities were calculated using BEAST v1 . 8 ( Drummond et al . , 2012 ) , running for 10 , 000 , 000 generations , with a burn-in of 1 , 000 , 000 generations . As the phylogenetic relationships within the GLDH clade were not well resolved , further unrooted phylogenetic analyses of GLDH were performed using an individual multiple sequence alignment to allow more positions to be used . Galdieria sulphararia 074G was grown in Galdieria Medium ( GM ) ( CCCryo , Potsdam-Golm , Germany ) at 30°C , light intensity 50 μmol m−2 s−1 . Porphyra umbilicalis was collected from Maer Rocks , Exmouth , UK ( 50° 36′ 31 . 6′′ N 3° 23′ 27 . 0′′ W ) . l-Galactose dehydrogenase activity was measured in ammonium sulphate ( 50% saturation ) precipitates of Porphyra thallus protein extracts ( Gatzek et al . , 2002 ) . To determine the impact of exogenous precursors on ascorbate , Galdieria cultures or slices of Porphyra thallus were incubated with sugars or aldonic acid lactones ( 25 mM Galdieria , 10 mM Porphyra ) for 24 hr in GM or artificial sea water ( Instant Ocean , Aquarium Systems , Sarrebourg , France ) . Galdieria cultures ( 15 ml ) were harvested by centrifugation and extracted with 0 . 5 ml 80% methanol containing 0 . 1% formic acid using sonication in the presence of glass beads . Porphyra thallus was powdered in liquid nitrogen followed by homogenisation in 80% methanol ( 0 . 1 g thallus in 0 . 5 ml extractant ) . Homogenates were centrifuged ( 10 min at 16 , 000×g , 4°C ) . Supernatants ( 100 µl ) were dried into glass vials , methoximated trimethylsilyl derivatives were prepared ( Lisec et al . , 2006 ) and analysed by accurate mass GC-EI-qToF MS ( Agilent 7200 , Agilent Technologies , Santa Clara , CA , USA ) . Derivatives were injected ( 0 . 2–0 . 4 µl , 1/2 to 1/50 split ratio ) onto a Zebron SemiVolatiles GC column ( 30 m analytical + 10 m guard length , 0 . 25 mm internal diameter , 0 . 25 µm film thickness , Phenomenex , Macclesfield , UK ) using He carrier gas ( 1 . 2 ml min−1 ) . Injector temperature was 250°C and column temperature program was 70°C for 4 min , followed by an increase to 310°C at 15°C/min . The column was held at the final temperature for 6 min . Compounds were fragmented at 70 eV and MS spectra were collected ( 50–600 amu at 5 spectra s−1 ) . Ascorbate is oxidised to dehydroascorbate ( DHA ) during derivatisation and DHA was identified by co-chromatography and comparison of accurate mass spectra with an ascorbate standard . The position of stable isotope incorporation from [1-13C]-D-glucose and [6-13C]-D-glucose into ascorbate was identified using mass spectra obtained from injection of [1-13C]-ascorbate derivatives . C6 of DHA was found in fragments with m/z values of 157 . 046 and 245 . 1029 , while m/z 316 . 1038 contained C1 and C6 . 13C enrichment was assessed by the relative abundance of m + 1 for each fragment relative to the 12C ascorbate standard . 13C-labelled compounds were obtained from Omicron Biochemicals ( South Bend , IN , USA ) and all other chemicals were from Sigma–Aldrich ( Dorset , UK ) . Reverse transcriptase PCR was used to verify the expression of GULO in C . paradoxa ( CCAP 981/1 ) and confirm its coding sequence . RNA was prepared using the TRIzol method ( Invitrogen , Paisley , UK ) from C . paradoxa cultures grown in standard medium ( MWC ) , at 20°C , 16:8 light:dark , light intensity 50 µmol m−2 s−1 . Reverse-transcriptase PCR was performed using a gene specific primer ( GGAACTCCTCGAACTTGGGG ) for reverse transcription , followed by amplification of a 1017 base pair region using the following PCR primers: GTCGCCCCTTCTGAGCATAG ( forward ) and CATGAGCGCGTCGAAGTCT ( reverse ) . NCBI accession number KJ957823 . Euglena gracilis ( strain Z ) was grown in Koren-Hutner medium ( KH ) under continuous illumination ( 24 μmol m−2 s−1 ) at 26°C . RNA was harvested by the TRIzol method and used to prepare cDNA . Paired end reads were generated by Illumina sequencing technology resulting in a total of 193 , 472 , 913 reads . De novo assembly was carried out using Trinity ( Haas et al . , 2013 ) , followed by further clustering with TGICL ( Pertea et al . , 2003 ) . The three major routes of ascorbate biosynthesis described in Figure 1 are well supported by biochemical and molecular evidence . However , there is some evidence to suggest that some classes of eukaryotes may use alternative routes to ascorbate or use multiple routes . These pathways are reviewed comprehensively elsewhere ( Loewus , 1999; Smirnoff , 2000; Linster et al . , 2007 ) but the implications for our findings are highlighted below . | Animals , plants , algae and other eukaryotic organisms all need vitamin C to enable many of their enzymes to work properly . Vitamin C also protects plant and algal cells from damage by molecules called reactive oxygen species ( ROS ) , which can be produced when these cells harvest energy from sunlight in a process called photosynthesis . Photosynthesis occurs inside structures called chloroplasts , and has evolved on multiple occasions in eukaryotes when non-photosynthetic organisms acquired chloroplasts from other algae and then had to develop improved defences against ROS . There are several steps involved in the production of vitamin C . In many animals , an enzyme called GULO carries out the final step by converting a molecule known as an aldonolactone into vitamin C; this reaction also produces ROS as a waste product . The GULO enzyme is missing in humans , primates and some other groups of animals , so these organisms must get all the vitamin C they need from their diet . Plants and algae use a different enzyme—called GLDH—to make vitamin C from aldonolactone . GLDH is very similar to GULO , but it does not produce ROS as a waste product . It is not clear how the different pathways have evolved , or why some animals have lost the ability to make their own vitamin C . Here , Wheeler et al . used genetics and biochemistry to investigate the evolutionary origins of vitamin C production in a variety of eukaryotic organisms . This investigation revealed that although GULO is missing from the insects and several other groups of animals , it is present in the sponges and many other eukaryotes . This suggests that GULO evolved in early eukaryotic organisms and has since been lost by the different groups of animals . On the other hand , GLDH is only found in plants and the other eukaryotes that can photosynthesize . Wheeler et al . 's findings suggest that GULO has been lost and replaced by GLDH in all plants and algae following their acquisition of chloroplasts . GDLH allows plants and algae to make vitamin C without also producing ROS , which could explain why vitamin C has been able to take on an extra role in these organisms . The results allow us to better understand the functions of vitamin C in photosynthetic organisms and the processes associated with the acquisition of chloroplasts during evolution . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"evolutionary",
"biology"
] | 2015 | Evolution of alternative biosynthetic pathways for vitamin C following plastid acquisition in photosynthetic eukaryotes |
Although it is often tacitly assumed that gene regulatory interactions are finely tuned , how accurate gene regulation could evolve from a state without regulation is unclear . Moreover , gene expression noise would seem to impede the evolution of accurate gene regulation , and previous investigations have provided circumstantial evidence that natural selection has acted to lower noise levels . By evolving synthetic Escherichia coli promoters de novo , we here show that , contrary to expectations , promoters exhibit low noise by default . Instead , selection must have acted to increase the noise levels of highly regulated E . coli promoters . We present a general theory of the interplay between gene expression noise and gene regulation that explains these observations . The theory shows that propagation of expression noise from regulators to their targets is not an unwanted side-effect of regulation , but rather acts as a rudimentary form of regulation that facilitates the evolution of more accurate regulation .
Many studies over the last decade have established that , even within homogeneous environments , gene expression varies across genetically identical cells due to thermodynamic fluctuations in the molecular events underlying gene expression and the small numbers of molecules involved ( Elowitz et al . , 2002; Rao et al . , 2002 ) . This phenomenon is commonly referred to as ‘expression noise’ ( Blake et al . , 2003; Raser and O'Shea , 2005 ) . Much progress has been gained in understanding the molecular mechanisms underlying noise in gene expression , and noise in transcription in particular ( see , for example , Sanchez and Golding , 2013 ) . In the simplest scenario , basic thermodynamic fluctuations and Brownian motion of the molecular players would cause transcription initiation at a given promoter to occur with a constant probability per unit time , and the corresponding mRNAs to decay with a constant probability per unit time , leading to a Poissonian steady-state distribution in the number of transcripts . Although such Poissonian fluctuations are observed for some genes , most genes exhibit much larger fluctuations in their mRNA copy number . It is generally believed that such increased variability originates in promoters stochastically switching between different states that are associated with different transcription initiation rates . In the simplest scenario , promoters stochastically switch between an ‘on’ and ‘off’ state , producing ‘bursts’ of transcript while in the on state , and this would lead to increased noise as has been well understood theoretically ( Kepler and Elston , 2001; Paulsson , 2005; Shahrezaei and Swain , 2008 ) . However , events such as the stochastic binding and unbinding of transcription factors ( TFs ) , or modifications of the local chromatin state , would generally cause most promoters to switch between a much larger number of different states . Moreover , the extent of promoter state switching would be expected to depend on the specific promoter architecture . Indeed , several studies have shown that different promoters show different amounts of gene expression noise , and that these differences are , at least to some extent , encoded in the promoter sequence ( Newman et al . , 2006; Hornung et al . , 2012; Silander et al . , 2012; Carey et al . , 2013; Jones , et al . , 2014 ) . Importantly , transcriptional noise is thus likely an evolvable trait that is subject to natural selection , but it is currently largely unclear how noise levels have been shaped by natural selection ( Raj and van Oudenaarden , 2008 ) . On the one hand , it can be argued that in each condition there is an optimal expression level for each protein , such that variations away from this optimal level are detrimental to an organism's fitness , implying that selection will act to minimize noise . Indeed , by investigating the association between expression noise and various statistics that can be considered proxies of organismal fitness , several studies have provided evidence that selection generally acts to minimize noise ( Newman et al . , 2006; Barkai and Shilo , 2007; Lehner , 2010; Lehner , 2008; Silander et al . , 2012 ) . In this interpretation , genes with lowest noise have been most strongly selected against noise , whereas high noise genes have experienced much weaker selection against noise . On the other hand , gene expression noise generates phenotypic diversity between organisms with identical genotypes , and there are well-established theoretical models showing that such phenotypic diversity can be selected for in fluctuating environments ( Bull , 1987; Kussell and Leibler , 2005 ) . In support of such theoretical models , a number of studies provided examples in which there is a positive association between expression noise and growth ( Blake et al . , 2006; Bishop et al . , 2007; Ackermann et al . , 2008; Zhang et al . , 2009 ) . It is thus possible that some of the genes with elevated noise may have been selected for phenotypic diversity .
In order to assess how natural selection has acted on the transcriptional noise of promoters , it is critical to determine what default noise levels would be exhibited by promoters that have not been selected for their noise properties . To address this , we evolved a large set of synthetic Escherichia coli promoters de novo in the laboratory using an experimental protocol in which promoter sequences were selected on the basis of the mean expression level they conferred , while experiencing virtually no selection on their noise properties ( Figure 1 ) . We synthesized a pool of random DNA sequences , 100–150 base pairs in length , and cloned these upstream of a sequence containing a strong ribosomal binding site and the open reading frame of green fluorescent protein ( GFP ) . Beginning with a library of more than 1 million random promoter clones , we used fluorescence activated cell sorting ( FACS ) to select cells expressing specific levels of GFP ( Figure 1A–C ) . After sorting , we used PCR mutagenesis to input more genetic variation into the library of promoters and repeated the sorting . After the initial FACS sort , this strategy of mutagenesis followed by FACS was repeated four times . The result was a genetically diverse collection of functional promoters that conferred expression close to a pre-specified target level . We selected a subset of 479 synthetic promoters from the third and fifth rounds of FACS selection , choosing equal numbers of promoters from each of six replicate lineages we evolved ( Figure 1; ‘Materials and methods’ ) . We then used flow cytometry , as described previously ( Silander et al . , 2012 ) , to measure the distribution of fluorescence levels per cell for each synthetic promoter , as well as for all native E . coli promoters ( Zaslaver et al . , 2006 ) . We used quantitative Western blotting to confirm that the mean fluorescence levels were directly proportional to GFP molecule numbers ( Figure 1—figure supplement 2 and Appendix 1 ) , which allowed us to express fluorescence levels in units of numbers of GFP molecules . 10 . 7554/eLife . 05856 . 003Figure 1 . Experimental evolution of functional promoters de novo . ( A ) We created an initial library of approximately 106 unique synthetic promoters by cloning random nucleotide sequences , of approximately 100–150 base pairs ( bp ) in length , upstream of a strong ribosomal binding site followed by an open reading frame for GFP , as used to quantify the expression of native E . coli promoters ( Zaslaver et al . , 2006 ) , and transformed this library into a population of cells ( ‘Materials and methods’ ) . We evolved populations of synthetic promoters by performing five rounds of selection and mutation on this library . In each round we used fluorescence activated cell sorting ( FACS ) to select 2 × 105 cells that lie within a gate comprising the 5% of the population closest in fluorescence to a given target level . Next , plasmids were isolated from the selected cells and PCR mutagenesis was used to introduce new genetic variation into the promoters . We then re-cloned the mutated promoters into fresh plasmids and transformed them into a fresh population of cells . We performed this evolutionary scheme on three replicate populations in which we selected for a target expression level equal to the median expression level ( 50th percentile ) of all native E . coli promoters and three replicate populations in which we selected for a target expression level at the 97 . 5th percentile of all native promoters ( referred to here as medium and high expression levels , respectively ) . ( B ) Changes in the fluorescence distribution for one evolutionary run selecting for medium target expression ( top ) and one evolutionary run selecting for high target expression ( bottom ) . The curves show the population's expression distributions before selection , with the numbers above each curve indicating the selection round . The colored bars at the top indicate the FACS gates that were used to select cells from the populations at each corresponding round . ( C ) Examples of fluorescence distributions for individual clones obtained after five rounds of evolution . Microscopy pictures of two individual clonal promoter populations are shown as insets . ( D ) For each native E . coli promoter ( blue ) and synthetic promoter ( red ) , the mean ( x-axis ) and variance ( y-axis ) of log-fluorescence intensities across cells were measured using flow cytometry . Fluorescence values are expressed in units of number of GFP molecules . The green curve shows the theoretically predicted minimal variance as a function of mean expression ( Appendix 1 ) . The insets show the log-fluorescence distributions for two example promoters ( corresponding to the larger dark blue and light blue dots ) . ( E ) Cumulative distributions of excess noise levels of native ( blue ) and synthetic ( red ) promoters . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00310 . 7554/eLife . 05856 . 004Figure 1—figure supplement 1 . Genetic diversity of 378 sequenced promoters , which were extracted from randomly selected clones from the populations that were obtained after three and five rounds of selection . Sequences were clustered using single-linkage based on 100% , 95% , or 90% sequence identity ( left , middle , and right panels ) and the bar plots show the corresponding histograms of cluster sizes . The results indicate that the promoters in the populations at the third and fifth rounds are highly diverse , deriving from many different initial random sequences in the initial library . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00410 . 7554/eLife . 05856 . 005Figure 1—figure supplement 2 . Mean log-fluorescence intensities as measured by FACS ( horizontal axis ) against estimated log GFP molecules per cell ( vertical axis ) as estimated from quantitative Westerns ( see Appendix 1 ) for eight selected promoters . Error bars were estimated from three replicates for the FACS measurements and six replicates for the GFP levels . The straight line shows the fit y = x + 1 . 06 , which is equivalent to: GFP molecules per cell = 2 . 88* mean FACS intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00510 . 7554/eLife . 05856 . 006Figure 1—figure supplement 3 . Relationship between log-protein levels as measured by GFP intensity in FACS ( vertical axis ) and log-mRNA levels ( horizontal axis ) . The mRNA levels are estimated relative to the mRNA level of reference gene IhfB . Error bars show ±1 standard deviation of the posterior probability distribution on mRNA levels ( Appendix 1 ) . Black data points correspond to native promoters and red data points to synthetic promoter . The straight line shows a linear fit with slope 1 , that is , the best fit to a model where the protein level p is directly proportional to the mRNA level m , log ( p ) = c + log ( m ) , with c = 7 . 06 ( Appendix 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00610 . 7554/eLife . 05856 . 007Figure 1—figure supplement 4 . Comparison of three biological replicate FACS measurements of means and excess noise of log-fluorescence for evolved E . coli promoters . The top three panels compare mean log-fluorescences across three replicates and the bottom three panels compare excess noise in log-fluorescences across three replicates . The Pearson squared correlation coefficients between pairs of replicate measurements are indicated at the top of each panel . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00710 . 7554/eLife . 05856 . 008Figure 1—figure supplement 5 . Relative noise levels ( variance of the log-expression distribution ) of five pairs of native promoters that have very similar mean expression levels . Each dot corresponds to one of the pairs of promoters and shows the ratio of the noise level of the highest noise promoter to that of the lower noise promoter as measured by FACS ( horizontal axis ) and by microscope ( vertical axis ) . The blue line shows the line y = x . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00810 . 7554/eLife . 05856 . 009Figure 1—figure supplement 6 . Mean log-fluorescence ( horizontal axis ) and excess noise levels ( vertical axis ) , that is , the difference between variance of log-fluorescence levels and the minimal variance at the corresponding mean , for all native ( blue dots ) and synthetic ( red dots ) promoters . Both axes are in units of number of GFP molecules . Note that , in contrast to raw variances in log-fluorescence that show a clear dependence on mean log-fluorescence , the excess noise levels show no dependence on mean . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 00910 . 7554/eLife . 05856 . 010Figure 1—figure supplement 7 . Cumulative distributions of excess noise levels for the native ( blue ) and synthetic promoters ( red ) . The left panel shows the cumulative distribution of excess noise for promoters whose mean log-expression was less than log ( 18 , 000 ) ( corresponding to the medium expressing synthetic promoters ) , and the right panel for promoters with mean log-expression more than log ( 18 , 000 ) ( corresponding to the high expressing synthetic promoters ) . High noise promoters are clearly enriched among native promoters for both medium and high expressing promoters . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 010 Observing that the fluorescence distributions across cells were well approximated by log-normal distributions ( Figure 1C ) , we characterized each promoter's distribution by the mean and variance of log-fluorescence , defining the latter as the promoter's noise level ( Figure 1D ) . This definition of noise is equivalent to the square of the coefficient of variation whenever fluctuations are small relative to the mean ( Appendix 1 ) , which applies to most promoters , that is , the variance is less than 0 . 25 for 75% of all promoters ( Figure 1D ) . Although our reporter constructs measure protein levels , that is , GFP , the differences in the noise levels of the reporters are likely dominated by differences in transcriptional noise of the promoters . First , the only differences between the different constructs are the promoter sequence inserts . Consequently , the mRNAs of the different reporters are almost identical , varying only by the short sequence segment between the transcription start site and the constant part of the construct . Second , the reporters were constructed specifically to measure transcription , and feature a constant 5′ UTR part upstream of the start codon of the GFP gene , including a strong ribosomal binding site ( Zaslaver et al . , 2006 ) . Using qPCR we confirmed that protein levels were determined primarily by mRNA levels ( Figure 1—figure supplement 3 and Appendix 1 ) . Because protein decay and dilution rates are identical for all reporters , this implies translation rates vary little across the reporters . Although we have not explicitly measured mRNA decay rates of the reporters , we presume that , because the mRNAs are nearly identical , and because translation rates vary little across the reporters , mRNA decay rates likely vary also only moderately across the reporters . Finally , we note that noise levels were reproducible across biological replicates ( Figure 1—figure supplement 4 ) , and noise levels estimated using microscopy were consistent with those measured by flow cytometry ( Figure 1—figure supplement 5 ) . Importantly , although the differences in noise levels are likely due to differences in transcriptional noise , fluctuations in translation and dilution rates will also contribute to total noise levels that we observe . Indeed , as expected ( Bar-Even et al . , 2006; Newman et al . , 2006 ) , we observed a systematic relationship between the mean and variance of expression levels of each promoter ( Figure 1D ) . In particular , we observed a strict lower bound on variance as a function of mean expression . This lower bound is well described ( Figure 1D , green curve ) by a simple model that incorporates background fluorescence , an intrinsic noise component which is proportional to the number of proteins produced per mRNA , and an extrinsic noise component which likely reflects overall fluctuations in transcription , translation , and dilution rates , that all reporters are subject to ( Taniguchi et al . , 2010 ) ( Appendix 1 ) . We defined the excess noise of a promoter as its variance above and beyond this lower bound , allowing us to compare the noise levels of promoters with different means ( Figure 1—figure supplement 6 ) . We found , surprisingly , that most of the synthetic promoters exhibited noise levels close to the minimal level exhibited by the native promoters ( Figure 1D ) . Additionally , a substantial fraction of native promoters exhibited excess noise levels significantly greater than the synthetic promoters ( Figure 1E and Figure 1—figure supplements 6 , 7 ) . For example , only 26 . 1% of the synthetic promoters exhibited excess noise above 0 . 05 , compared to 41 . 6% of the native E . coli promoters ( p < 7 . 7 × 10−10 , hypergeometric test ) . Given that the synthetic promoters were evolved from random sequence fragments and had not been selected on their noise properties ( Appendix 2 ) , we concluded that functional E . coli promoters should exhibit low excess noise levels by default . Importantly , this implies that the native promoters with elevated excess noise must have experienced selective pressures that caused them to increase their noise . To understand how selection might have acted to increase noise , we first investigated whether excess noise was associated with other characteristics of the promoters . Previous studies in Saccharomyces cerevisiae have shown that promoters with high noise tend to also show high expression plasticity , that is , large changes in mean expression level across environments ( Newman et al . , 2006 ) . Although we did not clearly observe this association in data from our previous study ( Silander et al . , 2012 ) , a recent re-analysis of this data did uncover a significant association between expression plasticity and noise ( Singh , 2013 ) , which we confirmed using our present data ( Figure 2A ) . In addition , we found that there is an equally strong relationship between excess noise and the number of regulators known to target the promoter ( Salgado et al . , 2013 ) ( Figure 2B ) . In particular , whereas the excess noise levels of promoters without known regulatory inputs are very similar to those of our synthetic promoters , promoters with one or more regulatory inputs have clearly elevated noise levels ( Figure 2C ) . The general association between elevated noise and gene regulation has recently been observed in eukaryotes as well ( Sharon et al . , 2014 ) , and mutations that lower gene expression noise typically target TF binding sites ( Hornung et al . , 2012 ) . 10 . 7554/eLife . 05856 . 011Figure 2 . Promoters with elevated noise exhibit high expression plasticity and large numbers of regulatory inputs . ( A ) Native promoters were sorted by their excess noise x and , as a function of a cut-off on x ( horizontal axis ) , we calculated the mean and standard error ( vertical axis ) of the variation in mRNA levels across different experimental conditions ( data from http://genexpdb . ou . edu/ ) of all promoters with excess noise larger than x . ( B ) Promoters were sorted by excess noise x as in panel A , and mean and standard error of the number of known regulatory inputs ( vertical axis , data from RegulonDB [Salgado et al . , 2013] ) for promoters with excess noise larger than x is shown . ( C ) Cumulative distributions of excess noise levels of synthetic promoters ( red ) and native promoters without known regulatory inputs ( black ) , with one known regulatory input ( green ) , and with two or more known regulatory inputs ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 011 Our results imply that native promoters with high noise must have experienced selection pressures that caused their noise levels to increase , and that there is a general association between high noise and gene regulation . We next aimed to develop a theoretical understanding of these two observations . Perhaps the simplest interpretation of the observation that natural selection must have increased the noise levels of some promoters is that these promoters were directly selected for increased noise . Several theoretical treatments have shown that phenotypic variability may be selectively beneficial when environments change in ways that cannot be accurately sensed or are too rapid for organisms to respond ( Bull , 1987; Haccou and Iwasa , 1995; Kussell and Leibler , 2005 ) , with the phenotypic variability acting as a ‘bet hedging’ strategy . Thus , it is conceivable that selection has directly selected for increased noise as a bet hedging strategy for a subset of promoters , and more recent theoretical work shows that increasing gene expression noise may indeed increase population growth rates in some scenarios ( Tanase-Nicola and ten Wolde , 2008 ) . However , this interpretation does not explain the association between noise and regulation . On the contrary , one would naively expect that bet hedging strategies function as an alternative to gene regulation , that is , when implementing sensing and regulation would be either too difficult or too costly to evolve ( Kussell and Leibler , 2005 ) . Regarding the general association between gene regulation and expression noise , using an analogy with the fluctuation-dissipation theorem from physics , it has been suggested that expression noise may be an unwanted but unavoidable side-effect of regulation ( Lehner and Kaneko , 2011 ) . Indeed , any regulator will have some noise in its expression or activity , and this noise will be propagated to its target genes . Consequently , this ‘noise-propagation’ effect will cause an increase in expression noise of the targets ( Thattai and van Oudenaarden , 2001 ) . Although noise-propagation is a plausible explanation for the general association between noise and regulation , its effects are detrimental to the accuracy of expression regulation , and one might thus expect natural selection to have acted to minimize its effects , for example , by minimizing the expression fluctuations in regulators . It would thus appear difficult to reconcile our observation that high noise promoters must have experienced selection to increase their noise levels with the assumption that selection has acted to minimize noise-propagation . Instead , our observations would be better explained by a scenario in which noise-propagation is positively selected for . To clarify these observations , we developed a general theoretical model for quantifying how selection acts on gene regulatory interactions . In particular , the model calculates the effect on fitness of evolving a new regulatory interaction between a given gene and a given regulator , as a function of properties of the regulator , and the way selection acts on the gene's expression levels . As explained in ‘Materials and methods’ and Appendix 3 , we derive that , under relatively mild assumptions , the fitness effects of a new regulatory interaction can be calculated analytically , and depend on only a few effective parameters . To explain this general model , we illustrate it using a simple scenario ( Figure 3 ) . 10 . 7554/eLife . 05856 . 012Figure 3 . A model of the evolution of gene expression regulation in a variable environment . ( A ) Expression distribution of an unregulated promoter ( blue curve ) and selected expression ranges in three different environments , that is , the red , gold , and green dashed curves show fitness as a function of expression level in these environments . Although our model applies more generally , for simplicity we here visualize selection as truncation selection ( i . e . , a rectangular fitness function ) . The fitness of the promoter in the gold environment is proportional to the shaded area . ( B ) Contour plot of the log-fitness change resulting from optimally coupling the promoter to a transcription factor ( TF ) with signal-to-noise ratio S and correlation R . Contours run from 7 . 5 at the top right to 0 . 5 at the bottom right . The three colored dots correspond to the TFs illustrated in panels C–H . The red curve shows optimal S as a function of R . ( C–E ) Each panel shows the expression distributions of an example TF across the three environments ( red , gold , and green curves ) . The corresponding values of correlation R and signal-to-noise S are indicated in each panel . ( F–H ) Each panel shows the expression distributions across the three environments for a promoter that is optimally coupled to the TF indicated in the inset . The shaded areas correspond to the fitness in each environment . The total noise levels of the regulated promoters are also indicated in each panel . The unregulated promoter has total noise σtot = 0 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 01210 . 7554/eLife . 05856 . 013Figure 3—figure supplement 1 . Phase diagram of the total noise σtot of a promoter with expression mismatch Y ( horizontal axis ) that is coupled ( at optimal coupling strength ) to a regulator whose regulatory activities have correlation R with the desired expression levels ( vertical axis ) and whose signal-to-noise ratio S has also been optimized . The colors indicate the value of σtot , running from σtot equal to the noise σ of the unregulated promoter ( red ) to σtot = 6σ ( blue ) . A phase boundary ( thick black curve ) separates solutions in a ‘basal noise regime’ at the top left , where the total noise equals the minimal noise σ2 , and solutions in an ‘environment-driven noise regime’ at the bottom right , where the total noise matches the variance in desired levels that is not tracked by the regulation , that is , σtot2= ( 1−R2 ) var ( μe ) −τ2 . The contours show optimal signal-to-noise ratios S* as a function of Y and R . Note that S* diverges at the phase boundary . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 01310 . 7554/eLife . 05856 . 014Figure 3—figure supplement 2 . Inferred noise-propagation strengths of individual E . coli transcription factors ( TFs ) . For all promoters p , the excess noise level Ep was modeled as a linear function Ep=∑r RprVr+noise , where Rpr = 1 when the regulator r is known to target promoter p and Rpr = 0 otherwise ( data from RegulonDB [Salgado et al . , 2013] ) , and Vr is the noise-propagation strength of regulator r . The noise-propagation strengths Vr are inferred by minimizing the squared deviation between the predicted and observed excess noise levels using a Gaussian prior and cross-validation to avoid over-fitting ( Balwierz et al . , 2014 ) . Each bar shows the inferred value of Vr for the TF indicated at the bottom of the bar , together with its error bar σ ( Vr ) . All TFs are shown for which Vr > σ ( Vr ) and are sorted from left to right by their significance Vr/σ ( Vr ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05856 . 014 We focus on a single gene and assume that the gene starts out unregulated , with an expression distribution characterized by a certain mean μ and variance σ2 ( Figure 3A , blue curve ) . In its natural habitat , the population experiences a number of different environments e that may require the gene to express at different levels and we assume that the fitnes of an individual cell , that is , its growth or survival rate , is a function of its gene expression state . Indeed , recent work has confirmed that expression fluctuations in single cells can affect their instantaneous growth rates ( Kiviet et al . , 2014 ) . In the simple scenario of Figure 3 , we assume there is an optimal level μe in each environment , and that cells with expression levels within a certain range τ around this optimum are selected . As an example , Figure 3A assumes there are three environments ( red , gold , and green ) , with the green environment requiring up-regulation of the expression and the red environment requiring down-regulation of the expression . The fitness in each environment corresponds to the fraction of cells with expression levels within the selected range , that is , the unregulated promoter has reasonably high fitness in the gold environment but very low fitness in the green and red environments . Since the overall fitness is the product of the fitness in each environment , a poor overlap between the expression distribution and selected range in any one environment leads to low overall fitness . In our model , the mismatch between the actual and desired expression levels is quantified by the ‘expression mismatch’ Y , where Y2 = var ( μe ) / ( σ2 + τ2 ) is the variance in the desired expression levels μe across environments relative to the sum of the variances of the fitness function τ2 and the expression distribution σ2 ( Y ≈ 4 for the example in Figure 3A ) . We now consider evolving a regulatory interaction between the promoter and a given regulator . We assume that , in each environment e , the regulator's expression , or more generally its activity , will have some average level re . Coupling the promoter to the regulator will have two effects . First , the mean expression of the gene in each environment will become correlated with the mean activity of the regulator . Our model assumes a linear interaction , that is , in each environment e the gene's mean expression becomes μ ( e ) = μ + cre , where c is the coupling constant between regulator and promoter . This is the typical way in which we think about gene regulation and we will call this effect on the gene's mean expression the ‘condition-response’ effect . Second , in each environment e the regulator's activity will also have some variance σr2 and this noise will be propagated to the target gene . Because of this noise-propagation effect , the target's noise level will increase by c2σr2 , becoming σ2+c2σr2 . We define the renormalized coupling constant X as the noise increase relative to the sum of the original noise levels and the variance of the fitness function , that is , X2=c2σr2/ ( σ2+τ2 ) . Our analysis shows that , besides the expression mismatch Y and coupling constant X , the fitness increase that results from coupling to a given regulator depends only on two effective parameters characterizing the regulator: First , the Pearson correlation coefficient R between the desired expression levels μe and the regulator's average activities re and , second , the signal-to-noise ratio of the regulator S , with S2=var ( re ) /σr2 defined as the ratio of the variance in the mean activities of the regulator across the environments and the noise level of the regulator in each environment . In terms of these parameters , the increase in log-fitness resulting from evolving a regulatory interaction becomes ( 1 ) d log[f ( X , Y , R , S ) ]=12 ( X2+R2 ) Y2− ( SX−RY ) 21+X2−12log[1+X2] . Notably , this equation applies independent of how the desired levels μe and regulator levels re vary across the environments and only depends on the assumption that the fitness function and expression distributions can be approximated by Gaussians . To illustrate the predictions of this theory , the contour plot of Figure 3B shows the log-fitness changes that can be obtained by optimally coupling the promoter to a regulator with a given correlation R and signal-to-noise S . We chose the range in S values such that d log[f] is positive over the parameter region shown , that is , d log[f] ≈ 0 in the lower right corner of Figure 3B . As intuitively expected , the highest fitness is obtained when coupling to an accurate regulator with high signal-to-noise S , whose activities correlate precisely with the desired expression levels ( cyan dot in Figure 3B ) . An example of such a TF is shown in Figure 3C . The resulting expression distributions of the promoter coupled to this TF accurately track the desired levels , with only moderately increased noise in the promoter's expression ( Figure 3F ) . In this parameter regime , the improvement in fitness is entirely due to the condition-response effect , and the increased noise of the target can indeed be considered a detrimental side-effect of the regulation . However , regulators that track the desired expression levels of the promoter with such high accuracy , that is , R = 0 . 95 , may often not be available . Interestingly , coupling to a noisy regulator whose activity is entirely uncorrelated with the desired expression levels ( blue dot in Figure 3B and Figure 3E , H ) also substantially increases fitness . In this regime , the increased fitness results exclusively from the noise-propagation mechanism , and by coupling to the regulator the promoter effectively implements a bet hedging strategy . Surprisingly , coupling to the uncorrelated noisy regulator ( blue dot in Figure 3B and Figure 3E , H ) outperforms coupling to a moderately correlated regulator ( magenta dot in Figure 3B and Figure 3D , G ) . To understand how coupling to a regulator with moderate correlation R = 0 . 64 can be outperformed by coupling to a regulator with no correlation R = 0 , we calculated the optimal signal-to-noise S as a function of its correlation R ( red curve in Figure 3B ) . This shows that the magenta regulator has too large an S for its correlation , that is , increasing the noise of this TF would result in an increase of the promoter's noise , and this would in turn lead to an increase in fitness in the green and gold conditions ( see Figure 3G ) . This illustrates that regulators may generally be under selection to become noisy themselves . To the left of the red curve in Figure 3B , noise-propagation is too large and the increased noise of the targets can be considered a detrimental side-effect of regulation . In contrast , to the right of the red curve , noise-propagation is too small , that is , increasing the noise of the regulator would improve fitness . Most interestingly , the red curve corresponds to a continuum of regulatory strategies in which the condition-response and noise-propagation effects are optimally acting in concert , going from being dominated by noise-propagation in the lower left , to being dominated by the condition-response in the top right . Importantly , this clarifies how accurate regulation can evolve smoothly from a state without regulation . Highly accurate regulation with high R and S can be reached by starting from coupling to a noisy regulator with low R and S , whose benefits come entirely from the noise-propagation , and then increasing both R and S in incremental steps along this continuum of regulatory strategies . We now discuss how this theoretical model helps us interpret our experimental observations . First , our model predicts that the selective pressure for a promoter to evolve regulatory interactions is determined by the expression mismatch Y . When Y < 1 , even a constitutively expressed promoter has a good overlap with the fitness function across all conditions , and there will be no selective pressure to evolve regulation . Our synthetic promoters , which were selected for expressing at a constant level , correspond to this situation , and our results show that such promoters have low noise by default . Thus , we interpret the observation that native promoters without known regulatory inputs have noise levels similar to those of our synthetic promoters ( Figure 2B ) as indicating that constitutive promoters have low noise by default . In the interpretation of our model , the promoters with elevated noise were those in which selection required their expression levels to vary significantly across environments , that is , for which Y ≫ 1 . How much expression noise a given promoter is likely to evolve depends on its value of its expression mismatch Y , and the values of the correlations R and signal-to-noise S of the regulators that are available in the genome . Since the precise environmental conditions that E . coli experiences in the wild , and how these determine optimal expression levels of its genes , are largely unknown , it is not possible to make quantitative predictions of the expected noise levels of specific promoters using our model . However , our model can be used to understand the general qualitative trends observed in Figure 2 . First , our model explains why there is a general correlation between expression noise and expression plasticity . Since regulators affect the expression of their targets in a linear manner , coupling a promoter to a combination of different regulators is equivalent to coupling the promoter to a single ‘effective regulator’ whose expression distribution is a linear combination of the expression distributions of the individual regulators . Assuming that coupling to such a linear combination of regulators can attain a correlation R with the desired expression levels of the gene , our model predicts that noise-propagation will be selected whenever Y2> ( 1−R2 ) −1; that is , whenever the expression mismatch Y is large enough , noise-propagation will be beneficial . If we additionally assume that selection has tuned the signal-to-noise of the regulators to optimize the amount of noise-propagation , then the final noise level of the promoter is predicted to equal σtot2= ( 1−R2 ) var ( μe ) −τ2 ( Figure 3—figure supplement 1 and Appendix 3 ) . This expression can be interpreted as saying that , of the original mismatch Y2 , a fraction R2Y2 is accounted for by the condition-response , whereas the remaining fraction ( 1 − R2 ) Y2 is accounted for by the noise-propagation . This implies that both the expression plasticity , which is given by the variance in the promoter's mean across conditions ( i . e . , R2Y2 ) , and the noise level ( i . e . , ( 1 − R2 ) Y2 ) are proportional to the original expression mismatch Y2 . Our model thus predicts that the expression plasticity and noise level should be correlated . Our model also predicts a general positive correlation between expression noise and the number of regulatory inputs . Starting from a high expression mismatch Y , each new regulatory interaction will reduce the mismatch from Y to Y′ < Y by a combination of the condition-response effect reducing the average deviations from the desired levels , and the noise-propagation increasing the overlap by virtue of increasing the expression noise . Whenever Y′ is still larger than 1 , the promoter will be under selective pressure to evolve further regulatory interactions . In this way , the higher the initial mismatch Y , the larger the expected number of regulatory interactions that will be necessary to reduce the mismatch below 1; that is , our model generally predicts that the number of regulatory interactions , the expression plasticity , and the final noise all correlate with the original mismatch Y . Finally , since in our model elevated noise levels are due to noise-propagation per definition , it trivially predicts that , the larger the number of regulatory inputs , the larger the final noise levels tends to be . More specifically , our model predicts that , in a given condition , the noise level of a promoter is determined by the noise levels of the TFs that regulate it . To test this prediction , we used a very simple linear model that assumes that the excess noise level of a gene is equal to the sum of the noise levels of the TFs that regulate it ( ‘Materials and methods’ ) . Although this simple model is very crude , that is , assuming noise-propagation to be of equal size for all targets of a given TF , and assuming that fluctuations in TF activities are all independent , it was nevertheless able to explain a substantial fraction of the variance in excess noise levels across promoters ( 17% ) . The top five TFs most significantly associated with elevated noise levels of their targets were CRP , H-NS , ArcA , ilvY , and GadX ( Figure 3—figure supplement 2 ) . Of these , GadX and H-NS were also identified , using a simpler method , in the data of our previous study ( Silander et al . , 2012 ) . The appearance of H-NS is interesting since it is a histone-like nucleoid-associated protein that acts as a silencer ( Dorman , 2004 ) , that is , somewhat analogous to the role of nucleosomes in eukaryotes , and in eukaryotes nucleosome positioning at promoters has been shown to be a major determinant of transcriptional noise ( Blake et al . , 2006; Tirosh and Barkai , 2008; Cairns , 2009 ) . The TFs ArcA and GadX are involved with responses to low oxygen and acid stress , respectively , and it is plausible that these TFs may be partially activated in the conditions in which our experiments are performed . Our cells are grown in M9 minimal media with glucose in micro-titer plates , and measurements are taken late in the exponential phase . It is well-known that in micro-titer plates oxygen limitation can become a major stress late in the exponential phase , and this may result in the activation of fermentation reactions which in turn cause acid stress . The appearance of CRP is consistent with our observation in Silander et al . ( 2012 ) that promoters of genes involved in carbon metabolism are over-represented among high noise promoters . In summary , modeling of excess noise levels in terms of known regulatory interactions shows that , in accordance with our model , a substantial amount of the variation in noise levels can be explained by noise-propagation from noisy regulators , and the regulators we identify as most significantly propagating noise are consistent with existing biological knowledge regarding our growth conditions .
Because genotype-phenotype relationships for complex phenotypic traits are poorly understood , it is often difficult to assess how observable variation in a particular trait has been affected by natural selection . Here we have shown that , by comparing naturally observed variation in a particular trait with variation observed in synthetic systems that were evolved under well-controlled selective conditions , definite inferences can be made about the selection pressures that have acted on the natural systems . In particular , by evolving synthetic E . coli promoters de novo using a procedure in which promoters are strongly selected on their mean expression and not on their expression noise , we have shown that native promoters must have experienced selective pressures that increased their noise levels , and that promoters with elevated noise are highly regulated by TFs . To account for this , we have developed a theoretical model that provides a simple mechanistic framework for understanding how selection acts on regulatory interactions . The key ingredient of the model is that it recognizes that a regulatory interaction affects the target's expression in two separate ways: the condition-response effect through which the mean expression of the target becomes a function of the mean activity of the regulator , and the noise-propagation effect through which the noise of the target is increased in proportion to the noise of the regulator . Our model elucidates that not only the condition-response effect but also the noise-propagation effect is often a functional consequence of the regulatory interaction; that is , instead of being just an unavoidable side-effect of regulation , noise-propagation is often beneficial and can be considered to act as a rudimentary form of regulation . Our framework vastly expands the evolutionary conditions under which novel regulatory interactions can evolve . Instead of assuming that regulators and their targets must evolve in a tightly coordinated fashion , noise-propagation alone may provide a sufficient benefit for a new regulatory interaction to evolve . This regulation can then be smoothly mutated along a continuum in which noise-propagation and condition-response are acting in concert , slowly lowering noise , and increasing the accuracy of the condition-response , eventually leading to highly accurate regulation . In this way our model provides a plausible scenario for how accurate regulatory interactions can evolve de novo from a state without regulation . Finally , our model shows that unless regulation is very precise , regulatory interactions that act to increase noise are beneficial . Thus , elevated levels of expression noise can be expected whenever the accuracy of regulation is limited .
We obtained chemically synthesized nucleotide sequences of random nucleotides 200 bp in length ( Purimex , Germany ) . Each sequence had defined 5′ and 3′ ends to allow PCR amplification . Within these constant regions , restriction sites for BamHI and XhoI were present . The intervening sequence was made up of 157 bp of random nucleotides ( 5′-CCTTTCGTCTTCACCTCGAG- ( N157 ) -GGGATCCTCTGGATGTAAGAAGG-3′ ) . However , as coupling of base pairs during oligonucleotide synthesis is not always successful and strand breaks can frequently occur in long oligonucleotides , many oligonucleotides were shorter than 200 bp in length . We used PCR to generate double-stranded DNA from the single-stranded oligonucleotides using forward and reverse primers matching the defined 5′ and 3′ ends . We gel-purified the double-stranded PCR product and double-digested it using BamHI and XhoI . After column purification , sequences were ligated into a version of the low-copy plasmid pUA66 , which contains a gfpmut2 open reading frame downstream of a strong ribosomal binding site ( Zaslaver et al . , 2006 ) . The vector was modified to remove a weak σ70 binding site present 24 bp upstream of the GFP open reading frame ( two point mutations , A → G and T → G , were introduced , changing the putative σ70 binding site from TAGATT to TGGATG , with the consensus σ70 binding site being TATAAT ) . The ligation was performed using T4 DNA ligase ( NEB ) at 16°C for 24 hr . The ligation product was then column purified and electroporated into E . coli DH10B cells . This protocol resulted in extremely high transformation yields ( approximately 106 individual clones per transformation ) . Cultures of transformed cells were regenerated for 1 hr in 1 ml SOC medium ( Super Optimal Broth supplemented with 20 mM glucose ) and afterwards 1 ml SOC containing 50 μg/ml kanamycin was added for overnight growth , ensuring that only cells containing the plasmid could grow . These cultures were then diluted 500-fold ( approximately 5 × 106 cells in total ) into M9 minimal media supplemented with 0 . 2% glucose and grown for 2 . 5 hr with shaking at 200 rpm . The distribution of GFP fluorescence levels was measured for each culture using FACS in a FACSAria IIIu ( BD Biosciences ) , with excitation at 488 nm and a 513/17 nm bandpass filter used for emission . We used this distribution of fluorescence values to designate a selection gate . The position of the gate was determined by measuring the mean fluorescence of two reference promoters ( Zaslaver et al . , 2006 ) : gyrB which exhibits a mean expression level that is at the 50th percentile all E . coli promoters; and rpmB , which exhibits a mean expression level that is at the 97 . 5th percentile of all E . coli promoters ( Silander et al . , 2012 ) . For each of these reference genes , the mean fluorescence level was measured and a selection gate was constructed , centered on this mean expression level , such that 5% of all clones in the population fell within the gate . For each round of selection , we sorted 200 , 000 cells contained within this gate . Sorted cells were then transferred to 4 ml Luria Broth ( LB ) media ( containing 50 μg/ml kanamycin ) and grown overnight . These cultures were stored supplemented with 7 . 5% glycerol at −80°C for subsequent analysis . For each expression level ( i . e . , reference gene ) we evolved three replicate populations . We refer to these as the medium expressers ( those promoters selected based on the gyrB reference gate ) and high expressers ( those promoters selected based on the rpmB reference gate ) . Following FACS-based selection on fluorescence , we introduced novel genetic variation into the populations using PCR mutagenesis . We first re-grew the cells overnight and used this culture to prepare plasmid DNA . We amplified the promoter sequences from these plasmids using the GeneMorph II Random Mutagenesis Kit ( Stratagene ) with the primers referred to previously that matched the defined regions of the promoters . We used 0 . 01 ng of DNA as starting material and 35 cycles for amplification . This resulted in a mutation rate of around 0 . 01 per bp ( such that we expect that , in 200 bp , 95% of the promoters will contain between zero and four mutations ) . These PCR products were then digested with XhoI and BamHI , ligated back into the vector , and again transformed into DH10B cells . After an initial round of selection on the initial library , this entire process ( PCR mutagenesis , transformation , and selection ) was repeated four times in total . At this point , the plasmid libraries of synthetic promoters were isolated and transformed into E . coli K12 MG1655 for comparison with a library of native E . coli promoters ( see below ) . To quantify fluorescence on a single-cell level , we used flow cytometry with a FACSCanto II ( BD Biosciences ) , with excitation at 488 nm and a 513/17 nm band-pass filter used for emission . We collected data for at least 50 , 000 events . We then gated this data as outlined in Silander et al . ( 2012 ) , identifying approximately 5000 cells most similar in forward scatter ( FSC ) and side scatter ( SSC ) . We then calculated the mean and variance in log-fluorescence using these cells , using a Bayesian procedure that accounts for outliers ( Appendix 1 ) . We randomly selected 479 promoters from the evolved set ( 72 medium expressers and 72 high expressers after three rounds of selection; 168 medium expressers and 167 high expressers after five rounds of selection ) and quantified mean and variance in fluorescence . We used the same measurement procedures to calculate mean and variance for all promoters contained in a library of E . coli promoters also placed upstream of the gfpmut2 open reading frame on the pUA66 plasmid ( Zaslaver et al . , 2006 ) . We refer to the promoters from this library as native E . coli promoters . For 288 promoters , we quantified fluorescence in three independent cultures and found that both mean and variance in expression were reproducible across replicate biological experiments ( Figure 1—figure supplement 4 ) . Additionally , we sequenced 378 sequences from our set of 479 promoter sequences , which showed that even after five rounds of selection , the promoters were quite diverse ( Figure 1—figure supplement 1 ) . To confirm the sensitivity and accuracy of the FACS measurements , we selected 10 promoters and used fluorescence microscopy to measure their mean and variance in fluorescence . The cells were grown in the same conditions described above , placed on 1% agarose pad , and images were obtained using a CoolSNAP HQ CCD camera ( Photometrics ) connected to a DeltaVision Core microscope ( Applied Precision ) with a UPlanSApo 100×/1 . 40 oil objective ( Olympus ) . Image-processing was done in soft-WoRx v3 . 3 . 6 ( Applied Precision ) and fluorescence values were extracted based on DIC image-mediated cell detection in MicrobeTracker Suite ( Sliusarenko et al . , 2011 ) . For each cell , we calculated fluorescence per cell volume by summing all pixel values and dividing by the volume of the cell as estimated by MicrobeTracker . Cells undergo substantial phenotypic changes when they are put on agar , including changes in the distribution of cell sizes . Consequently , it is problematic to compare absolute variance measurements directly between FACS and microscope . We therefore compared the relative noise levels of different promoters . The 10 selected native promoters consist of five pairs with almost identical mean expression values ( as measured by FACS ) but with noise levels that vary by different amounts . For each of the five pairs we calculated the ratio of the noise levels of the higher and the lower noise promoter as measured by both FACS and the microscope . As shown in Figure 1—figure supplement 5 , with the exception of one pair of promoters that showed almost equal noise levels in FACS but a 50% difference in noise in the microscope , all other pairs showed good correlation of the relative noise levels in FACS and in the microscope , confirming that relative noise levels are similar in FACS and microscope measurements . To determine the correspondence between fluorescence intensities and absolute GFP numbers per cell , eight individual promoter clones were grown in three biological replicates using the same media conditions as in the experimental evolution . The cells were then re-suspended in SDS sample buffer , heated for 5 min at 95°C , and proteins were resolved by 12% SDS-PAGE . Quantification was done by loading a standard curve consisting of 10 , 25 , 50 , 75 , and 100 ng of GFP ( #632373; Clonetech ) . Proteins were transferred to a Hybond ECL membrane ( GE Healthcare , Life Sciences ) , which was then blocked in TNT ( 20 mM Tris pH 7 . 5 , 150 mM NaCl , 0 . 05% Tween 20 ) with 1% BSA and 1% milk powder . Detection was performed with the ECL system after incubation with rabbit anti-GFP and polyclonal pig anti-rabbit . Western intensities for each sample were extracted using ImageJ ( Figure 1—figure supplement 2 ) . The number of cells loaded was estimated by calculating the relationship between OD600 and CFU counts . Details of the data analysis procedures are given in Appendix 1 . Native and evolved single-promoter populations were grown in three biological replicates by diluting overnight LB cultures 500-fold into M9 media supplemented with glucose . These cultures were grown for 2 . 5 hr , stabilized with an equal volume of RNA Later ( Sigma–Aldrich ) and RNA was extracted using the Total RNA Purification 96-Well Kit ( Norgen Biotek Corp ) with on-column DNAse I digestion . Reverse transcription was done using random hexamers and qPCR with TaqMan probes and performed by Eurofins Medigenomix GmbH ( Germany ) . Three technical replicates were performed . The efficiency of the primers and probes used were validated in a dilution series . Relative RNA levels per cell were obtained by normalizing to the reference gene ihfB using a Bayesian procedure for integrating data from the replicates and accounting for failed measurements ( Appendix 1 ) . The primers and probes used were: GFP forward primer: 5′-CCTGTCCTTTTACCAGACAA-3′; GFP reverse primer: 5′-GTGGTCTCTCTTTTCGTTGGGAT-3′; GFP probe: 5′-TACCTGTCCACACAATCTGCCCTTTCG-3′ , ihfB forward primer: 5′-GTTTCGGCAGTTTCTCTTTG-3′ , ihfB reverse primer: 5′-ATCGCCAGTCTTCGGATTA-3′ , ihfB probe: 5′-ACTACCGCGCACCACGTACCGGA-3′ ) . In a simple model of gene expression in which there are constant rates of transcription , translation , mRNA decay , and protein decay , the probability distribution for the number of proteins per cell is a negative binomial with variance proportional to the mean 〈n〉: var ( n ) = ( b+1 ) 〈n〉 , where the constant b is the ratio between the mRNA translation rate and the mRNA decay rate , which is often referred to as ‘burst size’ ( Shahrezaei and Swain , 2008 ) . However , in general there are also cell-to-cell fluctuations in the transcription , translation , and decay rates , which are proportional to these rates themselves . These fluctuations lead to an additional term in the variance var ( n ) which is proportional to the square of the mean: var ( n ) =β〈n〉+σab2〈n〉2 , where β is a renormalized burst size and σab2 is the relative variance of the product of transcription , translation , and decay rates across cells ( Appendix 1 ) . The total fluorescence in a cell ( measured in units equivalent to number of GFP proteins ) nmeas can then generally be written as: nmeas=nbg+〈n〉+ϵvar ( n ) , where nbg is background fluorescence and ϵ is a fluctuating quantity with mean zero and variance one . Assuming that the fluctuations are small relative to the mean , we then find for the variance of the logarithm of nmeas:var ( log[nmeas] ) =σab2 ( 1−nbg〈nmeas〉 ) 2+β〈nmeas〉 ( 1−nbg〈nmeas〉 ) . We fit this functional form to the minimum variance var ( log[nmeas] ) as a function of the mean , with σab2=0 . 025 and β = 450 . We defined the excess variance as the difference between the measured variance and this fitted minimal variance . A more detailed derivation is given in Appendix 1 . By comparing the distributions of the population's expression levels before and after rounds of selection ( without intervening mutation of the promoters ) , we found that the probability that a cell with expression level x is selected by the FACS is well-approximated by f ( x|μ* , τ ) =exp[− ( x−μ* ) 22τ2] , with μ* the desired expression level and τ the width of the selection window . For the last three rounds of selection for medium expression , the selection gates in the FACS were relatively constant , and we estimated τ ≈ 0 . 03 and μ* fluctuated slightly around an average value of μ* ≈ 8 . 1 for these selection rounds . With this selection function , a promoter genotype that exhibits a distribution of expression values with mean μ and standard deviation σ has a fitness ( fraction of cells selected in the FACS ) of ( 2 ) f ( μ , σ|μ* , τ ) =τ2τ2+σ2exp[− ( μ−μ* ) 22 ( τ2+σ2 ) ] . This estimated fitness function indicated that the fitness of promoter genotypes strongly depends on their mean μ and is almost independent of their excess noise ( Appendix 2—figure 3 and Appendix 2—figure 4 ) . In addition , applying additional rounds of selection of varying strengths to the population of evolved promoters did not systematically alter their distribution of excess noise levels . Details of the analysis of the FACS selection are given in Appendix 2 . Although the model we present can be extended to include the evolution of gene regulation for multiple genes , for simplicity we focused on the evolution of a single gene and its promoter . We assume that the population experiences a sequence of different environments and that , in each environment , the fitness of each organism is a function of its gene expression level . We characterized the fitness function in each environment by two parameters: the desired level μe that maximizes the fitness and a parameter τ that quantifies how quickly fitness falls away from this optimum . For simplicity and analytical tractability , we assumed a Gaussian form: f ( x|μe , τ ) =exp[− ( x−μe ) 22τ2] . Similarly , although it is straightforward to allow the variance τ2 to vary across conditions , the results are more transparent when we assume τ2 is the same in all environments . Note that this fitness function has the same form as the FACS selection function . Consequently , the fitness f ( μ , σ|μe , τ ) of a promoter with mean μ and variance σ2 is given by Equation 2 as well , with μe replacing μ* . The total number of offspring that a promoter will leave behind after experiencing all environments is given by the product of its fitness in each of the environments . Equivalently , the log-fitness of a promoter is proportional to its average log-fitness across all environments . For an unregulated promoter with fixed mean μ and variance σ2 in expression , we then find for the log-fitness:log[f ( μ , σ ) ]=− ( μ−〈μe〉 ) 2+var ( μe ) 2 ( τ2+σ2 ) +12log[τ2τ2+σ2] , where 〈μe〉 is the average of the desired expression levels across environments and var ( μe ) is the variance in the desired expression levels across environments . If we do not consider gene regulation but simply optimize the promoter's mean expression and noise level , then we find optimal log-fitness occurs when μ=〈μe〉 and σ2 = 0 ( when var ( μe ) < τ2 ) or σ2 = var ( μe ) − τ2 otherwise . That is , when the desired expression level varies more than the width of the selection window , fitness is optimized by increasing noise so as to ensure the distribution overlaps the desired levels across all conditions . This result is equivalent to previous results on the evolution of phenotypic diversity in fluctuating environments ( Bull , 1987 ) . To increase fitness , a promoter can evolve to become regulated by one of the regulators existing in the genome . Instead of having a constant mean expression μ , the promoter's mean expression will then become a function of the environment e: μ ( e ) = μ + cre , where re is the mean expression ( or more generally regulatory activity ) of the regulator in environment e , and c is the coupling strength . Note that , for simplicity , we thus assume a linear coupling between the means of regulator and target . Since any gene will have some variability in its expression , we assumed that the actual expression/activity of the regulator in each environment e is Gaussian distributed with a variance σr2 . As for the width of the fitness function τ , it would again be straightforward to allow σr2 to vary across conditions ( as it likely does in reality ) . However , the results are analytically more transparent and bring out the main features of the model better if we assume the regulator's noise σr2 is the same in all conditions . When coupled to the regulator , the promoter's total expression variance will become σtot2=σ2+c2σr2 and the log-fitness of the promoter becomes:log[f ( μ , σ , c ) ]=−〈 ( μ+cre−μe ) 2〉2 ( τ2+σ2+c2σr2 ) +12log[τ2τ2+σ2+c2σr2] . Assuming that the basal expression level μ is optimized to maximize log-fitness , that is , μ=〈μe〉−c〈re〉 , this log-fitness can be rewritten as:log[f ( X , Y , S , R ) ]=cons . −12Y2 ( 1−R2 ) + ( SX−RY ) 21+X2−12log[1+X2] . where X measures the coupling strength ( X2=c2σr2τ2+σ2 ) , Y is the expression mismatch that measures how much the desired expression level varies across environments ( Y2=var ( μe ) τ2+σ2 ) , S is the signal-to-noise of the regulator ( S2=var ( re ) σr2 ) , and R is the Pearson correlation between the desired expression levels μe and the activity levels re of the regulator . The change in log-fitness between the situation before and after adding of the regulatory interaction is obtained by subtracting log[f ( 0 , Y , S , R ) ] from log[f ( X , Y , S , R ) ] , yieldingd log[f ( X , Y , S , R ) ]=12Y2 ( R2+X2 ) − ( SX−RY ) 21+X2−12log[1+X2] . Note that this basic argument can be iterated . After the promoter has been coupled to a regulator , the residual deviations between the desired and actual expression levels are given by μ˜e=μe−cre and the new noise level of the promoter is given by σ˜2=σ2+c2σr2 . If we define a new expression mismatch Y˜2=var ( μ˜e ) / ( σ˜2+τ2 ) , then we can calculate the log-fitness changes associated with adding another regulatory interaction using exactly the same expressions as above , replacing Y by Y˜ . In addition , because the coupling between the activity of the regulator and the expression of the promoter is linear , coupling the promoter to an arbitrary linear combination of different regulators can be modeled as coupling the promoter to a single ‘effective’ regulator; that is , if a promoter is coupling to different regulators ri with coupling constants ci , then in environment e we have μ ( e ) =μ+∑i cirei , which is equivalent to coupling with constant c to a regulator with mean re=∑i cirei/c . If σi2 is the noise level of regulator i and Rij is the Pearson correlation in the fluctuations of regulators i and j , then this composite regulator has a total variance σr2=∑i ci2σi2+∑i≠j Rijσiσj . As can be easily seen from Equation 1 in the main text , if the best linear combination of regulators provides a correlation R with the promoter's desired levels , the optimal value S* of the signal-to-noise of this composite regulator is given by S* = RY/X . Substituting this back into Equation 1 , we find for the optimal coupling strength X*2=max[0 , ( 1−R2 ) Y2−1] . This function is plotted in Figure 3—figure supplement 1 , together with the values of S* as a function of Y and R . Note that ( 1 − R2 ) Y2 is the part of the expression mismatch that is not accounted for by the condition-response effect of the regulators . Whenever this remaining expression mismatch is less than 1 , noise-propagation is a detrimental side-effect of regulation and regulators will be selected to be as accurate as possible . However , when ( 1 − R2 ) Y2 > 1 , noise-propagation will be selected for , and the increase in the total noise is equal to the amount of expression mismatch not accounted for by the condition-response . Additional details on the derivation of our model and analysis of the behavior of the fitness function as a function of its parameters are given in Appendix 3 . We re-annotated the promoter fragments of Zaslaver et al . , 2006 by mapping the published primer pairs to the E . coli K12 MG1655 genome . Of the 1816 promoter fragments , 1718 could be unambiguously associated with a gene that was immediately downstream , and the 1718 promoter fragments were associated with 1137 different downstream genes ( for some genes there were multiple or repeated upstream promoter fragments ) . We used the operon annotations of RegulonDB ( Salgado et al . , 2013 ) to extract , for each promoter , the set of additional downstream genes that are part of the same operon as the first downstream gene . We obtained known regulatory interactions between TFs and genes from RegulonDB and counted , for each E . coli gene , the number of TFs known to regulate the gene . We defined the number of regulatory inputs of a promoter to equal the average of the number of inputs for all genes in the operon downstream of the promoter . We sorted promoters by their excess noise and , as a function of a cut-off on excess noise level , calculated the mean and standard error of the number of regulatory inputs for all promoters with excess noise level above the cut-off . We obtained genome-wide gene expression measurements from the Gene Expression Database ( http://genexpdb . ou . edu/ ) . For each E . coli gene , we obtained 240 log fold-change values x corresponding to the logarithm of the expression ratio of the gene in a perturbed and a reference condition . We defined the variance in expression of a gene as the average of x2 across the 240 experiments . We again sorted promoters by their excess noise and , as a function of a cut-off on excess noise level , calculated the mean and standard error of gene expression variances for all promoters with excess noise above the cut-off . Using the RegulonDB database ( Salgado et al . , 2013 ) , we constructed a binary matrix R of regulatory interactions , where the components Rpr = 1 when regulator r is known to target promoter p , and Rpr = 0 otherwise . Following previous work from our group in which we modeled gene expression patterns in mammals in terms of regulatory sites ( FANTOM Consortium et al . , 2009; Balwierz et al . , 2014 ) , we use a simple linear model to relate the excess noise Ep of each promoter p to the ( unknown ) noise-propagation strengths Vr of each regulator r:Ep=∑rRprVr+noise . We assume the noise is Gaussian distributed with unknown variance , and we use a Gaussian prior P ( Vr ) ∝e−λVr2/2 on the noise-propagation strengths Vr to avoid over-fitting . The hyper-parameter λ is chosen using a cross-validation , fitting the Vr on a random fraction of 80% of the promoters , and maximizing the quality of the predictions on the remaining 20% of the promoters . The quality of fit is quantified by the fraction of the variance in noise levels Ep that is explained by the fit . For our dataset , 17 . 1% of the variance of the overall dataset was explained by the fit . | Genes are stretches of DNA that contain the instructions needed to make proteins and other molecules . By changing how much protein is produced from each gene ( i . e . , its expression ) , many organisms—including humans—can produce a wide variety of cell types with very different behaviors . Similarly , single-celled organisms , such as bacteria , can adapt to survive and grow in different environments by changing gene expression levels . It is thus thought that gene expression must be precisely controlled . However , the molecular processes involved in gene expression are subject to random fluctuations , and so gene expression is inherently ‘noisy’ . This means that even groups of identical cells in identical environments will show variation in their gene expression patterns . Furthermore , different genes show different levels of noise . The DNA sequence of a part of each gene , called the promoter , has a big effect on these noise levels . Consequently , gene expression noise is a genetically encoded trait , and can therefore be shaped by natural selection . But it remains largely unclear how natural selection has affected gene expression noise . Now , Wolf et al . have carefully measured the gene expression noise of hundreds of synthetic promoters that were evolved in the laboratory from random DNA sequences , and a similar number of natural promoters in a bacterium called E . coli . These experiments revealed that , contrary to expectation , most lab-evolved promoters had low levels of noise . On the other hand , many natural promoters had high levels of noise . Wolf et al . also found that noisy promoters tend to be highly regulated by transcription factors: the proteins that control gene expression by binding to promoter regions . Together , these results imply that unregulated promoters start by having low noise as a default state . Selection pressures must then have caused some E . coli promoters to become regulated by transcription factors and raise their noise levels . But , what might these selection pressures have been ? Many genes need to be expressed at different levels in different conditions , and it is generally accepted that regulation by transcription factors evolves to ‘satisfy’ these requirements . However , transcription factors are themselves noisy , and this noise necessarily propagates to their target genes . Wolf et al . have now developed a general theory showing that this noise-propagation can often benefit an organism . This explains why natural selection can favor an increase in noise levels for regulated genes . Importantly , by showing that the main role of a transcription factor can be to increase the noise of its targets , it suddenly becomes very easy to see how new gene regulatory interactions can evolve from scratch . The next steps in understanding of how gene expression noise evolves will involve manipulating the expression noise of a gene , and measuring how selection acts on such changes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2015 | Expression noise facilitates the evolution of gene regulation |
Endocannabinoids are recently recognized regulators of brain development , but molecular effectors downstream of type-1 cannabinoid receptor ( CB1R ) -activation remain incompletely understood . We report atypical coupling of neuronal CB1Rs , after activation by endo- or exocannabinoids such as the marijuana component ∆9-tetrahydrocannabinol , to heterotrimeric G12/G13 proteins that triggers rapid and reversible non-muscle myosin II ( NM II ) dependent contraction of the actomyosin cytoskeleton , through a Rho-GTPase and Rho-associated kinase ( ROCK ) . This induces rapid neuronal remodeling , such as retraction of neurites and axonal growth cones , elevated neuronal rigidity , and reshaping of somatodendritic morphology . Chronic pharmacological inhibition of NM II prevents cannabinoid-induced reduction of dendritic development in vitro and leads , similarly to blockade of endocannabinoid action , to excessive growth of corticofugal axons into the sub-ventricular zone in vivo . Our results suggest that CB1R can rapidly transform the neuronal cytoskeleton through actomyosin contractility , resulting in cellular remodeling events ultimately able to affect the brain architecture and wiring .
The endocannabinoid ( eCB ) system is emerging as an important regulator of brain wiring during development with a variety of functions , ranging from lineage segregation of stem cells to refinement of synaptic functions in complex neuronal networks ( Williams et al . , 2003; Berghuis et al . , 2007; Harkany et al . , 2008; Mulder et al . , 2008; Vitalis et al . , 2008; Watson et al . , 2008; Wu et al . , 2010 ) . In both the embryonic and adult brains , eCB action is predominantly mediated by CB1 cannabinoid receptors ( CB1Rs ) , which is one of the most highly expressed neuronal G-protein-coupled receptors ( GPCRs ) , known to couple to Gi/o heterotrimeric proteins ( Howlett , 2005 ) , but the molecular mechanisms by which CB1R shapes developing neurons remain mostly unknown . The exact role of eCBs in shaping the neuronal architecture is also under debate , since several reports indicate neurite retraction , while others found the induction of neurite outgrowth following CB1R activation ( review in Gaffuri et al . , 2012 ) . Likewise , currently it is difficult to reconcile the locally repulsive effects of eCBs , reported at axonal growth cones ( Berghuis et al . , 2007; Argaw et al . , 2011 ) , and their role of mediating efficient directional axonal growth and shaping well-fasciculated axonal tracts ( Mulder et al . , 2008; Vitalis et al . , 2008; Watson et al . , 2008 ) . During neuronal development , an elaborate balance of positive and negative regulators is necessary to establish precise neuronal structure . This structure is stabilized by the cytoskeleton , which , similar to non-neuronal cells , is composed of two major polymers , the highly plastic filamentous-actin ( F-actin ) and the more stable microtubule ( MT ) networks . Actin filaments are often cross-linked to a molecular motor protein , the non-muscle myosin II ( NM II ) , whose contractile properties further endow the actomyosin network with highly dynamic control of cell behavior and architecture ( Vicente-Manzanares et al . , 2009 ) . The cytoskeleton is mainly regulated by Rho-like GTPases that control a wide variety of effector mechanisms such as actin polymerization and branching , actomyosin contractility , focal adhesions , microtubule dynamics , and membrane transport ( Kaibuchi et al . , 1999; Etienne-Manneville and Hall , 2002; Hall and Lalli , 2010 ) . Downstream protein kinases such as the Rho-associated , coiled coil-containing kinase ( ROCK ) are the key activator proteins of these convergent-signaling pathways . Interestingly , ROCK is associated with particular CB1R-induced phenotypes . In CB1R-over-expressing B103 cells , the endocannabinoid anandamide induces cell rounding via ROCK ( Ishii and Chun , 2002 ) , and CB1R activation results in RhoA- and ROCK-dependent repulsion of growth cones of cultured hippocampal neurons ( Berghuis et al . , 2007 ) , but neither the coupling mechanism of CB1R to ROCK nor the cytoskeletal targets downstream of CB1R-activated ROCK are identified yet . Since Rho-activated effectors operate over a large range of spatial and temporal scales , understanding of eCB-mediated structural plasticity requires the identification of the precise spatial and temporal dynamics of CB1R-mediated cytoskeletal modifications . In this study , by using highly resolved live imaging approaches , we report that CB1R-activation rapidly and reversibly contracts the neuronal actomyosin cytoskeleton through an unusual coupling to G12/G13 proteins that produce Rho- and ROCK-mediated NM II activation . In addition , we show that chronic CB1R-mediated activation of actomyosin contractility may mediate lasting changes in neuronal and cerebral morphology .
In order to investigate the spatio-temporal dynamics of cannabinoid-induced cytoskeletal modifications , we have established a sensitive , specific , and highly accessible experimental assay system to study neuronal remodeling downstream of CB1R activation . We have visualized highly dynamic neuronal growth cones in cultured hippocampal neurons , where the activation of endogenous CB1Rs results in repulsion ( Berghuis et al . , 2007 ) , by labeling endogenous F-actin with fluorescent LifeAct . This actin-binding peptide allows observation of the dynamic actin network without perturbing natural reorganization kinetics ( Riedl et al . , 2008 ) . Time-lapse microscopy of live neurons , expressing Flag-CB1R-eGFP and LifeAct-mCherry , showed numerous F-actin-rich dynamic growth cones ( Figure 1A ) advancing at individually variable velocities ( Figure 1A , B ) , but yielding a fairly constant mean growth rate of 20–30 µm/hr ( Figure 1D ) . In addition to growth cones , axonal F-actin was also present in filopodia and in isolated patches on the shaft of the distal axonal region ( Figure 1—figure supplement 1 ) . Strikingly , bath application of 100 nM WIN 55 , 212-2 ( WIN ) , a synthetic cannabinoid agonist , led to a rapid retraction of the F-actin-rich domain ( Figure 1A ) , with mean retraction amplitude of 62 . 2 µm ± 5 . 2 ( Figure 1C–E ) . Retraction was already detectable at 2 min after agonist exposure and typically reached a plateau between 10 and 20 min ( Figure 1C , D ) . The morphology of retracted axons was characterized by an F-actin-rich retraction bulb ( arrowheads on Figure 1A and Figure 1—figure supplement 1 ) and a thin membranous trailing remnant ( open arrowheads on Figure 1A and Figure 1—figure supplement 1 ) , the latter of which was not included in the length measurement . Pre-treatment with the CB1R selective antagonist/inverse agonist AM281 ( AM ) ( 1 µM ) inhibited retraction ( Figure 1D , E ) . 10 . 7554/eLife . 03159 . 003Figure 1 . CB1R activation induces retraction of actin-rich growth cones . Cultured DIV8 hippocampal neurons co-expressing Flag-CB1R-eGFP and LifeAct-mCherry on ( A–G ) and LifeAct-mCherry only on ( H and I ) . ( A ) Treatment with CB1R agonist WIN55 , 212-2 ( WIN , 100 nM , added at 0 min ) induces rapid retraction of the F-actin-rich domain ( arrowheads ) . Open arrowheads: growth cone position at 0 min . ( B ) Progression of individual growth cones in control conditions . ( C ) WIN-induced retraction of individual growth cones . ( D ) Mean values of growth cone progression in control condition or after treatment with WIN with or without pre-treatment with the CB1R-specific antagonist AM281 ( AM , 1 µM ) . WIN-induced growth cone retraction is effectively abolished by AM . ( E ) Amplitudes of growth cone retraction induced by different exo- and endocannabinoids , calculated as the net difference of mean growth cone position in the pre-treatment ( PRE on D ) and post-treatment ( POST on D ) time intervals from at least three independent experiments . ( F ) Concentration-response curve of WIN-induced retraction , 9 to 27 neurons per concentration from two independent experiments expressed as percentage of maximal retraction , Emax = 52 . 2 µm . ( G ) WIN-induced retraction ( 25 nM at 40 min ) is fully reversible after WIN-washout ( at 70 min ) , n = 9 . ( H ) Mean values of growth cone retraction downstream of endogenous CB1R activation , from four pooled independent experiments , outliers were removed in accordance with the Grubb's test . ( I ) Amplitudes of growth cone retraction downstream of endogenous CB1R activation after treatment with WIN ( 100 nM ) , 2-AG ( 1 µM ) , or with WIN ( 100 nM ) after pre-treatment with the CB1R-specific antagonist AM281 ( AM , 1 µM ) . WIN-induced growth cone retraction is effectively abolished by AM . Values in D , F , G , and H are mean ± SEM; values in E and I are presented as boxplots; n . s = p > 0 . 05 , ***p < 0 . 001 , calculated using Kruskal–Wallis one-way ANOVA followed by Dunn's post-tests on ( E and I ) and paired t-test on ( H ) . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 00310 . 7554/eLife . 03159 . 004Figure 1—figure supplement 1 . mCherry-LifeAct label ( red channel ) from Figure 1A . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 004 Further pharmacological characterization showed that several other chemically distinct CB1R agonists , the endocannabinoid 2-arachidonoylglycerol ( 2-AG ) ( 1 µM ) , the principal psychoactive marijuana constituent Δ9-tetrahydrocannabinol ( Δ9-THC ) ( 1 µM ) and the synthetic agonists CP55 , 940 ( 100 nM ) and HU-210 ( 100 nM ) also produced significant retraction ( Figure 1E ) . The retraction was saturable and concentration-dependent with a half-maximal effective concentration ( EC50 ) value of around 20 nM for WIN ( Figure 1F ) . When treatment with 25 nM WIN was followed by ligand-free wash-out , growth cone progression resumed normally showing the reversibility of cannabinoid-induced growth cone retraction ( Figure 1G ) . Finally , this retraction was not a result of CB1R over-expression since treatment with 100 nM WIN or 1 µM 2-AG induced significant retraction with similar kinetics in neurons transfected only with LifeAct-mCherry ( Figure 1H–I ) . However , the mean amplitude of retraction was lower and responses were more variable than in Flag-CB1R-eGFP-expressing neurons ( compare Figure 1C , D with Figure 1H , I ) , as expected in a heterogeneous neuronal population expressing endogenous CB1Rs at highly variable levels ( Leterrier et al . , 2006 ) . In addition , growth cone advance rapidly resumed even in the continued presence of 100 nM WIN ( Figure 1H ) . In conclusion , our results show that cannabinoids trigger a rapid , saturable , and reversible retraction of actin-rich growth cones downstream of both endogenous and overexpressed CB1Rs . First , we investigated which cytoskeletal elements act downstream of CB1Rs to induce rapid growth cone retraction . We expressed , in addition to LifeAct-mCherry , a GFP-tagged version of End-binding protein 3 ( EB3-eGFP ) , which binds to endogenous microtubule ( MT ) plus ends without changing MT growth parameters and thus allows the visualization of MT structure and dynamics ( Stepanova et al . , 2003 ) . Indeed , MTs in the entire neuron were labeled in green , with many bright comet-like fluorescent dashes in all the neuronal compartments , moving randomly in the cell body and directionally in axons and distal dendrites , representing dynamic MT plus ends ( Stepanova et al . , 2003 ) . During 100 nM WIN-induced retraction the dynamics of the two main cytoskeletal polymers , F-actin and MTs , was remarkably different ( Figure 2A ) . A significant portion of F-actin redistributed in the first 2–4 min after stimulation from its original location in growth cones into a more homogenous cable-like pattern on the distal axonal shaft ( Figure 2B and Figure 2—figure supplement 1 ) . In contrast , MTs bent during the same time frame forming periodic local loops ( Figure 2A , A’ , B and Video 1 ) before finally consolidating into a homogenously labeled retraction bulb . The F-actin cables ( bundles of F-actin filaments , which are not separately resolved here by diffraction-limited microscopy ) , often co-localized with regions displaying periodic bends in MTs ( Figure 2B' ) , suggesting that an F-actin-related force pulls strongly enough to bend MTs . This effect was not the result of the over-expression of the cytoskeletal markers EB3-eGFP or LifeAct-mCherry , since we could observe similar periodic MT bends , detected by post hoc immunohistochemistry , in neurons not expressing these markers ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 03159 . 005Figure 2 . CB1R-induced retraction is mediated by non-muscle myosin II dependent actomyosin contraction . Cultured hippocampal neurons co-expressing Flag-CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP at DIV6 were treated by WIN ( 100 nM ) at 0 min . ( A ) Microtubules ( MT ) bend and form small loops ( arrowhead on A′ ) in the first 4 min ( B ) F-actin is reorganized from the growth cone tips and isolated patches to homogenous cable-like distribution in distal axonal shaft . ( C–H ) Pre-treatment with: ( C and D ) MT polymerization inhibitor nocodazole ( 10 µM ) , ( E and F ) actin polymerization inhibitor cytochalasin D ( 1 µM ) , ( G and H ) Non-muscle myosin II-inhibitor blebbistatin ( 25 µM ) . Scale bars: 5 µm on ( A′ ) and ( B′ ) , 20 µm elsewhere . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 00510 . 7554/eLife . 03159 . 006Figure 2—figure supplement 1 . Averaged F-actin relocalization in the distal 60 µm in growth cones in the first 4 min after WIN treatment in five randomly chosen neurons from Figure 1C . Black , orange , and red curves represent the mean intensity of LifeAct-mCherry labeling at baseline , at 2 min and 4 min after addition of WIN ( 100 nM ) , respectively . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 00610 . 7554/eLife . 03159 . 007Figure 2—figure supplement 2 . CB1R-induced periodic microtubule bends are not due to EB3-eGFP and LifeAct-mCherry expression . Cultured hippocampal neurons were transfected at DIV6 with CB1R-eCFP and treated after 24 hr with 100 nM WIN55 , 212-2 for 10 min before fixation . Presence of periodic bends is shown by immunolabeling microtubules with anti-Tuj1 antibody . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 00710 . 7554/eLife . 03159 . 008Figure 2—figure supplement 3 . Concentration-response curve for the blebbistatin effect on the growth cone retraction essay after treatment with WIN ( 100 nM ) . *p < 0 . 05; ***p < 0 . 001 calculated using Kruskal–Wallis one-way ANOVA followed by Dunn's post-tests on ( I ) and using one-way ANOVA followed by Newman–Keuls post-tests . Scale bar: 20 µm on A , B , and B′ and 5 µm on C . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 00810 . 7554/eLife . 03159 . 009Video 1 . CB1R activation induces retraction of actin-rich growth cones . Dynamic , F-actin-rich growth cone of a cultured hippocampal neuron co-expressing CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP at DIV6 treated with 100 nM WIN at 10 min . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 009 To investigate the requirement for polymerized actin microfilaments and MTs in these retractions , we depolymerized MTs with nocodazole ( 10 µM ) and F-actin with cytochalasin D ( 1 µM ) ( Forscher and Smith , 1988 ) . Nocodazole pre-treatment stopped growth cone advance but WIN still induced significant retraction ( Figure 2C , D , Figure 3E and Video 2 ) . In contrast , cytochalasin D inhibited both growth cone advance and WIN-induced retraction ( Figure 2E , F , Figure 3E and Video 3 ) showing that while the presence of both F-actin and MTs is necessary for growth cone advance , as reported previously ( Dent et al . , 2003 ) , only F-actin is necessary for CB1R-induced retraction . 10 . 7554/eLife . 03159 . 010Figure 3 . CB1Rs activate non-muscle myosin II through heterotrimeric G12/G13 proteins , Rho GTPase , and ROCK . Cultured hippocampal neurons at DIV6 co-expressing a combination of LifeAct-mCherry , Flag-CB1R-eCFP , and EB3-eGFP as indicated and treated by WIN ( 100 nM ) at 0 min . ( A–B ) Representative LifeAct-mCherry expressing growth cones ( delimited with a dotted line ) at 2 min after treatment with vehicle ( A ) or WIN ( 100 nM , B ) , labeled with a phospho-Myosin Light Chain ( phosphoMLC ) antibody . Arrowheads show the distal axon adjacent to the F-actin-rich growth cone where WIN induces rapid and strong upregulation of myosin light chain phosphorylation . ( C ) pMLC labeling intensity at the distal 50–60 µm of the axon , adjacent to the actin-rich growth cone , from neurons expressing LifeAct-mCherry ( A ) or co-expressing LifeAct-mCherry and Flag-CB1R-eCFP ( B ) . The region-of-interest used to measure pMLC labeling intensity is delimited with a dotted line on a representative growth cone on Figure 3—figure supplement 1 . ( D ) Amplitude of 100 nM WIN-induced growth cone retraction in neurons co-expressing LifeAct-mCherry and EB3-eGFP pre-treated with 25 µM blebbistatin or 10 µM Y-27632 . ( E ) Amplitude of 100 nM WIN-induced growth cone retraction in neurons co-expressing LifeAct-mCherry , EB3-eGFP , and Flag-CB1R-eCFP pre-treated with: 1 µM cytochalasin D; 25 µM blebbistatin; 25 µM blebbistatin + 10 µM Y-27632; 10 µM Y-27632; 30 µM ML-7 + 10 µM Y-27632; 30 µM ML-7; 10 µM nocodazole; 100 ng/µl PTX . ( F ) Effect of siRNA-mediated knock-down of endogenous myosin IIA , IIB or of endogenous G12/G13 proteins on growth cone-retraction induced by 100 nM WIN in neurons co-expressing the three constructs , as compared to control ( luciferase ) siRNA . Results are pooled from at least two independent experiments , and outliers were removed in accordance with Grubb's test . Results in are expressed as boxplots . n . s p > 0 . 05; *p < 0 . 05; **p < 0 . 01; ***p < 0 . 001 calculated using Student's t-test on ( C ) , Kruskal–Wallis one-way ANOVA followed by Dunn's post-tests on ( D ) and ( E ) , and using one-way ANOVA followed by Newman–Keuls post-tests on ( F ) . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 01010 . 7554/eLife . 03159 . 011Figure 3—figure supplement 1 . Another representative LifeAct-mCherry expressing growth cone 2 min after treatment with WIN ( 100 nM ) , labeled with the phosphoMLC antibody , similarly to Figure 3B . The region of interest used for the quantification of phosphoMLC labeling intensity , approximately 50–60 µm on each image , is marked with dotted lines . This region was delimited using the LifeAct-mCherry image ( B ) and quantified on the raw pMLC image ( C ) . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 01110 . 7554/eLife . 03159 . 012Figure 3—figure supplement 2 . Amplitude of 100 nM WIN-induced growth cone retraction in neurons co-expressing LifeAct-mCherry , EB3-eGFP , and Flag-CB1R-eCFP with ( C3T ) or without ( WIN ) pre-treatement with 1 µg/ml C3T . Results are pooled from two independent experiments and outliers were removed in accordance with Grubb's test . Results in are expressed as boxplots . ***p < 0 . 001 calculated using Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 01210 . 7554/eLife . 03159 . 013Video 2 . Effect of microtubule depolymerization on CB1R-induced growth cone retraction . Dynamic , F-actin-rich growth cone of a cultured hippocampal neuron co-expressing Flag-CB1R-eGFP and LifeAct-mCherry at DIV6 , pre-treated with 10 µM Nocodazole at 20 min before treatment with 100 nM WIN at 40 min . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 01310 . 7554/eLife . 03159 . 014Video 3 . Effect of actin depolymerization on CB1R-induced growth cone retraction . Dynamic , F-actin-rich growth cone of a cultured hippocampal neuron co-expressing CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP at DIV6 pre-treated with 1 µM cytochalasin D at 20 min before treatment with 100 nM WIN at 40 min . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 014 A likely candidate for the generation of such rapid F-actin-related force , which is capable of bending microtubules , is non-muscle myosin II ( NM II ) , an ATPase protein with actin cross-linking and contractile properties , which is activated by the phosphorylation of its regulatory light chain . The two main activators of NM II are myosin light chain kinase ( MLCK ) and ROCK , the latter being already known to participate in CB1R-induced cytoskeletal modifications ( Ishii and Chun , 2002; Berghuis et al . , 2007 ) . This raises the possibility that ROCK- and/or MLCK-induced NM II contractility is responsible for the force-generation reported above . In order to directly investigate the implication of NM II , we pre-incubated neurons , for 20 min before WIN stimulation , with the highly selective NM II ATPase inhibitor blebbistatin ( 25 µM ) that blocks NM II in an actin-detached state without perturbing F-actin polymerization ( Kovacs et al . , 2004 ) . Blebbistatin pre-treatment induced substantial morphological changes of the growth cone , which continued to move forward in a rather disorganized fashion ( Figure 2G and Video 4 ) , typically transforming the growth cone lamellipodia into several dynamically advancing filopodia , as reported previously ( Rosner et al . , 2007 ) . Remarkably , blebbistatin completely abolished WIN-mediated retraction of these dynamically advancing F-actin-rich structures ( Figure 2G , H , Figure 3E and Video 4 ) , suggesting that the main force-generating factor downstream of CB1R activation is actomyosin contractility . This inhibitory effect of blebbistatin was concentration dependent with half-maximal value of inhibition ( EC50 ) of 116 nM ( Figure 2—figure supplement 3 ) . Immunocytochemical analysis of WIN-treated F-actin-rich growth cones at 2 min after the addition of WIN strikingly showed rapid and strong up-regulation of myosin light chain phosphorylation in the distal axon , adjacent to the F-actin-rich growth cone ( Figure 3A–C and Figure 3—figure supplement 1 ) , at the right place for the subsequent NMII-dependent contraction , both in neurons transfected only with LifeAct-mCherry ( Figure 3A–C ) and with Flag-CB1R-eCFP and LifeAct-mCherry ( Figure 3C ) . 10 . 7554/eLife . 03159 . 015Video 4 . Effect of NM II inhibition on CB1R-induced growth cone retraction . Dynamic , F-actin-rich growth cone of a cultured hippocampal neuron co-expressing Flag-CB1R-eGFP and LifeAct-mCherry at DIV6 pre-treated with 25 µM blebbistatin at 20 min before treatment with 100 nM WIN at 40 min . Only LifeAct-mCherry emission is visualized here . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 015 Next , we investigated the mechanism coupling CB1R to the ROCK/NM II pathway . First , we showed that NMII-dependent growth cone contraction is not a result of CB1R over-expression , since treatment with blebbistatin ( 25 µM ) or the ROCK inhibitor Y-27632 ( 10 µM ) ( Figure 3D ) significantly inhibited endogenous CB1R-induced retraction of growth cones , previously presented on Figure 1I , in neurons transfected only with LifeAct-mCherry and EB3-eGFP . Then we used neurons expressing Flag-CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP , our high-throughput experimental read-out , to characterize in detail the molecular mechanism of CB1R-induced actomyosin contractility . The amplitude of WIN-mediated retraction was significantly reduced by pre-treatment with the Rho inhibitor C3 transferase ( 1 µg/ml , Figure 3—figure supplement 2 ) , the ROCK inhibitor Y-27632 ( 10 µM ) ( Figure 3E ) , but not by the MLCK-specific inhibitor ML-7 ( 30 µM ) ( Figure 3E ) . Treatment with the inactive ( R ) - ( + ) -blebbistatin ( 25 µM ) stereoisomer was ineffective ( data not shown ) . The implication of neuronal NM II was further confirmed by siRNA knock-down of endogenous NM IIA and NM IIB ( Miserey-Lenkei et al . , 2010 ) , which resulted in significant reduction of WIN-mediated contractility as compared to control ( anti-luciferase ) siRNA ( Figure 3F ) . Next , we investigated which heterotrimeric G-protein family couples CB1Rs to Rho activation . Notably , treatment with pertussis toxin ( 100 ng/µl ) , a specific inhibitor of Gi/o heterotrimeric proteins , which are generally considered as the main signaling pathway of CB1Rs ( Howlett , 2005 ) , did not decrease significantly cannabinoid-induced growth cone retraction ( Figure 3E ) , similarly to a previously reported finding for ROCK-mediated induced cell rounding after anandamide treatment ( Ishii and Chun , 2002 ) . Another family of heterotrimeric G-proteins , G12/G13 , may mediate rapid growth cone collapse , neurite retraction , and cell rounding in neuronal cell lines in response to certain GPCR agonists such as lysophosphatidic acid ( LPA ) ( Katoh et al . , 1998; Kranenburg et al . , 1999 ) . Therefore , we inactivated endogenous G12/G13 proteins in our hippocampal neuronal cultures by using two pools of 4 different siRNAs directed against rat G12- or G13-alpha proteins , respectively . Used separately , neither pool decreased WIN-induced growth cone retraction as compared to control ( anti-luciferase ) siRNA ( Figure 3F ) . However , when we combined together 2 siRNAs of each pool , each resulting mixed pools efficiently inhibited WIN-mediated contractility ( Figure 3F ) . These results show that the presence of either G12 or G13 is necessary and sufficient for CB1R-induced actomyosin contraction . Finally , to verify that CB1R-induced retraction is not an artifact of altered adhesion properties of growth cones in vitro , we co-transfected Flag-CB1R-eCFP , EB3-eGFP , and LifeAct-mCherry into embryonic rat brains using in utero electroporation at embryonic day 16 ( E16 ) . In organotypic slices prepared from the offspring between postnatal day 4 and 6 ( P4–P6 ) , numerous corticofugal F-actin-rich growth cones from layer II–III pyramidal neurons could be visualized by video microscopy at 48 hr after slice preparation ( Figure 4A ) . Application of 1 µM WIN resulted in significant retraction of growth cones ( Figure 4B , D and Video 5 ) through activation of CB1R since this effect could be prevented by pre-treatment with 5 µM AM281 ( Figure 4D ) . This retraction displayed slower kinetics ex vivo than in vitro ( compare to Figure 1 ) probably due to limited diffusion of the highly hydrophobic WIN into the slice and/or into differences in adhesive and mechanistic properties within the organotypic brain slice . Previously , we have shown that at around P5 , cortical projection neurons still express CB1R , albeit at lower levels than at birth ( Vitalis et al . , 2008 ) , thus we have replicated these experiments by expressing only the cytoskeletal markers EB3-eGFP and LifeAct-mCherry . WIN-mediated activation of endogenous CB1Rs typically led to arrest or retraction of numerous growth cones ( Figure 4C , E ) . The relatively mild averaged effect is probably due to the variable level of endogenous CB1R expression in these neurons . Importantly , pre-treatment with blebbistatin ( 25 µM ) efficiently blocked this effect ( Figure 4E ) . 10 . 7554/eLife . 03159 . 016Figure 4 . Activation of exogenous or endogenous CB1Rs modifies growth cone dynamics ex vivo . Progression of dynamic , F-actin-rich corticofugal growth cones from organotypic slices cultured for 24 to 48 hr , prepared from P4-6 rat brains , previously electroporated in utero at E16 to express EB3-eGFP , LifeAct-mCherry , with or without Flag-CB1R-eCFP , was followed by time-lapse imaging . ( A ) Experimental design and illustration of a typical transfected cortical area ( A ) and of a typical labeled growing axon ( B ) . For the illustration , the organotypic section was fixed and EB3-eGFP signal was enhanced by incubation with an anti-GFP antibody . ( B–E ) Response to CB1R agonist WIN ( 1 µM , added at 0 min ) . The F-actin-rich growth cone is indicated by arrowheads . Open arrowheads indicate growth cone position at 0 min ( B , D ) WIN-induced retraction in growth cones expressing EB3-eGFP , LifeAct-mCherry , and Flag-CB1R-eCFP is abolished by pre-treatment with 5 µM CB1R-specific antagonist AM281 . ( C , E ) WIN-induced retraction in growth cones expressing EB3-eGFP and LifeAct-mCherry is abolished by pre-treatment with blebbistatin ( 25 µM ) . Results are pooled from at least two independent experiments and are expressed as mean ± SEM . *p < 0 . 05; **p < 0 . 01; ***p < 0 . 001 , calculated using Student's t-test . Scale bar: 100 µm on A , 20 µm elsewhere . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 01610 . 7554/eLife . 03159 . 017Video 5 . CB1R activation induces retraction of actin-rich growth cones in organotypic slices . Dynamic , F-actin-rich corticofugal growth cones from organotypic slices were cultured for 24 to 48 hr , prepared from P4-6 rat brains , previously electroporated with EB3-eGFP , LifeAct-mCherry , and Flag-CB1R-eCFP in utero ( See Figure 3 ) . Treatment with 1 µM WIN at 30 min induces retraction of the growth cone . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 017 In conclusion , we show that CB1R activation significantly reorganizes growth cones through MLCK/ROCK-mediated NM II activation . This large-scale actomyosin contractility ultimately leads to the remodeling of MT structure in the distal axonal segments . In the embryonic brain , developing corticofugal axons express high levels of CB1Rs ( Figure 5B’ , B’’ ) ( Vitalis et al . , 2008 ) . Genetic or pharmacological ablation of CB1Rs leads to axonal fasciculation deficits ( Mulder et al . , 2008; Watson et al . , 2008 ) . In order to investigate the importance of actomyosin contractility during the embryonic development of CB1R expressing axons , we inhibited in vivo the ATPase activity of NM II by in utero intra-cerebroventricular injection of rat embryos with blebbistatin ( Figure 5A ) . Notably , 100% of blebbistatin-injected embryos survived and developed without apparent gross anatomical brain defects , suggesting that neuronal NM II can be safely targeted in vivo . In embryos treated between E15 and E17 with active ( S ) - ( − ) -blebbistatin ( Figure 5D , D´ , G ) , but not with the inactive ( R ) - ( + ) stereoisomer ( Figure 5C , C´ , G ) , Tuj1-expressing axons showed important targeting errors , by invading the sub-ventricular zone , from which CB1R-expressing corticofugal axons are usually excluded ( Figure 5B , G and Figure 5—figure supplement 1 ) . Such a representative CB1R-expressing Tuj1-positive axon invading the SVZ from an embryon treated with ( S ) - ( − ) -blebbistatin is shown on Figure 5F . Treatment with the CB1R-specific antagonist AM251 ( 1 mM ) but not with its vehicle ( DMSO 2 . 8% ) led to similar developmental phenotype ( Figure 5E , E´ , G ) . 10 . 7554/eLife . 03159 . 018Figure 5 . Actomyosin contractility is required for the correct targeting of CB1R expressing corticofugal axons . ( A ) Experimental design . Left: in utero intracerebroventricular injection of E15 rat embryos . Right: analysis of axons in the lateral sub-ventricular zone ( SVZ , red ) . ( B ) E15 corticofugal axons starting from the cortical plate ( CP ) and progressing through the intermediate zone ( IZ ) highly co-express Tuj-1 ( green ) and CB1R ( magenta ) and mostly avoid the SVZ . ( C–G ) In embryos injected with 1 µl of the active NM II-ATPase inhibitor ( S ) - ( − ) -blebbistatin ( 250 µM ) ( D , D′ ) , or with AM251 ( 1 mM ) ( E , E′ ) , but not with the inactive ( R ) - ( + ) stereoisomer ( 250 µM ) or the vehicle of AM251 ( DMSO 2 . 8% ) ( C–C′ ) , there is a significant increase of mistargeted corticofugal axons in the lateral SVZ ( arrowheads , G ) . ( F ) Expression of endogenous CB1Rs in a representative Tuj1 positive axon invading the SVZ ( arrowheads ) from an embryon treated with active ( S ) - ( − ) -blebbistatin . Results are pooled from three independent experiments and are expressed as mean ± SEM , **p < 0 . 01 , ***p < 0 . 001 calculated using one-way ANOVA followed by Newman–Keuls post-tests . Scale bars: 100 μm on B and F ( left ) , 250 μm on C , D , and E and 25 μm on B′ , B″ , C′ , D′ , E′ , and F ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 01810 . 7554/eLife . 03159 . 019Figure 5—figure supplement 1 . Cortifugal origin of Tuj1-expressing axons in the SVZ . ( A ) Experimental design . Left: in utero transfection of cortical progenitors of E16 rat embryos with GFP . Right: analysis of GFP-expressing corticofugal axons of transfected and radially migrated neurons ( green ) at E20 . ( B ) Corticofugal axons starting from the cortical plate ( CP ) and progressing through the intermediate zone ( IZ ) co-express Tuj-1 ( magenta ) and GFP ( green ) and mostly avoid the SVZ . ( C and D ) Examples of the Tuj1 positive rare axons in the SVZ ( arrowheads ) that express GFP , suggesting their corticofugal origin . Scale bars: 200 µm on B and 20 µm on C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 019 Together with our above findings showing that activation of endogenous CB1Rs in organotypic slices leads to NM II-dependent arrest or contraction of axonal growth cones , these results suggest that both activation of endogenous CB1Rs and actomyosin contractility are required for correct path finding of corticofugal axons . Next , we asked whether the above reported CB1R-mediated effect on neuronal actomyosin contractility is restricted to growth cones , which are highly specialized mobile structures , or if we can also observe this phenomenon in other neuronal sub-compartments . The mouse neuroblastoma-derived Neuro2A cell line , a widely used model of neuronal physiology , presents simpler morphology than primary hippocampal neurons , enabling high-resolution quantitative measure of cellular structure and biomechanical characteristics . The Neuro2A cells grow neurites in culture , and we observed that F-actin accumulates in the shaft and extremity of these neurites as well as in highly dynamic filopodia and in patches of the cell cortex ( Figure 6A ) . CB1R-eGFP showed a characteristic distribution between the plasma membrane and endosomes , as described previously in various non-polarized cell-types ( Leterrier et al . , 2004 ) . Treatment with 100 nM WIN resulted in rapid rounding of the cell body and in retraction of neurites , leaving behind retraction-fiber-like remnants ( Figure 6A , B and Video 6 ) . F-actin was reorganized and accumulated at the end of the retraction bulb and under the plasma membrane of the cell body . This WIN-induced cell rounding could be blocked by blebbistatin treatment ( 25 µM ) ( Figure 6B ) , suggesting that the observed rapid morphological changes are due to a CB1R-induced general contraction of the actomyosin cell cortex , which is the association of plasma membrane lipids and the underlying actin filament network . 10 . 7554/eLife . 03159 . 020Figure 6 . CB1R-induced actomyosin contraction results in neurite retraction and transiently increased cell stiffness in Neuro2A cells . ( A and B ) Cells expressing Flag-CB1R-eGFP and LifeAct-mCherry . F-actin accumulates in the extremity and shaft of neurites ( arrowheads ) . Agonist WIN ( 100 nM ) induces retraction of neurites . Open arrowheads: neurite tip at 0 min . ( B ) Blebbistatin ( 25 µM ) significantly reduces 100 nM WIN-induced cell rounding . Results are expressed as mean ± SEM . ( C ) Phase-contrast image of a Neuro2A cell and the AFM cantilever . Stiffness response to 100 nM WIN at different cell locations ( crosses ) . Subsequent measurements were focused on the cell bodies , corresponding to positions 2 and 3 . ( D ) Blebbistatin ( 25 µM ) significantly reduces 100 nM WIN-induced increase of cell stiffness . Results are pooled from at least three independent experiments and are expressed as mean stiffness between 2 and 8 min after stimulation ±SEM . ( E ) 3D reconstruction shows neurite retraction , cell rounding , and transitory blebbing ( arrows ) following WIN treatment ( 100 nM ) . n . s p > 0 . 05; ***p < 0 . 001 , calculated using Student's t-test on B and using one-way ANOVA followed by Newman–Keuls post-tests on ( D ) . Scale bars: 10 µm on ( A ) and ( E ) , 15 µm on ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 02010 . 7554/eLife . 03159 . 021Video 6 . CB1R activation induces neurite retraction and cell rounding in Neuro 2A cells . Neuro2A cell expressing Flag-CB1R-eGFP and LifeAct-mCherry . Treatment with 100 nM WIN at 30 min induces neurite retraction and cell rounding . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 021 Comparable large-scale contraction of the actomyosin cortex was previously characterized in detail in cells entering division ( Théry and Bornens , 2008 ) , where a regulated balance between localized actomyosin-cortex-dependent surface tension and intracellular pressure allows dividing cells to control their volume , shape , and mechanical properties ( Stewart et al . , 2010 ) . When combined , these two effects result in an overall increase of cell cortex rigidity or stiffness ( Stewart et al . , 2010 ) , while local and temporary detachment of the plasma membrane from the actomyosin cortex results in characteristic blebbing ( Cunningham , 1995; Charras et al . , 2008 ) . Marked cell rounding , F-actin reorganization , and the presence of retraction fibers suggested that an analogous intracellular mechanism might be implicated in the above-reported cannabinoid-induced reorganization of the Neuro2A cells . We have performed two experiments to investigate this possibility . First , in order to directly measure putative contraction of the neuronal actomyosin cortex , we measured the cell cortex rigidity of isolated CB1R-expressing Neuro2A cells before and after cannabinoid treatment , by using atomic force microscopy ( AFM ) . Averaged AFM measurements in force mode with a 1-µm spherical bead attached to the cantilever ( Figure 6C ) indicated that the stiffness ( Young's modulus ) of unstimulated individual cells was approximately 300 Pa , close to values reported in acutely isolated hippocampal glial cells and neurons ( Lu et al . , 2006 ) . For this series of experiments , cells were grown on uncoated plastic , a highly adhesive substrate , in order to minimize the displacement of Neuro2A cells during the measurement . WIN stimulation led to an overall rapid and transient increase of cell stiffness at different locations on the same cell , with the exception of the trailing edge ( Figure 6C4 ) . As the contribution of the underlying coverslip was significant in neurites ( compare ordinate scale of Figure 6C1 with Figure 6C2–4 ) , in the following experiments we have centered our force measurements on the cell bodies , corresponding to the positions 2 and 3 on Figure 6C . The important transient WIN-induced increase in cell stiffness was absent during incubation with vehicle solution and could be prevented by pre-treatment with blebbistatin ( 25 µM ) ( Figure 6D ) , showing that activation of NM II is necessary to induce the measured changes . Next , in order to follow morphological changes at the plasma membrane in detail , high-resolution time-lapse image stacks of retracting WIN-treated Neuro2A cells were acquired , deconvoluted , and reconstructed in three dimensions ( Figure 6E and Video 7 ) . Prior to remodeling , the shape of the cells suggested a large degree of reinforcement probably owing both to the intracellular actomyosin cortex directly beneath the plasma membrane , and to the attachment of the cell to its substrate . After CB1R agonist application , the cell changed shape drastically , acquiring a more spherical morphology . Moreover , we observed localized blebbing behavior of the cell membrane , starting at the early stages ( ∼2 min ) of the contraction and ceasing after 6–8 min ( Figure 6E and Video 7 ) . 10 . 7554/eLife . 03159 . 022Video 7 . CB1R activation induces neurite retraction , cell rounding , and temporary blebbing in Neuro 2A cells . 3D reconstruction of a Neuro2A cell expressing Flag-CB1R-eGFP and DsRed2 . Treatment with 100 nM WIN at 7 min ( 420 s ) induces neurite retraction , cell rounding , and transitory blebbing . Scale bar: 10 μmDOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 022 In conclusion , our results show that CB1R activation leads to rapid and NM II-dependent neurite retraction and rounding of the cell body in Neuro2A cells , which is accompanied by formation of retraction fibers and by transient increase in cell stiffness and blebbing behavior . Collectively , these findings suggest that global CB1R activation results in large-scale contraction of the neuronal cytoskeleton , which is mechanistically similar to the molecular machinery engaged in mitotic cell rounding . Previously , we have reported that chronic in vitro activation of CB1Rs leads to significant inhibition of dendritic development in cultured hippocampal neurons , while genetic or pharmacological inhibition of CB1Rs leads to more numerous and longer dendrites ( Vitalis et al . , 2008 ) . Similarly , genetic or pharmacological inhibition of CB1R leads to more complex somatodendritic morphology in septal cholinergic neurons ( Keimpema et al . , 2013 ) . These data suggest that , in addition to axons , where CB1Rs are naturally targeted through transcytotic targeting ( Leterrier et al . , 2006 ) , the transitory presence of CB1Rs on the somatodendritic membrane may allow efficient coupling to growth inhibitory signaling pathways . It was reported that increased NM II activity , through constitutively active MLCK or RhoA , decreases both the length and number of neurites and , consequently , delays or abolishes the development of neuronal polarity in cultured hippocampal neurons ( Kollins et al . , 2009 ) . We thus studied whether a long-term effect of the above described rapid , CB1R-activation dependent and NM II-mediated contraction of the neuronal cytoskeleton could explain the negative regulatory effects of CB1R activation at a longer time scale ( ∼24 hr ) . First , we verified the presence of the rapid structural effects of CB1R activation in the somatodendritic region . Neurons expressing CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP at DIV9 responded to 100 nM WIN with rapid morphological reorganization of the somatodendritic compartment , characterized by retraction of distal dendritic regions and broadening of the proximal portion of dendrites ( Figure 7A , A' and Video 8 ) . While the overall dynamics of EB3-eGFP comets was not apparently modified , the MTs in individual dendrites often displayed a characteristic bent morphology , parallel to the appearance of straight cable-like F-actin bundles ( Figure 7A , A' and Video 8 ) suggesting the presence of a rapid CB1R-activation-dependent actomyosin contraction . Overnight treatment with WIN ( 100 nM ) resulted in a significant decrease in the number of dendrites of developing hippocampal neurons , expressing Flag-CB1R-eGFP and the soluble cytoplasmic marker DsRed2 at DIV4 ( Figure 7B , C ) , as reported previously ( Vitalis et al . , 2008 ) . This effect was abolished in the presence of both Y-27632 ( 10 µM ) or blebbistatin ( 25 µM ) ( Figure 7B , C ) . Notably , treatment with both inhibitors led to more developed dendrites also in control conditions , confirming previous reports on the constitutive inhibition of neurite development through ROCK and NM II ( Kollins et al . , 2009 ) . 10 . 7554/eLife . 03159 . 023Figure 7 . Acute and chronic effects of CB1R-mediated actomyosin contraction on somatodendritic morphology . ( A ) Cultured hippocampal neurons expressing CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP at DIV8 . Application of 100 nM WIN results in rapid and significant reorganization of somatodendritic morphology , characterized by retraction of distal dendritic parts ( arrowheads ) , and broadening of the proximal part of dendrites ( arrows ) . ( A′ ) In dendrites , characteristic microtubule bending ( arrowheads ) and appearance of straight cable-like F-actin bundles ( arrowheads ) are accompanied by CB1R endocytosis after agonist activation ( arrows ) . ( B and C ) Chronic inhibition of ROCK or NM II abolishes CB1R-activation induced changes structure of the cultured hippocampal neurons expressing Flag-CB1R-eGFP and the structural marker DsRed2 at DIV4 . Cells were fixed at 24 hr after treatment with inhibitors of ROCK ( Y-27632 , 10 µM ) or NM II ( blebbistatin , 25 µM ) in the presence of vehicle ( VE ) or CB1R agonist WIN ( 100 nM ) . A representative cell is shown for each condition . ( C ) Results are pooled from at least two independent experiments and are expressed as mean ± SEM . n . s p > 0 . 05; **p < 0 . 01 , calculated using one-way ANOVA followed by Newman–Keuls post-tests . Scale bars: 20 µm on ( A ) , 5 µm on ( A′ ) , and 50 µm on ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 02310 . 7554/eLife . 03159 . 024Video 8 . CB1R activation induces rapid remodeling of the somatodendritic region in cultured hippocampal neurons . Somatodendritic region of a cultured hippocampal neuron co-expressing CB1R-eCFP , LifeAct-mCherry , and EB3-eGFP at DIV8 . The axon , whose initial segment is typically strongly labeled with EB3-GFP , exits the frame in the upper-left corner . The F-actin-rich growth cone , such as shown in Video 1 , is at the growing end of the axon , typically hundreds of microns away from the soma at DIV8 . Treatment with 100 WIN at 10 min induces retraction of distal dendrites and broadening of proximal dendrites . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03159 . 024 In conclusion , our results show that the chronic activation of CB1Rs reshapes somatodendritic morphology through enhancement of the naturally present ROCK- and NM II-mediated contractile tone of neurons .
Our results show that acute CB1R activation results in rapid contraction of the neuronal actomyosin cytoskeleton . CB1R acts through heterotrimeric G12/G13 proteins , Rho GTPase , and ROCK to induce the contractile interaction of NM II with F-actin . This contraction triggers the retraction of the actin-rich growth cone of the most distal 60–70 µm of the axon in cultured hippocampal neurons and in cortical neurons in organotypic slices . Pharmacological inhibition of either CB1Rs or NM II during brain development leads to excessive growth of corticofugal axons in vivo , suggesting that CB1R-induced actomyosin contractility is necessary for the correct pathfinding by mediating their repulsion from the sub-ventricular zone . This contractile behavior is not limited to the growth cone since CB1R-induced actomyosin contraction leads to neurite retraction , cell rounding , and a significant elevation in cell rigidity in the Neuro2A cells . Similarly , in the somatodendritic region of cultured hippocampal neurons , distal regions of dendrites retract while proximal parts broaden . Finally , ROCK and NM II mediate the inhibitory effect of chronic CB1R activation on dendrite development , by increasing the natural contractile tone of neurons . Owing to its position downstream of convergent signaling pathways , the NM II protein plays a pivotal role in the control of tissue architecture through its participation in processes that require cell reshaping and movement , such as cell adhesion , cell migration , and cell division ( Vicente-Manzanares et al . , 2009 ) . In neurons , NM II is also involved in diverse aspects of cell movement , such as neuronal migration and the structural organization and efficient extension of the growth cone , which requires an intricate balance between dynein , microtubules , actin , and different myosin II isoforms ( Vallee et al . , 2009 ) . Remarkably , previous in vitro results have shown that NM II is important for turning in response to boundaries of substrate-bound laminin-1 ( Turney and Bridgman , 2005 ) and that pharmacological inhibition or genetic silencing of NM II leads to disorganization of the growth cone , allowing rapid axon extension over inhibitory substrates ( Hur et al . , 2011 ) . In the present study , we report a comparable in vivo effect , by showing that blebbistatin treatment leads to elevated axonal invasion of the embryonic sub-ventricular zone ( SVZ ) . This territory , which is populated by proliferating neuronal progenitors , is typically avoided by corticofugal axons during their progression towards sub-cortical target zones . To our knowledge , these results show for the first time the existence of NM II-mediated axonal repulsion in vivo , suggesting that mobilization of NM II participates in the correct guidance of corticofugal axonal projections . Since pharmacological CB1R blockade has similar effects to NM II inhibition ( i . e . , excessive axonal growth ) , a likely scenario suggests that the endocannabinoid 2-AG , whose synthesizing enzyme DAGLα is specifically expressed at high levels by proliferating progenitor cells of the SVZ ( Goncalves et al . , 2008 ) , acts through CB1Rs to repulse invading corticofugal axons through NM II-mediated growth cone retraction . CB1Rs also rapidly modify the morphology of Neuro2A cells and cultured hippocampal neurons through enhanced actomyosin contractility , leading to large-scale reorganization of neuronal compartments that contain F-actin . Notably , CB1Rs not only alter the internal organization of the growth cone , as suggested previously ( Berghuis et al . , 2007; Argaw et al . , 2011 ) , but cause the retraction of the distal axon over several tens of microns , both in cultured hippocampal neurons and in cortical neurons in organotypic slices . This NM II-dependent contraction leads to the characteristic periodic bending of microtubules . This particular phenotype was similarly observed during strong NM II-mediated retraction in DRG neurons after the activation of the LPA receptor ( Bouquet et al . , 2007 ) or after treatment with Sema 3A ( Gallo et al . , 2002; Wylie and Chantler , 2003; Gallo , 2006; ) . In addition , nitric oxide , widely recognized to induce axonal retractions during development ( Cramer et al . , 1998; Ernst et al . , 2000 ) , was reported to induce similar rapid axonal retraction accompanied by periodic bends ( He et al . , 2002 ) . In addition , RhoA , ROCK , and NM II are known constitutive inhibitors of neurite development ( Kollins et al . , 2009 ) . By activating the CB1R/G12/G13/Rho GTPase/ROCK/NM II axis characterized in our study , endo- or exogenous cannabinoids are likely to mobilize a widely employed myosin-activating machinery that is involved in growth cone navigation and in the establishment and maintenance of neuronal morphology . Interestingly , similar G12/G13-dependent signaling mechanism is mobilized downstream of at least two developmentally implied neuronal GPCRs , the LPA receptor ( Katoh et al . , 1998; Kranenburg et al . , 1999 ) and GPR55 , a putative ‘atypical’ cannabinoid receptor ( Ryberg et al . , 2007; Sharir and Abood , 2010 ) , through activation by lysophosphatidylinositols but not through genuine cannabinoid ligands ( Obara et al . , 2011 ) . Therefore , coupling of a bona fide neurotransmitter and drug receptor , such as CB1R , to this major developmental pathway may open interesting research and therapeutic perspectives . NM II is also involved in integrin-mediated cell adhesion; in turn , the adhesive properties of the substrate also control NM II activation ( Vicente-Manzanares et al . , 2009 ) . However , CB1R-mediated morphological effects reported in the present study may not result from reduced neuronal adhesion , considering the time-scale of the rapid neuronal retraction . Instead , retracting the Neuro2A cells and growth cones of hippocampal neurons both in vitro and ex vivo leave behind thin membranous fibers , which contain F-Actin and are still attached to the adhesive substrate . The formation and morphology of these fibers are similar to those of retraction fibers reported in mitotic cells ( Cramer and Mitchison , 1995 ) also generated by rapid contraction of the cellular actomyosin cortex ( Théry and Bornens , 2008 ) . The ensemble of these results combined with the bulk of the available experimental data ( reviewed in Gaffuri et al . , 2012 ) suggests that endocannabinoids acting through CB1Rs exert a general negative effect on cell spreading and neurite growth . Basal cell-autonomous or paracrine activation of CB1Rs would yield relatively weak tonic inhibition of growth in the majority of developmental settings in all neuronal sub-regions where F-actin is present . Local growth-promoting effectors at the growth cone , such as the self-amplifying autocrine promoter BDNF ( Cheng et al . , 2011 ) or netrin-1 ( Argaw et al . , 2011 ) , may locally surmount this weak negative tone . The resulting ‘channeling’ effect of widespread CB1R-mediated weak inhibition would help the neuron to focus its resources to a limited amount of growth locations , leading to more efficient polarized growth . Such weak inhibition may serve also to coordinated guidance of axonal fascicles in the brain where moderate production of eCBs by nearby axons would be used as a repulsion cue that helps axons to grow straightly towards their target . However , when growth cones reach a region highly enriched with eCBs , such as the sub-ventricular zone , enhanced eCB signaling could result in growth cone arrest , repulsion , or collapse , efficiently steering out CB1R-expressing axons from these areas . Finally , CB1R-induced actomyosin contractility may also contribute to establish functionally adequate somatodendritic morphology , acting as a negative regulator of dendritic growth . In our study , we were able to characterize in detail CB1R-induced actomyosin contraction , which is rather subtle and transitory downstream of endogenous GPCRs , by using high-resolution time-lapse imaging , atomic force microscopy , and moderate over-expression of CB1Rs . Consequently , the amplitudes of reported cytoskeleton changes are likely dramatic compared to CB1R-induced remodeling in typical physiological settings . Nevertheless , the results concerning endogenous CB1Rs , obtained in cultured neurons , in organotypic slices , and in vivo suggest physiological relevance for our findings . In conclusion , we identify NM II-mediated actomyosin contraction as a mechanism conveying a wide-ranging inhibitory role for cannabinoids in neuronal expansion and growth , downstream of CB1R coupled to G12/G13 proteins and the Rho-associated kinase ROCK . Such modulation of the neural actomyosin cytoskeleton has not yet been reported downstream of neurotransmitter GPCRs , therefore our results open previously unexpected perspectives in the study and comprehension of brain function .
CB1R agonists WIN55 , 212-2 , CP-55940 , HU-210 , 2-arachidonoylglycerol ( 2-AG ) , and CB1R-specific antagonist/inverse agonists AM281 and AM251 were acquired from Tocris Bioscience ( Bristol , UK ) . Rho-associated kinase inhibitor Y-27632 and nocodazole were purchased from Calbiochem ( San Diego , CA ) . Blebbistatin , cytochalasin D , ML-7 , Δ9-Tetrahydrocannabidiol solution ( THC ) , and pertussis toxin ( PTX ) were brought from Sigma ( Saint-Louis , MO ) . The Rho-GTPase inhibitor C3 transferase ( C3T ) was purchased from Cytoskeleton , Inc . Mouse anti-neuron-specific beta III tubulin ( Tuj-1 ) antibody was obtained from Sigma ( Catalog Number T8660 ) , rabbit anti-myosin phospho S19/phospho S20 antibody was obtained from Rockland ( Gilbertsville , PA , Cat . no . 600-401-416 ) , rabbit anti-CB1R antibody was produced by Eurogentec ( Seraing , Belgium ) and described previously ( Thibault et al . , 2013 ) , and chicken anti-GFP antibody was from AVES ( Tigard , OR ) . Alexa-Fluor-conjugated secondary antibodies were purchased from Life Technologies ( Carlsbad , CA ) . All culture media and additives were from PAA Laboratories ( Pasching , Austria ) . The DsRed2 encoding plasmid was produced by Clontech ( Mountain View , CA ) . The CB1R-eCFP ( Enhanced Cyan Fluorescent Protein ) and Flag-CB1R-eGFP constructs have previously been described elsewhere ( Leterrier et al . , 2004 , 2006 ) . LifeAct-mCherry was a kind gift of G Montagnac and P Chavrier ( Institut Curie , Paris , France ) . pEGFP-N3–EB3 plasmid was a kind gift of M Piel ( Institut Curie , Paris , France ) . pCAG-Cre and pCALNL-GFP in which GFP was replaced by Flag-CB1R-eCFP , LifeAct-mCherry , or eGFP-EB3 sequences for in utero electroporation experiments were a kind gift from T Matsuda and C Cepko ( Harvard Medical School ) . All constructs were verified by full-length sequencing . For silencing rat non-muscle myosin IIA , rat non-muscle myosin IIB , rat G12- and rat G13- specific SMARTpools were chemically synthesized by Dharmacon Research ( Lafayette , CO ) and siRNA targeting luciferase ( CGUACGCGGAAUACUUCGA , Proligo-Sigma ) was used as a control , as described previously ( Miserey-Lenkei et al . , 2010 ) . Neuro2A cells ( ATCC CCL-131 ) were grown in DMEM ( Life Technologies ) supplemented with 4 . 5 g/l glucose , GlutaMAX I ( Life Technologies ) , 10% fetal bovine serum , 10 U/ml penicillin G and 10 mg/ml streptomycin . Neuronal cultures were prepared as described previously ( Carrel et al . , 2011 ) . Briefly , hippocampi of rat embryos were dissected at embryonic days 17–18 . After trypsinization , tissue dissociation was achieved with a Pasteur pipette . Cells were plated on poly-D-lysine-coated coverslips at a density of 60 , 000–75 , 000 cells per 15 mm coverslip and cultivated in complete Neurobasal ( Life Technologies ) medium supplemented with B27 ( Life Technologies ) , containing 0 . 5 mM L-glutamine , 10 U/ml penicillin G , and 10 mg/ml streptomycin containing conditioned medium obtained by incubating glial cultures ( 70–80% confluency ) for 24 hr . Experiments were performed in agreement with the institutional guidelines for the use and care of animals and in compliance with national and international laws and policies ( Council directives no . 87-848 , 19 October 1987 , Ministère de l'Agriculture et de la Forêt , Service Vétérinaire de la Santé et de la Protection Animale ) . Neuro2A cells were transfected , in 6-well plates for ATF and deconvolution or in 12-well plates for video microscopy , with 0 . 8 μg of plasmid DNA using Effectene reagent ( Qiagen , Venlo , NL ) and processed 24 hr after transfection . Hippocampal neurons were transfected on DIV3 ( for morphometry ) or DIV5-8 ( for videomicroscopy ) as follows: for each coverslip , plasmid DNA ( 2 μg ) and Lipofectamine 2000 ( 1 . 25 μl , Life Technologies ) in Neurobasal medium were combined and incubated for 30 min . After the addition of complete Neurobasal medium containing B27 supplement , the mix was applied onto the neuronal culture for 3 hr at 37°C . Receptor expression was allowed in growth medium for 24 to 72 hr after transfection . Immediately after transfection , DIV3 transfected hippocampal neurons were incubated with different pharmacological treatments and fixed after 24 hr . Our transfection protocol leads to moderate over-expression of CB1Rs and we imaged only low-expressing neurons in which sub-neuronal traffic and targeting of transfected receptors is similar to that of endogenous CB1Rs ( Leterrier et al . , 2006; Vitalis et al . , 2008; Thibault et al . , 2013 ) . For siRNA transfections , two different mixes were prepared: one with Lipofectamine ( 2 µl ) and plasmid DNA ( 1 . 25 µg of Flag-CB1-eGFP and 1 . 25 µg of LifeAct-mCherry ) in 50 µl of Neurobasal medium and one with Lipofectamine and siRNA ( 2 . 4 µl of each siRNA at 50 µM alone or combined with other siRNAs were mixed in 50 µl of Neurobasal medium ) . In controls , appropriate volumes of anti-luciferase siRNA ( 50 µM ) were used to match the total amount of transfected siRNAs . After 30 min of incubation , the two mixes were combined , completed to 250 µl with conditioned complete Neurobasal medium containing B27 supplement and applied to the neuronal culture for 3 hr at 37°C . At the end of incubation , the mix was replaced by fresh complete Neurobasal medium and neurons were used 48 to 72 hr later . Animals were housed individually with free access to food and water and maintained in a temperature-controlled environment on a 12 hr light/dark cycle . Experiments were performed in agreement with the institutional guidelines for the use and care of animals and in compliance with national and international laws and policies ( Council directives no . 87-848 , 19 October 1987 , Ministère de l'Agriculture et de la Forêt , Service Vétérinaire de la Santé et de la Protection Animale ) . Pregnant Sprague–Dawley rats at gestation day 16 were anesthetized with Ketamine/Xylazine ( 75/10 mixture ) . The abdominal cavity was opened to expose the uterine horns . 1–3 μl of plasmids ( 0 . 5 µg/µl for pCAG-Cre , 1 µg/µl for pCALNL-LifeAct-mCherry and pCALNL-EB3-eGFP , and 1 . 5 µg/µl for pCALNL-Flag-CB1R-CFP ) with 1 mg/ml Fast Green ( Sigma ) were microinjected through the uterus into the lateral ventricles of embryos by pulled glass capillaries ( Drummond Scientific , Broomall , PA ) . Electroporation was performed by placing the heads of the embryos between tweezer-type electrodes . Square electric pulses ( 65 V , 50 ms ) were passed five times at 1 s intervals using a CUY21 EDIT electroporator ( Nepa Gene , Chiba , Japan ) . For axonal localization analysis , rat brains ( E20 ) were dissected and fixed for 48 hr in 4% paraformaldehyde ( PFA ) in PBS at 4°C . Brains were then cryoprotected in 30% sucrose in PBS , frozen in OCT compound ( Sakura , Tokyo , Japan ) , and sectioned coronally at 16 µm using a cryostat . For organotypic slice preparation , rats were sacrificed at P4-6 by decapitation under deep anesthesia with pentobarbital . Brains were dissected and transferred into liquid 3% low-melting agarose ( 38°C ) and placed on ice . Embedded brains were cut coronally ( 300 μm ) with a VT1000S vibratome ( Leica , Nussloch , Germany ) at 4°C . Slices were transferred onto sterilized culture plate inserts ( 0 . 4-μm pore size , Millicell-CM , Millipore , Billerica , MA ) and cultured in semidry conditions in a humidified incubator at 37°C under 5% CO2 atmosphere in wells containing Neurobasal medium ( Life Technologies ) supplemented with 1% B27 ( vol/vol ) , 1% N2 ( vol/vol ) , 1% GlutaMAX I ( vol/vol ) , and 1% penicillin/streptomycin ( vol/vol , Life Technologies ) . Slices were cultured for 24–48 hr before videomicroscopy . For illustration of electroporated cortical area , some organotypic slices cultured for 24 hr were fixed for 2 hr in 4% PFA in PBS . Pregnant Sprague–Dawley rats at gestation day 15 were prepared as for in utero electroporation . Then , 1 μl of a solution containing active NM II ATPase inhibitor ( S ) - ( − ) -blebbistatin ( 250 μM ) , inactive ( R ) - ( + ) stereoisomer ( 250 μM ) , AM251 ( 1 mM ) , or 2 . 8% DMSO ( vehicle for AM251 ) , mixed with 1 mg/ml Fast Green were microinjected through the uterus into the lateral ventricles of embryos by pulled glass capillaries . Embryos were allowed to develop in utero for 2 days . E17 brains were then dissected , fixed for 48 hr in 4% paraformaldehyde ( PFA ) in PBS at 4°C , cryoprotected in 30% sucrose in PBS , frozen in OCT compound , and sectioned coronally at 20 µm using a cryostat . For immunohistochemical staining of brain sections or fixed organotypic brain slices , sections were incubated with a combination of mouse anti-Tuj1 antibody , C-Ter rabbit antibody , and chicken anti-GFP antibody ( each diluted at 1:1000 ) overnight at room temperature in PBS ( 0 . 02 M ) containing 0 . 3% Triton and 0 . 02% sodium azide ( PBS-T-azide ) . For immunofluorescence detection of phosphoMLC , cultured neurons were fixed for 15 min in 4% PFA with 4% sucrose , permeabilized with PBS-T-azide and incubated for 90 min with the anti-phosphoMLC antibody ( 1:1000 ) diluted in 2% Bovin Serum Albumin and 3% Normal Goat Serum . Following washes , sections or coverslips were incubated with the appropriate secondary antibodies for 2 hr at room temperature and coverslipped with Mowiol mounting medium . For time-lapse microscopy , coverslips were placed in a Ludin chamber ( Life Imaging Services , Basel , Switzerland ) filled with imaging buffer ( 120 mM NaCl , 3 mM KCl , 2 mM CaCl2 , 2 mM MgCl2 , 10 mM glucose , 10 mM HEPES , and 2% B27 , pH 7 . 35 , 250 mOsm to match culture growth medium ) ( Lu et al . , 2007 ) . Wide-field images were taken on a motorized Nikon Eclipse Ti-E/B inverted microscope with the Perfect Focus System ( PFS ) in a 37°C chamber , using an oil immersion CFI Plan APO VC 60x , NA 1 . 4 objective ( Nikon , Melville , NY ) , equipped with a Polychrome V monochromator ( Till Photonics , Gräfelfing , Germany ) and an Intensilight light source ( Nikon ) , a CoolSnap HQ2 camera ( Photometrics , Tucson , AZ ) , and piloted by Metamorph 7 . 7 ( Molecular Devices , Sunnyvale , CA ) . All filter sets were purchased from Semrock ( Rochester , NY ) and the absence of cross-talk between different channels was checked with selectively labeled preparations . For the evaluation of neurite retraction in vitro , neurons or Neuro2A cells co-expressing CB1R-eCFP/EB3-eGFP/LifeAct-mCherry or Flag-CB1R-eGFP/LifeAct-mCherry or expressing LifeAct-mCherry alone were imaged every 2 min in each corresponding detection channel , and the mCherry detection channel was used for quantification . Treatments with inhibitors were applied on transfected cells 20 min before stimulation with the agonist . Blebbistatin treated cells were only illuminated through the mCherry excitation channel , in order to avoid phototoxic effects of lower illumination wavelengths . For the evaluation of axon retraction ex vivo , neurons co-expressing Flag-CB1-CFP/EB3-eGFP/LifeAct-mCherry or expressing LifeAct-mCherry were imaged every 3 min for 150–240 min , and inhibitor treatments were applied 30 min before agonist stimulation . For pharmacological treatments , ligands dissolved in dimethylsulfoxide were added directly to the culture medium . The highest final concentration reached was 0 . 2% DMSO; control experiments with up to 0 . 5% DMSO have shown the absence of effects on neuronal morphology and on the cellular distribution of CB1Rs . For the analysis of cortifugal axon development , images of labeled rat brain sections were taken on a Zeiss AxioImager M1 microscope using a 40× 0 . 75 numerical aperture ( NA ) objective . In each experiment , all acquisitions were performed using strictly identical exposure conditions . For the analysis of the images the SVZ was delimited and the corticofugal axons present in it were counted in blind . Between 5 and 9 embryos were employed per condition analyzing a mean of 9 brain slices per animal . For morphological analysis , widefield images were taken on a Zeiss Imager M1 microscope with dry 20× NA 0 . 75 and 40× NA 0 . 75 and oil-immersion 100× NA 1 . 3 objectives ( Zeiss , Oberkochen , Germany ) . In all cases , emission and excitation filters proper to each fluorophore were used sequentially and the absence of cross-talk between different channels was checked with selectively labeled preparations . Neurites were outlined and measured using an assisted semiautomatic method ( NeuronJ ) ( Meijering et al . , 2004 ) . For neurons at DIV4 , primary and secondary dendrites were outlined and their number and length were measured . Retraction of neurites was determined using Metamorph . For the CB1/Tuj1 and Tuj1/GFP co-localization experiments , and for illustration of the electroporated cortical area , images were taken on a Nikon A1 laser-scanning confocal microscope with dry 10× NA 0 . 30 and 20× NA 0 . 75 and oil-immersion 60× , NA 1 . 4 objectives . Novascan ( Ames , IA ) cantilevers with attached SiO2 spherical beads ( 1-µm diameter ) and nominal spring constant 0 . 06 N/m were used . Photodiode sensitivities of each cantilever were calibrated before and after measurements on the stiff surface region of culture dishes . The cantilever spring constant was determined using the thermal fluctuations method implemented in the Nanoscope 8 software ( Hutter , 1993 ) . Measurements were carried out 1 day after seeding Neuro2A cells at 37°C on a commercial AFM ( Catalyst , Bruker , Billerica , MA ) mounted on an inverted optical microscope ( Olympus , Tokyo , Japan ) . We obtained force–distance ( F–z ) curves of ∼3 µm peak-to-peak amplitude at 0 . 5 Hz , ∼3 µm/s . The maximum relative deflection ( d ) was controlled to reach an indentation depth of <400 nm . We placed the cantilever tip around the center of the cells with the help of optical images of the tip and samples acquired with a CCD camera ( Hamamatsu , Shizuoka Pref . , Japan ) . Three different cells were probed in a single AFM session by acquiring force curves at time intervals of <1 min . The same spherical tip was used in all measurements . Measurements were carried out before and after addition of vehicle or 100 nM of WIN at time point 0 . Blebbistatin was applied 20 min before time point 0 . Each experiment approaching F–z curve was fitted by the Hertz model of a sphere indenting an elastic half space ( Rico et al . , 2005 ) : F=43E1−ν2Rδ3/2 , where E being the Young's modulus , v the Poisson ratio ( 0 . 5 ) , R , the radius of the sphere , and , δ the indentation , which was calculated in terms of the point of contact ( zc ) and deflection offset ( d0 ) as δ = z − zc − ( d − d0 ) . To follow shape change of retracting Neuro2A cells co-expressing Flag-CB1R-eGFP and the structural marker DsRed2 , high-resolution images were acquired as a three-dimensional time series . For 61 time frames separated by 30 s , 51 z-slices of dimensions 149 . 64 μm × 111 . 8 μm ( 1392 pixels × 1040 pixels ) and separated by 0 . 5 μm in height were captured . The signal to noise ratio of the images was improved by deconvoluting the z-stacks at each time frame by iteratively computing the maximum-likelihood deconvolved image using the Richardson–Lucy algorithm ( Huygens Professional , Huygens , Inc . , Hilversum , Netherlands ) . The stopping criteria for the algorithm was determined using a conservative estimate of the image quality improvement at each iteration , and approximately 60 iterations of the algorithm were required in order to significantly improve the image quality without introducing artefacts into the deconvolved image . The surface of the deconvolved image stacks was computed at each time frame using a surface-rendering algorithm ( FreeSFP , Huygens , Inc . ) . Data were analyzed using Prism ( GraphPad Software , La Jolla , CA ) . Kolmogorov–Smirrow and Shapiro–Wilk tests were used to verify the normal distribution of the data . If the hypothesis of normality was confirmed , the significance of differences in mean was calculated using Student's t-test or one-way ANOVA followed by Newman–Keuls post-tests for p-value adjustment , elsewhere Kruskal–Wallis one-way ANOVA followed by Dunn's post-tests was used . For significance symbols , ‘ns’ means p ≥ 0 . 05 , one symbol means p ≤ 0 . 05 , two symbols mean p ≤ 0 . 01 , and three symbols mean p ≤ 0 . 001 . Outliers were removed when appropriate by applying Grubbs's test ( ESD method [extreme studentized deviate] ) available at the GraphPad QuickCalcs website: http://www . graphpad . com/quickcalcs/ConfInterval1 . cfm ( April 2014 ) or at NIST/SEMATECH e-Handbook of Statistical Methods , http://www . itl . nist . gov/div898/handbook/eda/section3/eda35h1 . htm , April 2014 . | Our brains are full of cells called neurons , which are connected to each other in complex networks that send messages around the brain . The way the neurons connect to each other , known as brain wiring , differs widely between individuals . Moreover , our brain wiring changes in response to our environment and experiences throughout our lives , from developing embryo to old age . One way this happens is through the action of chemicals called cannabinoids . Produced naturally in the body , cannabinoids are also found in the popular recreational drug cannabis that is increasingly being used in medicine to treat chronic pain and other conditions . However , cannabis misuse can have negative side effects on the brain leading to memory loss and mental illness , especially in young people . Cannabinoids can be detected by a group of proteins called cannabinoid receptors , but it is not clear how this leads to changes in brain wiring . Roland et al . now show that detection of cannabinoids by a type-1 cannabinoid receptor triggers a series of events that change how neurons grow and connect with each other . Detection of the cannabinoid by the receptor leads to the activation of an enzyme called ROCK . This , in turn , activates a motor protein called non-muscle myosin II that inhibits the growth of neurons . Roland et al . suggest that this prevents the neurons from reaching their neighbors and forming new connections . Investigating how this works in individuals with medical conditions that alter brain function could help inform us how cannabis could be used more safely . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2014 | Cannabinoid-induced actomyosin contractility shapes neuronal morphology and growth |
Calcium channel blockers ( CCBs ) are prescribed to patients with Marfan syndrome for prophylaxis against aortic aneurysm progression , despite limited evidence for their efficacy and safety in the disorder . Unexpectedly , Marfan mice treated with CCBs show accelerated aneurysm expansion , rupture , and premature lethality . This effect is both extracellular signal-regulated kinase ( ERK1/2 ) dependent and angiotensin-II type 1 receptor ( AT1R ) dependent . We have identified protein kinase C beta ( PKCβ ) as a critical mediator of this pathway and demonstrate that the PKCβ inhibitor enzastaurin , and the clinically available anti-hypertensive agent hydralazine , both normalize aortic growth in Marfan mice , in association with reduced PKCβ and ERK1/2 activation . Furthermore , patients with Marfan syndrome and other forms of inherited thoracic aortic aneurysm taking CCBs display increased risk of aortic dissection and need for aortic surgery , compared to patients on other antihypertensive agents .
Marfan syndrome is a systemic connective tissue disorder caused by mutations in FBN1 , the gene encoding extracellular matrix protein fibrillin-1 . A major cause of mortality in Marfan patients is aortic dissection and rupture . In Marfan mice , multiple phenotypic manifestations , including aortic aneurysm , developmental lung emphysema , mitral valve disease , and skeletal muscle myopathy , correlate with enhanced transforming growth factor beta ( TGFβ ) signaling , while treatment with either TGFβ neutralizing antibody ( NAb ) or the angiotensin-II type 1 receptor ( AT1R ) blocker ( ARB ) losartan can ameliorate these phenotypes , in association with evidence of reduced TGFβ signaling ( Neptune et al . , 2003; Ng et al . , 2004; Habashi et al . , 2006; Cohn et al . , 2007; Cook et al . , 2015 ) . Both canonical ( Smad2/3 ) and noncanonical ( ERK1/2 ) TGFβ-dependent signaling cascades have been shown to be activated in the aortas of Marfan mice , while selective inhibition of extracellular signal-regulated kinase ( ERK1/2 ) activation using RDEA119 ( refametinib ) rescues aortic growth and aortic wall architecture in Marfan mice ( Habashi et al . , 2011; Holm et al . , 2011 ) . Calcium channel blockers ( CCBs ) are a class of blood pressure lowering medications that block movement of calcium into cells from the extracellular space . There are several sub-classes of CCB based on chemical structure , including dihydropyridines ( e . g . , amlodipine ) and non-dihydropyridines , which include both phenylalkylamines ( e . g . , verapamil ) and benzothiazepines ( e . g . , diltiazem ) . CCBs are currently considered an alternative therapeutic strategy for Marfan patients intolerant of β-blockers , due to their ability to reduce contractility of the heart ( negative ionotropy ) and to lower blood pressure ( Milewicz et al . , 2005; von Kodolitsch and Robinson , 2007; Hartog et al . , 2012 ) . In doing so , they theoretically reduce stress on the aortic wall during cardiac systole . However , there is limited empirical evidence for their safety or utility in patients with Marfan syndrome and other forms of inherited thoracic aortic aneurysm ( Williams et al . , 2008 ) . To address this , we assessed the effect of two major classes of CCB in a mouse line heterozygous for a cysteine substitution in an epidermal growth factor-like domain in fibrillin-1 ( Fbn1C1039G/+ ) , representative of the most common class of mutations causing Marfan syndrome . This mouse model has been shown previously to closely recapitulate many of the phenotypic manifestations seen in patients with the disorder , including aortic aneurysm ( Habashi et al . , 2006; Holm et al . , 2011; Cook et al . , 2015 ) .
Wild-type ( WT ) and Marfan mice were treated with either placebo or amlodipine from 2 to 5 months of age . They were treated with a dose of amlodipine ( 12 mg/kg/day ) that resulted in a similar reduction in blood pressure as is obtained using losartan in our Marfan mouse model ( Figure 1—figure supplement 1S1 ) . They underwent unsedated in vivo echocardiography at baseline prior to treatment and then every month thereafter . Between 2 and 4 months of age , placebo-treated Marfan mice showed greater aortic root growth than WT littermates , which was unexpectedly exacerbated by amlodipine . This exacerbation was seen in both WT and Marfan mice but was of greater magnitude in the Marfan animals , indicating a specific interaction between the drug and the genotype of the mice ( genotype effect: p < 0 . 0001 , amlodipine effect: p < 0 . 001 , interaction effect: p < 0 . 01 , Figure 1A ) . Amlodipine-treated WT and Marfan mice also showed striking dilatation of the ascending aorta , an aortic segment just distal to the aortic root , which is less commonly affected in people and mice with Marfan syndrome . Amlodipine-induced growth in the ascending aorta was several fold greater than that seen in the aortic root , with Marfan mice again appearing particularly susceptible ( genotype effect: p < 0 . 0001 , amlodipine effect: p < 0 . 0001 , interaction effect: p < 0 . 0001 , Figure 1A ) . Furthermore , while no WT or placebo-treated Marfan mice died during the 3-month drug trial , more than 40% of amlodipine-treated Marfan mice died secondary to aortic rupture , as evidenced by hemothorax or hemopericardium ( p < 0 . 01 , Figure 1B ) . 10 . 7554/eLife . 08648 . 003Figure 1 . Effect of amlodipine in wild-type ( WT ) and Marfan mice . ( A ) Echocardiography data showing mean ( ±2 SEM ) growth in the aortic root and ascending aorta from 2 to 4 months of age . Number of mice per group ( male/female ) = WT placebo 11 ( 5/6 ) , WT amlodipine 10 ( 6/4 ) , Marfan placebo 14 ( 7/7 ) , Marfan amlodipine 11 ( 6/5 ) . Mean ( ±2 SEM ) weight per group ( in grams ) at 4 months = 27 . 4 ± 2 . 4 g ( WT placebo ) , 27 . 6 ± 2 . 8 g ( WT amlodipine ) , 28 . 1 ± 3 . 0 g ( Marfan placebo ) , 27 . 9 ± 2 . 8 g ( Marfan amlodipine ) . ( B ) Survival curve from 2 to 5 months of age . Number of mice per group ( male/female ) = WT placebo 11 ( 5/6 ) , WT amlodipine 10 ( 6/4 ) , Marfan placebo 14 ( 7/7 ) , Marfan amlodipine 19 ( 9/10 ) . ( C ) Representative VVG staining ( upper panel ) and Masson's trichrome staining ( lower panel ) of the proximal ascending aorta in 5-month-old male mice . Scale bar: 40 µm . ( D ) Mean ( ±2 SEM ) aortic wall architecture score of the proximal ascending aorta in 5-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Scale: 1 ( normal ) to 5 ( extensive damage ) . Plac , placebo; Aml , amlodipine . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 00310 . 7554/eLife . 08648 . 004Figure 1—figure supplement 1 . Effect of amlodipine in wild-type ( WT ) and Marfan mice . ( S1 ) Mean ( ±2 SEM ) systolic and diastolic blood pressure , and heart rate , in 3-month-old mice . Number of mice per group = 8 ( 4 male; 4 female ) . ( S2 ) Representative latex-injected images showing ascending aortic size ( distance between arrowheads ) in 5-month-old male mice . Scale bar: 2 mm . ( S3 ) Representative VVG staining ( upper panel ) and Masson's trichrome staining ( lower panel ) of the descending thoracic aorta in 5-month-old male mice . Scale bar: 40 µm . ( S4 ) Mean ( ±2 SEM ) aortic wall architecture score of the descending thoracic aorta in 5-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Scale: 1 ( normal ) to 5 ( extensive damage ) . ( S5 ) Mean ( ±2 SEM ) aortic root and ascending aortic growth from 2 to 4 months of age in mice treated with 3 mg/kg/day amlodipine . Number of mice per group ( male/female ) = WT placebo 5 ( 3/2 ) , WT amlodipine 5 ( 2/3 ) , Marfan placebo 10 ( 6/4 ) , Marfan amlodipine 8 ( 4/4 ) . Plac , placebo; Los , losartan; Aml , amlodipine . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 004 Following death or sacrifice , latex injection of the vasculature illustrated the enlargement of the ascending aorta in amlodipine-treated Marfan mice ( Figure 1—figure supplement 1S2 ) . Histologic analyses revealed greater ascending aortic wall thickening , elastic fiber fragmentation , reduced elastin content , and increased collagen deposition in Marfan animals compared to WT littermates , all of which were exacerbated by amlodipine treatment ( Figure 1C ) . Aortic architecture was graded quantitatively on a scale of 0–5 , by 4 observers blinded to both genotype and treatment arm , as described previously ( Holm et al . , 2011; Cook et al . , 2015 ) . While amlodipine had no effect on aortic architecture in WT animals ( p = 0 . 16 ) , it induced significant histological damage in Marfan mice ( interaction effect: p < 0 . 001 , Figure 1D ) . Interestingly , these deleterious histological effects of amlodipine were not observed in the descending thoracic aorta of Marfan mice ( Figure 1—figure supplement 1S3; p = 0 . 86 , Figure 1—figure supplement 1S4 ) , an aortic segment that lacks predisposition for dilatation or evidence of increased TGFβ signaling in Marfan mice ( Haskett et al . , 2012 ) . Prior studies assessing the effects of CCBs in other mouse models of aortic disease , such as the angiotensin-II infusion model , have typically used lower doses of amlodipine , in the range of 1 to 5 mg/kg/day ( Chen et al . , 2013; Takahashi et al . , 2013 ) . We utilized a dose of 12 mg/kg/day in order to obtain a similar reduction in blood pressure as is achieved using losartan in our Marfan mouse model , as we wished to eliminate blood pressure as a confounding variable when comparing the differential physiological and biochemical effects of the two agents . To ensure that the deleterious effect we observed was not simply due to toxicity of the drug , we repeated the trial using a dose of 3 mg/kg/day . Even at this lower dose , amlodipine still exacerbated both aortic root and ascending aortic growth in WT mice and Marfan littermates , with greater accentuation again occurring in the ascending aorta and in Marfan animals ( aortic root interaction effect: p < 0 . 01 , ascending aorta interaction effect: p < 0 . 01 , Figure 1—figure supplement 1S5 ) . To assess the generalizability of our observations with amlodipine , we treated WT and Marfan mice with verapamil from 2 to 6 months of age . Compared to placebo-treated WT and Marfan mice , verapamil-treated animals showed enhanced growth in both the aortic root and ascending aorta , with greater accentuation again occurring in Marfan animals , indicating a specific interaction between the drug and the genotype of the mice ( aortic root interaction effect: p < 0 . 01 , ascending aorta interaction effect: p < 0 . 01 , Figure 2A; Figure 2—figure supplement 1S1 ) . While verapamil had no effect on aortic architecture in WT mice , it significantly exacerbated the histological changes seen in the ascending aortic wall of Marfan mice , both qualitatively ( Figure 2B ) and quantitatively ( interaction effect: p < 0 . 0001 , Figure 2C ) . As with amlodipine , verapamil had no effect on aortic architecture in the descending thoracic aorta of Marfan mice ( Figure 2—figure supplement 1S2; p = 0 . 72 , Figure 2—figure supplement 1S3 ) . 10 . 7554/eLife . 08648 . 005Figure 2 . Effect of verapamil in wild-type ( WT ) and Marfan mice . ( A ) Mean ( ±2 SEM ) growth in the aortic root and ascending aorta from 2 to 6 months of age . Number of mice per group ( male/female ) = WT placebo 8 ( 4/4 ) , WT verapamil 10 ( 4/6 ) , Marfan placebo 9 ( 5/4 ) , Marfan verapamil 11 ( 6/5 ) . Mean ( ±2 SEM ) weight per group ( in grams ) at 6 months = 30 . 3 ± 2 . 6 g ( WT placebo ) , 30 . 6 ± 2 . 2 g ( WT verapamil ) , 31 . 3 ± 3 . 3 g ( Marfan placebo ) , 31 . 5 ± 3 . 2 g ( Marfan verapamil ) . ( B ) Representative VVG staining ( upper panel ) and Masson's trichrome staining ( lower panel ) of the proximal ascending aorta in 6-month-old male mice . Scale bar: 40 µm . ( C ) Mean ( ±2 SEM ) aortic wall architecture score of the proximal ascending aorta in 6-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Scale: 1 ( normal ) to 5 ( extensive damage ) . Plac , placebo; Ver , verapamil . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 00510 . 7554/eLife . 08648 . 006Figure 2—figure supplement 1 . Effect of verapamil in wild-type ( WT ) and Marfan mice . ( S1 ) Representative latex-injected images showing aortic size ( distance between arrowheads ) in 6-month-old male mice . Scale bar: 2 mm . ( S2 ) Representative VVG staining ( upper panel ) and Masson's trichrome staining ( lower panel ) of the descending thoracic aorta in 6-month-old male mice . Scale bar: 40 µm . ( S3 ) Mean ( ±2 SEM ) aortic wall architecture score of the descending thoracic aorta in 6-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Scale: 1 ( normal ) to 5 ( extensive damage ) . Plac , placebo; Ver , verapamil . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 006 To interrogate the mechanism underlying this detrimental CCB effect in Marfan mice , we performed Western blot analyses on the aortas of 5-month-old animals ( Figure 3A ) . Treatment with amlodipine accentuated signaling changes previously observed in Marfan mice , including enhanced activation of both canonical ( Smad ) and noncanonical ( ERK1/2 ) TGFβ-dependent signaling cascades , when normalized to either β-Actin or their respective total proteins ( Smad3 treatment effect: p < 0 . 0001 , ERK1/2 treatment effect: p < 0 . 01 , Figure 3A ) . Again , amlodipine had a greater effect in Marfan mice than WT littermates , indicating a specific interaction between the drug and the genotype of the animals ( Smad3 interaction effect: p < 0 . 001 , ERK1/2 interaction effect: p < 0 . 01 , Figure 3A ) . 10 . 7554/eLife . 08648 . 007Figure 3 . CCB effect is ERK1/2- and AT1R-dependent in wild-type ( WT ) and Marfan mice . ( A ) Western blot analysis of the aortic root and ascending aorta in 5-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . ( B ) Mean ( ±2 SEM ) ascending aortic growth from 2 to 4 months of age . Number of mice per group ( male/female ) = WT placebo 9 ( 5/4 ) , WT amlodipine 8 ( 4/4 ) , WT amlodipine + RDEA119 7 ( 3/4 ) , Marfan placebo 9 ( 4/5 ) , Marfan amlodipine 10 ( 6/4 ) , Marfan amlodipine + RDEA119 11 ( 6/5 ) . ( C ) Survival curve from 2 to 4 months of age . ( D ) Western blot analysis of the aortic root and ascending aorta in 4-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . ( E ) Mean ( ±2 SEM ) ascending aortic growth from 2 to 4 months of age . Number of mice per group ( male/female ) = WT placebo 11 ( 6/5 ) , WT amlodipine 11 ( 6/5 ) , WT amlodipine + losartan 8 ( 4/4 ) , Marfan placebo 7 ( 4/3 ) , Marfan amlodipine 6 ( 3/3 ) , Marfan amlodipine + losartan 9 ( 4/5 ) . ( F ) Western blot analysis of the aortic root and ascending aorta in 4-month-old mice . Number of mice per group = 3 ( 2 male , 1 female; or 1 male , 2 female ) . Plac , placebo; Aml , amlodipine; RDEA , RDEA119; Los , losartan; Geno , genotype; Treat , treatment; I/A , interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 00710 . 7554/eLife . 08648 . 008Figure 3—figure supplement 1 . CCB effect is ERK1/2- and AT1R-dependent in wild-type ( WT ) and Marfan mice . ( S1 ) Representative latex images , VVG and trichrome staining in 4-month-old male mice . Latex scale bar: 2 mm . VVG and trichrome scale bar: 40 μm . ( S2 ) Mean ( ±2 SEM ) aortic architecture score in 4-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Plac , placebo; Aml , amlodipine; RDEA , RDEA119 . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 008 Combined treatment with amlodipine and the selective inhibitor of ERK1/2 activation RDEA119 prevented amlodipine-induced ascending aortic enlargement ( treatment effect: p < 0 . 0001 ) , with the magnitude of effect being significantly greater in Marfan mice ( interaction effect: p < 0 . 001 , Figure 3B ) . Furthermore , RDEA119 rescued the premature lethality ( p < 0 . 05 , Figure 3C ) , and alterations in aortic wall architecture ( Figure 3—figure supplement 1S1; interaction effect: p < 0 . 0001 , Figure 3—figure supplement 1S2 ) , seen in amlodipine-treated Marfan mice , in association with abrogated ERK1/2 activation ( treatment effect: p < 0 . 0001 , Figure 3D ) . RDEA119 had no significant effect on Smad3 activation ( p = 0 . 26 , Figure 3D ) , suggesting that rescue of CCB-induced aortic aneurysm exacerbation in Marfan mice can occur independent of Smad3 activation status . In contrast , these data suggest that enhanced ERK1/2 activation is a critical mediator of CCB-mediated aortic aneurysm progression in Marfan mice . ERK1/2 activation in Marfan mice has previously been shown to also be AT1R dependent , since the AT1R blocker ( ARB ) losartan can rescue aortic root growth in Marfan animals in association with normalization of ERK1/2 activation ( Holm et al . , 2011 ) . Losartan also prevented amlodipine-induced ascending aortic aneurysm growth ( treatment effect: p < 0 . 01 ) , with the magnitude of effect being significantly greater in Marfan mice ( interaction effect: p < 0 . 001 , Figure 3E ) . This correlated with reduced ERK1/2 activation in these animals ( treatment effect: p < 0 . 0001 , Figure 3F ) . Hence the deleterious effect of CCBs in Marfan mice also appears to be AT1R-dependent . There is evidence that TGFβ-dependent gene expression can be mediated by protein kinase C ( PKC ) in both human fibroblasts and aortic vascular smooth muscle cells ( VSMCs ) ( Mulsow et al . , 2005; Ryer et al . , 2006 ) , and PKC blockade can inhibit TGFβ-dependent phenotypes and gene expression in vitro and in vivo ( Li and Jimenez , 2011; Lee et al . , 2013 ) . Furthermore , angiotensin-II has been shown to activate ERK1/2 via PKC ( Shah and Catt , 2002; Olivares Reyes et al . , 2005; Olson et al . , 2008 ) , with TGFβ potentially serving as an intermediary in this process ( Uchiyama-Tanaka et al . , 2001 ) . Western blot analysis of the aortic root and ascending aorta showed that PKCβ activation was significantly greater in Marfan mice than WT littermates ( genotype effect: p < 0 . 0001 ) , and both TGFβ NAb and losartan significantly reduced it ( treatment effect: p < 0 . 0001 for both , Figure 4A , B ) . This closely paralleled their effects on ERK1/2 activation ( Figures 3F , 4A ) . Furthermore , amlodipine treatment accentuated PKCβ activation in Marfan mice , in association with increased activation of phospholipase C ( PLCγ ) , its upstream activator , both of which were rescued by treatment with losartan ( treatment effect: p < 0 . 0001 for both , Figure 4B ) . These data suggest that PKCβ and its upstream activator PLCγ are both TGFβ- and AT1R-dependent in Marfan mice . 10 . 7554/eLife . 08648 . 009Figure 4 . PKC activation in placebo- and CCB-treated wild-type ( WT ) and Marfan mice . ( A ) Western blot analysis of the aortic root and proximal ascending aorta in 4-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . ( B ) Western blot analysis of the aortic root and proximal ascending aorta in 4-month-old mice . Number of mice per group = 3 ( 2 male , 1 female; or 1 male , 2 female ) . ( C ) Mean ( ±2 SEM ) ascending aortic growth from 2 to 4 months of age . Number of mice per group ( male/female ) = WT placebo 8 ( 4/4 ) , WT amlodipine 9 ( 4/5 ) , WT amlodipine + enzastaurin 8 ( 3/5 ) , Marfan placebo 12 ( 7/5 ) , Marfan amlodipine 8 ( 5/3 ) , Marfan amlodipine + enzastaurin 8 ( 5/3 ) . ( D ) Mean ( ±2 SEM ) aortic root growth from 2 to 4 months of age . Number of mice per group ( male/female ) = WT placebo 8 ( 4/4 ) , WT enzastaurin 6 ( 3/3 ) , Marfan placebo 12 ( 7/5 ) , Marfan enzastaurin 8 ( 4/4 ) . ( E ) Western blot analysis of the aortic root and ascending aorta in 4-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Plac , placebo; NAb , neutralizing antibody; Los , losartan; Aml , amlodipine; Enz , enzastaurin; Geno , genotype; Treat , treatment; I/A , interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 00910 . 7554/eLife . 08648 . 010Figure 4—figure supplement 1 . PKC activation in placebo- and CCB-treated wild-type ( WT ) and Marfan mice . ( S1 ) Representative latex-injected images of 4-month-old male mice . Scale bar: 2 mm . ( S2 ) Representative VVG staining ( upper panel ) and Masson's trichrome staining ( lower panel ) of the proximal ascending aorta in 4-month-old male mice . Scale bar: 40 µm . ( S3 ) Mean ( ±2 SEM ) aortic architecture score of the aortic root and proximal ascending aorta in 4-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Scale: 1 ( normal ) to 5 ( extensive damage ) . Plac , placebo; Aml , amlodipine; Enz , enzastaurin . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 010 Enzastaurin is a PKC inhibitor with relative selectivity for PKCβ and has been shown previously to inhibit PKC-mediated ERK1/2 activation ( Ruvolo et al . , 2011; Wu et al . , 2012 ) . It competes with ATP for the nucleotide triphosphate-binding site of PKC , thereby blocking its activation ( Graff et al . , 2005 ) . In keeping with the hypothesis that PKCβ mediates CCB-induced aortic aneurysm exacerbation in Marfan mice , treatment with enzastaurin led to a significant reduction in ascending aortic growth in amlodipine-treated animals ( treatment effect: p < 0 . 0001 ) , with the magnitude of effect being significantly greater in Marfan mice ( interaction effect: p < 0 . 0001 , Figure 4C ) . Latex-injected images show the rescue achieved by enzastaurin on ascending aortic aneurysm in amlodipine-treated Marfan mice ( Figure 4—figure supplement 1S1 ) . Enzastaurin also significantly rescued the deleterious histological changes imposed on the Marfan aorta by amlodipine , both qualitatively ( Figure 4—figure supplement 1S2 ) and quantitatively ( treatment effect: p < 0 . 0001 , Figure 4—figure supplement 1S3 ) . Cumulatively , these data suggest that PKCβ mediates amlodipine-induced aortic aneurysm exacerbation in Marfan mice . To confirm that PKCβ also mediates aortic aneurysm progression in placebo-treated Marfan mice , aortic root growth was measured in WT and Marfan animals over a 2-month treatment period ( Figure 4D ) . This showed that enzastaurin was indeed able to significantly reduce aortic root growth in both WT and Marfan mice ( treatment effect: p < 0 . 0001 ) . While the magnitude of effect was greater in Marfan mice than WT littermates , this difference trended towards significance but did not quite reach it ( interaction effect: p = 0 . 09 ) . Western blot analysis of the aortic root and ascending aorta confirmed that the protection conferred on aortic growth by enzastaurin correlated with a significant reduction in PKCβ activation in both placebo- and amlodipine-treated Marfan mice ( treatment effect: p < 0 . 0001 , Figure 4E ) . Furthermore , enzastaurin was able to significantly rescue ERK1/2 activation in these animals ( treatment effect: p < 0 . 0001 , Figure 4E ) , inferring that PKCβ may mediate ERK1/2 activation in this setting . We became interested in the clinically available antihypertensive agent hydralazine , not only because it reduces blood pressure ( a desirable effect in Marfan syndrome ) but also because it has been shown to inhibit PKC-mediated ERK1/2 activation in vivo ( Deng et al . , 2003; Gorelik et al . , 2007 ) . Therefore , we performed a trial of WT and Marfan mice treated with hydralazine from 2 to 6 months of age , at a dose of 16 mg/kg/day ( Doblinger et al . , 2012; Shinmura et al . , 2015 ) . This dose of hydralazine reduced systolic and diastolic blood pressure by roughly 10–15% in our mice ( Figure 5—figure supplement 1S1 ) . Compared to WT mice , placebo-treated Marfan animals showed greater aortic root growth , which was fully rescued by hydralazine ( genotype effect: p < 0 . 001 , treatment effect: p < 0 . 0001 , Figure 5A ) . While the magnitude of rescue was greater in Marfan mice than WT littermates , this difference trended towards significance but did not quite reach it ( interaction effect: p = 0 . 07 ) . 10 . 7554/eLife . 08648 . 011Figure 5 . Effect of hydralazine in wild-type ( WT ) and Marfan mice . ( A ) Mean ( ±2 SEM ) aortic root growth from 2 to 6 months of age . Number of mice per group ( male/female ) = WT placebo 9 ( 4/5 ) , WT hydralazine 12 ( 7/5 ) , Marfan placebo 15 ( 6/9 ) , Marfan hydralazine 12 ( 6/6 ) . Mean ( ±2 SEM ) weight per group ( in grams ) at 6 months = 31 . 4 ± 2 . 4 g ( WT placebo ) , 31 . 2 ± 3 . 1 g ( WT hydralazine ) , 31 . 0 ± 2 . 7 g ( Marfan placebo ) , 30 . 4 ± 3 . 5 g ( Marfan hydralazine ) . ( B ) Western blot analysis of the aortic root in 6-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . ( C ) Diagram illustrating key nodal points in Marfan mouse aortic disease pathogenesis . Drugs shown in red ameliorate aneurysm progression , while manipulations shown in blue exacerbate it . Plac , placebo; Hyd , hydralazine; Geno , genotype; Treat , treatment; I/A , interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 01110 . 7554/eLife . 08648 . 012Figure 5—figure supplement 1 . Effect of hydralazine in wild-type ( WT ) and Marfan mice . ( S1 ) Mean ( ±2 SEM ) systolic and diastolic blood pressure , and heart rate , in 3-month-old mice . Number of mice per group = 8 ( 4 male; 4 female ) . ( S2 ) Representative parasternal long-axis in vivo echocardiography images of 6-month-old male mice . Scale bar: 1 mm . ( S3 ) Representative VVG staining ( upper panel ) and Masson's trichrome staining ( lower panel ) of the aortic root in 6-month-old mice . Scale bar: 40 µm . ( S4 ) Mean ( ±2 SEM ) aortic wall architecture score of the aortic root in 6-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Scale: 1 ( normal ) to 5 ( extensive damage ) . ( S5 ) Western blot analysis of the aortic root in 4-month-old mice . Number of mice per group = 4 ( 2 male , 2 female ) . Plac , placebo; Hyd , hydralazine; Aml , amlodipine; RDEA , RDEA119; Geno , genotype; Treat , treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 012 Representative parasternal long-axis echocardiography images show the rescue achieved by hydralazine on aortic root aneurysm in Marfan mice ( Figure 5—figure supplement 1S2 ) . Hydralazine treatment also led to a significant rescue of aortic wall architecture , with the magnitude of effect being significantly greater in Marfan mice ( Figure 5—figure supplement 1S3; interaction effect: p < 0 . 001 , Figure 5—figure supplement 1S4 ) . This correlated with a reduction of both PKCβ and ERK1/2 activation in the aortas of these animals ( treatment effect: p < 0 . 0001 for both , Figure 5B ) , with the magnitude of effect again being greater in Marfan mice ( interaction effect: p < 0 . 001 and p < 0 . 0001 , respectively ) . While PKCβ activation was greater in the aortas of Marfan mice compared to WT littermates ( genotype effect: p < 0 . 0001 , Figure 5—figure supplement 1S5 ) , and this was exacerbated by amlodipine ( treatment effect: p < 0 . 0001 ) , RDEA119 had no significant effect on PKCβ activation ( post-hoc analysis: p = 0 . 44 ) . This suggests that PKCβ lies upstream of ERK1/2 in the pathogenic sequence of events driving aortic aneurysm progression in Marfan mice and suggests that both enzastaurin and hydralazine achieve their beneficial effect in Marfan animals via inhibition of PKCβ-mediated ERK1/2 activation ( Figure 5C ) . Finally , we wished to determine whether these observations made in a mouse model of Marfan syndrome hold relevance for patients . In collaboration with other members of the Genetically Triggered Thoracic Aortic Aneurysms and Cardiovascular Conditions ( GenTAC ) consortium ( www . gentac . rti . org ) , we conducted a case-control study to assess the effect of CCBs on aortic dissection or aortic surgery in patients with Marfan syndrome and other forms of inherited thoracic aortic aneurysm ( iTAA; Table 1 ) . The other forms of iTAA primarily included Loeys-Dietz , Turner and Ehlers-Danlos syndromes , familial thoracic aortic aneurysm , and bicuspid aortic valve ( BAV ) with aneurysm . 10 . 7554/eLife . 08648 . 013Table 1 . GenTAC human dataDOI: http://dx . doi . org/10 . 7554/eLife . 08648 . 013Aortic dissectionAortic surgeryMarfanOtherMarfanOthern = 531n = 1819n = 531n = 1819Odds in CCB5 . 1%0 . 57%28 . 1%10 . 70%Odds in non-CCB0 . 41%0 . 12%5 . 1%4 . 4%Odds ratio12 . 54 . 75 . 52 . 4 p-value0 . 032NS<0 . 001<0 . 01Odds ratio ( BP ) *12 . 75 . 65 . 42 . 2 p-value0 . 06NS<0 . 0010 . 016Odds ratio ( Aortic Size ) *11 . 24 . 15 . 02 . 2 p-value0 . 08NS<0 . 010 . 017Odds ratio ( β-blocker ) *15 . 93 . 75 . 72 . 0 p-value0 . 045NS<0 . 010 . 026Odds of ‘aortic dissection’ and ‘aortic surgery’ in patients with Marfan syndrome ( ‘Marfan’ ) and other forms of inherited thoracic aortic aneurysm ( ‘Other’ ) . Odds ( written as % ) = number of people who incurred an event ( i . e . , dissection or surgery ) divided by the number who did not . Odds of aortic dissection or surgery were calculated separately for patients who had used CCBs ( ‘Odds in CCB’ ) compared to those who had not ( i . e . , ‘Odds in non-CCB’ ) . Odds Ratio = Odds of aortic dissection or surgery in patients who had taken CCBs divided by the odds in patients who had not taken CCBs . *The Odds Ratio was then adjusted for blood pressure ( ‘BP’ ) , aortic size ( ‘Aortic Size’ ) , and β-blocker use ( ‘β-blocker’ ) at enrollment , with corresponding p-values . Marfan patients with native aortic roots at the time of enrollment ( n = 531 ) who received CCBs ( as compared to other antihypertensive agents ) had an increased odds of aortic dissection ( odds ratio ( OR ) 12 . 5 , p = 0 . 032 ) . Strong trends were maintained after correction for either systolic blood pressure ( OR 12 . 7 , p = 0 . 06 ) or aortic root size ( OR 11 . 2 , p = 0 . 08 ) at enrollment . A more profound detrimental effect of CCBs on aortic dissection could be masked by the practice of performing prophylactic aortic surgery when the aorta reaches a dimension that confers risk for dissection . This is generally considered correct clinical management , since the rate of morbidity and mortality with aortic dissection is higher when surgery is performed on an emergent rather than an elective basis . In such a scenario , one might see an increased rate in the need for aortic surgery in the absence of an increased rate of aortic dissection . In keeping with this hypothesis , we found that CCB-treated Marfan patients had an increased odds of needing aortic surgery ( OR 5 . 5 , p < 0 . 001 ) when compared to patients on other antihypertensive agents , which remained significant when corrected for either blood pressure ( OR 5 . 4 , p < 0 . 001 ) or aortic size ( OR 5 . 0 , p < 0 . 01 ) at enrollment . For patients with other forms of iTAA and native aortic roots at the time of enrollment ( n = 1819 ) , there was suggestion of an increased odds for aortic dissection in those who had taken CCBs , although this did not reach statistical significance ( OR 4 . 7 , p = 0 . 26 ) . This was again most likely secondary to prophylactic surgical intervention , given that CCB-treated patients did have an increased odds of needing aortic surgery ( OR 2 . 4 , p = 0 . 004 ) , which remained significant when corrected for either blood pressure ( OR 2 . 2 , p = 0 . 016 ) or aortic size ( OR 2 . 2 , p = 0 . 017 ) at enrollment . Given that β-blocker therapy is a first-line therapy for patients with Marfan syndrome and other related iTAA syndromes , and CCBs are typically given to patients who cannot tolerate β-blockers , one could hypothesize that the observations herein may relate more to an absence of protective β-blockade rather than use of CCBs . This interpretation is directly at odds with our mouse model data , where disease acceleration occurred due to the presence of a deleterious factor ( i . e . , CCB therapy ) , not simply due to the absence of a protective agent ( i . e . , β-blockade ) . If the latter were the case , then one would expect CCB-treated Marfan mice to have shown the same growth rate as placebo-treated Marfan animals , which was not the case . To further assess this , we controlled for β-blocker use in our GenTAC analysis ( Table 1 ) and found that it did not fundamentally alter the conclusions of the study . The odds of aortic dissection and aortic surgery remained significantly increased in Marfan patients on CCBs ( OR 15 . 9 , p = 0 . 045; OR 5 . 7 , p < 0 . 01; respectively ) , while the odds of aortic surgery remained significantly increased in other iTAA syndrome patients on CCBs ( OR 2 . 0 , p = 0 . 026 ) .
Any class of drug that lowers blood pressure and/or reduces cardiac contractility may theoretically be beneficial in Marfan and related inherited TAA syndromes , since a reduction in hemodynamic stress placed upon an inherently weakened aortic wall should hypothetically limit aortic expansion and delay or prevent aortic dissection . Prior studies in mouse models of Marfan syndrome and related disorders have shown that β-blockers ( e . g . , propranolol ) ( Habashi et al . , 2006; Gallo et al . , 2014 ) or angiotensin converting enzyme inhibitors ( ACEi , e . g . , enalapril ) ( Habashi et al . , 2011 ) achieve a relatively modest inhibition of aortic aneurysm growth and fail to preserve aortic wall architecture when compared to the ARB losartan , despite an equivalent reduction in blood pressure . Notably , in Marfan mice the extent of relative protection of these agents correlates closely with suppression of TGFβ-dependent canonical ( Smad2/3 ) or noncanonical ( ERK1/2 ) signaling ( Holm et al . , 2011; Cook et al . , 2015 ) . While these different classes of antihypertensive agent show varying degrees of protective effect in Marfan mice , they all confer at least some degree of protection against aneurysm progression . To our knowledge , this is the first time that a class of clinically available blood pressure lowering agents has been shown to exacerbate aortic aneurysm and cause dissection in a mouse model of Marfan syndrome . This is despite the fact that amlodipine lowered blood pressure equally as much as losartan , an effect which should provide some degree of protection against aneurysm progression . This deleterious effect on aneurysm growth was not limited to the dihydropyridine amlodipine , but also extended to another class of CCB , namely the phenylalkylamine verapamil . Given that a significant number of patients with Marfan syndrome and related conditions are prescribed CCBs , a previously unidentified deleterious role for these drugs has important clinical ramifications . It is interesting to note that amlodipine and verapamil showed the same trend , namely a small deleterious effect in WT mice and a much greater accentuation of growth in Marfan mice , with the effect being greater in the ascending aorta than the aortic root . In terms of absolute growth , amlodipine had a significantly greater effect than verapamil , even at its lower dose . While the two drugs both target L-type calcium channels on the cell membrane , verapamil is generally considered to have greater selectivity for cardiac tissue , while amlodipine is considered to have stronger tropism for aortic VSMCs . It may be for this reason that amlodipine has a relatively greater detrimental effect on the aorta compared to verapamil . We have further elucidated that PKCβ-mediated ERK1/2 activation contributes to the deleterious effect of CCBs in Marfan mice , while inhibition of this TGFβ- and AT1R-dependent pathway using either a PKC inhibitor ( enzastaurin ) or the clinically available antihypertensive agent hydralazine is able to prevent aortic aneurysm progression in Marfan mice , in association with blunted PKCβ and ERK1/2 activation . While ARBs such as losartan and ACE inhibitors such as enalapril are teratogenic and hence not appropriate for use in pregnancy , hydralazine is well tolerated , making it an appealing alternative therapeutic strategy for Marfan syndrome , particularly in pregnant women . It is also notable that while CCBs accelerate growth of the aortic root in Marfan mice , they have an even more pronounced effect on the more distal ascending aorta , an aortic segment that is less frequently affected in Marfan syndrome . The molecular basis of regional predisposition for aortic aneurysm is not yet fully understood but may relate to the distinct developmental origins of the two aortic segments . While vascular smooth muscle cells in the aortic root primarily derive from the second heart field ( SHF ) , those in the more distal ascending aorta derive from the cardiac neural crest ( CNC ) . Prior work has shown that L-type calcium channel function is critical for correct CNC migration , differentiation , and morphogenic patterning , as well as maintenance of an appropriate post-developmental differentiated state ( Moran , 1991 ) . Furthermore , mutations in L-type calcium channels result in cellular hypertrophy and hyperplasia in neural-crest-derived tissues ( Ramachandran et al . , 2013 ) . This may explain why CCBs appear to have a greater deleterious effect on the ascending aorta compared to the aortic root . Interestingly , the ascending aorta is characteristically involved in patients with BAV with aneurysm . Despite being the most common developmental cardiovascular abnormality in humans , the genetic cause ( s ) and molecular mechanisms underlying BAV and the associated aortopathy are not well understood . This work provides rationale and incentive to elucidate whether altered calcium , PKC and/or ERK1/2 signaling may play a role in this condition . There is only one prior study assessing the effect of CCBs on aortic growth in Marfan patients ( Rossi-Foulkes et al . , 1999 ) . The number of CCB-treated patients was small ( n = 6 ) , and the study combined data on patients treated with either CCBs or β-blockers . Given that β-blockers are known to confer protection against aortic growth in Marfan patients ( Salim et al . , 1994; Shores et al . , 1994; Silverman et al . , 1995 ) , and nearly 80% of patients in the study were on β-blockers but not CCBs , a detrimental effect of CCBs could easily have been masked . Interestingly in the study , the complication rate was more than twice as high in CCB-treated patients as compared to those on β-blockers ( 33% vs 15% ) , although numbers were too small to draw firm conclusions . While randomized double-blinded prospective clinical trials are the preferred approach to analyze the therapeutic efficacy of a drug , this is not really feasible for agents that are found to be severely detrimental in pre-clinical models , since ethical approval for a prospective human trial is unlikely to be granted . Retrospective analyses of large data sets are a more realistic solution in such scenarios . An inherent limitation to any analysis of rare Mendelian disorders is the number of patients available . This is even more challenging when stratifying patients based on drugs that have been prescribed . The GenTAC consortium was established to try to overcome some of these challenges . However , it needs to be recognized that the human data contained herein is still limited by relatively small sample size ( more so for aortic dissection than aortic surgery ) and an inability to control for the dose or number of medications that patients were receiving . We were able to control for aortic size at the time of enrollment , which is an indicator of baseline aortic disease severity and a predictor of future risk of aortic dissection and/or need for aortic surgery . We were also able to control for blood pressure , which is a known risk factor for aneurysm progression . Hence while the conclusions that can be drawn are not definitive , the data suggest that CCBs should be used with caution in Marfan patients . They also suggest that CCBs may be deleterious in other Marfan-related conditions , although a larger sample size will be needed to assess risk after stratification by individual disorders .
All mice were cared for under strict compliance with the Animal Care and Use Committee of the Johns Hopkins University School of Medicine . The Fbn1C1039G/+ line was maintained on a pure C57BL/6J background ( backcrossed for >12 generations ) , allowing for valid comparisons . Mice were sacrificed with an inhalation overdose of halothane ( Sigma–Aldrich , St . Louis , MI , United States ) . Mice underwent immediate laparotomy , descending abdominal aortic transection , and phosphate-buffered saline ( PBS ) ( pH 7 . 4 ) was infused through the right and left ventricles to flush out the blood . Mouse aortic root and ascending aortas ( aortic root to origin of right brachiocephalic trunk ) were harvested , snap-frozen in liquid nitrogen and stored at minus 80° centigrade until processed . Protein was extracted using the reagents and protocol from a Total Protein Extraction Kit containing protease inhibitor and Protein Phosphatase Inhibitor Cocktail ( Millipore , MA , United States ) . Aortas were homogenized using a pellet pestle motor ( Kimble-Kontes , NJ , United States ) as per the extraction kit protocol . Samples were then stored once more at minus 80° centigrade until Western blot analysis was performed . Mice that were analyzed for aortic histology had latex infused into the left ventricle at a pressure between 70 and 80 mmHg , as confirmed using a handheld digital manometer ( Fisher Scientific , Pittsburgh , PA , United States ) . Mice were then fixed for 24 hr in 10% buffered formalin , after which time the heart and aorta were removed and stored in 70% ethanol until histological analysis was performed . Mice were started on medication at 8 weeks of age . Losartan was dissolved in drinking water and filtered to reach a concentration of 0 . 6 g/l , giving an estimated daily dose of 60 mg/kg/day ( based on a 30 g mouse drinking 3 mls per day ) . Amlodipine was dissolved in drinking water and filtered to reach a final concentration of 0 . 12 g/l , giving an estimated daily dose of 12 mg/kg/day . Verapamil was dissolved in drinking water and filtered to reach a concentration of 1 . 44 g/l , giving an estimated daily dose of 144 mg/kg/day . Hydralazine was dissolved in drinking water and filtered to reach a concentration of 0 . 16 g/l , giving an estimated daily dose of 16 mg/kg/day . Placebo-treated animals received drinking water . RDEA119 ( 25 mg/kg ) and enzastaurin ( 15 mg/kg ) were reconstituted in 10% 2-hydroxypropyl-beta-cyclodextrin ( Sigma–Aldrich ) dissolved in PBS and administered twice daily by oral gavage . Treatment for both was initiated at 8 weeks of age and continued for 8 weeks . 10% 2-hydroxypropyl-beta-cyclodextrin dissolved in PBS was used as the placebo control . Mouse monoclonal TGFβ NAb ( 1d11; R&D Systems , Minneapolis , MN , United States ) was reconstituted in PBS and administered via intraperitoneal injection three times a week at a dose of 5 mg/kg . Treatment was initiated at 1 month of age and continued for 2 months . IgG ( Zymed Laboratories Inc , San Francisco , CA , United States ) was reconstituted in PBS , and administered at a dose of 10 mg/kg as a control . Nair hair removal cream was used on all mice the day prior to echocardiograms . All echocardiograms were performed on awake , unsedated mice using the Visualsonics Vevo 2100 and a 30 MHz transducer . Mice were imaged at baseline and every 2 months after treatment until the time of sacrifice . The aorta was imaged using a parasternal long axis view . Three separate measurements of the maximal internal dimension at the sinus of Valsalva and proximal ascending aorta were made from distinct captured images and averaged . All imaging and measurements were performed blinded to both genotype and treatment arm . Blood pressures and heart rates were measured by tail cuff plethysmography the week prior to completion of a study . Mice were habituated to the system for 5 days and then on the final day 3–5 measurements were obtained and averaged . 8 mice for each treatment group were analyzed . Latex-infused ascending aortas were transected just above the level of the aortic valve , and 3-mm transverse sections were mounted in 4% bacto-agar prior to paraffin fixation . Five micrometer aortic sections underwent Verhoeff-van Giesen ( VVG ) and Masson's Trichrome staining and were imaged at 40× magnification , using a Nikon Eclipse E400 microscope . Wall architecture of 4 representative sections for each mouse was assessed by 4 blinded observers and graded on an scale of 1 ( indicating normal histology ) to 5 ( indicating diffuse elastic fiber fragmentation and histological damage ) , and the results were averaged . Aortic tissue homogenates were dissolved in sample buffer , run on a NuPAGE Novex 4–12% Bis-Tris Gel ( Invitrogen , CA , United States ) , and transferred to nitrocellulose membranes using the iBlot transfer system ( Invitrogen ) . Membranes were washed in PBS and blocked for 1 hr at room temperature with 5% instant non-fat dry milk dissolved in PBS containing 1% Tween-20 ( Sigma , MO , United States ) ( PBS-T ) . Equal protein loading of samples was determined by a protein assay ( Bio-Rad , CA , United States ) and confirmed by probing with antibodies against β-Actin ( Sigma ) . Membranes were probed overnight at 4° centigrade with primary antibodies against pSmad3 ( 1880-1; Millipore ) , Smad3 ( #9513; Cell Signaling ) , pERK1/2 ( #4370; Cell Signaling ) , ERK1/2 ( #4695; Cell Signaling ) , pPKCβ ( #75837; Abcam , United Kingdom ) , PKCβ ( #32026; Abcam ) , pPLCγ ( #2821; Cell Signaling ) , and PLCγ ( #2822; Cell Signaling ) , dissolved in PBS-T containing 5% milk . Blots were then washed in PBS-T and probed with HRP-conjugated anti-rabbit secondary antibody ( GE Healthcare , United Kingdom ) dissolved in PBS-T containing 5% milk at room temperature . Blots were then washed in PBS-T , developed using SuperSignalWest HRP substrate ( Pierce Scientific , IL , United States ) , exposed to BioMax Scientific Imaging Film ( Sigma ) , and quantified using ImageJ analysis software ( NIH , MD , United States ) . This was performed using clinical data from the GenTAC registry . In short , patients were recruited from 5 regional clinical centers that cover a wide geographic catchment area within the United States , including Johns Hopkins University , Oregon Health & Science University , University of Pennsylvania , University of Texas Health Science Center at Houston/Baylor College of Medicine , and Weill Cornell Medical College . Each site obtained approval to conduct the study from their respective institutional review boards , and informed consent was obtained at each site . Longitudinal observational data were collected on adults and children diagnosed with one of 12 thoracic aortic aneurysm-related conditions , primarily including Marfan , Loeys-Dietz , Ehlers-Danlos and Turner syndromes , BAV , and familial thoracic aortic aneurysm . Designated registry investigators at each enrolling site confirmed diagnostic classifications of genetically associated aortic conditions . Demographic indices were abstracted from medical records at the time of registry enrollment by site-specific research coordinators , and all samples were de-identified to preserve patient confidentiality . The current analysis was limited to aortic dissection or aortic surgery that occurred in the aortic root , ascending aorta and/or aortic arch , and excluded patients who had had aortic dissection or surgery in these regions prior to enrollment into the study . CCB use was considered positive if patients had taken them prior to , or were taking them at the time of , enrollment into the study . Aortic dissection or surgery was assessed prospectively in the follow-up period after enrollment . Mean follow-up was 50 . 8 ± 1 . 6 months for the Marfan patients and 43 . 4 ± 0 . 8 months for all TAA syndrome patients . Statistical analyses were performed using SAS Version 9 . 3 ( SAS Institute , Cary , NC , United States ) . Exact ORs and two-sided p-values were calculated using PROC LOGISTIC . Models were adjusted for potential cofounders identified a priori ( i . e . , systolic blood pressure measurement and aortic root size measurement at the time of enrollment ) . Interaction terms were determined not to be statistically significant ( p > 0 . 05 ) , so only main effects were included in the final models . All quantitative data are shown as bar graphs produced using Excel ( Microsoft , Redmond , WA , United States ) . Mean ±2 standard errors of the mean ( SEM ) are displayed . Statistical analysis was performed using two-way ANOVA for all continuous data with three or more groups and two potentially interacting terms ( e . g . , echocardiography , Western blot ) ; Kruskal–Wallis ANOVA was used to analyze categorical data with three or more groups and two potentially interacting terms ( e . g . , aortic architecture score ) ; one-way ANOVA was used to analyze continuous data with three or more groups but no interaction ( e . g . , blood pressure ) ; two-tailed t tests were used to analyze data comparing two groups , or to make selective planned comparisons between individual groups within a larger study . Significance values for the effects of genotype , treatment , and/or any interaction between two variables have been included in each figure , where appropriate . If only placebo treatment for WT mice was included in an analysis , no interaction between drug treatment and genotype could be assessed , so it is not included in the figure . A p value < 0 . 05 was considered statistically significant in all analyses . | Marfan syndrome is a disorder that affects the body's connective tissues , which maintain the structure of the body and support organs and other tissues . People with Marfan syndrome have connective tissues that can stretch more than those of other people , which put them at increased risk of a life-threatening tear in their aorta ( the main artery in the body ) , muscle weakness and other problems . A cell communication pathway called TGFβ signaling is involved in cell growth and many other important processes . TGFβ signaling is more active in patients with Marfan syndrome due to mutations in a gene called FBN1 . Drugs that block TGFβ signaling—which are also used to treat high blood pressure—can reduce the symptoms of the disorder . Unfortunately , not all people with Marfan disease can tolerate these drugs and other medications called calcium channel blockers , which also lower blood pressure , are often used as an alternative . It is thought that calcium channel blockers help reduce stress on blood vessels , but there is little data to show whether these drugs are safe and helpful for patients with Marfan syndrome . Now , Doyle , Doyle et al . studied the effect of two different calcium channel blockers on mice that have a mutation in Fbn1—the mouse equivalent of FBN1—that is similar to those found in humans with Marfan syndrome . The experiments show that the aortas of these mice grew more quickly and were more likely to tear when compared to mice that did not receive these drugs . Many of these aortic tears were fatal . The calcium channel blockers increased the activity of two signaling molecules that are regulated by TGFβ signaling . Treating the Marfan mice with other drugs that lower the activity of these signaling molecules protected the aorta , even if they were also treated with the calcium channel blockers . Doyle , Doyle et al . examined a registry of human patients . This revealed preliminary evidence that aortic tears and aortic repair surgery were more common in patients with Marfan syndrome who had received calcium channel blockers than patients who had been treated with other drugs . Together , these findings suggest that it may be dangerous to treat patients with Marfan syndrome with calcium channel blockers . Additional work will be needed to confirm this risk , to find out if it extends to other similar conditions , and to explore the therapeutic potential of drugs that target the two enzymes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"medicine"
] | 2015 | A deleterious gene-by-environment interaction imposed by calcium channel blockers in Marfan syndrome |
Bathymodiolus mussels live in symbiosis with intracellular sulfur-oxidizing ( SOX ) bacteria that provide them with nutrition . We sequenced the SOX symbiont genomes from two Bathymodiolus species . Comparison of these symbiont genomes with those of their closest relatives revealed that the symbionts have undergone genome rearrangements , and up to 35% of their genes may have been acquired by horizontal gene transfer . Many of the genes specific to the symbionts were homologs of virulence genes . We discovered an abundant and diverse array of genes similar to insecticidal toxins of nematode and aphid symbionts , and toxins of pathogens such as Yersinia and Vibrio . Transcriptomics and proteomics revealed that the SOX symbionts express the toxin-related genes ( TRGs ) in their hosts . We hypothesize that the symbionts use these TRGs in beneficial interactions with their host , including protection against parasites . This would explain why a mutualistic symbiont would contain such a remarkable ‘arsenal’ of TRGs .
Mussels of the genus Bathymodiolus dominate deep-sea hydrothermal vents and cold seeps worldwide . The key to their ecological and evolutionary success is their symbiosis with chemosynthetic bacteria that provide them with nutrition ( von Cosel , 2002; Van Dover et al . , 2002 ) . Bathymodiolus mussels host their symbionts inside specialized gill epithelial cells called bacteriocytes ( Cavanaugh et al . , 2006; Petersen and Dubilier , 2009 ) . Their filtering activity exposes Bathymodiolus mussels to a plethora of diverse microbes in their environment . Despite this , they are colonized by only one or a few specific types of chemosynthetic symbionts . Some mussel species associate exclusively with sulfur-oxidizing ( SOX ) symbionts that use reduced sulfur compounds and sometimes hydrogen as an energy source , and carbon dioxide as a carbon source . Some have only methane-oxidizing ( MOX ) symbionts that use methane as an energy source and carbon source . Some mussel species host both types in a dual symbiosis ( Fisher et al . , 1993; Distel et al . , 1995; Duperron et al . , 2006; Dubilier et al . , 2008; Petersen et al . , 2011 ) . In all species except one , a single 16S rRNA phylotype for each type of symbiont ( SOX or MOX ) is found in the gills ( Dubilier et al . , 2008 ) . There are more than 30 described Bathymodiolus species , and most associate with a characteristic symbiont phylotype , which is not found in other species ( Duperron et al . , 2013 ) . Although these associations are clearly very specific , the molecular mechanisms that underpin this specificity are still unknown . No chemosynthetic symbiont has ever been obtained in pure culture . Therefore , molecular methods for investigating uncultured microbes have been essential for understanding their biodiversity , function , and evolution ( reviewed by Dubilier et al . , 2008 ) . The Bathymodiolus symbionts are assumed to be horizontally transmitted , which means that each new host generation must take up their symbionts from the surrounding environment or co-occurring adults ( Won et al . , 2003b; Kadar et al . , 2005; DeChaine et al . , 2006; Fontanez and Cavanaugh , 2014; Wentrup et al . , 2014 ) . To initiate the symbiosis , hosts and symbionts must have evolved highly specific recognition and attachment mechanisms . Once they have been recognized , the symbionts need to enter host cells and avoid immediate digestion , just like other intracellular symbionts such as Burkholderia rhizoxinica and Rhizobium leguminosarum , or pathogens such as Legionella , Listeria , or Yersinia ( Hentschel et al . , 2000; Moebius et al . , 2014 ) . Indeed , like many intracellular pathogens , the Bathymodiolus symbionts seem to induce a loss of microvilli on the cells they colonize ( Cossart and Sansonetti , 2004; Bhavsar et al . , 2007; Haglund and Welch , 2011; Wentrup et al . , 2014 ) . Finally , the symbionts achieve dense populations inside the host cells ( e . g . , Duperron et al . , 2006; Halary et al . , 2008 ) . Therefore , they must be able to avoid immediate digestion by their hosts . Although the mechanisms of host cell entry and immune evasion have been extensively studied in pathogens and plant–microbe associations such as the rhizobia-legume symbiosis , far less is known about the mechanisms beneficial symbionts use to enter and survive within animal host cells . The symbiosis between Vibrio fisheri bacteria and Euprymna scolopes squid is one of the few beneficial host-microbe associations where the molecular mechanisms of host-symbiont interaction have been investigated . A number of factors are involved in initiating this symbiosis such as the symbiont-encoded ‘TCT toxin’ , which is related to the tracheal cytotoxin of Bordetella pertussis ( McFall-Ngai et al . , 2013 ) . A few studies of intracellular insect symbionts have shown that they use type III and type IV secretion systems to establish and maintain their association with their host ( reviewed by Dale and Moran , 2006; Snyder and Rio , 2013 ) . These secretion systems are commonly used by intracellular pathogens to hijack host cell processes , allowing their entry and survival within host cells ( e . g . , Hueck , 1998; Steele-Mortimer et al . , 2002; Backert and Meyer , 2006 ) . An example is the Sodalis symbionts of aphids and weevils , which use a type III secretion system for entry to the host cell and are thought to have evolved from pathogens ( Dale et al . , 2001; Clayton et al . , 2012 ) . The virulence determinants of their pathogenic ancestors might therefore have been co-opted for use in beneficial interactions with their insect hosts . In contrast to the Sodalis symbionts of insects and the Vibrio symbionts of squid , the Bathymodiolus SOX symbionts are not closely related to any known pathogens . Moreover , because they fall interspersed between free-living SOX bacteria in 16S rRNA phylogenies , they are hypothesized to have evolved multiple times from free-living ancestors ( Figure 2—figure supplement 1 ) ( Petersen et al . , 2012 ) . Comparative genomics is a powerful tool for identifying the genomic basis of beneficial and pathogenic interactions , particularly if the symbionts or pathogens have close free-living relatives that do not associate with a host ( e . g . , Galagan , 2014; Ogier et al . , 2014; Zuleta et al . , 2014 ) . Genomes of closely related free-living and symbiotic relatives of Bathymodiolus SOX symbionts were recently published . Their closest free-living relatives are marine SOX bacteria called SUP05 , which are abundant in the world's oceans , particularly in oxygen minimum zones ( OMZs ) and hydrothermal plumes ( Sunamura et al . , 2004; Lavik et al . , 2009; Walsh et al . , 2009; Anantharaman et al . , 2012; Wright et al . , 2012; Petersen and Dubilier , 2014 ) . The Bathymodiolus SOX symbionts and SUP05 bacteria form a monophyletic clade together with the SOX symbionts of vesicomyid clams based on 16S rRNA gene phylogenies ( Figure 2—figure supplement 1 ) ( Distel et al . , 1995; Petersen et al . , 2012 ) . Closed genomes are available for the symbionts of two clam species ( Kuwahara et al . , 2007; Newton et al . , 2007 ) . All members of this monophyletic group ( the mussel and clam symbionts , and SUP05 ) share similar core metabolic features . They are all capable of autotrophic growth , and all use reduced sulfur compounds as an energy source ( Newton et al . , 2008; Walsh et al . , 2009 ) . They can differ in auxiliary metabolic capabilities such as hydrogen oxidation , nitrate reduction , or mixotrophy ( Newton et al . , 2008; Petersen et al . , 2011; Anantharaman et al . , 2012; Murillo et al . , 2014 ) . However , the major difference between these organisms is their lifestyle: SUP05 bacteria are exclusively free-living . The clam symbionts are exclusively host-associated , are vertically transmitted , and have reduced genomes . The Bathymodiolus symbionts appear to have adapted to both niches , as they have a host-associated stage and are assumed to also have a free-living stage . The goal of this study was to identify the genomic basis of host-symbiont interactions in Bathymodiolus symbioses . We used high-throughput sequencing and binning techniques to assemble the first essentially complete draft genomes of the SOX symbionts from Bathymodiolus mussels . We used comparative genomics of the symbionts' genomes to those of their close free-living and obligate symbiotic relatives to reveal genes potentially involved in Bathymodiolus host-symbiont interactions . We used phylogenetics and bioinformatic prediction of horizontally acquired genes to investigate the origins of these genes . Finally , we used transcriptomics and proteomics to determine whether potential host-symbiont interaction genes are being expressed by the symbionts in their host .
We sequenced the genomes of the SOX symbionts from three Bathymodiolus individuals: two were Bathymodiolus azoricus from the Menez Gwen vent field on the northern Mid-Atlantic Ridge ( MAR ) ( Figure 1 ) . We refer to these as BazSymA and BazSymB . The third mussel individual was an undescribed Bathymodiolus species ( BspSym ) , from the Lilliput hydrothermal vent on the southern MAR ( SMAR ) ( Figure 1 ) . Symbiont draft genomes from each individual were almost complete ( see ‘Materials and methods’ ) . Despite different sequencing and assembly strategies , the draft genomes were 90 . 7–97 . 7% complete ( Table 1 ) . The total assembly sizes were between 1 . 7 and 2 . 3 Mbp , on 52 to 506 contigs ( Table 1 ) . Each draft genome contained one copy of the 16S rRNA gene . The BazSymB assembly only contained an 829 bp fragment of the 16S rRNA gene; however , we PCR amplified and sequenced this gene from the DNA used to generate the metagenome . The 16S rRNA genes from the two B . azoricus symbionts were 100% identical and were 99 . 3% identical to BspSym . The core metabolic potential of the Bathymodiolus SOX symbionts is described in Appendix 1-Symbiont metabolism . A detailed description of the genomes is beyond the scope of this article and will be published elsewhere . 10 . 7554/eLife . 07966 . 003Figure 1 . Sampling sites . Map showing the sampling sites of Bathymodiolus mussels at hydrothermal vents along the Mid-Atlantic Ridge ( red stars ) . B . sp . is found at Lilliput ( BspSym ) , Bathymodiolus azoricus at Menez Gwen ( BazSymA and BazSymB ) and Lucky Strike . The details of the sampling sites are described in Supplementary file 1E . The map was produced with GeoMapApp 3 . 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 00310 . 7554/eLife . 07966 . 004Table 1 . Overview of the genomes compared in this study: SOX symbiont of B . sp , two individual SOX symbionts of B . azoricus , SOX symbiont Candidatus Vesicomyosocious okutanii , SOX symbiont of Calyptogena magnifica ( Candidatus Ruthia magnifica ) , and free-living SUP05DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 004GenomeCollection siteContigsGC content ( % ) Length/Span ( Mbp ) †Number of CDSsHGTEstimated completeness‡Coverage§Separation method#ReferencesB . sp symbiont ( BspSym ) Lilliput5238 . 231 . 8/2 . 3222533%95 . 39%22XFiltrationPetersen et al . , 2011 , this studyB . azoricus symbiont ( BazSymB ) Menez Gwen23938 . 201 . 5/1 . 7180230%90 . 60%8XGradient centrifugation/ binningThis studyB . azoricus symbiont ( BazSymA ) *Menez Gwen50637 . 581 . 85/1 . 85200835%97 . 70%59XBinningThis studyCa . V . okutaniiSagami Bay131 . 591 . 0/1 . 098026%93 . 58%–Whole genome assemblyKuwahara et al . , 2007Ca . R . magnificaEast Pacific Rise , 9°N134 . 031 . 2/1 . 2121023%94 . 84%–Whole genome assemblyNewton et al . , 2007SUP05Saanich Inlet9739 . 291 . 4/2 . 5158630%85 . 76%–BinningWalsh et al . , 2009SOX , sulfur-oxidizing . *SOX symbiont sequences recovered from metagenome of adductor muscle . HGT = Genes that potentially originated from horizontal gene transfer . †Length is the total length of sequence information on contigs without Ns , and span is the entire length of scaffold assembly including Ns . ‡The completeness of the genome was estimated with CheckM using a set of lineage-specific genes for proteobacteria ( Parks et al . , 2015 ) . §Median coverage . #Separation method indicates the experimental separation of symbionts from host tissue and co-occurring symbionts ( filtration or gradient centrifugation ) , or the in silico separation of genomic information from hosts and co-occurring bacteria ( binning ) . Bathymodiolus symbiont genomes were more similar to the free-living SUP05 than to the clam symbionts in terms of size and GC content ( Table 1 ) . Analysis of codon usage showed that all three Bathymodiolus SOX symbiont genomes had a greater proportion of genes that may have been acquired through recent horizontal gene transfer ( HGT ) compared to the clam symbionts Candidatus Vesicomyosocius okutanii , and Candidatus Ruthia magnifica ( Table 1 ) . BspSym , the genome that assembled into the fewest contigs , lacked synteny compared to SUP05 and the clam symbionts , as shown by whole genome alignment ( Figure 2—figure supplement 2 ) . Genome alignment of BazSymA and BazSymB was not attempted because the assemblies were highly fragmented . The possibility of incorrect genome assembly for BspSym was ruled out for four regions by PCR amplification of sequences spanning the regions without synteny . For confirmation , one PCR product was Sanger-sequenced and found to be identical to the draft genome assembly of BspSym . These regions without genome synteny therefore most likely represent true genome reshuffling in Bathymodiolus symbionts . The Bathymodiolus symbiont genomes had more mobile elements compared to their closest relatives ( Supplementary file 1A ) . Bathymodiolus symbionts had between 13 and 23 transposases and three to five integrases . SUP05 had 14 transposases and one integrase . We did not find any transposases or integrases in the clam symbiont genomes . The Bathymodiolus symbionts were highly enriched in restriction-modification system genes ( between 10 and 22 genes ) , whereas SUP05 only had one , and the clam symbionts had none . This large difference raises the possibility that restriction-modification systems are involved in genome reshuffling in the Bathymodiolus symbionts . Between 2 . 3 and 7 . 6% of the genes found only in the Bathymodiolus SOX symbionts but not in the clam symbionts and SUP05 , genomes were annotated as toxin or virulence genes ( Figure 2 ) . Most were related to genes from one of three toxin classes: ( 1 ) the RTX ( repeats in toxins ) toxins , ( 2 ) MARTX ( multifunctional autoprocessing RTX toxins ) , a sub-group of RTX toxins , and ( 3 ) YD repeat toxins ( also called rhs genes as they were initially described as ‘recombination hotspots’ ) . Representatives from all three toxin-related genes ( TRGs ) classes were found in each of the three-draft genomes , except for RTX , which were not found in BazSymA . The number of genes from each class varied between the three genomes . We found the largest number of TRGs in the genome with the fewest contigs , BspSym , which had at least 33 YD repeat genes , eight RTX genes , and 19 MARTX-like genes . In the BazSymB genome , 14 YD repeat genes , two RTX genes , and up to 10 MARTX genes were found . BazSymA had 16 YD repeat genes , and one MARTX ( Figure 2 , Supplementary file 1B ) . This indicates that these toxin-related classes are common to the SOX symbionts of both B . sp . and B . azoricus . In the BspSym genome , which assembled into the largest contigs , 22 out of 88 TRGs were found directly upstream or downstream of mobile elements . 10 . 7554/eLife . 07966 . 005Figure 2 . Genes shared between the Bathymodiolus and vesicomyid SOX symbionts and free-living SUP05 . Protein-coding sequences from the Bathymodiolus sulfur-oxidizing ( SOX ) symbiont were compared to the clam symbiont genomes and to the SUP05 metagenome from Walsh et al . ( 2009 ) with BLAST score ratios ( BSR ) . ( A ) Venn diagram of the shared and unique gene content in the clam symbionts , mussel symbionts , and SUP05 bacteria . Predicted protein sequences of each mussel symbiont were compared to a combined data set of the clam symbionts ( Rma and Vok ) and SUP05 . Similarly , protein sequences of each clam symbiont were compared to a combined data set of mussel symbionts ( BspSym , BazSymB , and BazSymA ) . Depending on the reference genome , the number of shared genes varies slightly and possibly reflects the presence of paralogous genes and redundant sequence information in these draft genomes . Abbreviations are explained in detail in Table 1 . The BLAST score ratio ( BSR ) threshold was 0 . 4 . ( B ) Venn diagram of mussel symbiont toxin-related genes ( TRGs ) , calculated with a BSR threshold of 0 . 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 00510 . 7554/eLife . 07966 . 006Figure 2—figure supplement 1 . Maximum likelihood 16S rRNA phylogeny of the close relatives of the Bathymodiolus SOX symbionts . The tree was estimated from an alignment of 1653 nucleotide positions and was rooted with four sequences from Thiomicrospira species . The number of sequences per collapsed group is shown next to the gray blocks . Diagonal lines in the out-group branch indicate that the branch is not to scale . B . = Bathymodiolus; A . = Adipicola; I . = Idas . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 00610 . 7554/eLife . 07966 . 007Figure 2—figure supplement 2 . Whole genome alignment . Each colored block is a region of the genome that aligned to part of another genome because it is homologous and the genes are arranged in the same order . Lines crossing represent conflicting information when compared to other genomes . These are the sites where lack of synteny was observed . Red vertical lines represent contig boundaries . BspSym = SOX symbiont of Bathymodiolus sp . , Vok = SOX symbiont Candidatus Vesicomyosocious okutanii , Rma = SOX symbiont of Calyptogena magnifica ( Ca . Ruthia magnifica ) , SUP05 = free-living marine sulfur oxidizers . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 00710 . 7554/eLife . 07966 . 008Figure 2—figure supplement 3 . Metabolic reconstruction of the Bathymodiolus symbiont . Key metabolic pathways were inferred from genomic information using Pathway Tools ( Caspi et al . , 2014 ) . Red stars indicate that the gene was not found in the B . sp . symbiont genome , and blue stars indicate that the gene was not found in BazSymB , but was found in BazSymA , both symbionts of B . azoricus . Red arrow indicates a missing enzyme that could be replaced with an alternative reaction . Green arrow indicates an inorganic pyrophosphate-dependent step in the modified version of the Calvin cycle . Nar = nitrate reductase; Nir = nitrite reductase; Nor = nitric oxide reductase; Hup = membrane-bound hydrogenase; SOX = sulfur oxidation; rDSR = reverse dissimilatory sulfite reductase; Sqr = sulfide-quinone reductase; Apr = adenylsulfate reductase; SAT = sulfate adenyltransferase; P = phosphate; BP = biphosphate; COX = cytochrome c oxidase; Gln = glutamine; Arg = arginine; Pro = proline; Met = methionine; Lys = lysine; Thr = threonine; Ile = isoleucine; PPi = inorganic pyrophosphate; PPase = soluble pyrophosphatase; SS = secretion system . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 008 Toxin genes are known to have unusually high substitution rates due to an ‘evolutionary arms race’ with their targets ( Jackson et al . , 2009; Linhartová et al . , 2010 ) . Accordingly , many of the Bathymodiolus symbiont TRGs were highly variable between the symbionts of B . azoricus and B . sp . , but also between the symbionts from the two B . azoricus individuals , and even between different copies within one genome ( see ‘Variability of TRGs within Bathymodiolus SOX symbiont populations’ ) . We therefore searched for homologs of the TRGs in the clam symbionts and SUP05 with a lower BSR of up to 0 . 25 , but even with this reduced stringency , no hits were found ( see ‘Materials and methods’ ) . We ruled out the possibility that the TRGs were not found in SUP05 draft genome because of its incompleteness ( ∼89% complete ) , by searching for homologs of the Bathymodiolus symbionts TRGs in unbinned metagenomes and metatranscriptomes from hydrothermal plumes and OMZs that are enriched in SUP05 . If free-living SUP05 also encoded these TRGs , we would expect to find them regularly in SUP05-enriched metagenomes and metatranscriptomes . Instead , no hits were found in four out of these six data sets . In a metagenome from the Lost City hydrothermal vent , we found one weak hit to a YD repeat gene ( 31% similarity ) . In a metagenome from the Guaymas Basin hydrothermal vent , we found one weak hit to an RTX gene ( 34% similarity ) . However , both of these metagenomes were from sites colonized by either Bathymodiolus mussels ( Lost City ) , or Riftia pachyptila tubeworms ( Guaymas Basin ) , whose symbionts also encode a hemolysin gene of the RTX class ( Gardebrecht et al . , 2012 ) . These rare hits might therefore come from contamination by symbionts in the environment ( Harmer et al . , 2008 ) . Considering the almost complete absence of genes similar to the TRGs of Bathymodiolus in SUP05-enriched next-generation sequence data sets , we conclude that these genes are specific to the Bathymodiolus SOX symbionts and are not found in their close symbiotic or free-living relatives . Most symbiont TRGs were so divergent that they could not be confidently aligned . One exception was the YD repeat genes , a few of which contained a conserved repeat region . We reconstructed the phylogeny of this conserved region . YD repeat sequences from the symbionts of both Bathymodiolus species formed a distinct cluster , distant from all other sequences in public databases ( Figure 3 , Figure 3—figure supplement 1 ) . The Bathymodiolus symbiont sequences did not cluster according to their host species , but instead were intermixed , suggesting gene duplication events prior to the divergence of the B . azoricus and B . sp . symbiont lineages . The Bathymodiolus symbiont sequences fell into a cluster that contained mostly pathogenic bacteria such as Yersinia pestis and Burkholderia pseudomallei . This cluster was well supported by Bayesian analysis ( 0 . 91 posterior probability ) . This cluster also contained a number of beneficial symbionts such as B . rhizoxinica , which is an intracellular symbiont of the fungus Rhizopus , and the Photorhabdus and Xenorhabdus symbionts of soil nematodes ( Waterfield et al . , 2001; Goodrich-Blair and Clarke , 2007; Moebius et al . , 2014 ) . 10 . 7554/eLife . 07966 . 009Figure 3 . Phylogeny of YD repeat-containing proteins . The tree is a consensus of bayesian and maximum likelihood analyses , result of an alignment of 536 amino acids . Black circles represent branches with posterior probability >0 . 8 and bootstrap value >80 . White circles represent branches with either posterior probability >0 . 8 or bootstrap value >80 . The number of sequences per collapsed group is shown next to the gray bloks . Purple: organism found in intestinal microflora or in close association with another organism; green: free-living; red: pathogen . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 00910 . 7554/eLife . 07966 . 010Figure 3—figure supplement 1 . Consensus of bayesian and maximum likelihood phylogeny of YD proteins with identifiers . Trees were estimated from an alignment of 536 amino acids . Circles represent branches with posterior probability higher than 0 . 8 and bootstrap values higher than 80/100 . If both reconstruction methods are significant , the circle is black , otherwise it is white . Purple: found in intestinal microflora or in close association with other organisms; green: free-living; red: pathogen . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 010 To overcome the difficulties in aligning these highly divergent TRGs , we constructed gene sequence similarity networks based on BLAST to depict relationships among and between the symbiont TRGs , and those in public databases . This analysis revealed distinct sequence clusters that contained genes with >25% similarity over at least half of the length of the gene ( Figure 4 ) . If a cluster contained at least one gene that was similar to at least one other gene in another cluster ( similarity cut-offs as above ) , then these clusters were joined to create a larger network . They were also joined if both clusters contained genes that had similarity to another gene in the database ( i . e . , if they could be joined by at most two steps ) . This allowed us to identify distinct sub-groups within the three toxin-related classes , and to identify toxin sequences from public databases that were most similar to the Bathymodiolus symbiont genes . 10 . 7554/eLife . 07966 . 011Figure 4 . Protein similarity network of toxin-related proteins in the Bathymodiolus symbionts . Each node corresponds to a protein sequence and the links between nodes represent BLAST hits . The length of the edges is inversely proportional to the sequence similarity . Protein clusters containing RTX or multifunctional autoprocessing RTX ( MARTX ) proteins are shown in the red panel on the left , and sequence clusters containing YD repeats are shown in the gray panel on the right . Arrowheads are proteins from B . azoricus symbionts , and triangles are proteins from B . sp . symbionts . The symbols are colored in green if they were identified in the Bathymodiolus symbionts as YD repeat-containing genes , red if they were identified as RTX genes , and purple for MARTX genes . Some protein sequences were similar to the TRGs but not annotated as such as these are partial genes that did not have any conserved domain . If the clusters contained mostly genes with a particular annotation , we named the clusters after these annotations , for example , cluster ‘TcB/TcC’ contained proteins annotated as TcB or TcC . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 01110 . 7554/eLife . 07966 . 012Figure 4—figure supplement 1 . Network of toxin-related proteins in the Bathymodiolus symbionts with BLAST hits from Vibrio , Photorhabdus , Xenorhabdus , and Pseudomonas highlighted . Each node corresponds to a protein sequence and the links between nodes represent BLAST hits . The length of the link is proportional to the sequence similarity . Protein clusters containing RTX or MARTX are shown in the red panel on the left . Sequence clusters containing YD repeats are shown in the gray panel on the right . Arrowheads are proteins from B . azoricus symbionts , and triangles are proteins from B . sp . symbionts . The symbols are colored in green if they could be identified in the Bathymodiolus symbionts as YD repeat-containing proteins , red if they could be identified as RTX proteins , and purple for MARTX . If the clusters contained mostly proteins with a particular annotation , we named the clusters after these annotations , for example , cluster ‘TcB/TcC’ contained proteins annotated as TcB or TcC . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 01210 . 7554/eLife . 07966 . 013Figure 4—figure supplement 2 . Genomic architecture of MARTX regions . The two MARTX regions in BspSym are shown . Operons identified by assembling transcriptome data are indicated in yellow boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 013 TRGs from the Bathymodiolus symbionts clustered together with toxin and TRGs from phylogenetically diverse organisms including characterized toxins of gammaproteobacterial Vibrio and Pseudomonas , and TRGs of the gammaproteobacteria Shewanella , the actinobacterial Rhodococcus , the cyanobacterium Trichodesmium , and the firmicute Caldicellulosiruptor ( see e . g . , Figure 4—figure supplement 1 ) . The RTX genes clustered into two separate networks , one that had similarity to RTX secretion and activation genes , and one that had similarity to RTX toxins . Five distinct networks contained MARTX genes . One of these included genes from the symbiont MARTX1 cluster , and genes from other organisms that were annotated as MARTX or filamentous hemagglutinin . One network contained some genes that we classified as RTX and some we classified as MARTX , reflecting their shared features such as the RTX repeats . Eight MARTX genes had no significant hits to any other gene in public databases . The YD repeat genes formed five distinct networks . Sequences from the first three had structural similarity to TcB and TcC , two subunits of the ABC toxins of Photorhabdus and Xenorhabdus , the beneficial symbionts of entomopathogenic nematodes . The B and C subunits form a cage-like structure that encapsulates the toxic domain ( an adenosine diphosphate ( ADP ) ribosylation domain , located at the C terminus of the C subunit ) . The A subunit forms a syringe-like structure , which delivers the toxin to the insect cell ( Meusch et al . , 2014 ) . The genes in the fourth YD network had structural similarity to TcA genes that encode the syringe-like A subunit . The fifth YD network had similarity to genes annotated as RhsB , which was shown to play a role in bacteria–bacteria competition in Escherichia coli ( Poole et al . , 2011 ) . The Bathymodiolus symbiont genomes encoded more YD repeat and MARTX genes than any other genome that we compared them to ( Figure 5 , Figure 5—figure supplements 1–3 ) . This is remarkable considering the relatively small size of their genomes , and the fact that they are still incomplete . A few published genomes encoded more RTX genes , but these were much larger ( >5 Mbp ) ( Figure 5—figure supplement 3 ) . 10 . 7554/eLife . 07966 . 014Figure 5 . Distribution of the three major TRGs classes according to lifestyle . Each dot represents one sequenced genome . The sum of TRGs is on the Y axis , and the total number of genes predicted in each genome is on the X axis . Free-living bacteria are shown in red , host-associated bacteria that live outside of host cells are in green , and host-associated bacteria that can live inside host cells are shown in blue . The positions of the Bathymodiolus SOX symbionts are indicated . A detailed overview of all organisms that had similar TRGs to the SOX symbiont with the number of TRGs is shown in Supplementary file 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 01410 . 7554/eLife . 07966 . 015Figure 5—figure supplement 1 . YD genes per genome , normalized to the total gene count . Each dot is colored by the category to which they belong . Bathymodiolus SOX symbionts are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 01510 . 7554/eLife . 07966 . 016Figure 5—figure supplement 2 . MARTX genes per genome , normalized to the total gene count . Each dot is colored by the category to which they belong . Bathymodiolus SOX symbionts are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 01610 . 7554/eLife . 07966 . 017Figure 5—figure supplement 3 . RTX genes per genome , normalized to the total gene count . Each dot is colored by the category to which they belong . Bathymodiolus SOX symbionts are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 017 The vast majority of RTX , MARTX , and YD repeat proteins have not been functionally characterized . The few proteins whose function has been studied in detail are from bacteria that are known pathogens or cultured strains that can form biofilms . Because of this , it is generally assumed that RTX , MARTX , and YD repeat proteins function in host-microbe interactions , in microbe–microbe antagonism , or in biofilm formation , but this has not been extensively tested . To further investigate the functional role of the TRGs encoded by the Bathymodiolus symbionts , we tested whether similar genes are more likely to be found in bacteria that live in a particular niche ( extracellular host-associated , intracellular host-associated , or free-living ) , or that express a particular phenotype ( pathogenesis or biofilm formation ) . First , we used the Kruskal–Wallis one-way analysis of variance to determine whether the distribution of the three TRGs classes differed significantly ( I ) between biofilm-forming vs non-biofilm-forming bacteria , ( II ) between pathogenic and non-pathogenic bacteria , and ( III ) between free-living bacteria , host-associated intracellular bacteria , and host-associated extracellular bacteria , without considering whether the bacteria were pathogenic . There was no significant enrichment of any TRG category in bacteria known to form biofilms vs those that do not ( Table 2 ) . One class , MARTX was significantly enriched in the genomes of pathogenic vs non-pathogenic bacteria ( p-value = 0 . 007 , Kruskal–Wallis test ) . There was also a significant bias in the distribution of genes encoding MARTX and YD repeat genes in bacteria according to their lifestyle ( extracellular host-associated , intracellular host-associated , or free-living ) . 10 . 7554/eLife . 07966 . 018Table 2 . p-values obtained with Kruskal–Wallis rank sum testDOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 018B/NB df = 1P/NP df = 1Ext/Int/FL df = 2YD0 . 0970 . 52170 . 010*RTX0 . 7150 . 7930 . 308MARTX0 . 7730 . 007*3 . 21e−06*The three main lifestyle categories were tested against each toxin-related class . Number of TRGs per genome was normalized to the total gene count . FL = free-living , Ext = extracellular host-associated , Int = intracellular host-associated , P = pathogen , NP = non-pathogen , B = found in biofilms , NB = not found in biofilms , df = degrees of freedom , TRG , toxin-related gene , MARTX , multifunctional autoprocessing RTX . *p-value was considered to be significant ( p < 0 . 05 ) . When three categories are tested , such as in ( III ) above , the Kruskal–Wallis test does not identify which category the bias is associated with . To tease apart which of these three niche categories was most enriched in TRGs , we did Mann–Whitney–Wilcoxon tests ( Table 3 ) . These showed that both YD repeat and MARTX genes were enriched in the genomes of host-associated microbes ( YD repeat: p-value = 0 . 026 , MARTX: p-values = 2 . 125e−6 , 1 . 618e−6 , Mann–Whitney–Wilcoxon test ) . While MARTX genes were enriched in host-associated bacteria regardless of their location , YD repeat genes were only significantly enriched in intracellular bacteria . In contrast to YD repeat and MARTX genes , RTX did not show any enrichment in the three defined categories . RTX are therefore widely distributed among bacteria and are just as likely to be found in free-living and host-associated bacteria ( Appendix 2 ) . 10 . 7554/eLife . 07966 . 019Table 3 . p-values obtained with Mann–Whitney–Wilcoxon test for enrichment of YD and MARTX genes similar to those from the SOX symbiontDOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 019FL/ExtFL/IntExt/IntYD0 . 1290 . 0260 . 006*MARTX2 . 125e−06*1 . 618e−06*0 . 751FL = free-living , Ext = extracellular host-associated , Int = intracellular host-associated , MARTX , multifunctional autoprocessing RTX , SOX , sulfur-oxidizing . *p-value was considered to be significant ( p < 0 . 05 ) . Bacteria that are closely related often have similar genomic and physiological features . However , toxin genes are commonly gained through HGT , which may weaken the phylogenetic signal in their distribution patterns ( reviewed by Dobrindt et al . , 2004; Gogarten and Townsend , 2005 ) . To tease apart the possible phylogenetic influence on the TRGs distribution , we used Permanova to test whether any of the three classes was enriched in particular phylogenetic groups at the class , order , and family levels . Only RTX genes were significantly enriched , and only at the order level ( p-value = 0 . 0159 ) ( Supplementary file 1C ) . Therefore , phylogeny is not the main driver in the toxin-related distribution of YD repeats genes and MARTX . Toxin genes often have unusually high substitution rates , making them highly variable compared to non-toxin genes ( e . g . , Ohno et al . , 1997; Davies et al . , 2002 ) . We compared the substitution rates of all genes in the Bathymodiolus SOX symbiont genomes within the population of symbionts associated with each mussel species . This was done for each of the two species , B . azoricus and B . sp . by mapping transcriptome reads from three Bathymodiolus individuals to the draft genomes of their respective symbionts ( see below for transcriptomes ) . We calculated the number of single nucleotide polymorphisms ( SNPs ) per kb per gene . The number of SNPs in most of the TRGs was not significantly higher than the genome-wide average ( Figure 6 ) . However , we found 22 TRGs that did have significantly more SNPs than most of the other genes in the genomes . Among the 22 highly variable genes , we found representatives of each TRG class: YD repeats , RTX , and MARTX ( Figure 6 ) . 10 . 7554/eLife . 07966 . 020Figure 6 . Single nucleotide polymorphisms per gene . The number of single nucleotide polymorphisms ( SNPs ) per gene was normalized according to the length minus regions of unknown sequence for genes containing N's . Genes smaller than 150 bp were not considered . The dotted line represents the median plus one standard deviation of the number of SNPs per gene per kb . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 02010 . 7554/eLife . 07966 . 021Figure 6—source data 1 . Variability in TRGs encoded by the Bathymodiolus SOX symbionts . DOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 021 Transcriptome sequencing revealed that all predicted TRGs of the SOX symbionts in B . azoricus and B . sp gills were expressed . Reads mapping to TRGs accounted for 0 . 67–1 . 71% of mRNA in B . azoricus symbionts and 0 . 58–3 . 14% in B . sp . symbionts . All TRGs were found in the transcriptomes of at least one of the three individuals that we sequenced per species ( Supplementary file 1D ) . The expression levels of some genes from the RTX , MARTX , and YD repeats classes were in some cases higher than the expression of the essential Calvin cycle gene ribulose bisphosphate carboxylase/oxidase ( RuBisCO ) , which accounted for 0 . 03–0 . 5% of mRNA in B . azoricus symbionts and 0 . 11–1 . 04% in B . sp . symbionts . We could also confirm expression of some TRGs at the protein level in B . azoricus mussels ( samples of B . sp . were not available for proteomics ) . We analyzed ( I ) gradient centrifugation fractions that were enriched in the Bathymodiolus SOX symbiont and ( II ) whole gill tissue . Some of the characterized toxin proteins are associated with membranes . We therefore also analyzed a fraction enriched in membrane proteins . In these proteomes , we found 12 YD repeat proteins and one RTX . We also found a number of toxin-associated proteins such as one RTX activator and two RTX toxin transporters ( Table 4 ) . Nine of the 12 identified YD repeat proteins were present in the membrane proteome . Our method allowed us to identify symbiont-encoded proteins that were found in higher relative abundance in the whole gill fraction compared to the symbiont-enriched fraction ( see Appendix 4 ) . These proteins are potentially exported to the host tissue , indicating that they play a role in direct host-symbiont interactions . One RTX activator and one YD repeat protein were significantly enriched in the host tissue ( Table 4 ) . 10 . 7554/eLife . 07966 . 022Table 4 . Toxin-related proteins found in the proteome of the SOX symbiont from B . azoricusDOI: http://dx . doi . org/10 . 7554/eLife . 07966 . 022ProteomeIdentifierCategoryAnnotationMolecular weight ( kDa ) Max . number of unique peptides*SMHost_EST_000107YDIPR006530|YD repeat4311SMHost_EST_000115YDIPR006530|YD repeat4212NHost_EST_000248YDIPR006530|YD repeat377MHost_EST_002123YDIPR006530|YD repeat245SM†Thio_BAZ_1943_contig360420_0RTX ( activator ) Hemolysin-activating lysine-acyltransferase ( Hemolysin C ) 193SMTox_BAZ_119_contig00027_0YDRHS repeat-associated core domain-containing protein20217SMTox_BAZ_120_contig00027_1YDVirulence plasmid 28 . 1 kDa A protein6211SMTox_BAZ_1734_contig02141_2RTX ( transporter ) Secretion protein HlyD family protein4310SMTox_BAZ_2494_contig00030_0YDVirulence plasmid 28 . 1 kDa A protein18333MTox_BAZ_3202_scaffold00038_7RTXHemolysins and related proteins containing CBS domains352SMTox_BAZ_525_contig104979_0YDVirulence plasmid 28 . 1 kDa A protein522S†ToxAzor_892893YDRhs1142SMToxSMAR_1260BAT01109YD[weak similarity to] Toxin complex/plasmid virulence protein3218NToxSMAR_2052BAT01788 , Thio_BAZ_1733_contig02141_1 or Thio_BAZ_2580_scaffold00010_8RTX ( transporter ) Toxin secretion ATP-binding protein795SToxSMAR893-894YDRhs family protein1031SMToxAzor_890891YDRhs family protein676S = soluble proteome , M = membrane-enriched proteome , SM = found in both proteomes , SOX , sulfur-oxidizing . *The highest number of unique peptides detected in one sample . †Proteins that are potentially exported by the symbiont to the host gill tissue .
Two scenarios could explain the origin of the large complement of TRGs in the Bathymodiolus symbionts and not in their close relatives: firstly , the TRGs could have been in the genome of their last common ancestor but were all subsequently lost in both the clam symbionts and SUP05 . Alternatively , the Bathymodiolus symbionts could have acquired these genes via HGT after their divergence from the clam symbionts and SUP05 . Toxins are often found in ‘genomic islands’ that have been acquired through HGT ( reviewed by Lindsay et al . , 1998; Ochman et al . , 2000; Dobrindt et al . , 2004; Soucy et al . , 2015 ) . Several observations point towards HGT of the TRGs into the Bathymodiolus symbionts rather than their loss by the clam symbionts and SUP05 . Firstly , 63–68% of the TRGs could be identified as potentially horizontally acquired based on codon usage analysis , in contrast to 30–35% predicted for all coding sequences ( Table 1 , Figure 6—source data 1 ) . This means that their transfer was relatively recent , as the codon usage of these genes has not yet adapted to one typical of the symbiont genomes . Secondly , the content of the Bathymodiolus symbiont genomes attest to the major role that HGT has played in their evolution . They are enriched in mobile elements such as transposases and restriction-modification systems compared to their closest relatives ( Supplementary file 1A ) . The lack of synteny we observed in the symbiont genomes is consistent with the presence of mobile elements and major HGT events ( Kobayashi , 2001; Rocha and Danchin , 2002; Achaz et al . , 2003 ) . Thirdly , the TRGs from the Bathymodiolus symbionts are similar to genes from distantly related bacteria . Finally , mobile elements were regularly found directly upstream or downstream of the TRGs in the Bathymodiolus symbiont genomes . The linkage of mobile elements with some of these genes could explain the mechanism of their transfer into the Bathymodiolus symbiont genomes . This could also explain why each genome contained multiple copies of TRGs , as mobile elements are also prone to duplication ( Reams and Neidle , 2004 ) . Considering these observations , and the absence of these TRGs in their close relatives , we consider it most likely that they were acquired through HGT . Genes from the second major class of TRGs , MARTX , that were similar to those from Bathymodiolus symbiont , were significantly enriched in beneficial and pathogenic host-associated bacteria . One of the regions encoding MARTX genes from the Bathymodiolus SOX symbiont has a domain structure similar to the filamentous hemagglutinin FhaB from B . pertussis and Bordetella bronchiseptica ( MARTX1 , see Appendix 3 ) . FhaB is involved in attachment of Bordetella to their human host and suppression of the immune response ( reviewed by Melvin et al . , 2014 ) . B . bronchiseptica has two distinct phenotypic stages: an infective stage , where fhaB is upregulated , and a non-infective stage , where fhaB is downregulated . The non-infective stage is necessary for its survival in the environment outside of the host . The lifestyle of the Bathymodiolus symbiont has striking similarities to B . bronchiseptica . The symbionts must also survive in the environment to be transmitted from one host generation to the next . Our transcriptomes showed that these MARTX genes are expressed by Bathymodiolus symbionts within the host tissue . Unfortunately , we do not have samples to test whether they are downregulated in environmental symbiont stages . If so , it may have a similar function to its homologs in pathogenic Bordetella . MARTX-like genes also mediate cell–cell attachment in the symbiotic bacterial consortium ‘Chlorochromatium aggregatum’ ( Vogl et al . , 2008; Liu et al . , 2013 ) . Like the MARTX , YD repeat proteins were also significantly more enriched in host-associated compared to free-living bacteria . Some characterized YD repeat proteins function in competition between closely related bacterial strains ( Waterfield et al . , 2001; Kung et al . , 2012; Koskiniemi et al . , 2013 ) . In its intracellular niche , why would the Bathymodiolus symbiont need to compete against other bacteria ? Multiple strains of SOX symbionts can co-occur in Bathymodiolus mussels ( Won et al . , 2003a; DeChaine et al . , 2006; Duperron et al . , 2008 ) . These may compete with each other for nutrients and energy , or for space within host cells . Bacteria that express toxins to inhibit their close relatives need an immunity protein to protect them from each toxin ( Zhang et al . , 2012; Benz and Meinhart , 2014 ) . These immunity proteins are encoded immediately downstream or upstream of the toxin protein . The toxin-immunity pair is usually linked to genes encoding a type VI secretion system , which is the mechanism of toxin delivery for all so far described YD proteins involved in bacteria–bacteria competition ( Zhang et al . , 2012; Benz and Meinhart , 2014 ) . None of the YD repeat proteins in the Bathymodiolus SOX symbiont genome was found in an operon with an identifiable immunity protein , and no type VI secretion system gene was found in any of our draft genomes . The arrangement of YD repeat genes in the Bathymodiolus symbiont genomes and the lack of genes encoding type VI secretion systems are therefore inconsistent with a role in competition between closely related symbiont strains . Bathymodiolus mussels are also infected by bacterial intranuclear pathogens called Candidatus Endonucleobacter bathymodioli , which are related to the genus Endozoicomonas ( Zielinski et al . , 2009 ) . Ca . E . bathymodioli invades the mussel cell nuclei where it multiplies , eventually bursting the infected cell . Intranuclear bacteria are never found in the nuclei of symbiont-containing cells , which led Zielinski et al . ( 2009 ) to hypothesize that the symbionts can protect their host cells against infection . Consistent with this hypothesis , growth inhibition assays showed that B . azoricus gill tissue homogenates inhibit the growth of a broader spectrum of pathogens compared to the symbiont-free foot tissues ( Bettencourt et al . , 2007 ) . The mechanisms of protection against cultured bacterial pathogens and Ca . E . bathymodioli are unknown . There is no evidence from their genomes that the SOX symbionts of B . sp . and B . azoricus produce antibiotics , but they do have genes for bacteriocin production that were expressed in the six transcriptomes analyzed in this study . Expression of some of the TRGs discovered here , for example , those related to toxins involved in bacteria–bacteria competition , or the production of bacteriocins by symbionts could explain the absence of intranuclear bacteria from symbiont-hosting cells . Symbionts of other marine invertebrate hosts are able to recognize , enter , and survive within host cells without a large number of TRGs . For example , the genomes of the SOX symbionts of hydrothermal vent Riftia tubeworms and the heterotrophic symbionts of whale-fall Osedax worms are virtually complete , but they only contain one or a few RTX genes , and no YD repeat genes ( Robidart et al . , 2008; Gardebrecht et al . , 2012; Goffredi et al . , 2014 ) . There is overwhelming evidence that SOX symbionts are beneficial for their Bathymodiolus mussel hosts ( Appendix 1 ) . It is therefore highly unlikely that the Bathymodiolus SOX symbionts are pathogens that have been mistaken for beneficial symbionts . The remarkably large number of TRGs in the Bathymodiolus SOX symbionts bears striking similarity to the arsenal of toxins encoded by Candidatus Hamiltonella defensa , which is a facultative symbiont of aphids ( Oliver et al . , 2003 , 2005 ) . Both symbionts encode multiple copies of RTX and YD repeat proteins ( Degnan and Moran , 2008 ) . Ca . Hamiltonella defensa is a defensive symbiont that protects its host from attack by parasitic wasps ( Oliver et al . , 2003 ) . Its protective effect is linked to its complement of RTX and YD repeat toxins ( Degnan and Moran , 2008; Oliver et al . , 2010 ) . Based on our phylogeny of YD repeats , those from the Bathymodiolus symbionts cluster together with sequences from both pathogens and beneficial symbionts . Of the beneficial symbionts in this cluster , all except the Bathymodiolus symbionts have been shown to produce exotoxins that damage the organisms their host parasitizes , either a plant in the case of B . rhizoxinica or an insect in the case of Photorhabdus/Xenorhabdus ( Partida-Martinez et al . , 2007 ) . This raises the intriguing possibility that some of the TRGs in the SOX symbiont genomes might function in protecting the mussel hosts against eukaryotic parasites . Compared to our knowledge of parasitism in shallow-water bivalves , little is known about parasitism in deep-sea Bathymodiolus mussels . This is surprising considering that these incredibly dense communities would be ideal habitats for parasites ( Moreira and López-García , 2003 ) . Two studies have investigated parasitism in Bathymodiolus mussels based on the microscopic identification of unusual ‘inclusions’ in mussel tissues ( Powell et al . , 1999; Ward et al . , 2004 ) . The most abundant parasites resembled Bucephalus-like trematodes of the phylum Platyhelminthes , which are common in shallow-water mussels ( e . g . , Wardle , 1988; Lauenstein et al . , 1993; da Silva et al . , 2002; Minguez et al . , 2012 ) . Bucephalus trematodes infect the gonads of their mussel hosts , which often results in sterilization ( Hopkins , 1957; Coustau et al . , 1991 ) . Like their shallow-water relatives , the Bucephalus-like trematodes were abundant in the gonads of Bathymodiolus childressi from cold seeps in the Gulf of Mexico ( Powell et al . , 1999 ) . Powell et al . ( 1999 ) estimated that due to this heavy infection , up to 40% of B . childressi populations are reproductively compromised . The distribution of these trematode parasites has not yet been systematically investigated in Bathymodiolus . However , of the three species so far studied , only B . childressi was infected by trematodes ( Powell et al . , 1999; Ward et al . , 2004 ) . B . childressi is one of the few Bathymodiolus species that only associates with MOX symbionts , but not with SOX symbionts . If many of the TRGs encoded by the Bathymodiolus SOX symbiont are being used to defend its host against parasites , as is hypothesized for Ca . H . defensa , then this could help to explain why B . childressi is so heavily infected by trematodes . The MOX symbionts of B . azoricus and B . childressi do not encode the abundant TRGs of the Bathymodiolus SOX symbiont ( Antony CP , personal communication , May 2015 ) . Our SNP analysis provides further support for the hypothesis that some TRGs may be used for direct beneficial interactions , and some may be used for indirect interactions such as protection against parasites . Genes involved in direct host-symbiont interactions such as recognition and communication are expected to be conserved within the symbiont population of one host species ( Jiggins et al . , 2002; Bailly et al . , 2006 ) . Consistent with this , eight out of nine genes in the MARTX1 region , which we hypothesize may be involved in attachment to the host , do not have a significantly larger number of SNPs per kb compared to the rest of the genome ( Figure 6—source data 1 ) . In the ninth gene , annotated as a hypothetical protein , SNPs per kb were slightly above average . In contrast to genes involved in direct host-symbiont interactions , those involved in indirect interactions such as defense against parasites are expected to be highly diverse ( see Appendix 2 ) . The large sequence variability in 22 of the TRGs is therefore consistent with a role for these genes in protecting the host against parasites . The genomes of the uncultured Bathymodiolus SOX symbionts encode a unique arsenal of TRGs , unexpected for a beneficial , nutritional symbiont . We hypothesize that the Bathymodiolus SOX symbiont has ‘tamed’ these genes for use in beneficial interactions with their host . Some of the TRGs may benefit the symbiosis by protecting the symbionts and their hosts from their natural enemies . In most cases , symbionts are either nutritional , that is , their primary role is to provide their host with most or all of its nutrition , or they are defensive ( Douglas , 2014 ) . The Bathymodiolus symbiont is therefore unusual , as it may play an essential role in both nutrition and defense . The TRGs were most likely acquired by HGT , and this may be the mechanism by which its free-living ancestors acquired the ability to form an intimate relationship with marine animals . Remarkably , the Bathymodiolus SOX symbionts encode a larger complement of these TRGs than any so far sequenced pathogen , suggesting that these ‘toxins’ , although initially discovered in pathogens , may in fact belong to larger protein families that function in both beneficial and pathogenic host-microbe interactions . An alternative to the hypothesis that toxins may be tamed for use in beneficial interactions would be that ‘symbiosis factors’ may be commandeered for use in harmful interactions . Given our recent recognition of the ubiquity and vast natural diversity of mutualistic interactions between bacteria and eukaryotes ( McFall-Ngai et al . , 2013 ) , it is possible that many of the genes that are currently annotated as toxins may have first evolved through beneficial host-microbe associations .
We collected Bathymodiolus mussels in Lilliput , Menez Gwen , and Lucky Strike vent sites on the MAR . Bathymodiolus sp . from Lilliput on the SMAR were sampled and processed for genome sequencing as in Petersen et al . ( 2011 ) . For transcriptomics , we sampled mussels from the SMAR at 09°32 . 85′S , 13°12 . 64′W . Specimens of B . azoricus were collected in three cruise expeditions to Menez Gwen at ( i ) 37°45 . 5777′N , 31°38 . 2611′W during the MOMARETO cruise , ( ii ) 37°50 . 68′N , 31°31 . 17′W during the RV Meteor cruise M82-3 , and ( iii ) Lucky Strike at 37°16′58 . 5′′N , 32°16′32 . 2′′W during the Biobaz Cruise . The adductor muscle was dissected from samples of the MOMARETO cruise , while the gill tissue was dissected from mussels collected during the RV Meteor cruise . For samples from the RV Meteor cruise , we used a combination of differential and rate-zonal centrifugation to enrich Bathymodiolus SOX symbiont from gill tissue for genomic and proteomic analyses . Samples for transcriptomics were fixed on board in RNAlater ( Sigma , Germany ) according to the manufacturer instructions and stored at −80°C . An overview of the samples used in this study is shown in Supplementary file 1E . DNA extraction and genome sequencing of the B . sp . SMAR SOX symbiont was described in Petersen et al . ( 2011 ) . Briefly , the gill tissue of a single individual was ground in a glass tissue homogenizer and frozen until further processing . In the home laboratory , the homogenate was diluted in phosphate-buffered saline ( PBS ) 1× , centrifuged at 400×g for one minute , and the supernatant filtered through a 12-µm GTTP filter ( Millipore , Germany ) . Centrifugation and filtration was repeated 20 times . The filtrates were passed through GTTP filters of 8 µm , 5 µm , 3 µm , and 2 µm . Cells collected on the 0 . 2-µm filter were used for DNA extraction after Zhou et al . ( 1996 ) with an initial incubation overnight at 37°C in extraction buffer and proteinase K . 6 kb mate-paired reads were sequenced with 454-Titanium and 36 bp Illumina reads . 454 reads were assembled with Newbler v2 . 3 ( 454 Life Sciences Corporation ) and 569 pyrosequencing errors were corrected using the Illumina reads . DNA from the B . azoricus SOX symbiont enrichments from three individuals ( gradient pellets , see Appendix 4 ) was extracted according to Zhou et al . ( 1996 ) . Genomic DNA was extracted from adductor muscle using a CTAB/PVP extraction procedure ( 2% CTAB , 1% PVP , 1 . 4 M NaCl , 0 . 2% beta-mercaptoethanol , 100 mM Tris HCl pH 8 , 0 . 1 mg ml−1 proteinase K ) . After complete digestion of tissues ( 1 hr at 60°C ) , the mixture was incubated with 1 µl of RNase for 30 min at 37°C . An equal volume of chloroform-isoamyl alcohol ( 24:1 ) was added and tubes were slowly mixed by inversion for 3 min before a 10 min centrifugation at 14 , 000 rpm and 4°C . The supernatant was collected in a fresh tube , and DNA was precipitated with 2/3 volume of cold isopropanol ( 1 hr at −20°C ) . The DNA pellet was recovered by centrifugation ( 14 , 000 rpm at 4°C for 20 min ) , washed with 75% cold ethanol , air-dried , and suspended in 100 µl of sterile water . 454 sequencing was done by Genoscope to sequence the gradient pellet from gill tissue , and by OIST to sequence the adductor muscle of B . azoricus . For the gradient pellet , a 3 kb insert 454 library was prepared according to manufacturer protocols for mate-pair sequencing . 630752 reads were generated on a Titanium FLX sequencing machine and assembled using Newbler software ( version 08172012 ) . 1310 contigs larger than 500 bp were obtained , forming 130 scaffolds of a total length of 1668565 bp . The assembly from the adductor muscle was done with Newbler v . 2 . 7 ( 454 Life Sciences Corporation ) as described in Takeuchi et al . ( 2012 ) resulting in 644000 contigs of a total length of 510449434 bp . The adductor muscle and gradient pellet metagenomes of B . azoricus were binned to separate the SOX symbiont from the MOX symbiont and host genomes with Metawatt V . 1 . 7 , which uses tetranucleotide frequencies , coverage , GC content , and taxonomic information for binning ( Strous et al . , 2012 ) . Only sequences longer than 800 bp were considered for further analyses . Since we could only recover an 829 bp fragment of the 16S rRNA from BazSymB , the same DNA that was used for 454 sequencing was used as template for PCR amplification with the universal primers GM3f/GM4r ( Muyzer et al . , 1995 ) . The PCR product was directly sequenced with Sanger and assembled using Geneious V7 ( Kearse et al . , 2012 ) . The 16S rRNA of BazSymB can be found under the accession number ( LN871183 ) . We annotated the genomes of the Bathymodiolus symbionts ( BspSym: PRJNA65421 , BazSymA: PRJEB8263 , and BazSymB: PRJEB8264 ) , Candidatus V . okutanii ( NC_009465 ) , Candidatus R . magnifica ( CP000488 ) , and SUP05 metagenome ( ACSG01000000; GQ351266 to GQ351269 and GQ369726 ) with the following workflow: we used Glimmer ( Delcher et al . , 2007 ) for open reading frame ( ORF ) prediction . Ribosomal RNA genes were detected with RNAmmer ( Lagesen et al . , 2007 ) and tRNAs with tRNAscan-SE ( Lowe and Eddy , 1997 ) . Annotation was done with GenDB 2 . 4 ( Meyer , 2003 ) and supplemented by JCoast 1 . 7 ( Richter et al . , 2008 ) to integrate the results of BLASTp ( cut-off e-value of 10 . 0 ) against sequence databases NCBI-nr ( Altschul et al . , 1997 ) SwissProt ( Boeckmann , 2003 ) , KEGG ( Kanehisa et al . , 2011 ) , COG ( Tatusov , 2000 ) , Pfam ( Bateman et al . , 2004 ) , and InterPro ( Hunter et al . , 2009 ) . TMHMM ( Krogh et al . , 2001 ) was used for transmembrane helix analysis and SignalP ( Emanuelsson et al . , 2007 ) for signal peptide predictions . Sequences of the cytochrome c oxidase subunit 1 of the Bathymodiolus mussels were submitted to the European Nucleotide Archive when available ( Supplementary file 1E ) . We designed primers to amplify four regions covering region with lack of synteny . The primer sequences and annealing temperatures are listed in Supplementary file 1F . The PCR program consisted of an initial denaturation step of 98°C for 30 s , followed by 35 cycles at 98°C for 10 s , specific annealing temperature for 30 s , 72°C for 2 min , and a final extension at 72°C for 10 min . We obtained a PCR product of the expected size based on our assembly of all four targeted regions . We sequenced one of these by Sanger sequencing using ABI BigDye v3 . 1 and the ABI PRISM 3100 genetic analyzer ( Applied Biosystems , Foster City , CA ) . We used CheckM to evaluate the completeness of our draft genomes with a set of single-copy marker genes that are specific to proteobacteria ( lineage-specific marker set of CheckM p_Proteobacteria , UID3880 ) ( Parks et al . , 2015 ) . We estimated the similarity of BazSymB and BazSymA draft genomes with the mean and standard deviation of genes with bi-directional best BLAST hits . The initial comparison of gene content between the clam symbionts , mussel symbionts draft genomes , and SUP05 metagenome was done by BLAST score ratio ( BSR ) with a BSR cut-off of 0 . 4 ( Rasko et al . , 2005 ) . Since toxin genes are expected to have a higher mutation rate , we compared the toxin distribution among the three mussel symbiont draft genomes and their closest relatives with a BSR cut-off of 0 . 2 . Because Ca . R . magnifica is the largest vesicomyid genome available , a whole genome comparison with SUP05 was done using a Dotplot produced by Ugene ( http://ugene . unipro . ru ) with a minimum length of 50 bp and 90% similarity . Since a high gene synteny could be observed , Ca . R . magnifica was used as scaffold to reorder the contigs from SUP05 and the B . sp . symbiont using Mauve ( Rissman et al . , 2009 ) . We did not do this analysis for the B . azoricus symbiont genomes because they were highly fragmented . To compare the influence of HGT on genome evolution , we used the method of Davis and Olsen ( 2010 ) to identify genes with different codon usage patterns , indicating that they may have been recently acquired via HGT . The initial genome annotation contained many genes annotated as toxins ( see Appendix 4 for details of the automatic annotation ) . Most were related to toxins from three broad previously defined classes , the YD repeats , RTX , and MARTX toxins . Most of the YD repeat and RTX genes could be identified accurately by automatic annotation due to the presence of signature repeat regions . Some of the predicted YD genes appeared to be truncated . Alignment with the closest hits was used to extend the partial ORFs that were not correctly predicted . To curate the annotation of MARTX genes , we built Hidden Markov Model ( HMM ) profiles for the characterized MARTX domains in Vibrio cholerae: actin cross-linking domain , Rho GTPase inactivation , and cysteine protease domain ( Satchell , 2011 ) . The profiles were used to scan the SOX symbiont genomes with HMMER ( Finn et al . , 2011 ) . We searched for functional domains in the proteins identified as TRGs with SMART ( Ponting et al . , 1999 ) . To analyze the diversity and redundancy of the TRGs in the SOX genomes , we constructed a protein similarity network . All protein sequences of the SOX symbionts were searched against Uniprot with BLAST ( coverage >50% , similarity >25% , e-value < e−5 ) . We also used BLAST of all against all sequences of the Bathymodiolus SOX symbiont to recover partial TRGs that could not be identified by automatic annotation . The two searches were combined to produce a sequence similarity network based on transitivity clustering using the plugin Blast2SimilarityGraph in Cytoscape ( Srinivasan and Moon , 1999; Shannon et al . , 2003; Wittkop et al . , 2010 ) . Only the sub-networks that were connected in a maximum of two steps to the TRGs were considered . All TRGs of the SOX symbionts were submitted to Phyre2 ( Kelley and Sternberg , 2009 ) to predict the secondary structure of the protein . We looked for clusters in the protein similarity network that could be associated with a subunit of a toxin complex or an active domain . None of the three toxin-related classes has characteristic patterns , profiles , or domains that can be searched with standard tools . Often , their only shared feature is a repeat region , such as the RTX repeat in RTX and MARTX , which is a calcium-binding site containing G and N residues . However , the number of G and N residues and the length of the repeat is highly variable , which makes profile searches impossible . Moreover , these repeat regions can be shared by other calcium-binding proteins such as integrins and fibronectin , which do not act as toxins . The sequence and length of the YD repeat is similarly variable . To identify homologs of the Bathymodiolus symbiont TRGs in published metagenomes and metatranscriptomes , we therefore used BLASTp ( coverage >50% , similarity >25% , e-value < 0 . 001 ) ( Supplementary file 1G ) . Best hits were blasted against the SOX symbiont proteins to search for signatures of the TRGs . We looked for bacterial genomes that had similar TRGs to the Bathymodiolus SOX symbiont , using the Integrated Microbial Genomes ( IMG ) database ( Markowitz et al . , 2011 ) with BLASTp ( >50% coverage , > 25% similarity , e-value 0 . 001 ) . 122 genomes were retrieved and manually curated for the potential of pathogenicity and biofilm formation , as well as the lifestyle categories extracellular host-associated , intracellular , and free-living bacteria . Kruskall–Wallis and Mann–Whitney–Wilcoxon tests were used to estimate if the increased number of a class of TR is biased in a certain lifestyle . All statistical analyses were done in R . The 16S rRNA gene sequences of the Bathymodiolus symbionts were imported into the Silva database ( release Ref 119 ) and initially aligned with the SINA aligner ( Pruesse et al . , 2012 ) . The final alignment was refined with MAFFT ( Katoh et al . , 2002 ) . A maximum likelihood tree was estimated from the alignment of 1653 nucleotide positions using RaxML with 100 bootstrap replicates . To construct the YD phylogeny , the data set from BspSym was used to obtain related sequences from GenBank and B . azoricus . Because two metagenomes of B . azoricus were analyzed , we used CD-hit to remove redundancy at 100% similarity ( Li and Godzik , 2006 ) . YD repeat proteins are very variable in the C and N terminus . Therefore , the selection criterion we used was the presence of the conserved rhs domain . Only this ‘core’ region was used for phylogeny ( Jackson et al . , 2009 ) . YD proteins were aligned with MAFFT with BLOSUM30 ( Katoh et al . , 2002 ) . Phylogenetic analyses were done in ARB with maximum likelihood and bayesian reconstructions using a filter of 10% similarity , which resulted in 536 amino acid positions ( Ronquist and Huelsenbeck , 2003; Ludwig et al . , 2004; Stamatakis , 2006 ) . Bootstrap support was calculated with 100 replicates in maximum likelihood . MrBayes was run for 9 million generations and two independent runs of four heated chains . A consensus tree of both methods was constructed . Polytomies were introduced when both methods did not agree . Genomes with curated metadata are available through IMG ( Markowitz et al . , 2011 ) . Nevertheless , most entries contained no information in categories we were interested in such as biofilm formation and intracellular lifestyle . IMG was used as starting point to search for curated genomes that had genes with similarity to the TRGs of the mussel symbionts ( >10% similarity , e-value < e−5 ) . The protein sequences of these genomes were retrieved from NCBI based on the genome name ( modified Python script from Sixing Huang , Max Planck Institute Bremen ) . BLASTp was used to retrieve protein sequences related to the SOX TRGs ( >50% coverage , > 25% similarity , e-value of 0 . 001 ) . 122 genomes had at least one gene that was similar to at least one TRG . We considered the genomes of different strains as independent events for statistical analysis , as even closely related strains can have different lifestyles . These genomes were manually curated based on a literature search and classified according to the following categories: ( 1 ) lifestyle: divided into three sub-categories ( a ) associated: if the organism is at any stage host-associated ( based on IMG metadata ) , ( b ) intracellular: includes obligate and facultative symbionts ( searched in Google scholar with the keywords ‘intracellular bacteria’ and references read for more details when the abstract was not sufficient ) , ( c ) free-living: bacteria that are not host-associated and not intracellular; ( 2 ) pathogen: bacteria that can produce disease ( information obtained from IMG ) ; ( 3 ) biofilm: bacteria found in biofilms ( Google scholar keywords ‘biofilm’ and ‘microbial mat’ with literature analyses when not clear ) . The data were formatted and merged with self-written Perl scripts and R Development Core Team ( 2011 ) . Genomes with similar toxins or TRGs to the Bathymodiolus symbionts are shown in Supplementary file 1B . The sum of the number of TRGs belonging to YD , RTX , and MARTX was normalized with the total gene count for each genome and multiplied by a factor of 1000 . To compare the number of TRGs against the lifestyle categories , we used Kruskall–Wallis test . A post hoc analysis was carried out on significant p-values for the associated category with the Mann–Whitney–Wilcoxon test . Statistical analyses were done in R . We tested whether the bacteria were enriched in any of the three toxin-related classes at the class , order , or family by using one-way Permanova with 9999 permutations and Euclidean distances . p-values were corrected with Bonferroni correction for multiple testing . To extract the total RNA of three individuals of B . azoricus , the gill tissue was incubated overnight in RNAlater ( Sigma ) at 4°C . A fragment of the gill was dissected and homogenized . RNA was extracted with RNeasy Plus MicroKit ( Qiagen , Hilden , Germany ) according to the manufacturer's instructions . To remove cell debris and to improve RNA yield , we used QIAshreder Mini Spin Columns ( Qiagen ) . The quality of the RNA was assessed with Agilent 2100 Bioanalyzer . The RNA was used for cDNA synthesis with the Ovation RNA-Seq System V2 ( NuGEN , San Carlos , CA ) . To extract the total RNA of three individuals of B . sp . , the gill tissue was placed separately on liquid nitrogen , homogenized , and stored overnight at 4°C in self-made RNAlater ( 10 mM ethylenediaminetetraacetic acid ( EDTA ) , 25 mM tri-sodiumcitrate-dihydrate , 5 . 3 M ammonium sulfate , adjusted to pH 5 . 2 ) . After removal of RNAlater , samples were incubated in 600 µl RLT-β-mercaptoethanol buffer ( 1:100 ) for 10 min and homogenized on QIAshredder columns ( Qiagen ) . Total RNA was extracted with RNeasy Mini Kit ( Qiagen ) . We applied the DNA-free DNAse Treatment and Removal Kit according to the manufacturer's instructions ( Invitrogen , Carlsbad , CA/Ambion , Austin , TX ) . RNA quality was checked on the Experion Automated Electrophoresis Station using the RNA StdSens Analysis protocol ( BioRad , Hercules , CA ) . Libraries of B . spp . were generated with the Illumina TruSeq RNA Sample Preparation Kit and sequenced 2 × 100 paired-end on an Illumina HiSeq 2000 platform at the Institute of Clinical Molecular Biology ( Kiel ) . A total of 32 . 9 , 38 . 2 , and 38 . 4 million reads were sequenced per individual of B . azoricus . Libraries of B . azoricus were generated with DNA library prep kit for Illumina ( BioLABS , Frankfurt am Main , Germany ) and sequenced as single 100-bp reads on an Illumina HiSeq 2500 platform at the Max Planck Genome Centre ( Cologne ) . A total of 4 . 3 , 4 . 8 , and 6 . 9 million reads were sequenced per individual of B . azoricus . Adaptor removal and quality trimming was done with Nesoni ( http://www . vicbioinformatics . com/software . nesoni . shtml ) using a quality threshold of 20 . To remove ribosomal sequences from the data , we mapped the reads against the SILVA 115 SSU database with Bowtie2 and kept those reads that failed to align ( Langmead and Salzberg , 2012; Quast et al . , 2013 ) . The abundance of the transcripts per gene was estimated with Rockhopper that uses upper quartile normalization ( McClure et al . , 2013 ) . The expression values of the TRGs were normalized to the expression of RubisCO ( Supplementary file 1D ) . Transcriptome reads that mapped to the SOX symbionts with Bbmap ( http://bbmap . sourceforge . net/ ) were deposited in the European Nucleotide Archive under the accession numbers PRJEB7941 for B . azoricus and PRJEB7943 for B . sp . We calculated SNPs to compare substitution rates of genes in the SOX symbiont genomes using three transcriptomes of B . sp and three of B . azoricus ( Supplementary file 1E ) . Reads were normalized to a coverage of maximum 200 and minimum of five with BBNorm v33 ( http://sourceforge . net/projects/bbmap/ ) . Transcriptomes of Bathymodiolus sp . were mapped against the draft genome BspSym and transcriptomes of B . azoricus were mapped to the draft genomes of BazSymA and BazSymB with BBMap v33 . We only considered those reads that mapped to a single position in the reference genome and that had higher than 90% identity alignments . SNPs were called independently for the draft genomes BspSym , BazSymA , and BazSymB with the Genome Analysis ToolKit as described by De Wit et al . ( 2012 ) with some modifications ( McKenna et al . , 2010 ) . In summary , regions needing realignment were identified and realigned over intervals . SNPs and insertions or deletions ( InDels ) were called with the haplotype caller with a minimum confidence of 20 . A filter was applied around InDels with a mask extension of 5 . SNPs per gene were obtained with BEDTools ( Quinlan and Hall , 2010 ) . The number of SNPs per gene was normalized according to the gene length minus regions of unknown sequence for genes containing Ns . We did not consider genes shorter than 150 bp nucleotides or outlier genes at the ends of scaffolds that had an unusual number of SNPs . Soluble proteins were extracted from SOX symbiont enrichments , host cytosolic fractions , and whole gill and foot tissue in biological duplicates ( Appendix 4 ) . All proteome samples were obtained from the M82-3 cruise . Membrane proteins were extracted from the SOX symbiont enrichments and the whole gill tissue samples . We used 1D-PAGE followed by liquid chromatography ( 1D-PAGE-LC ) to separate proteins and peptides as described previously with minor modifications ( Heinz et al . , 2012 ) . MS/MS spectra were acquired with a LTQ Orbitrap Velos mass spectrometer ( Thermo Fisher , Bremen , Germany ) for soluble proteins and a LTQ Orbitrap Classic ( Thermo Fisher Scientific Inc . , Waltham , MA ) for membrane proteins ( Appendix 4 ) . MS/MS data were searched against two databases using the SEQUEST algorithm ( Eng et al . , 1994 ) . The first database , designated ‘reduced’ database , contained protein sequences from the SOX and MOX symbionts from Bathymodiolus , as well as from the host ( see Appendix 4 ) . The second database contained in addition sequences from host-related bivalves and symbiont-related bacteria ( Supplementary file 1H ) . False discovery rates were determined with the target-decoy search strategy as described by Elias and Gygi ( Elias and Gygi , 2007; Kleiner et al . , 2012a ) . | Although bacteria are commonly associated with causing illness , many are actually beneficial to the organism they live in or on . The phenomenon of one species helping another to survive is known as symbiosis . Animals thrive at hydrothermal vents in the deep sea because of their partnerships with symbiotic bacteria . The bacteria use the geochemical energy found at hydrothermal vents to convert carbon into sugars , thus providing their animal hosts with essential nutrients . Unlike the symbiotic communities that associate with humans and other mammals , in which thousands of bacterial species co-exist , deep-sea mussels associate with just one or two species of symbiotic bacteria . This relative simplicity is ideal for investigating how the intimate associations between animals and bacteria work . Genes contain the instructions cells and organisms need to survive , and so one way that researchers investigate symbiosis is by studying the genes of the organisms involved . Such studies of beneficial bacteria are beginning to reveal that the molecular mechanisms involved in symbiosis are remarkably similar to those responsible for the harmful effects produced by some bacteria . By performing genetic sequencing on the symbiotic bacteria from deep-sea mussels , Sayavedra et al . have discovered that the bacteria have an unusually large number of toxin-like genes , and that all of these genes are active in the bacteria when they are inside host mussels . This was unexpected , as the bacteria are known to benefit their mussel hosts . The toxin-like genes from the symbiotic bacteria are similar to toxins found in the bacteria that cause diseases such as cholera and the plague in humans and other animals . Sayavedra et al . suggest that the symbiotic bacteria have ‘tamed’ these toxins to use them in beneficial interactions with their host . For example , some of the toxins could help the bacteria and mussels to recognize and interact with each other , and others could help to protect the mussel host from its natural enemies . The next step will be to test these ideas , which will be challenging as the mussels cannot be bred in the laboratory . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"genetics",
"and",
"genomics"
] | 2015 | Abundant toxin-related genes in the genomes of beneficial symbionts from deep-sea hydrothermal vent mussels |
Von Hippel-Landau ( VHL ) protein is a potent tumor suppressor regulating numerous pathways that drive cancer , but mutations in VHL are restricted to limited subsets of malignancies . Here we identified a novel mechanism for VHL suppression in tumors that do not have inactivating mutations . Using developmental processes to uncover new pathways contributing to tumorigenesis , we found that Daam2 promotes glioma formation . Protein expression screening identified an inverse correlation between Daam2 and VHL expression across a host of cancers , including glioma . These in silico insights guided corroborating functional studies , which revealed that Daam2 promotes tumorigenesis by suppressing VHL expression . Furthermore , biochemical analyses demonstrate that Daam2 associates with VHL and facilitates its ubiquitination and degradation . Together , these studies are the first to define an upstream mechanism regulating VHL suppression in cancer and describe the role of Daam2 in tumorigenesis .
Tumor suppressor and oncogenic pathways function in part by subverting existing cellular programs to promote cancer ‘hallmark’ properties that engender malignant growth ( Hanahan and Weinberg , 2011 ) . This corruption of normal cellular physiology is mediated by aberrant activities of these tumorigenic pathways , which are predominantly driven by genetic mutation . Importantly , genetic mutation is not the sole source of this dysregulation , as changes in gene expression via promoter methylation or protein turnover can phenotypically resemble driver mutations and contribute to tumorigenesis ( Esteller et al . , 1999; Hegi et al . , 2004; Pineda et al . , 2015; Reinstein and Ciechanover , 2006; Semenza , 2003; Shen et al . , 2005; Zöchbauer-Müller et al . , 2001 ) . While these broad regulatory processes have been linked to cancer , the underlying molecular mechanisms that regulate expression of key components of tumorigenic pathways are very poorly characterized . VHL is a key tumor suppressor that is mutated in Von Hippel-Landau disease , a hereditary cancer predisposition syndrome that often manifests as clear-cell renal carcinoma ( ccRCC ) ( Chen et al . , 1995; Gossage et al . , 2015; Kim and Kaelin , 2004; Maher et al . , 1990 ) . VHL functions by binding to HIF1α and hydroxylated Akt and modulating their degradation and activity , respectively . Mutant forms of VHL associated with ccRCC are incapable of binding HIF1α or pAkt , resulting in stabilized expression and activation of these proteins , which ultimately facilitates tumorigenesis ( Guo et al . , 2016; Ivan et al . , 2001; Jaakkola et al . , 2001; Maxwell et al . , 1999; Min et al . , 2002; Ohh et al . , 2000 ) . While dysregulated HIF1α and pAkt are associated with most forms of cancer , mutations in VHL are found predominately in ccRCC . This dichotomy suggests that additional regulatory mechanisms oversee VHL dysregulation or inactivation in other malignancies . Indeed recent studies in glioma have shown that ID2 can interfere with VHL activity in cell lines ( Lee et al . , 2016 ) and that it’s subject to regulation by microRNAs ( Li et al . , 2017 ) . Nevertheless , the upstream mechanisms that directly regulate VHL expression and protein turnover in cancer remain poorly defined . One potential mode of tumor suppressor gene regulation is through developmental mechanisms . Developmental processes directly contribute to all forms of malignancy and are utilized by tumorigenic pathways to maintain cells in an undifferentiated and proliferative state ( Jackson et al . , 2006; Kesari et al . , 2005; Stiles and Rowitch , 2008 ) . Given these established molecular and functional interactions , it stands to reason that expression of tumorigenic pathways may be reciprocally regulated by developmental mechanisms . However , whether such reciprocal regulation of tumor suppressor pathways by developmental factors contributes to tumorigenesis is poorly defined . To investigate the interface between developmental programs and the regulation of tumor suppressor pathways , we used malignant glioma as a model system . As a molecular entry point for these studies , we focused on Daam2 , a key developmental regulator that suppresses glial differentiation and also contributes to dorsal patterning in the developing CNS ( Lee et al . , 2015; Lee and Deneen , 2012 ) . Here we found that Daam2 promotes tumorigenesis in mouse and human models of malignant glioma . Bioinformatics analysis revealed that Daam2 and VHL expression is inversely correlated across a host of human malignancies . These in silico observations are corroborated by in vivo functional studies , which revealed that Daam2 promotes tumorigenesis by suppressing VHL expression . Mechanistically , we found that Daam2 associates with VHL and facilitates its degradation by the ubiquitin pathway . Together , these studies represent the initial characterization of Daam2 function in glioma and define , for the first time , an upstream regulatory mechanism that controls VHL protein expression in cancer . Moreover , because mutations in VHL are restricted to a limited set of malignancies , we have identified a new mechanism for VHL inactivation in tumors that do not have inactivating mutations .
Previously , we identified Daam2 as a component of the Wnt receptor complex that directly contributes to dorsal patterning in the embryonic spinal cord and oligodendrocyte differentiation during development and after injury ( Lee et al . , 2015; Lee and Deneen , 2012 ) . To further explore its role in neurological diseases , we sought to investigate its role in malignancies of the CNS . Towards this we took advantage of the TCGA pan-cancer expression data set ( Akbani et al . , 2014; Weinstein et al . , 2013 ) and evaluated Daam2 expression across a spectrum of 34 malignancies , finding that it’s most highly expressed in low-grade glioma ( LGG ) and glioblastoma multiforme ( GBM ) ( Figure 1A ) . To validate DAAM2 expression in human LGG and GBM , we used in situ hybridization ( ISH ) across a cohort of 35 LGG and 40 GBM primary human samples , finding that DAAM2 demonstrates heterogeneous expression within each glioma sub-type ( Figure 1B–C; Figure 1—figure supplement 1 ) . Notably , the majority of tumors exhibited staining intensity scores that exceeded normal brain samples , indicating that Daam2 expression is elevated within glioma tumors . In addition , we assessed DAAM2 expression via qRT-PCR in primary human glioma samples and non-tumor white matter , and consistent with the tissue microarray data , we found that Daam2 expression is elevated in a majority of these primary samples ( Figure 1—figure supplement 1 ) . Next , we evaluated Daam2 expression in our mouse model of malignant glioma , where we combine in utero electroporation ( IUE ) , with CRISPR-mediated deletion of Nf1 , Pten , and Trp53 ( herein CRISPR/IUE ) ( Figure 2—figure supplement 1 ) ( Chen et al . , 2016; Chen and LoTurco , 2012; Chen et al . , 2012; Cong et al . , 2013; John Lin et al . , 2017 ) . This model closely resembles the genetics of GBM ( Alcantara Llaguno et al . , 2009; Kwon et al . , 2008; Zhu et al . , 2005 ) and begins producing tumors detectable around post-natal week 8 . Combining this model , with our the Daam2-LacZ mouse line , we performed immunostaining on the resultant tumors , finding that Daam2 exhibits elevated expression levels in tumors , compared to normal brain tissue ( Figure 1D; Figure 2—figure supplement 1 ) . Finally , we evaluated Daam2 expression in xenograft tumors generated from primary human GBM cell lines , finding that Daam2 is also highly expressed in these human cell line models ( Figure 1—figure supplement 1 ) . Put together , these studies indicate that Daam2 expression is elevated in both human LGG and GBM and is expressed in the associated model systems . The expression analysis in human glioma and associated mouse models led us to examine whether Daam2 contributes to glioma tumorigenesis . To assess its role in glioma , we performed overexpression , gain-of-function ( GOF ) studies in human GBM cell lines , finding that Daam2 accelerates the rate of cell growth in vitro ( Figure 2A–B , F ) . Next , we determined how Daam2 influences anchorage independent growth , via soft agar assay , finding that it also accelerates colony formation ( Figure 2C–E ) . Together , these in vitro studies , indicate that Daam2 promotes cell proliferation and growth in human GBM cell lines To evaluate the role of Daam2 in tumorigenesis , we turned to mouse models of malignant glioma . The first model combines targeted PiggyBac overexpression of oncogenic Ras-V12 ( PB-Ras ) in glial precursors with IUE , to generate malignant glioma in mice around post-natal day 14 ( Figure 2—figure supplement 1 ) . Combining PiggyBac-mediated overexpression of Daam2 with this Ras-driven model resulted in an acceleration of tumorigenesis ( Figure 2—figure supplement 2 ) . To further substantiate these findings in additional mouse models , we next used our CRISPR/IUE model ( Figure 2—figure supplement 1 ) . Consistent with the Ras model , combined overexpression of Daam2 in the CRISPR/IUE model , also resulted in accelerated tumorigenesis ( Figure 2G–I ) . BrdU labeling analysis of the resulting tumors from both models revealed that overexpression of Daam2 results in an increase in the number BrdU-expressing cells ( Figure 2J–L; Figure 2—figure supplement 2 ) . These data , combined with our in vitro studies , indicate that overexpression of Daam2 promotes glioma cell proliferation and tumorigenesis . To further delineate the role of Daam2 in glioma , we next performed a series of complementary loss-of-function ( LOF ) studies in human and mouse glioma models . In human GBM cell lines we performed shRNAi-mediated knockdown of human Daam2 , finding that decreased expression of Daam2 inhibited their rate of growth in vitro ( Figure 3A–B; Figure 3—figure supplement 1 ) . Next , we assessed the tumorigenic potential of these GBM cell lines , finding that knockdown of Daam2 resulted in a significant decrease in tumor growth in vivo ( Figure 3C–E ) . These knockdown studies across both in vitro and in vivo systems , complement the overexpression studies , and further substantiate the role of Daam2 in glioma cell proliferation and tumorigenesis . To further evaluate the necessity for Daam2 in glioma tumorigenesis , we used CRISPR/IUE model to generate malignant glioma in Daam2+/- and Daam2−/− mice ( Lee et al . , 2015 ) . As shown in Figure 3F–H , mice lacking Daam2 demonstrated decreased rates of tumor formation in this model compared to the heterozygote control . Moreover , BrdU labeling revealed substantial decreases in the number of proliferating cells in Daam2−/− tumors ( Figure 3I–K ) . Together , our LOF , and complementing GOF ( Figure 2 ) studies in human and mouse models of glioma indicate that Daam2 promotes cell proliferation and tumorigenesis . Having established that Daam2 functions to promote glioma tumorigenesis , we next sought to uncover the mechanism by which it operates . Previously , we found that Daam2 functions as a positive regulator of Wnt-signaling in the developing CNS ( Lee and Deneen , 2012 ) , suggesting that it may also function in this manner in glioma . To evaluate Wnt-activity we used an established Wnt-reporter ( TOP-FLASH ) and found that modulation of Daam2 expression has a modest effect on Wnt activity in glioma cell lines and did not impact Wnt activity in our mouse model of glioma ( Figure 4—figure supplement 1 ) . That Wnt-activity is not affected by changes in Daam2 expression raises the question of how Daam2 promotes glioma tumorigenesis . To identify changes in protein expression associated with Daam2-mediated tumorigenesis we performed reverse phase protein lysate microarray ( RPPA ) on FACS-isolated mouse glioma samples that overexpress Daam2 . Analysis revealed a cohort of proteins that are strongly downregulated in glioma samples that overexpress Daam2 ( Figure 4A; Figure 4—figure supplement 1; Supplementary file 1 ) . To further substantiate these potential relationships , we leveraged existing TCGA RNA-Seq and RPPA data to determine whether there is an inverse correlation between Daam2 expression and this cohort of proteins across a spectrum of human malignancies ( Figure 4—figure supplement 2 ) . This analysis identified von Hippel Landau ( VHL ) as one of the proteins from this cohort with the most significant inverse correlation score across this spectrum of malignancies , with lung adenocarcinoma and GBM demonstrating the strongest negative correlations between Daam2 and VHL ( Figure 4B; Figure 4—figure supplement 2 ) . VHL is an established tumor suppressor that is frequently mutated in clear cell renal carcinoma ( ccRCC ) and functions by facilitating the degradation of HIF1α and pAkt ( Guo et al . , 2016; Ivan et al . , 2001; Jaakkola et al . , 2001; Maxwell et al . , 1999; Min et al . , 2002; Ohh et al . , 2000 ) . Given that Daam2 expression is inversely correlated with VHL , we next assessed whether Daam2 demonstrates congruent correlation with HIF1α and pAkt in the panel of human cancers . Analysis of these data revealed that Daam2 expression was positively correlated with HIF1α and pAkt protein expression across this cohort of malignancies , with GBM and lung adenocarcinoma also showing the strongest positive correlation ( Figure 4B ) . Together , our RPPA screen and associated bioinformatics analysis of human malignancies indicate that Daam2 expression is inversely correlated with VHL expression and positively correlated with its downstream signaling axis . The RPPA data suggests that Daam2 modulates expression of VHL in glioma . To directly test this hypothesis , we evaluated expression of VHL and its signaling axis in our Daam2-GOF and Daam2-LOF mouse tumor models . Immunostaining for VHL in these models corroborated our RPPA data , where VHL expression was dramatically reduced in Daam2-GOF tumors , and increased in the Daam2−/− tumors ( Figure 4C–F ) . In addition , we also observed elevated levels of VHL protein in the brains of Daam2-/- mice , further reinforcing these expression dynamics ( Figure 4—figure supplement 2 ) . Next we assessed expression of VHL’s downstream effectors , pAkt and HIF1α , finding that both of these proteins demonstrated elevated levels of expression in the Daam2-GOF tumors , and pAkt having decreased levels in the Daam2−/− tumors ( Figure 4G–K ) . One prediction of elevated HIF1α is a concomitant increase in the expression of its target genes . Therefore we assessed the expression of HIF1α targets Glut1 and VEGFA in the Daam2-GOF tumors , finding that these proteins also demonstrate increased expression ( Figure 4K ) . Finally , one consequence of these changes in gene expression is elevated levels of angiogenesis ( Fan et al . , 2014; Keith and Simon , 2007; Semenza , 2004 ) , which we detected in Daam2-GOF tumors via staining for the endothelial marker , CD31 ( Figure 4—figure supplement 3 ) . These data , combined with our bioinformatics analysis , indicate that Daam2 suppresses VHL-signaling in malignant glioma . These observations raise the question of whether the effects of Daam2 on glioma tumorigenesis are mediated through its suppression of VHL expression . To determine whether an epistatic relationship exists , we overexpressed VHL in the presence of Daam2 overexpression in human GBM cell lines , finding that VHL suppresses the increased rate of cell growth mediated by Daam2-alone ( Figure 5A–C ) . Next , we extended these studies to our mouse models of glioma , finding that combined overexpression of VHL with Daam2 similarly suppressed the accelerated rate of tumorigenesis and proliferation mediated by Daam2-alone ( Figure 5D–O; T-V ) . Moreover , overexpression of VHL resulted in concordant suppression of pAkt and angiogenesis in these tumors ( Figure 5P–S; Figure 4—figure supplement 3 ) . Put together , these data indicate that Daam2 promotes tumorigenesis by suppressing VHL expression in glioma . Analysis of human cancer data , along with our functional studies in mouse models , strongly suggests that Daam2 expression results in the loss of VHL protein . To determine the mechanism by which Daam2 influences the levels of VHL protein , we next performed a series of immunoprecipitation experiments to evaluate the biochemical relationship between Daam2 and VHL . Using protein lysates extracted from our PB-Ras model of glioma , we found that Daam2 co-immunoprecipitates with VHL ( and vice versa ) , suggesting that these proteins associate ( Figure 6A ) . Given that Daam2 associates with VHL , and expression of VHL protein is negatively correlated with Daam2 , we hypothesized that these expression dynamics are the result of Daam2 promoting the degradation of VHL . To test this we co-transfected Daam2 and VHL in 293 cells , and measured the rate of cyclohexamide-mediated degradation , finding that VHL degradation is enhanced in the presence of Daam2 ( Figure 6B , D ) . The ubiquitination pathway is a central mechanism that oversees protein degradation , ( Pickart and Eddins , 2004; Kerscher et al . , 2006; Ulrich and Walden , 2010 ) , therefore , we next examined whether Daam2 facilitates the ubiquitintion of VHL . Using in vitro systems , we found that in the presence of elevated levels of Daam2 , the extent of VHL ubiquitination is substantially increased and levels of VHL protein demonstrate a concomitantly decrease ( Figure 6C ) . Together , these mechanistic studies reveal that the underlying biochemical relationship between Daam2 and VHL is mediated by ubiquitin-driven protein degradation .
Leveraging existing TCGA expression data across a broad spectrum of malignancies , we found that Daam2 is most highly expressed in glioma and melanoma . These expression characteristics in glioma were confirmed using tissue arrays and functional studies revealed that Daam2 promotes cell proliferation and tumorigenesis in human and mouse glioma models . These are the first studies to describe the role of Daam-family proteins in tumorigenesis . Daam1 and Daam2 have highly conserved formin domains yet exhibit non-overlapping expression patterns in the developing CNS ( Habas et al . , 2001; Kida et al . , 2004; Lee and Deneen , 2012; Nakaya et al . , 2004 ) , suggesting distinct functions during development . Indeed , Daam2 functions in this context via canonical Wnt-signaling , while Daam1 operates via the non-canonical Wnt , planar cell polarity ( PCP ) pathway ( Habas et al . , 2001; Lee and Deneen , 2012; Li et al . , 2011; Liu et al . , 2008; Zhu et al . , 2012 ) . It will be important to discern whether this functional diversity between Daam1 and Daam2 is also applicable to CNS malignancies . Given that the PCP pathway has been implicated in tumor cell invasion and migration ( Anastas and Moon , 2013; Paw et al . , 2015; Weeraratna et al . , 2002 ) , it is possible that Daam-family proteins also contribute to these features of tumorigenesis . Recent studies have implicated Daam1 , Daam2 and other formin-family genes in the migration of breast and neuroblastoma cell lines ( Luga et al . , 2012 ) , respectively , suggesting that the Daam-family proteins may also contribute to invasion and metastasis in other malignancies . Our previous studies have shown that Daam2 potentiates canonical Wnt signaling through the clustering of existing Wnt receptor complexes ( Lee and Deneen , 2012 ) . While the Wnt pathway has been implicated in several forms of cancer , including medulloblastoma , mutations in key components of the Wnt-pathway and dysregulated Wnt-signaling has not been widely linked to low- or high-grade glioma ( Bienz and Clevers , 2000; de La Coste et al . , 1998; Klaus and Birchmeier , 2008; Lustig and Behrens , 2003; Morin et al . , 1997; Rubinfeld et al . , 1997; Zurawel et al . , 1998 ) . This , coupled with the fact that Daam2 acts upon existing Wnt receptor complexes , may explain why we did not witness any overt changes in Wnt activity when we manipulated Daam2 expression in our glioma models . Nevertheless , we cannot formally rule out a possible role for Daam2 in canonical or non-canonical Wnt signaling in glioma or in other malignancies driven by Wnt dysregulation . Finally , our observations that Daam2 expression is increased in a majority of glioma tumors raises the question how it becomes dysregulated . One possibility is that it is actively regulated by transcription factors or microRNAs that are present in glioma . Currently , these mechanisms associated with Daam2 remain undefined and identifying them are key areas of future investigation that will shed new light on how developmental dysregulation influences tumorigenesis . Another possibility to consider is that Daam2 dysregulation is a passive by-product of its expression in a cell lineage that is over-represented in glioma . Daam2 is expressed in both glial lineages ( oligodendrocytes and astrocytes ) , in both precursor and mature stages of development . Both of these lineages are present in one form or another within the bulk tumor , therefore its possible that elevated Daam2 expression is the result of its expression in analogous cell populations that comprise the primary , bulk tumor . Deciphering which scenario ( passive or active ) is responsible for Daam2 dysregulation in glioma is an important area of future investigation . Our observation that manipulation of Daam2 expression promoted tumorigenesis , but not canonical Wnt signaling , points to the possibility that it functions through a Wnt-independent mechanism . RPPA and bioinformatics approaches revealed that Daam2 expression is inversely correlated with a cohort of genes ( Figure 4—figure supplement 2 ) , including VHL , across a broad spectrum of cancers . Moreover , the RPPA data analysis identified a larger cohort of proteins that were downregulated in the presence of Daam2 in our mouse model of glioma ( Figure 4—figure supplement 1 ) . Thus , future studies may also be geared towards delineating the links between Daam2 and these other downregulated candidates ( i . e . Claudin-7 and Syk , etc . ) in the context of glioma and glial development . These observations , coupled with our findings that Daam2 associates with- and facilitates- the ubiquitination of VHL , suggest that it may also play a role in the ubiquitination pathway . The nature of this prospective relationship between Daam2 and the existing ubiquitination machinery , and whether these relationships are specific to malignancy or can be extended to development are areas of future investigation and represent potentially new parallels between development and cancer . Indeed dysregulation of ubiquitination is linked to protein aggregation associated with several neurological disorders , functioning through both neuronal and glial populations ( Jansen et al . , 2014 and Hegde and Upadhya , 2007 ) . Another feature of Daam2 is that it functions to suppress the differentiation of oligodendrocyte progenitor cells ( OPC ) during development and after injury ( Lee et al . , 2015 ) . Studies in mouse models suggest that OPCs may serve as a cell of origin for glioma , while NG2-positive OPC populations are endowed with tumor initiating properties ( Ligon et al . , 2007; Liu et al . , 2011; Persson et al . , 2010; Stiles and Rowitch , 2008; Sugiarto et al . , 2011; Yadavilli et al . , 2015 ) . Moreover , OPCs have latent proliferative capacity that is essential for injury responses , suggesting parallels between the processes that drive injury responses and tumorigenesis . Indeed recent studies have linked the VHL-HIF1α signaling axis to OPC development and neonatal hypoxic injury responses ( Yuen et al . , 2014 ) . That Daam2 promotes tumorigenesis via suppression of VHL and also regulates OPC development and injury responses , suggests that it may also utilize these tumorigenic mechanisms in the context of injury . Together these findings add to the emerging evidence that OPC associated signaling networks play critical roles in glioma pathophysiology and further reinforce the parallels between neurological disorders and cancer biology . Amongst the candidates identified in our RPPA screen , we chose to focus our studies on VHL because it’s a potent effector of tumorigenesis across many forms of cancer . Our mechanistic studies revealed that Daam2 promotes glioma tumorigenesis via suppression of VHL , a classic tumor suppressor that promotes HIF1α degradation and inhibits Akt activity . Importantly , inactivating mutations in VHL are predominately found in ccRCC and are very rare in most other forms of cancer , including glioma . These observations raise the critical question of how VHL becomes dysregulated in cancers that do not have these mutations . Our studies define a novel regulatory mechanism that operates upstream of VHL in cancer and provides additional insight into how its expression is extinguished in tumors that do not have inactivating mutations . Moreover , because Daam2-VHL expression demonstrates a robust , inverse correlation across a spectrum of cancers , this is likely to be a generalized mechanism of VHL suppression in cancer . Future areas of investigation include structure/function mapping of the Daam2-VHL domains in the context of ubiquitination , tumorigenesis , and HIF1α signaling . Interestingly , prior studies in yeast have shown that VHL can be degraded via the Hsp70 and Hsp110 chaperone system ( McClellan et al . , 2005; Melville et al . , 2003 ) . Given that Hsp’s are expressed across a host of cancers ( García-Morales et al . , 2007; Nylandsted et al . , 2002; Sauvageot et al . , 2009 ) , it will be important to determine whether Daam2 engages the Hsp chaperone system to facilitate VHL degradation in glioma and other malignancies . Another intriguing possibility is whether VHL can reciprocally degrade Daam2 protein . Given that VHL functions as an E3 ligase and associates with Daam2 , this may be another mechanism by which this complex operates . While we did not observe overt decreases in Daam2 protein levels in the presence of VHL ( not shown ) , such a mechanism may be context or cell lineage specific , requiring additional co-factors or environmental conditions ( i . e . hypoxia , etc . ) . Future studies in other cancer models and cell systems will be required to determine if this is the case . Given its role as the central regulator of HIF1α , surprisingly little is known about the regulatory biology surrounding VHL and its role in glioma formation . Studies in glioma cell lines suggest that ID2 interference with VHL activity can deregulate HIF1α expression and promote tumorigenesis in xenograft models ( Lee et al . , 2016 ) . Interestingly , ID2 has also been linked to the suppression of OPC differentiation via direct regulation of cell cycle kinetics ( Wang et al . , 2001 ) . These observations further reinforce the relationship between OPC development and glioma tumorigenesis and highlight key parallels between ID2 and Daam2 function across these systems , as both genes suppress VHL , inhibit OPC differentiation , and promote glioma tumorigenesis . Because both proteins associated with- and regulate- VHL , albeit via distinct mechanisms , it’s possible that these suppressive functions are coordinated and contribute tumorigenesis . From this , a model emerges where ID2 displaces VHL from the ubiquitin complex , which facilitates its association with Daam2 and its subsequent ubiquitination and degradation . It will be important to decipher whether ID2 and Daam2 functions are in fact coordinated , and the extent to which Hsp proteins participate in this mechanism . Finally , understanding whether and how these prospective relationships are also applicable to OPC development and injury responses may also reveal new insights in CNS repair mechanisms .
To detect the expression of human Daam2 , we generated mRNA probe and performed in situ hybridization on human gliomas . Human glioma tissue microarray ( US Biomax , GL803b ) contains 35 LGG and 40 GBM and five adjacent normal brain tissue , single core per case . To generate human Daam2 probe , pCMV-SPORT6 containing huDaam2 mRNA sequence ( BC078153 ) was purchased from Openbiosystems ( EHS1001-9143512 ) . Antisense probe was produced by T7 RNA Polymerase followed by Sal1 enzyme digestion . The sense probe was generated by SP6 RNA Polymerase followed by Not1 enzyme digestion . To apply ISH on human TMA , tissue array was firstly de-paraffinized by Xylene , 100% EtOH , 90% EtOH , 70% EtOH , 50% EtOH , two times of 5 min for each step . Then deproteinization was conducted by proteinase K ( 10 ug/ml ) for 5 min , 4% PFA for 15 min , 0 . 25% acetic anhydride in 0 . 1M triethanolamine for 5 min , washing with PBS in between for 5 min . Hybridization of the RNA probe at 65°C for overnight after pre-hybridization at 65°C for 1 hr . Chromogenic detection was finished the following day . Three times 2xSCC washing at 65°d for 15 min each first , then rinsing with PBT , 1 hr blocking ( 20% lamb serum in PBT ) , 2 hr antibody incubation ( DIG-AP , 1:2000 ) , washing with PBT , and 10 min AP blocking . The slide was developing in NBT/BCIP in the dark for overnight . The reaction was stopped by AP buffer once the signal was developed . The intensity grades ( 0 , 1 , 2 , 3 ) is scored by the pathologist , Dr . Carrie Mohila . To detect the expression of Daam2 in mouse gliomas , we used mRNA probes ( Lee et al . , 2015 ) To generate mouse gliomas , we used pregnant wild type and Daam2 knockout mice for IUE surgeries . The pregnant wild-type mice were purchased from The Center for Comparative Medicine ( CCM ) at Baylor College of Medicine ( BCM ) . The Daam2 knockout mice were generated previous studies ( Lee et al . , 2015 ) . We also purchased SCID ( Taconic ) mice for human glioma transplantation assay . All procedures were approved by the Institutional Animal Care and Use Committee at Baylor College of Medicine and conform to the US Public Health Service Policy on Human Care and Use of Laboratory Animals . To generate mouse gliomas , we performed in utero electroporation ( IUE ) . Briefly , the uterine horns were exposed , and DNA combination was injected into the embryonic lateral ventricles along with Fast Green dye as the indicator . Then electroporation was accomplished by BTW tweezertrodes connected with the pulse generator ( BTX 8300 ) in the setting of 33V , 55 ms per pulse for six times , 100 ms intervals . In CRISPR-IUE model , the DNA combination is composed of the ‘helper-plasmid’ pGLAST-PBase ( 2 . 0 ug/ul ) and all the other DNA ( 1 . 0 ug/ul ) , including pbCAG-GFP , pbCAG-Luciferase , crNF1 , crPTEN , crp53 ( Chen and LoTurco , 2012; John Lin et al . , 2017 ) . In the HRas-IUE model , the DNA combination is composed of pGLAST-PBase ( 2 . 0 ug/ul ) and others ( 1 . 0 ug/ul ) , including pbCAG-GFP2aHRas and pbCAG-Luciferase . For the IUE/Daam2/VHL overexpression study , pbCAG-FlagDaam2 ( 1 . 0 ug/ul ) , pbCAG-HA-VHL ( 1 . 0 ug/ul ) were co-injected with the CRISPR-IUE or the Ras-IUE master mix described above . Six-week-old SCID male mice ( Taconic ) were used for human GBM cell line transplantation . 4 × 104 luciferase-infected primary GBM cells were injected into each mouse brain , at the location of the 1 mm front , 2 mm right , 3 mm deep from the Bregma . Animals were euthanized and perfused at six weeks after transplantation of GBM cells . Brains were fixed in 4% paraformaldehyde and 70% EtOH overnight , respectively . After fixation , brains were embedded in paraffin , sectioned and subjected to molecular and pathological analysis via immunostaining or hematoxylin and eosin staining . To measure the tumor growth after manipulation of Daam2 , mice are subjected to bioluminescent imaging before harvesting . Mice were monitored once a week . D-Luciferin ( Perkin Elmer , #122799 ) was diluted to 15 mg/ml with PBS and injected into each mouse at a dose of 10 ul/g body weight . After 10 min , mice were placed inside a Bruker FX Pro Imager for 2 min Bioluminescence Imaging and 10 s X-ray Imaging . X-ray image was transparently overlapped with the bioluminescence image . A free circle surrounded the region of interest ( ROI ) was selected for the quantification of the luciferase intensity . Relative luciferase intensity was measured as the intensity of control group was normalized as 1; experiment group = actual value of experiment group/the actual value of control group . To evaluate tumor proliferation , mice were subjected intraperitoneal injection with BrdU ( 200 μg/g body weight ) 4 hr before harvesting . Brains were perfused and fixed overnight in 4% paraformaldehyde and 70% EtOH overnight for paraffin embedding . For the BrdU immunohistochemistry , additional 2N HCl treatment was performed . Tumors derived from either IUE or transplantation were dissected from whole mouse brains . Tumor regions were visualized as GFP positive tissue under a fluorescence dissection microscope . For paraffin sections , the brain was fixed in 4% Formalin then 70% Ethanol overnight . For western blot and RPPA , GFP-positive tissues were dissociated from the whole brain then homogenized with lysis buffer . For H&E staining , slides were firstly deparaffinized by sequentially three times of 3 min Xylene , 100% EtOH , 95% EtOH , 80% EtOH , 70% EtOH . Hematoxylin was stained for 5 min , then rinsing with tap water till no more color changing . 2–3 dips in the acidic solution ( 1% HCl in 70% EtOH ) reduced the background . After rinsing with tap water , Eosin was staining for 1 min , following by tap water rinsing . Finally , dehydration was accomplished by three times of 5 min 95% EtOH , 100% EtOH , and Xylene . For IHC , deparaffinization process is the same as H&E staining . Antigen retrieval was performed by 10 min microwaving using Na-Citrate pH6 . 0 . Endogenous peroxidases were blocked using 3% H2O2 . After 1 hr serum blocking , slides were incubated with antibody overnight in cold room . The other day , slides were rinsed off in PBS and incubated with secondary antibody for 1 hr . Then DAB and hematoxylin were applied for the color matrix and counterstain . Finally , dehydration was the same as the H&E process . CD31 staining was quantified using approaches described in Tian et al . ( 2017 ) . The following antibodies were used for IHC and western blot: BrdU ( anti-rat , Abcam , ab6326 , 1:200 ) , VHL ( anti-rabbit , Santa Cruz , sc5575 , 1:20 ) , VHL ( anti-mouse , #556347 , 1:500 ) , CD31 ( anti-rabbit , Abcam , ab28364 , 1:100 ) , pAKT ( Ser473 ) ( anti-rabbit , Cell Signaling , #9271 , 1:1000 ) , HIF1a ( anti-mouse , Novus , NB100-105 , 1:500 ) , Flag ( anti-mouse , Sigma , F1804 , 1:1000 ) , HA ( anti-rat , Roche , clone 3F10 , 1:500 ) , GAPDH ( anti-mouse , Millipore , AB2302 , 1:1000 ) , and LacZ ( anti-rabbit , Cappel , 55976 , 1:1000 ) . GFP+ tumor tissues were dissociated under a fluorescence dissection microscope and sorted with FACS machine . Tumor cells were homogenized with ice-cold lysis buffer . Supernatants ( tumor lysates ) were transferred after centrifuging at 4°C , 14 , 000 rpm for 10 min . The lysates were mixed with 4X SDS sample buffer . Samples were boiled for 5 min at a final protein concentration of 1–1 . 5 ug/ul and a total volume of 40 ul . The lysis buffer was composed of 1% Triton X-100 , 50 mM HEPES , pH 7 . 4 , 150 mM NaCl , 1 . 5 mM MgCl2 , 1 mM EGTA , 100 mM NaF , 10 mM Na pyrophosphate , 1 mM Na3VO4 , 10% glycerol , and contained freshly added protease and phosphatase inhibitors from Roche ( # 05056489001 and 04906837001 , respectively ) . Samples were probed with 287 antibodies . The RPPA was performed and analyzed by Functional Proteomics RPPA Core Facility at MD Anderson Cancer Center . To test VHL degradation , 1 × 105 293 T cells were plated into the 12 well plate one day before transfection . 100 ng Myc-tagged Daam2 and GFP-tagged VHL was transfected by iMFectin DNA Transfection Reagent ( GenDEPOT ) according to the protocol . 30 hr post-transfection , cells were treated with 50 ug/ml CHX for 0 , 2 , 4 , 6 hr respectively . Lysis and Western blot process is the same as previously described . To quantify the protein abundance , the western blot band intensity in 0 hr sample of both VHL -/+D2 groups are set as ‘1' , then the intensity of 2 , 4 , 6 hr degradation western blot band is normalized and plotted using Nonlinear Regression Curve by GraphPad . RNA expression data underlying the results presented in Figure 1 were generated by TCGA Research Network ( http://cancergenome . nih . gov/ ) . All data used in this study were publicly available , e . g . from The Broad Institute’s Firehose pipeline ( http://gdac . broadinstitute . org/ ) , doi:10 . 7908/C11G0KM9 . From TCGA , we collected molecular data on 10224 tumors of various histological subtypes ( ACC project , n = 79; BLCA , n = 408; BRCA , n = 1095; CESC , n = 304; CHOL , n = 36; COAD/READ , n = 623; DLBC , n = 48; GBM , n = 161; HNSC , n = 520; KICH , n = 66; KIRC , n = 533; KIRP , n = 290; LAML , n = 173; LGG , n = 516; LIHC , n = 371; LUAD , n = 515; LUSC , n = 501; MESO , n = 87; OV , n = 262; PAAD , n = 178; PCPG , n = 179; PRAD , n = 497; SARC , n = 259; SKCM , n = 469; TGCT , n = 150; THCA , n = 503; THYM , n = 120; UCEC , n = 545; UCS , n = 57; UVM , n = 80 ) from TCGA , for which RNA-seq data ( v2 platform alignment ) were available . Correlation between VHL and AKT pS473 protein on TCGA samples , were obtained from The Cancer Proteome Atlas , or TCPA , using level four data from the portal ( Akbani et al . , 2015; Li et al . , 2013 ) . One-way ANOVA was used to analyze bioluminescent intensity and BrdU-positive cell counts to determine the differences between Ctrl , D2 and D2/VHL groups , followed by Tukey’s test to compare between individual groups , which is demarcated by an asterisk in the graphs . Independent t-test was used to analyze the differences in bioluminescent intensity and BrdU-positive cell counts between Ctrl vs . D2 , Het vs . KO , Scrambled-shRNAi vs . D2-shRNAi . | Glioblastoma is the deadliest form of brain cancer , and the rate of patient survival has not significantly improved over the past 70 years . This cancer arises when glial cells , which provide support and insulation to nerve cells , develop mutations that alter the activity of certain genes or alter the role they play in cells . However , there are also several key genes linked to glioblastomas that don’t exhibit mutations , such as the gene that encodes the Von Hippel Landau protein ( or VHL for short ) . This protein normally helps to protect us from developing cancer , but it is not clear how this protein is prevented from performing this role in glioblastomas . One possibility is that proteins that regulate how cells grow and develop may control VHL . For example , a protein called Daam2 plays a critical role in a signaling pathway that is required for glial cell development . Zhu et al . used biochemical techniques to study Daam2 and VHL in both human cells and mouse models of glioblastoma . The experiments show that glioblastoma cells have lower levels of VHL compared to normal cells . This decrease is caused by Daam 2 , which interacts with VHL and promotes its degradation . Further experiments found that in several different types of cancer , higher levels of Daam2 are linked with the presence of lower levels of VHL . These findings indicate that the interaction between Daam2 and VHL could be a new target for drugs to treat glioblastoma and possibly other forms of cancer . Daam2 and VHL have also been linked to multiple sclerosis , cerebral palsy and other diseases that affect the nervous system . Therefore , understanding how these proteins interact may also help to develop new treatments for these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2017 | Daam2 driven degradation of VHL promotes gliomagenesis |
The control principles behind robust cyclic regeneration of hair follicles ( HFs ) remain unclear . Using multi-scale modeling , we show that coupling inhibitors and activators with physical growth of HFs is sufficient to drive periodicity and excitability of hair regeneration . Model simulations and experimental data reveal that mouse skin behaves as a heterogeneous regenerative field , composed of anatomical domains where HFs have distinct cycling dynamics . Interactions between fast-cycling chin and ventral HFs and slow-cycling dorsal HFs produce bilaterally symmetric patterns . Ear skin behaves as a hyper-refractory domain with HFs in extended rest phase . Such hyper-refractivity relates to high levels of BMP ligands and WNT antagonists , in part expressed by ear-specific cartilage and muscle . Hair growth stops at the boundaries with hyper-refractory ears and anatomically discontinuous eyelids , generating wave-breaking effects . We posit that similar mechanisms for coupled regeneration with dominant activator , hyper-refractory , and wave-breaker regions can operate in other actively renewing organs .
Featuring prominent growth cycles , the hair follicle ( HF ) is a model system of choice for studying tissue regeneration . At the level of cellular activities , the hair growth cycle consists of three consecutive phases: anagen , phase of active proliferation; catagen , apoptotic involution phase; and telogen , relative quiescence phase ( Al-Nuaimi et al . , 2010; Paus and Foitzik , 2004; Schneider et al . , 2009; Stenn and Paus , 2001 ) . Cyclic regeneration is sustained by the bulge stem cells , located at the base of the permanent HF portion ( Cotsarelis et al . , 1990 ) . During anagen initiation , signals from the niche , including the dermal papilla ( DP ) , stimulate bulge stem cells and adjacent hair germ ( HG ) progenitors to proliferate ( Enshell-Seijffers et al . , 2010; Greco et al . , 2009; Legrand et al . , 2016 ) . Activated progenitors generate all lower HF structures , including the outer root sheath ( ORS ) and hair matrix . During catagen , a widespread apoptotic program remodels the HF back toward a telogen state ( Botchkarev et al . , 2001b; Fessing et al . , 2006; Foitzik et al . , 2000; Lindner et al . , 1997; Mesa et al . , 2015 ) . Conceptually , since the bulge produces downward migrating progeny ( Hsu et al . , 2011 ) , it effectively serves as a progenitor source , while the matrix functions as a sink , and the ORS as a channel for progenitors transiting between them . The signaling mechanisms that time these coordinated cellular activities during hair regeneration remain incompletely understood ( Al-Nuaimi et al . , 2014; Bernard , 2012; Lin et al . , 2009; Paus et al . , 1999 ) . The putative ‘hair cycle clock’ is thought to be composed of one or several activator/inhibitor pairs acting to time key cycle phase transitions at set thresholds of their activities . Accordingly , cycle pace will depend on the speed at which activators and inhibitors reach their respective thresholds ( Chen et al . , 2015 ) . Importantly , HFs exist as large populations and at least in the dorsal skin they interact to coordinate growth cycles ( Hodgson et al . , 2014; Plikus et al . , 2011 , Plikus and Chuong , 2008a , Plikus et al . , 2008b ) . Such coordination implies that at least some of the activators and inhibitors should be present between HFs , in the so-called skin macro-environment . Previous work on dorsal skin indicates that BMP and WNT pathways constitute important components of the hair cycle clock . Indeed , defects in either of these pathways can dramatically change hair cycle progression ( Botchkarev et al . , 2001a; Botchkarev and Sharov , 2004; Choi et al . , 2013; Enshell-Seijffers et al . , 2010; Kandyba and Kobielak , 2014; Kandyba et al . , 2013; Kobielak et al . , 2003 , 2007; Sharov et al . , 2005 , 2006 ) , and ligands and antagonists for both pathways mediate macro-environmental coordination between HFs ( Chen et al . , 2014; Plikus et al . , 2011; Plikus and Chuong , 2014 ) . Additionally , FGF , PDGF , TGFβ , TNFα and other pathways can modulate hair cycle ( Chen et al . , 2015; Festa et al . , 2011; Higgins et al . , 2014; Ito et al . , 2003; Kimura-Ueki et al . , 2012; Leishman et al . , 2013; Oshimori and Fuchs , 2012; Plikus , 2012; Rivera-Gonzalez et al . , 2016 ) . Importantly , the combined signaling activities for the above pathways partition the hair cycle in the dorsal skin into four functional phases , each with its distinct activator/inhibitor profile: propagating anagen ( P ) , autonomous anagen ( A ) , refractory telogen ( R ) and competent telogen ( C ) ( Hodgson et al . , 2014; Plikus and Chuong , 2014; Plikus et al . , 2008b ) . Interactions between HFs enable hair regeneration across dorsal skin to self-organize into dynamic patterns . Critical for this self-organization are the following HF-to-HF interactions: P-phase HFs can induce neighboring C-phase HFs to enter anagen via diffusible activators , leading to hair growth coupling , while A-phase anagen or R-phase telogen HFs cannot couple due to high levels of inhibitors ( Murray et al . , 2012; Plikus et al . , 2011; Plikus and Chuong , 2014; Plikus et al . , 2008b ) . It remains unknown , however , whether this self-organization mechanism and its underlying WNT/BMP signaling activities is a general feature of all body skin or a special case for dorsal skin only . Integrative understanding of large-scale hair regeneration requires a systems biology approach . Previous modeling on HFs include cellular automaton models ( Halloy et al . , 2000; Plikus et al . , 2011 ) , feedback-control model ( Al-Nuaimi et al . , 2012 ) and the FitzHugh-Nagumo ( FHN ) excitable medium model ( Murray et al . , 2012 ) . Here , we present a unified three-dimensional and stochastic modeling framework for the HF that captures: ( i ) activator/inhibitor signaling dynamics in a single HF , ( ii ) cyclic growth of a single HF , and ( iii ) coupling between multiple HFs through diffusive signals . Using this model , we reveal that skin as a whole behaves as a heterogeneous regenerative field , where: ( a ) dominant hair cycle waves start in the ventrum , ( b ) propagate dorsally in a bilateral pattern , ( c ) stop at the boundary with hyper-refractory ear skin , and ( d ) break at non-propagating anatomical landmarks , such as eyelids and ears . We also show that WNT and BMP serve as a universal activator/inhibitor signaling pair , whose varying activities underlie distinct hair regeneration dynamics in all anatomical locations studied . These results provide new understanding of how the entire skin of the animal manages all of its hair regeneration .
First , in modeling the geometry of a single HF , we considered four key expression sites for activator/inhibitor ligands , antagonists , and receptors along the HF axis: bulge , HG , matrix , and DP ( Figure 1A ) . During the cycle , the bulge ( assigned as Region I ) remains relatively static , whereas the DP moves up and down along the HF axis . Also dynamic are the HG and matrix . The former only exists during telogen , while the latter only exists during anagen . The HG grows down to make matrix during anagen onset , whereas during catagen , the matrix collapses , and a new HG reforms . Simplistically , cyclic HG→matrix→HG dynamics are coordinated with the DP; thus , in the model we identify them jointly as Region II . Next , we considered that both regions produce signaling factors . Although a biological simplification , we assumed that Region I does so at a rather constant rate , while Region II shows distinct temporal dynamics ( Appendix 2—table 4 ) . We also assumed that Region II is essential for sending hair cycle-promoting signal ( s ) , while Region I is the primary signal target . In short , we hypothesized that the essential temporal molecular dynamics in the HF operate as follows: Region II generates a signaling ligand ( L ) gradient; Region I detects it and transmits it into ligand-bound receptors ( LR ) that then , through a series of intermediate signaling steps not captured in the model directly ( such as activities of the downstream signaling pathways and involvement of additional cell populations ) , regulates cyclic HF growth ( Figure 1A ) . The molecular signaling events , either activating or inhibitory , can be summarized as: ( 1 ) ∂∂t[L] = Diffusion+Production+Reaction of L and R ( 2 ) ∂∂t[LR] = Reaction of L and R +Degradation +Extra Source10 . 7554/eLife . 22772 . 003Figure 1 . Model recapitulates hair cycling and its associated activator and inhibitor signaling dynamics . ( A ) Schematic depiction of HF growth dynamics during telogen and anagen . Telogen and anagen HFs are shown on the left and in the center , respectively . In both hair cycle phases , Region I ( purple ) represents bulge and Region II ( orange ) represents DP with HG during telogen phase , and DP with matrix during anagen phase . On the right , schematic drawing of diffusive activator ( Act . L . in green ) and inhibitor ( Inh . L . in red ) interactions with their corresponding receptors ( Act . R and Inh . R , not depicted ) that form ligand-bond-receptors ( Act . LR and Inh . LR ) and their coupling with physical growth of the HF ( blue ) is shown . ( B ) Typical noise-free dynamics of the activator ( green ) and inhibitor ( red ) and cyclic HF growth ( blue ) are shown . X-axis is time in simulated days . Y-axis for activator and inhibitor shows simulated signaling levels , and for HF growth – simulated length of the HF . Grey area demarcates one modeled hair growth cycle . ( C ) The duration of ~anagen and ~telogen phases as the function of inhibitor signaling strengths . X-axis shows modeled inhibitor levels with ‘0’ being an arbitrary baseline levels . Y-axis shows time in simulated days . Upon stronger inhibitory signaling ( high Inh . L level ) ~anagen shortens ( yellow ) and ~telogen lengthens ( purple ) . The entire cycle ( blue ) becomes longer either with stronger or weaker inhibitory signaling . When inhibitory signaling becomes either very strong or very weak , the excitability of the system breaks down and HFs equilibrate in one state ( grey regions ) . Also see Appendix 2—tables 1 , 2 and 4 . ( D–E’’ ) A total of 236 putative activator genes ( green ) and 122 putative inhibitor genes ( red ) available from a whole skin microarray dataset were identified to recapitulate temporal dynamics of the simulated activator ( D ) and inhibitor ( E ) , respectively . Multiple WNT pathway members are in the putative activator gene set ( D’ , D’’ ) , while BMP pathway members are among the putative inhibitor genes ( E’ , E’’ ) . See gene list in Dataset 1 . For all genes log-transformed , zero-mean expression profile values were calculated using colorimetric ratio-scale algorithm as reported in ( Lin et al . , 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 003 where L , R and LR stand for ligands , receptors , and ligand-bound receptors , respectively . In the dynamics of LR ( Equation 2 ) , the ‘Extra Source’ describes stochastic signaling effects due to noise , and potential signaling contributions from Region I ( Appendix 2-Governing equations for activators and inhibitors ) . As Equations 1 and 2 show , ligand-receptor interactions in the model take place only for the same signaling pathway , and no direct pathway cross-talk is set to occur . This , again , is a biological simplification . Recently , evidence for pathway interactions have emerged ( Kandyba et al . , 2013 ) , and its effect is explored in Appendix 2-Possible interactions between the activator and inhibitor pathways do not qualitatively alter the HF dynamics . Our model integrates key signaling features of the hair growth cycle: strong activator signals enhance HF growth , while strong inhibitor signals prevent it . We modeled HF growth through the spatial average of LR concentration differences between the levels of activator and inhibitor in Region I ( Equation 7 in Appendix 2-Modeling HF phases by concentration difference ) . We assumed the hair cycle has two critical ‘checkpoints’: ( i ) the event in late competent telogen , when production of activator starts to increase ( Chen et al . , 2014; Greco et al . , 2009; Oshimori and Fuchs , 2012; Plikus et al . , 2008b ) , and ( ii ) the event of anagen termination , when the HF starts to involute . Thus , our model recognizes two phases determined by these checkpoints: ~anagen , starting from the moment of activator amplification until anagen termination , and ~telogen , lasting until the next activator amplification event . In the context of the conventional hair growth cycle , ~anagen incorporates the late portion of competent telogen and the entire anagen , while ~telogen includes catagen , refractory telogen and the remainder of competent telogen ( Plikus et al . , 2011; Plikus and Chuong , 2014; Plikus et al . , 2008b ) ( Appendix 2-Modeling HF phases by concentration difference; Appendix 2—figure 2 ) . Model simulations produce several emergent behaviors . The cycle becomes autonomous – that is , it displays stable periodicity and excitability emerges naturally without a built-in ‘clock’ ( Figure 1B ) . Cycling is maintained within a range of parameter values , allowing testing for various intrinsic and extrinsic signaling scenarios ( Figure 1C ) . Associated with these dynamics are periodic changes in the system’s geometry – the signaling source in Region II moves cyclically . Simulations indicate that the moving HF geometry in the model is critical , greatly contributing to the regulation of the cycle . In a single HF model , activator/inhibitor diffusion occurs only along the HF axis . When a HF population is modeled , hair-to-hair communication emerges naturally as ligand diffusion from neighbors supplements intrinsic HF ligand levels . As such , hair cycle pace depends on interactive signaling between neighboring HFs – a feature that we explore below . Our model predicts that HF cycling occurs only within a certain range of signal strengths , that is the excitable regime ( Figure 1C , white region ) . Within this regime , activator and inhibitor are predicted to inversely modulate duration of both ~telogen and ~anagen phases . At certain , either too high or too low signal strengths , the excitability is predicted to break down and the HF is expected to enter a non-cycling state of equilibrium ( Figure 1C , grey regions ) . For example , when inhibitor levels are very high , the HF is predicted to equilibrate in an extended telogen ( Appendix 2—figure 5A ) , while extended anagen is predicted for the opposite signaling condition ( Appendix 2—figure 5B ) . Next , we used bioinformatic and experimental approaches to validate the model’s key prediction that the same activator or inhibitor pathway can inversely modulate telogen and anagen phase duration . Considering the established roles for BMP and WNT as respective inhibitor and activator pathways regulating telogen duration in the dorsal skin , we explored if they can also regulate anagen duration in the same skin region in a model-predicted fashion . First , we found that model-predicted temporal dynamics for inhibitor and activator during ~anagen ( Figure 1D and 1E ) match the actual anagen expression dynamics for multiple BMP and WNT pathway members established on a highly temporally resolved whole-tissue dorsal skin microarray dataset ( Lin et al . , 2009 ) ( Figure 1D–E’’; Appendix 1-Identifying model predicted hair cycle activators and inhibitors ) . We also show that perturbing BMP ( for details see Appendix 1-Validating model-predicted roles for BMP signaling in hair cycle control ) and WNT in transgenic mice ( for details see Appendix 1-Validating model-predicted roles for WNT signaling in hair cycle control ) alters dorsal anagen phase duration and leads to hair length defects in a way that is consistent with the model’s predictions . Overall , this data shows that our model generates biologically meaningful outcomes and that its predictive power is robust . Next , we set out to explore novel aspects of hair regeneration at the population level . For this purpose , we modeled a linear array of HFs ( i . e . two-dimensional organization; Appendix 2—figure 4A ) and a grid of HFs ( i . e . three-dimensional organization; Appendix 2—figure 4B ) . In both cases , the diffusion of activators and inhibitors accompanying each HF during growth naturally led to HF coupling ( Appendix 1-Validating model-predicted roles for BMP signaling in hair cycle control ) and emergence of several known features of collective hair growth behavior , including spontaneous anagen initiation and anagen wave spreading ( Appendix 2—figures 11 , 12 ) . We then focused on the phenomenon of bilaterally symmetric hair growth that is prominent in young mice ( Plikus et al . , 2008b ) yet remains unexplained . Conventionally , first anagen in the dorsal skin of newborn mice is considered synchronous . On the other hand , adult mice display fully asynchronous and asymmetric dorsal hair growth patterns ( Chen et al . , 2014; Plikus and Chuong , 2008a; Plikus et al . , 2009 ) . This , however , is preceded by prominent bilateral symmetry , which often persists into the fourth hair cycle ( Plikus and Chuong , 2008a ) . We now show that in the three-dimensional model where all HFs are assumed to be identical , full asynchrony evolves within just one cycle , and bilateral symmetry cannot be achieved ( Appendix 2-Dorsal and ventral HF patterns; Appendix 2—figures 11 , 12; Appendix 2—video 1 ) . Therefore , we hypothesized that first anagen is inherently asynchronous as a result of spatially patterned HF development . Indeed , spatial distribution of early anagen HFs in the dorsal skin of newborn mice ( Figure 2A–D ) reveals head-to-tail and subtle lateral-to-medial asynchronies . We modeled the impact of these asynchronies on hair growth pattern evolution . Simulations reproduced head-to-tail asynchrony ( Appendix 2-Dorsal and ventral HF patterns; Appendix 2—figures 14 , 15; Appendix 2—video 3 ) ; however , it persisted for at least 10 cycles , which is far more than the 3–4 cycles observable in mice . Moreover , prominent bilateral symmetry failed to form . 10 . 7554/eLife . 22772 . 004Figure 2 . Spatiotemporal patterning of early hair cycles . ( A–D ) Analysis of the whole mount dorsal skin samples from P0 ( n = 3 ) ( A ) and P1 WT mice ( n = 3 ) ( A’ ) reveals subtle head-to-tail and lateral-to-medial hair cycle asynchronies . Asynchronies were inferred from examining the size of pigmented HFs . Larger HFs result from earlier anagen onset . ( B ) Heatmaps of skin samples from A and A’ built based on black pixel density ( reflecting pigmented anagen HFs ) . ( C ) Quantification of anagen HFs at different phases confirms head-to-tail pattern asynchrony . Morphological definition of anagen phases used for this analysis is provided at the bottom on the panel . ( D ) Analysis of the whole mount dorsal skin samples from P0 220bpMsx2-hsplacZ mice , where lacZ reporter activates in anagen HFs starting from phase IIIb , confirms head-to-tail asynchrony . ( E , F ) Modeling rapid hair growth pattern evolution in the context of two heterogeneous domains . ( E ) Schematic depiction of the modeling conditions with central Dorsal domain flanked by two lateral Ventral sub-domains with coupling between Dorsal and Ventral HFs . ( F ) Compared to Dorsal domain HFs , Ventral domain HFs were assigned with higher levels of total available activator and inhibitor receptors , allowing shorter ~anagen and ~telogen duration . Furthermore , hair cycle asynchrony was introduced into Dorsal domain to model the initial head-to-tail asynchrony . In simulations , interactions between HFs across domain boundaries result in bilateral symmetry during the second cycle ( simulated time t68-78 . 5 ) . Also , initial asynchrony breaks down in the cycle 3 ( t130 ) , and partial bilateral symmetry maintains into the late cycles ( see Appendix 2—video 4 ) . Scale bars: A , A’ , D – 5 mm . Images on A , A’ and D are composites . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 004 We note that the above and previous simulations ( Murray et al . , 2012; Plikus et al . , 2011 ) were performed on homogenous HF populations , where all HFs are assumed to be identical . We then considered that novel patterns might develop upon interaction of two or more HF populations , whose activator/inhibitor signaling levels are inherently different . In principle , dorsal skin HFs can interact with HFs from other body regions , such as ventral skin , where hair cycle dynamics are potentially distinct . Because all skin is continuous and forms an approximation of a cylinder , we modeled it as an unrolled sheet , where two Ventral sub-domains flank a rectangular Dorsal domain ( Figure 2E ) . For initial modeling conditions ( Appendix 2-Dorsal and ventral HF patterns ) , we considered that: ( i ) the first cycle on the dorsal skin has built-in head-to-tail asynchrony , and that ( ii ) ventral HFs develop with a 3- to 4-day delay relative to dorsal HFs ( Appendix 1—figures 7–9 ) . Because ventral HFs are known to produce distinctly shorter hairs ( Candille et al . , 2004 ) , in the model we assumed that they have faster cycle dynamics compared to dorsal HFs ( Appendix 2-Changes in the total amount of activator and inhibitor receptors results in different sensitivity of ~anagen and ~telogen lengths to signaling changes , Appendix 2-Dorsal and ventral HF patterns; Appendix 2—figures 8 , 9 ) . Indeed , in this configuration , our model readily reproduced patterns with aspects of bilateral symmetry already in the second cycle as the result of dominant waves spreading from the Ventral to the Dorsal domain ( Figure 2F , t68-78 . 5 ) . Importantly , after the second cycle , the effect of the initial built-in head-to-tail asynchrony started to disappear . Instead , the interaction between Ventral and Dorsal HFs continued to produce prominent bilateral symmetry in the third ( Figures 2F , t130-145 ) and later cycles ( Appendix 2—video 4 ) . Taken together , the model predicts that rapid hair growth pattern evolution requires interaction of two or more skin domains with distinct hair cycle parameters . Next , we imaged Flash mice , whose luciferase reporter produces skin-specific WNT activity signal and allows to sensitively and non-invasively determine the location and percentage of anagen HFs across the entire body ( Hodgson et al . , 2014 ) ( Figure 3A–C ) . Luminescence levels were measured both dorsally and ventrally and mice were followed up until day P119 , encompassing up to five hair growth cycles . Combined analysis from multiple mice reveals prominent phase advancement in ventral over dorsal anagen , specifically during the second , third , and fourth hair cycles ( Figure 3B , blue area ) . Additionally , the spatial luminescence signal mapping reveals distinct ventral-to-dorsal anagen propagation with features of bilateral symmetry during second ( Figure 3C; Appendix 1—figure 6 ) and third cycles ( Figure 3C’ ) , supporting the patterning mechanism predicted by the model ( Figure 2F ) . We also mapped body-wide hair growth patterns on the basis of anagen HF pigmentation between days P0-P55 ( Figure 3D–G; Appendix 1—figures 7–12 ) . This analysis confirms ventral over dorsal anagen phase advancement starting from the second cycle and provides the following additional insights:10 . 7554/eLife . 22772 . 005Figure 3 . Dorsal-ventral HF interactions produce bilateral symmetry . ( A ) Time-lapse bioluminescence in dorsal and ventral skin of the representative Flash mouse between days P5-P48 . Bioluminescent signal is color-coded according to the colorimetric scale on the right . ( B ) Combined temporal dynamics ( from 6 Flash mice ) of the bioluminescent signal-based anagen measurements over four hair cycles ( days P5-P119 ) . Dorsal skin dynamics are in brown and ventral skin dynamics are in blue . Prominent temporal advancement of ventral over dorsal anagen initiation can be seen during second , third and fourth cycles ( light blue areas ) . This advancement is accompanied by dominant ventral-to-dorsal anagen wave spreading . ( C , C’ ) Mapping of Flash-based anagen reveals ventral-to-dorsal hair growth wave propagation and bilateral pattern symmetry . New anagen areas for each time point are color-coded . Second anagen initiation is shown on panel C , and third anagen initiation on panel C’ . Also see Appendix 1—figure 6 . ( D–G ) Hair growth distribution patterns on P17 ( D ) , P21 ( E ) , P39 ( F ) and P55 ( G ) . Three mice were analyzed at each time point . Inverted whole mount skin samples from representative mice are shown . Schematic pattern maps are provided with color-coded anagen ( green ) , catagen ( yellow ) and telogen ( red ) regions . Also see Appendix 1—figures 7–12 . ( H , I ) HF cycling dynamics in chin skin grafts remain faster compared to dorsal skin grafts . After transplantation , first anagen initiated similarly in both chin and dorsal skin grafts , however , second anagen initiated significantly faster in chin grafts . Representative chin and dorsal skin grafts are shown on ( H ) . Combined temporal dynamics of skin grafts in anagen and telogen are shown on ( I ) . Dorsal graft dynamics are in brown and chin graft dynamics are in blue . Temporal advancement of chin over dorsal second anagen initiation is highlighted with light blue color . Also see Appendix 1—figure 13 . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 005 ( i ) Ventrally , anagen phase is the shortest in the ‘chin domain’ , ending around P10 . It is longer in the ‘ventral domain’ proper , ending in the genital area around P14 and in the chest area around P17 ( Figure 3D ) . ( ii ) Dorsally , anagen is the shortest in the ‘cranial domain’ , ending around P14 . In the ‘dorsal domain’ proper it ends as a head-to-tail wave between P15-P20 ( Figure 3D; Appendix 1—figures 10 , 11 ) . ( iii ) First ventral telogen is shorter than dorsal telogen . Second anagen initiates in the chin and ventral domains already between P21-24 and then spreads toward ventral-dorsal boundaries in form of two bilaterally symmetric waves ( Figure 3E; Appendix 1—figure 12 ) . Second anagen also ends faster in the ventral skin , maintaining ventral-dorsal asynchrony and bilateral symmetry ( Figure 3F ) . ( iv ) Third anagen initiates the fastest in the chin domain , as early as P42 ( Figures 3G and 4H ) . When transplanted onto the back of pigmented SCID mice , chin skin grafts ( n = 8 ) showed faster cycling compared to dorsal skin grafts ( n = 8 ) . While first post-transplantation anagen started with similar timing in both chin and dorsal grafts , consecutive anagen started significantly faster in chin grafts ( Figure 3H and I; Appendix 1—figure 13 ) . Furthermore , in many instances , grafts induced anagen in the surrounding dorsal host skin . Taken together , these data support that dominant ventral-to-dorsal hair wave spreading drives rapid hair growth pattern evolution and bilateral symmetry . Underlying this behavior are faster hair growth cycle dynamics in chin and ventral HFs , a property that is partially maintained upon skin grafting . Next , we asked if faster hair cycle dynamics in chin and ventral domains correlate with distinct molecular dynamics in putative activators and inhibitors . We performed RNA-seq profiling of whole skin from chin , ventral and dorsal domains at six hair cycle time points: first ( aka competent ) telogen , early anagen , mid-anagen , late anagen , catagen and early second ( aka refractory ) telogen . Analysis revealed non-overlapping transcriptomic trajectories of the hair cycle between the three domains ( Figure 4A–B’’ ) and domain-specific expression patterns for multiple putative activator and inhibitor genes at all hair cycle time points ( Appendix 1—figures 14–19; Dataset 2 ) . We then asked if refractory properties of early telogen differ between the domains . Differential gene expression analysis ( Figure 4C–D ) revealed enrichment in chin and ventral domains for gene ontologies related to macrophage function and lipid storage , and enrichment in chin domain for muscle-related genes ( Figure 4E ) . Consistently , chin skin shows more contractile cells around HFs , and chin and ventral skin have thicker dermal adipose tissue and substantially more CD11b+;F4/80+ macrophages as compared to dorsal skin ( Appendix 1—figures 20 , 21 ) . Furthermore , dorsal early telogen skin shows gene expression changes consistent with higher refractivity – it is enriched for several BMP ligands , and depleted for BMP antagonists and WNT ligands ( Figure 4F ) . Consistently , in Axin2-lacZ WNT reporter mice , many more HFs with WNT-active DPs are seen in chin and ventral as compared to dorsal skin at P36 ( Figure 4G; Appendix 1—figure 22A ) . WNT activity increases in dorsal skin to the levels of ventral skin only by P42 ( Figure 4H; Appendix 1—figure 22B ) . Furthermore , in P42 BRE-gal BMP reporter mice , many more HFs with BMP-active bulges are seen in dorsal as compared to chin and ventral skin ( Figure 4I; Appendix 1—figure 22C ) . In Krt14-Wnt7a mice , spontaneous anagen initiation sites in dorsal skin overrun ventral-to-dorsal wave dominance ( Figure 4J; Appendix 1—figure 23 ) . In contrast , in Krt14-Bmp4 mice , ventral-dorsal hair growth waves stall and asymmetric anagen patches form instead ( Figure 4K ) . Together , this data confirms that lower refractivity and the underlying differences in BMP and WNT activities form the bases for ventral-dorsal hair growth dominance . 10 . 7554/eLife . 22772 . 006Figure 4 . BMP and WNT signaling differences underlie regionally specific telogen phase duration . ( A–B’’ ) PCA analysis reveals largely non-overlapping transcriptomic trajectories across six hair cycle stages in chin ( B ) , ventral ( B’ ) and dorsal domains ( B’’ ) . Combined , deconstructed PCA plots are shown with all data points marked as grey dots and domain-specific data points outlined and color-coded . Color-coding is based on the hair cycle timeline from Appendix 1—figure 14A; transcriptomic trajectories are drawn with dark lines . ( C ) Deconstructed PCA plot for refractory ( early second ) telogen is shown with domain-specific data points highlighted and color-coded based on the schematic drawing on A . ( D–F ) Analysis of refractory telogen data identified 1407 differentially expressed genes across the three domains ( D ) , with each domain showing enrichment for distinct gene ontologies ( E ) . Multiple putative hair cycle activator and inhibitor genes show domain-specific differential expression ( F ) . Putative activators are in green and putative inhibitors are in red . For each gene , relative fold changes for ventral over chin and dorsal over chin expression levels are indicated . Genes that show cyclic expression patterns are highlighted with blue . See additional expression data analysis on Appendix 1—figures 14–19 and in Dataset 2 . Asterisk marks non-canonical WNT ligand . ( G , H ) Analysis of Axin2-lacZ skin during second telogen reveals faster activation of WNT signaling in chin and ventral HFs over dorsal HFs . At P36 majority of HFs in chin and ventral skin have WNT-active DPs . In dorsal skin , the number of HFs with WNT-active DPs is low at P36 , but increases by P42 . ( I ) Analysis of P42 BRE-gal skin shows that many more dorsal HFs have BMP-active bulges as compared to chin and ventral HFs . Also see Appendix 1—figure 22 . ( J ) Overexpression of Wnt7a results in disruption of ventral-to-dorsal hair growth wave dominance and spontaneous anagen appears in the dorsal domain at P60 . ( K ) Overexpression of Bmp4 results in stalled ventral-to-dorsal hair growth wave spreading and patchy , asymmetric hair growth at P57 . Scale bars: G-I – 200 um , J-K – 500 um . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 006 Our model also predicts conditions when hair cycling stops and HFs equilibrate in an extended telogen , such as due to high levels of inhibitors ( Appendix 2-Hyper-refractory domain; Appendix 2—figure 5A , 16 ) . We profiled mouse skin for the existence of such behavior and found ears to match such prediction . In the ear skin , HF morphogenesis begins between days P2-P4 , and HFs remain in anagen until about P15 ( Figure 5A ) . After first anagen , and for at least three months , they remain in an extended telogen , while at the same time dorsal HFs have already reached their third cycle ( Figure 5B and C ) . Seldom , solitary anagen HFs can be found , but no coordinated hair growth waves , characteristic to other skin regions , are observed ( Figure 5B , day P95 ) . Moreover , anagen waves spreading from cranial skin could not propagate into ear skin ( Figure 7E ) . These observations are consistent with the possibility that ear skin is hyper-refractory . Next , we examined ear HFs’ responses to several potent anagen inducers: cyclosporin A ( Maurer et al . , 1997; Paus et al . , 1989 ) , smoothened agonist ( SAG ) ( Paladini et al . , 2005 ) and hair plucking ( Chen et al . , 2015 ) . We show that while dorsal telogen HFs readily respond to cyclosporin A ( Appendix 1—figure 27B ) , ear HFs remain quiescent even 3 weeks after treatment ( Figure 5F ) . Anagen can be induced in response to SAG; however , this response occurs late , after 3 weeks , and remains restricted to the medial side of the ear ( Figure 5E ) . This is contrasted by rapid SAG-induced anagen in dorsal skin ( Appendix 1—figure 27A ) . Plucking induces anagen along the medial side of the ear; however , there is no anagen wave spreading into the unplucked region , a feature common in dorsal skin ( Chen et al . , 2015 ) ( Figure 5D; Appendix 1—figures 25A-B , 26 ) . Furthermore , whole ear plucking experiments reveal very sparse anagen activation along the lateral side ( Appendix 1—figure 25C ) . These data demonstrate that physiologically , adult ear HFs equilibrate in a hyper-refractory telogen state , yet in principle remain capable of regenerative cycling in response to selective external stimuli . 10 . 7554/eLife . 22772 . 007Figure 5 . Ear skin shows hyper-refractory properties with telogen arrested HFs . ( A–C ) Morphogenesis and physiological cycling of ear HFs . ( A ) Analysis of ear tissue histology shows that developing HFs first appear on day P4 , and progress toward mature anagen by P7 . They enter catagen around P15 and first telogen by P17 ( based on three mice for each time point ) . ( B , C ) Whole mount ear skin analyses show that ear HFs fail to enter second coordinated anagen and , instead , remain in an extended telogen . Seldom , isolated anagen HFs can be found ( see P95 sample on B ) . Data are based on three mice for each time point . ( D ) HFs along medial side of the ear re-enter anagen after plucking ( also see Appendix 1—figure 25 ) . Experiment is based on five mice for each time point analyzed . Representative ear skin image and accompanying heatmap is shown . Heatmap criteria are shown at the bottom . ( E ) Unlike in dorsal skin ( see Appendix 1—figure 27A ) , ear HFs poorly respond to topical SAG treatment . Anagen induction is limited to the medial edge of the ear . ( F ) Unlike in dorsal skin ( see Appendix 1—figure 27B ) , ear HFs fail to re-enter anagen in response to topical cyclosporin A treatment . Experiments on E and F are based on three mice for each time point analyzed . Representative ear skin images and accompanying heatmaps are shown . Scale bars: A – 100 um . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 007 To understand how ear HF hyper-refractivity relates to activator and inhibitor signaling levels , we compared on RNA-seq refractory telogen dorsal skin with telogen ear skin and , additionally , cartilage/muscle complex , a structure unique to ears . We show that , transcriptionally , these three tissue types are distinct ( Figure 6A ) , containing large number of differentially expressed genes ( Figure 6B; Dataset 3 ) enriched for distinct gene ontologies ( Figure 6C ) . Analysis of the signaling pathways implicated in the hair cycle control revealed a number of differentially expressed WNT and BMP pathway ligands and antagonists ( Figure 6D ) . Compared to dorsal skin , ear skin is enriched for transcripts for several WNT antagonists , including Dkkl1 , Dkk2 and Sfrp2 , as well as collagen Col17a1 , implicated in HF stem cell maintenance ( Matsumura et al . , 2016 ) . Cartilage/muscle complex is prominently enriched for Bmp5 , and multiple WNT antagonists , including Frzb , Sfrp2 , Sfrp5 and Wif1 . Additionally , it showed upregulated expression of other known hair cycle inhibitors Fgf18 ( Kimura-Ueki et al . , 2012; Leishman et al . , 2013 ) and Ctgf ( Liu and Leask , 2013 ) . 10 . 7554/eLife . 22772 . 008Figure 6 . WNT and BMP signalings modulate ear HF hyper-refractory state . ( A–C ) Transcriptomes of first telogen ear skin , first telogen dorsal skin and ear cartilage/muscle complex are distinct , as revealed by PCA analysis ( A ) . They contain 1334 differentially expressed genes ( B ) , spanning distinct gene ontologies ( C ) . ( D , E ) These tissues show differential expression of multiple ligands and antagonists for several major signaling pathways , prominently WNT and BMP . Putative activators are in green and putative inhibitors are in red . For each gene , relative fold changes for ear skin over dorsal skin and cartilage/muscle complex over dorsal skin expression levels are indicated . Select genes are highlighted . ( F , F’ ) BRE-gal reporter reveals high BMP activity in telogen ear HFs and in the adjacent cartilage/muscle complex ( n = 8 ) . ( G , G’ ) Axin2-lacZ reporter reveals near absence of WNT activity in ear HFs and cartilage/muscle complex . Seldom , sites of dermal reporter activity can be found ( n = 8 ) . ( H–H’’ ) Compared to wild type mice ( n = 4 ) ( H ) , ears of Krt14-Noggin ( n = 4 ) ( H’ ) and Krt14-Wnt7a mice ( n = 4 ) ( H’’ ) show prominent increases in spontaneous anagen frequency . Cumulative heatmaps from four individual ear samples are shown . Also see Appendix 1—figure 28 . Scale bars: F , G – 500 um; F’ , G’ – 100 um . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 00810 . 7554/eLife . 22772 . 009Figure 7 . Hair growth waves distort at the non-propagating boundaries . ( A , B ) Introduction of a non-propagating barrier ( A ) or an aperture ( B ) into the model produces simulations with distorted anagen spreading wave front ( green ) . ( C , E ) Distortions in the geometry of hair growth waves are commonly seen in the head region at the boundaries with the hyper-refractory ears and eyelids , the physical breaks in the skin . Seldom , similar distorted patterns can be seen around limb skin ( D ) . Hair growth patterns on C-E are accompanied by color-coded schematic drawings . Colors are defined at the bottom . Hair growth distortion patterns shown were documented in ten mice each . DOI: http://dx . doi . org/10 . 7554/eLife . 22772 . 009 We validated WNT and BMP changes from RNA-seq by studying relevant pathway reporters and measuring changes in ear hair cycling in mutant mouse models . Using Axin2-lacZ reporter mice , we show isolated sites of WNT activity in ear skin dermis , and a lack of activity in telogen HFs as well as in the cartilage and muscle ( Figure 6G and G’ ) . Using BRE-gal reporter mice , we show high levels of BMP activity in telogen ear HFs ( in the bulge ) , as well as in the cartilage and muscle ( Figure 6F and F’ ) . Overexpression of the BMP antagonist Noggin in Krt14-Noggin mice partially rescued the hyper-refractory state – substantially more spontaneous anagen HFs can be found in Krt14-Noggin ears as compared to wild-type control ( Figure 6H and H’; Appendix 1—figure 28B , D ) . Wnt7a overexpression in Krt14-Wnt7a mice also reactivated anagen in ear skin , albeit to a lesser extent compared to Noggin overexpression ( Figure 6H’’; Appendix 1—figure 28C , D ) . Together , these results support that hyper-refractivity of ear HFs depends on higher levels of BMP ligands and WNT antagonists , in part produced by the cartilage/muscle complex ( Figure 6E ) . Lastly , our model predicts that hair growth waves can form distorted patterns around non-propagating skin regions , such as hyper-refractory hair-bearing skin or hairless skin ( Figure 7A and B; Appendix 2-Hyper-refractory domain and Wave breaker; Appendix 2—figure 17; Appendix 2—video 5 ) . We considered that pattern distortion could occur in the cranial skin at the boundaries with hyper-refractory ears and eyelids – naturally occurring physical breaks in the skin . Indeed , we observe that hair growth waves prominently break around the eyelids and ears – anagen waves propagate faster through the hair-bearing skin around the eyelids and ears , and then distort into the spaces in front of these anatomical structures ( Figure 7C and E ) . Similar patterns are also observed for the ventral-to-dorsal hair growth wave around the limbs ( Figure 7D ) . We conclude that distortions of hair growth waves around anatomical structures with temporary or permanent non-propagating properties contribute to rapid body-wide hair growth pattern evolution .
Previous mathematical models have recapitulated cycling of a single HF ( Al-Nuaimi et al . , 2012; Halloy et al . , 2000 ) or in HF populations in two dimensions ( Murray et al . , 2012; Plikus et al . , 2011 ) . Here , we developed a multiscale model where coupling of activator and inhibitor signals with the movements of a HF in a three-dimensional space simulates cyclic growth and communication between neighboring HFs . In a single HF regime , our model faithfully predicts the effects that changes in WNT and BMP signaling can exert on the length of the anagen phase of the hair cycle . Similar to the FHN generic excitable media model ( Murray et al . , 2012 ) , our model also recapitulates several known population-level features of the HF system such as spontaneous hair growth initiation and hair wave spreading . Importantly , however , only our model allows incorporation of differential HF growth in space , a feature required for simulating heterogeneous skin properties such as interactions between skin domains with different hair cycle frequencies or the hair wave distortion effect . Thus , while the multiscale nature and non-linearity make our model more difficult to derive analytical results , its heterogeneous domain feature allows studying complex skin-wide hair growth dynamics ( see Appendix 2-Comparison with FitzHugh-Nagumo ( FHN ) model ) . Hair growth in newborn mice is commonly thought to occur simultaneously across the entire skin . In fact , we show that the first cycle is already distinctly patterned: at birth , anagen HFs in dorsal skin have head-to-tail and lateral-to-medial asynchronies , while first anagen entry by ventral HFs is delayed by approximately 3 days and proceeds as a concentric lateral-to-midline wave . Similarly delayed by 6 days are ear HFs . First anagen naturally follows the process of HF morphogenesis , which is known to be temporarily asynchronous , and to occur , at least in the dorsal skin , in three successive waves ( reviewed in Clavel et al . , 2012 ) . Pattern-wise , development of HFs relies on reaction-diffusion ( Sick et al . , 2006 ) and on space-filling expansion-induction mechanisms ( Cheng et al . , 2014 ) . Importantly , models for both mechanisms assume spatially synchronous HF morphogenesis . Our findings of spatial asynchrony of the first anagen indicate spatial asynchrony of HF morphogenesis . Future studies will be required to understand the modeling and signaling aspects of such phenomenon . Our data reveal prominent regional differences in hair cycle dynamics and show that interaction between HFs across domain boundaries drives rapid evolution of complex hair growth patterns . Specifically , we show that during early postnatal cycles , chin and ventral domains become the dominant sources of skin-wide anagen waves . Such dominant behavior of chin and ventral domains is accompanied by distinct activity dynamics for WNT and BMP , putative hair cycle activators and inhibitors , respectively . Transgenic mouse studies further confirm the functional importance of differential WNT and BMP activities in setting distinct hair growth pace across discrete anatomical skin regions . Admittedly , an in-depth follow-up study will be necessary to identify and verify the major site-specific cellular sources for WNT and BMP ligands and antagonists . We also show that ear skin behaves as a hyper-refractory domain , where telogen HFs are resistant to anagen-inducing stimuli and cannot participate in hair growth wave propagation . We reveal that such hyper-refractivity relates to high levels of BMP ligands and WNT antagonists , in part produced by the cartilage/muscle complex , a structure unique to the ear skin . Thus , novel behaviors can be produced by the cooption of signals from new tissue modules , rather than by the modification of preexisting ones . This finding parallels the modulatory effects of non-HF cell types on the dorsal skin hair cycle , including adipose progenitors ( Festa et al . , 2011; Rivera-Gonzalez et al . , 2016 ) , mature adipocytes ( Plikus et al . , 2008b ) , and resident macrophages ( Castellana et al . , 2014; Chen et al . , 2015 ) . Finally , we show that anatomically defined structures that cannot propagate hair growth waves , namely ears and eyelids , can generate a ‘wave-breaker’ effect . Similar distortion effects are likely to occur around other anatomical structures , such as the tail and genitals , and around skin defects , such as scars , and can jointly contribute to rapid hair growth pattern evolution . Taken together , our study reveals that the skin as a whole functions as a complex regenerative landscape with regions of fast , slow , and very slow hair renewal ( Appendix 1—figure 29 ) . We show that this behavior produces a fur coat with variable hair density , which likely serves an adaptive role , such as in thermoregulation . Mechanistically , we show that the WNT/BMP activator/inhibitor signaling pair modulates hair regeneration in all skin regions studied . This suggests that the WNT/BMP ‘molecular language’ for hair growth is general , rather than a special case for a specific body site . Its generality allows for hair-to-hair communications to arise across anatomic domain boundaries , which , in turn , enables novel hair growth dynamics not obvious from prior work – fast cycling skin regions ( such as chin skin ) function as a kind of hair growth pacemaker . Furthermore , our findings on ear hair cycle expand the repertoire of tissues with signaling macro-environment function to include any closely-positioned anatomic structures with signaling properties , such as cartilage . We posit that some of the newly found hair regeneration features can have analogs in other organs . For instance , dominant anatomically defined pacemakers are common in the electrically coupled muscle-based tissues , including heart and stomach , where they generate directional contractile rhythmicity . Other actively regenerating organs , such as the intestines and bone marrow , can likely contain anatomic regions of faster and slower regeneration and , conceivably , they can be coupled to work in coordination . Knowledge learned from the skin system in the current study can guide the search for regenerative landscapes in these and other organs . Because coordination principles observed in the skin may be universal , the likelihood of them operating in other organs is substantial despite prominent anatomical differences .
The modeling framework is based on a hybrid approach , with individual HFs modeled as an expanding or contracting one-dimensional line and with the diffusive molecules described in reaction-diffusion equations ( Appendix 2 , Equations 1–4 ) . The latter are solved using a finite difference scheme with the standard central difference approximation on the diffusion ( see Appendix 2-1-dimensional ( 1D ) HF model to Numerical methods in Appendix 2 ) . Krt14-Noggin ( Plikus et al . , 2004 , 2005 ) , Krt14-Bmp4 ( Guha et al . , 2004 ) , Krt14-Cre;Wnt7bfl/fl ( Kandyba and Kobielak , 2014 ) , Krt14-Wnt7a ( Plikus et al . , 2011 ) , Krt5-rtTA;tetO-Dkk1 ( Chu et al . , 2004 ) , 220bpMsx2-hsplacZ reporter ( Brugger et al . , 2004 ) , BRE ( Bmp response element ) -gal BMP reporter ( Javier et al . , 2012 ) , Axin2-lacZ ( Lustig et al . , 2002 ) , and Flash WNT reporter mice ( Hodgson et al . , 2014 ) were used . For Dkk1 induction , P30 Krt5-rtTA;tetO-Dkk1 mice were placed on 2 mg/ml Doxycycline-containing water ab libitum , and skin was collected at P44 for histology and at P50 for hair length measurements . 5 × 5 mm skin grafts from chin and dorsal domains of P21 C57BL/6J male mice were transplanted onto the dorsum of gender-matched pigmented P50 SCID recipients . At the time of grafting , donor skin was in first telogen and recipient skin was in second telogen . In dorsal skin , club hairs were plucked from 5 × 5 mm areas . In the ear pinna , plucking was done on the caudal skin . For quantitative plucking , approximately 500 club hairs we plucked along the medial ear side . Cyclosporin A: for the dorsal skin , 100 ul of Cyclosporin A solution ( 1 , 5 , and 10 mg/ml ) was applied topically once a day for 7 days . For the ear pinna , caudal skin was treated with 100 ul of 10 mg/ml of Cyclosporin A once a day for 7 days . Smoothened agonist ( SAG ) : for the dorsal skin , 120 uM of SAG in DMSO/acetone was applied topically once a day for 4 days as described ( Paladini et al . , 2005 ) . For the ear pinna , caudal skin was treated with 25 ul of SAG solution once a day for 4 days . Guard , awl , auchene and zigzag club hair types were photographed , traced and calibrated using Adobe Illustrator software . See Appendix 1—table 1 . Club hair density was evaluated on whole-mount telogen skin samples that were pre-treated with 1 mg/mL Collagenase/Dispase and counterstained with hematoxylin . Histology was performed on 4% PFA-fixed sections . For BRE-gal and Axin2-lacZ specimens , whole mount lacZ staining was performed first followed by histology . The primary antibodies used were rabbit anti-keratin Krt5 ( 1:250 , Abcam , UK ) , rabbit anti-perilipin ( 1:750; Cell Signaling ) , rabbit anti-αSMA ( 1:200; Abcam ) . Actin was detected with phalloidin ( Alexa Fluor 488; Molecular Probes ) . Whole body imaging of Flash mice was performed as previously described ( Hodgson et al . , 2014 ) . Briefly , mice were injected with 150 mg/kg of firefly D-luciferin substrate and imaged with the Xenogen IVIS Spectrum system . Second telogen skin from C57BL/6J male mice was treated with Dispase to separate epidermis from dermis . Epidermis was digested with Accutase and dermis with Collagenase . Epidermal and dermal cell suspensions were combined and stained with anti-CD11b ( eBioscience ) and anti-F4/80 antibodies ( eBioscience ) . Due to small tissue size , chin skin cells from three mice were combined for each experiment . FACS data were analyzed using FlowJo . Total RNA was isolated using the RNeasy Mini Kit ( Qiagen ) . RNA samples with RIN >8 . 0 were considered for cDNA library preparation . Full-length cDNA library amplification and tagmentation was performed as previously described ( Picelli et al . , 2014 ) . Libraries were multiplexed and sequenced as paired-end on an Illumina Next-Seq500 platform . Paired-end reads were aligned to the mouse genome ( mm10/gencode . vM8 ) and quantified using the RNA-seq by Expectation-Maximization algorithm ( RSEM ) with standard parameters ( version 1 . 2 . 25 ) ( Li and Dewey , 2011 ) . Samples were batch-effect corrected . EdgeR ( version 3 . 14 . 0 ) was employed to identify differentially expressed genes ( DEGs ) across samples of interest . FPKM values were taken as inputs for PCA analysis and DEG analyses . Data is available at GEO: GSE85039 . | Skin includes hundreds of thousands of hair follicles that cycle through different stages of activity . Each follicle grows hair , sometimes ( in the case of long hairs like human head hair and horse tail hairs ) for several years , before losing it . The follicle then goes through a resting stage before starting to grow another hair . To achieve high hair density , the follicles need to coordinate their hair-making activities . If they all worked independently from one another , bald patches would inevitably form that would compromise how effectively the skin works . Groups of cells can communicate using a variety of chemical signals . It was not known whether cells in hair follicles from different regions of the skin rely on the same signals to communicate , and whether follicles in neighboring regions are able to ‘understand’ one another . Through a combination of mathematical modeling and experimental results from mice , Wang , Oh et al . now show that hair follicles across the body use a common signaling system . This system consists of a pair of signals: ‘activators’ that stimulate hair growth , and ‘inhibitors’ that prevent it . The balance between these two signals affects the pattern of hair growth . For example , higher levels of activators allow fur to grow thickly on the belly of the mouse , likely to protect against heat loss and injuries from the ground . By contract , higher levels of inhibitors make the hairs on the ear sparse , which may prevent them from interfering with hearing . There is little evidence that hair follicles on the scalp communicate in adult humans . Learning to activate and control communication between these follicles could provide a way to treat male pattern baldness and similar conditions . Understanding how hair follicles communicate may also help researchers to develop ways of regenerating other fast-renewing organs , such as the gut and bone marrow . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2017 | A multi-scale model for hair follicles reveals heterogeneous domains driving rapid spatiotemporal hair growth patterning |
Recent sequencing studies have extensively explored the somatic alterations present in the nuclear genomes of cancers . Although mitochondria control energy metabolism and apoptosis , the origins and impact of cancer-associated mutations in mtDNA are unclear . In this study , we analyzed somatic alterations in mtDNA from 1675 tumors . We identified 1907 somatic substitutions , which exhibited dramatic replicative strand bias , predominantly C > T and A > G on the mitochondrial heavy strand . This strand-asymmetric signature differs from those found in nuclear cancer genomes but matches the inferred germline process shaping primate mtDNA sequence content . A number of mtDNA mutations showed considerable heterogeneity across tumor types . Missense mutations were selectively neutral and often gradually drifted towards homoplasmy over time . In contrast , mutations resulting in protein truncation undergo negative selection and were almost exclusively heteroplasmic . Our findings indicate that the endogenous mutational mechanism has far greater impact than any other external mutagens in mitochondria and is fundamentally linked to mtDNA replication .
All cancers result from somatic mutations in their genomes . Beyond the ∼3200 Mb of nuclear genomic DNA , human cells have hundreds to thousands of mitochondria present in every cell , each carrying one or a few copies of the 16 , 569 bp circular mitochondrial genomes ( Smeitink et al . , 2001; Legros et al . , 2004; Koppenol et al . , 2011 ) . In addition to their role in cellular energy balance through oxidative phosphorylation , mitochondria are involved in many essential cellular functions including modulation of oxidation–reduction status , contribution to cytosolic biosynthetic precursors , and initiation of apoptosis . Mitochondria in eukaryotic cells evolved by endosymbiosis from a free-living α-proteobacterium ( Gray et al . , 1999 ) . Over 2 billion years of co-evolution , many ancestral mitochondrial genes have transferred to the nucleus ( Falkenberg et al . , 2007; Calvo and Mootha , 2010; Wallace , 2012 ) . What remains in the mitochondrial genome is distinctive for the striking asymmetry between the two complementary mtDNA strands in terms of nucleotide content and gene distribution ( Andrews et al . , 1999 ) . The heavy ( H ) strand is guanine-rich ( C/G = 0 . 4 ) and is the template from which most mitochondrial proteins ( 12 out of 13 ) are transcribed , whereas only one protein-coding gene , MT-ND6 , is transcribed from the correspondingly cytosine-rich light ( L ) strand . Mutations in the mitochondrial genome cause inherited disease ( Chinnery , 1993 ) , with a maternal inheritance pattern because only eggs contribute mitochondria to the zygote . The penetrance of inherited mitochondrial disease is determined stochastically by both the random assortment of mutated vs wild-type mitochondrial genomes during meiosis and random drift during the early cell divisions after fertilization . In cancer , the role of somatically acquired mtDNA mutations is controversial . Although cancer-specific mutations have been previously reported ( Polyak et al . , 1998; Brandon et al . , 2006; Chatterjee et al . , 2006; He et al . , 2010; Larman et al . , 2012 ) , the limited sample size or poor sensitivity of capillary sequencing for heteroplasmic mutations has not allowed a comprehensive analysis of the mutational signatures of mitochondrial mutations nor their likely functional significance . It has long been proposed that mitochondria might contribute to cancer development given their fundamental importance to cellular biology ( Wallace , 2012 ) . Previous reports suggested that mitochondrial somatic mutations might be under positive selection and thus contribute to cancer development , but the small number of reported mutations renders this conclusion uncertain ( Brandon et al . , 2006; Chatterjee et al . , 2006; Larman et al . , 2012; Schon et al . , 2012 ) . Nonetheless , the hypothesis of functionally relevant mitochondrial mutations is an appealing one because cancer cells have greatly increased energy demands over normal cells and demonstrate a switch from aerobic glycolysis in mitochondria to lactic acid fermentation in the cytosol ( the Warburg effect ) ( Hanahan and Weinberg , 2011; Koppenol et al . , 2011 ) . In each cell cycle , the replicating genome is at risk of de novo mutations , which can promote the development of cancer . These mutations may be generated by intrinsic cellular errors during DNA replication or repair or through exposure to mutagens , such as reactive oxygen species , tobacco smoke , and ultraviolet light ( Pleasance et al . , 2010a , 2010b ) . Recently , >20 mutational signatures operative in cancers have been identified in the nuclear genome ( Alexandrov et al . , 2013 ) . Whether any of these mutational processes also affect the mitochondrial genome has not been studied . Furthermore , whether there are mtDNA-specific mutational processes in somatic cells remain unclear , although the many unique features of mtDNA replication and repair , coupled with the high concentration of reactive oxygen species generated by the electron transport chain , could be associated with distinctive mutation signatures . In this study , we compare 1675 cancer and paired normal mtDNA sequences across 31 tumor types using massively parallel DNA sequencing technologies to obtain a systematic and unbiased catalog of somatic mitochondrial mutations . We find that mtDNA mutations are almost exclusively the product of a mutational process that is specific to mitochondria and probably linked to the unique mechanism of genome replication these organelles employ . We find no evidence for positive selection of mitochondrial mutations during oncogenesis , suggesting that they confer no clonal advantage on the nascent cancer cells .
We extracted the mtDNA sequences from 704 whole-genome and 971 whole-exome sequencing data generated on primary cancers and compared them with mtDNA sequences from their matched normal samples . Given the abundance of mtDNA per cancer cell , a standard coverage of 30–40× in the nuclear genome provides significantly greater coverage of the mitochondrial genome ( average read depth = 7901 . 0× ) , enabling accurate identification of somatic mutations including rare heteroplasmic variants . We also assessed whether whole-exome sequencing could be used to identify mtDNA mutations from off-target reads derived from the mitochondrial genome . We found an average read depth of 92 . 1× across the mitochondrial genome in exome studies . From 139 samples in which we had both exome and whole-genome sequencing data , the overall read depths correlated strongly ( R2 = 0 . 59 , Figure 1—figure supplement 1 ) as did variant allele fractions for mtDNA somatic mutations ( R2 = 0 . 97 , Figure 1—figure supplement 2 ) . Validation experiments suggested the sensitivity of whole-exome sequencing for detection of mtDNA somatic mutations to be 71 . 4% compared to whole-genome sequencing ( Figure 1—figure supplement 3 and ‘Materials and Methods’ , ‘Off-target mtDNA reads in whole-exome sequencing’ and ‘DNA cross-contamination’ ) . To reduce potential false-positive calls of mtDNA somatic mutations , we only report variants called with an allele fraction of >3% . This eliminates the risk of miscalls due to mtDNA-derived pseudogenes in the nucleus ( NuMTs ) because mtDNA copy numbers are 100–1000 times higher than nuclear genomes in human somatic cells , and the sequence homology between mtDNA and NuMTs presented in the human reference genome is generally <95% ( in 96 out of 101 NuMTs with length greater than 300 bp ) . Furthermore , pairwise comparison between cancer and matched normal mtDNAs from the same individual further minimizes the contamination of NuMTs in the mutation calling . In total , 1675 tumor–normal pairs across 31 tumor types were analyzed ( Table 1 and Supplementary file 1 ) . For 61 of these patients , we had sequencing data available from multiple sites of the primary cancer , several time points or matched primary cancers , and metastases ( a total of 73 such cancer samples ) , allowing us to study the timing of mtDNA mutations in cancer evolution ( Supplementary file 1 ) . We identified 1907 somatic mtDNA substitutions ( Figure 1 and Supplementary file 2 ) . In contrast to inherited polymorphisms ( n = 38 , 706 , available at Supplementary file 2 ) , which were almost always homoplasmic in both the cancer and counterpart normal , the variant allele fractions ( VAFs ) of these somatic substitutions were highly variable in the cancer , ranging from our detection threshold ( 3% ) to homoplasmy ( 100% ) . Of these 1907 somatic substitutions , 1209 ( 63 . 4% ) were not registered in the databases of mtDNA common polymorphism ( Ingman and Gyllensten , 2006; Levin et al . , 2013 ) . In comparison , when we examined substitutions found in both the tumor and the normal samples from a patient , only 21 ( 0 . 05% ) were not registered in the polymorphism databases , a significantly different fraction from the tumor-only variants ( p < 10−10; Chi-squared test ) . We found 595 ( 31 . 2% ) recurrent mutations that can be collapsed onto 246 mtDNA positions , which is a 6 . 9-fold higher level of recurrence than expected by chance ( p < 10−10 ) . This suggests that the generation or fixation of mtDNA mutations is not random , but influenced by factors such as the underlying mutational process or positive selection . 10 . 7554/eLife . 02935 . 003Table 1 . Summary statistics of mtDNA sequence dataDOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 003WGSWXSAverage mt RD ( WGS ) Average mt RD ( WXS ) TotalWGSWXSAverage mt RD ( WGS ) Average mt RD ( WXS ) TotalBreast2849811594 . 352 . 7382Meningioma012-42 . 512Colorectal17534916 . 9276 . 676Ependymoma1910323 . 752 . 710Lung6002798 . 1-60Prostate80017810 . 6-80MPD121381517 . 010 . 9150Hepatocellular047-205 . 847MDS3755648 . 744 . 578Melanoma1313513 . 9353 . 526ALL646886 . 635 . 970Gastric013-184 . 113CLL605002 . 2-6Cholangiocarcinoma08-143 . 98AML166783 . 627 . 47Mesothelioma06-106 . 36Multiple myeloma069-43 . 269Bladder540646 . 2-54AMKL09-24 . 29Renal023-35 . 423Lymphoma04-99 . 54Ovarian038-58 . 938Uterine2723736 . 0149 . 550Osteosarcoma38909525 . 5119 . 2128Cervical052-85 . 252Chondrosarcoma047-99 . 147Adenoid cystic ca . 160714 . 775 . 661Ewing sarcoma027-69 . 527Head & Neck4331369 . 118 . 846Kaposi sarcoma09-181 . 09Chordoma16111240 . 082 . 127Total; 31 cancer types7049711675WGS , whole-genome sequencing; WXS , whole-exome sequencing; mt RD , mitochondrial read depth; MPD , myeloproliferative disease; MDS , myelodysplastic syndrome; ALL , acute lymphoblastic leukemia; CLL , chronic lymphoblastic leukemia; AML , acute myeloid leukemia; AMKL , acute megakaryoblastic leukemia . 10 . 7554/eLife . 02935 . 004Figure 1 . Mitochondrial somatic substitutions identified from 1675 Tumor–Normal pairs . mtDNA genes and intergenic regions are shown . The strand of genes is shown based on mtDNA strand containing equivalent sequences of transcribed RNA . Substitution categories ( silent , non-silent ( missense and nonsense ) , non-coding ( tRNA and rRNA ) , and intergenic ) are shown by the shapes of each substitution . Six classes of substitutions are presented color-coded . The substitutions on the H , and L strand ( when six substitutional classes were considered ) are shown outside and inside of mtDNA genes , respectively . Vertical axes for H and L strand substitutions represent the VAF of each variant . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 00410 . 7554/eLife . 02935 . 005Figure 1—figure supplement 1 . Correlation in amount of mtDNA reads between whole-genome and whole-exome sequencing . 139 DNA samples , either from tumors or bloods , sequenced by whole-genome sequencing were additionally sequenced by whole-exome sequencing . We compared the amount of mtDNA reads between whole-genome and whole-exome sequencing . As shown in this figure , we found strong positive correlation . * CGP; Cancer Genome Project , Wellcome Trust Sanger Institute , WUGSC; Washington University Genome Sequencing Center . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 00510 . 7554/eLife . 02935 . 006Figure 1—figure supplement 2 . Correlation of heteroplasmy levels between whole-genome and whole-exome sequencing . To validate the sensitivity and specificity of variant calling in this study , 19 tumor and normal pairs ( which were originally whole-genome sequenced ) were whole-exome sequenced and mtDNA variants were assessed independently . We correlated the heteroplasmic levels of 20 mutations detected in common . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 00610 . 7554/eLife . 02935 . 007Figure 1—figure supplement 3 . Validation of mtDNA somatic substitutions . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 00710 . 7554/eLife . 02935 . 008Figure 1—figure supplement 4 . Amount of off-target mtDNA reads across four sequencing centers . * CGP; Cancer Genome Project , Wellcome Trust Sanger Institute ( n = 855 ) , WUGSC; Washington University Genome Sequencing Center ( n = 140 ) , BCM; Baylor College of Medicine ( n = 85 ) , BI; Broad Institute ( n = 435 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 00810 . 7554/eLife . 02935 . 009Figure 1—figure supplement 5 . Filtering samples of potential DNA contaminations . ( A ) A histogram presenting potential sample swaps in tumor–sample pairs . ( B ) A histogram presenting potential minor DNA cross-contamination in tumor samples . Cross-contamination levels were considered in filtering substitutions ( see “Minor cross-contamination of DNA samples” section in Materials and Methods ) . ( C ) Histograms showing number of somatic substitutions overlapping with known inherited polymorphisms and ( D ) number of back mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 009 Of the 1675 cancer samples , 976 ( 58 . 3% ) harbored at least one somatic substitution and 521 ( 31 . 1% ) had multiple substitutions , ranging from 2 to 7 ( Figure 2A ) . In those with multiple substitutions , 72 pairs of mutations were sufficiently close to phase ( Nik-Zainal et al . , 2012b ) such that we could determine whether they were linked on the same mtDNA genome or were on different copies . We found that 45 ( 62 . 5% ) pairs of mutations were linked on the same mtDNA genome ( Supplementary file 3 and Figure 2—figure supplement 1 ) . Furthermore , of these linked mutations , 33 showed a clear temporal order: that is , one mutation was demonstrably sub-clonal to the other . This is rather unexpected , since each somatic cell has 100–1000 copies of the mitochondrial genome , and we might anticipate that random mutations would , on average , affect different copies . That many pairs of mutations are phased on the same mtDNA genome and yet show a clear sub-clonal relationship suggests that they occur sufficiently separated in time to allow the mitochondrial genome carrying the earlier mutation to drift towards a substantial fraction of all genomes in that cell before the second mutation occurs , consistent with a previous report ( De Alwis et al . , 2009 ) . 10 . 7554/eLife . 02935 . 010Figure 2 . mtDNA somatic substitutions of human cancer . ( A ) Number of somatic substitutions in a tumor sample . ( B ) Average number of somatic substitutions per sample across 31 tumor types . ( C ) Age of diagnosis and number of mtDNA somatic substitutions in breast cancers . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01010 . 7554/eLife . 02935 . 011Figure 2—figure supplement 1 . VAFs of phased somatic mtDNA substitutions . This figure presents VAF pairs between co-clonal , sub-clonal , and different strand mtDNA substitutions . We expect similar VAFs for co-clonal pairs; lower VAF in sub-clonal mutations compared to clonal ones; and sum of a VAF pair is equal or less than 1 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 011 The number of somatic mtDNA substitutions varied significantly according to tumor type ( p = 4 . 4 × 10−52 ) after correcting for confounding variables such as sequencing coverage: gastric , hepatocellular , prostate , and colorectal cancers had the highest number of mtDNA substitutions ( Figure 2B ) . In contrast , hematologic cancers ( acute lymphoblastic leukemia , myeloproliferative disease , and myelodysplastic syndrome ) had fewer mutations . Several possible explanations could underpin these differences across tumor types . It could be that the mutation rates differ across cell lineages; it could be that selection pressures shape the number of mutations; or the number of mtDNA genome generations could differ across cell lineages . Of these explanations , we believe that the second is unlikely because , as we shall see , positive selection is not a major component of mitochondrial mutations . Interestingly , we find a positive correlation between the number of mtDNA somatic mutations and age at diagnosis in breast cancers ( p = 0 . 0004; Figure 2C ) , in keeping with the idea that the number of mitochondrial generations is linked to mutation burden . The mutational burden of an established cancer represents the accumulated variation acquired in the lineage of cell divisions from fertilized egg to transformed cell and will include events acquired in normal development and homeostasis as well as those acquired during tumorigenesis ( Stratton et al . , 2009 ) . Interestingly , mtDNA mutations have been found at high rates in normal colonic crypt cells ( Taylor et al . , 2003; Ericson et al . , 2012 ) . Given that we find high burdens of mutations in colonic tumors as well , the differences we see across tumor types may arise from pre- or post-transformation differences in mtDNA burden across tissues . With respect to signatures of somatic substitutions , C > T and T > C transitions constituted 90 . 9% of all the 1907 substitutions ( Figure 1 ) among the six classes of possible base substitutions . To characterize this aggregated signature of mtDNA cancer specific mutations in more detail , we looked for the presence of mtDNA strand bias between the complementary H and L strands of mtDNA . The two main substitution classes showed an extreme level of mtDNA strand bias . 84 . 1% of the C > T transitions were on the H strand . This level of strand bias occurred despite the fact that cytosine is 2 . 4-fold less common on the H than the L strand , so the C > T substitution rate is 12 . 6-fold higher on the H strand . By contrast , 76 . 8% of the T > C transitions were on the L strand despite its lower thymine content ( 1 . 3-fold less than the H strand ) . This implies that the T > C mutation rate on the L strand is 4 . 2-fold higher than on the H strand . We then examined the sequence context in which these mutations occurred by examining the bases immediately 5′ and 3′ to the mutated bases . This generates 96 possible mutation classes ( the 6 substitution classes multiplied by the 16 combinations of immediate 5′ and 3′ nucleotides ) . Both C > T and T > C mutations showed highly distinctive sequence contexts . CH > TH substitutions ( i . e . C > T mutations on the H strand ) were enriched for the NpCpG trinucleotide context ( 8- to 15-fold more frequent than expected by chance; Figure 3A ) . By contrast , TL > CL substitutions ( i . e . T > C mutations on the L strand ) showed 5- to 8-fold enrichment in NpTpC . This strand-asymmetric mutational signature is not similar to any of the 21 cancer-associated mutational signatures recently identified from the nuclear DNA of 30 different cancer types ( Alexandrov et al . , 2013 ) . 10 . 7554/eLife . 02935 . 012Figure 3 . Replicative strand bias for mtDNA somatic substitutions . ( A ) Replicative strand-specific substitution rate ( # of observed/# of expected ) by 96 trinucleotide context . Substitutions in a specific mtDNA segment ( from Ori-b to OH ) are not included , because they present a different substitutional signature . ( B ) Mutational signature across tumor types . Eighteen tumor types , which include at least 25 mtDNA mutations , were shown . ( C ) Inverted substitution signature in the Ori-b–OH . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01210 . 7554/eLife . 02935 . 013Figure 3—figure supplement 1 . Replicative strand bias observed in mtDNA substitutions . ( A ) Mutational signature of mtDNA somatic substitutions on the 12 L strand genes by replicative strand ( L/H strand ) . It agrees very well with the background mutational signature . ( Chi-square p = 0 . 99999 ) . ( B ) Mutational signature of mtDNA somatic substitutions on the H strand gene ( MT-ND6 ) by replicative strand . It is very close to the background very close to the expected background signature ( Chi-square p = 0 . 027 ) . If we consider signature by transcriptional strand , the signature difference is very clear ( Chi-square p = 1 × 10−21 ) . These suggest the strand bias not to be transcription-coupled , but replication coupled . ( C ) Mutational spectrum of mtDNA somatic substitutions on the 22 tRNA genes by replicative strand . Again , it agrees very well with the background mutational signature ( Chi-square p = 0 . 71 ) . ( D ) Mutational spectrum of mtDNA somatic substitutions on the 22 tRNA genes by non-transcribed ( coding ) and transcribed ( non-coding ) strand . Strand bias was greatly subsided because somatic substitutions on 14 L strand and 8 H strand tRNAs neutralize the strand bias ( CH > TH and TL > CL ) each other . As a result , this signature of tRNA mutations by transcriptional strand is significantly different from the background one ( Chi-square p = 3 . 3 × 10−12 ) . Taken all together , we concluded that the cause of strand bias is not transcription-coupled but is replicative . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 013 Of the 18 tumor types that presented at least 25 mtDNA somatic substitutions in this study , the mutational signatures were broadly consistent across tumor types ( Figure 3B ) , with the exception that multiple myeloma had a somewhat higher rate of TH > CH changes than other histologies ( p = 8 . 1 × 10−6 ) . Thus , in contrast to the mutational signatures found in nuclear genomes , where there is striking heterogeneity both across tumor types and across individuals within a tumor type ( Alexandrov et al . , 2013 ) , the mutational profile in the mitochondrial genome of somatic cells is remarkably homogeneous . The major known cause of mutational strand bias in nuclear DNA is transcription-coupled nucleotide excision repair , where DNA lesions on the transcribed ( non-coding ) strand are more frequently repaired ( Alexandrov et al . , 2013 ) . However , we find that the strand bias always favors CH > TH and TL > CL whether the gene is transcribed from the H strand or from the L strand ( Figure 3—figure supplement 1 ) . This is not compatible with transcription-coupled repair , for which the direction of strand bias is fundamentally dictated by which strand is transcribed . Instead , the mtDNA mutational strand bias reported here appears to be driven by differences in replication between the two strands . mtDNA replication harbors substantial strand asymmetry between the H and L strands: mtDNA replication initiates from an origin of replication ( OH ) in the D-loop , with the nascent H and the L strand replicating as leading and lagging strand , respectively ( Clayton , 1982; Falkenberg et al . , 2007; Holt and Reyes , 2012 ) . We observed that C > T substitutions were prevalent in the leading ( heavy ) strand , whereas T > C substitutions were found in the lagging ( light ) strand ( Figure 1 ) . Remarkably , this strand bias was reversed in the D-loop itself ( Figures 1 and 3C ) , further suggesting that the mtDNA somatic mutations are replication-coupled: according to a recently proposed bidirectional model of mtDNA replication ( Yasukawa et al . , 2005 , 2006; Holt and Reyes , 2012 ) , mtDNA replication is also able to initiate from the so-called Ori-b site , typically located around genomic position 16 , 197 and proceeds on both strands away from the origin ( Figure 1 ) . Replication of the nascent H strand continues unimpeded like the traditional model , but the nascent L strand terminates at the so-called OH site , typically around mtDNA position 191 bp . Under this model , then , the leading and lagging strand are reversed in the few hundred base-pairs of the D-loop , which is consistent with the reversed mutational signature in this region ( Figures 1 and 3C ) . It is not entirely straightforward to infer the mutational signatures operating on the mitochondrial genome in the germline . De novo mutations are generally rare and often discovered because they cause disease; distinguishing the ancestral base and the derived base is challenging for single nucleotide polymorphisms; and comparative mtDNA genomics across species extends over considerable evolutionary time . In contrast , because ancestral and derived states are defined for tumor–normal pairs , a much clearer picture emerges of the somatic mtDNA mutation signature . We therefore assessed whether the signature that emerges for somatic mitochondrial mutations could extend to explain sequence composition of the human mtDNA genome . It appears that the mutational mechanism which has generated the CH > TH and TL > CL signature in cancer mtDNA is equivalent to the one that has been operating during evolution of human germline mtDNA ( Nikolaou and Almirantis , 2006 ) . This manifests as the depletion of certain codons in the reference human mtDNA sequence through the action of the CH > TH and TL > CL mutational process over time ( Figure 4A ) . For example , the GCG triplet codon ( Alanine ) appears to have been replaced by its synonymous GCA codon ( due to CH > TH ( GL > AL ) ) , with the former being 15 . 8-fold less frequently observed in the 12 mtDNA protein-coding genes that are transcribed from the H strand ( and encoded on the L strand ) . All 32 synonymous codon pairs present the same tendency . Consistent with this interpretation , the gene transcribed from the L strand ( MT-ND6 ) demonstrates the opposite direction of skew . Further analyses of mtDNA codon usage from seven animal species suggest that the CH > TH and TL > CL mutational pressure may be characteristic of vertebrates , and primates in particular ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 02935 . 014Figure 4 . Mutational signature similar to processes shaping human mtDNA sequence over evolutionary time . ( A ) Triplet codon depletion in human mtDNA by equivalent ( CH > TH and TL > CL ) mutational pressure . Relative frequency of each triplet codon within synonymous pairs ( NNT–NNC or NNA–NNG ) is shown by color . The arrows beside the box highlight the T > C ( red ) and G > A ( blue ) substitutional pressures on the L strand in germline mtDNA . ( B ) Correlation of triplet codon frequencies between from observed and from simulated evolutions of a random sequence mtDNA by the mtDNA somatic mutational signature with constraining mitochondrial protein sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01410 . 7554/eLife . 02935 . 015Figure 4—figure supplement 1 . TC and GA skew for L strand mtDNA genes across 8 animal species . C . elegans ( a nematode ) and D . melanogaster ( fruit fly ) mtDNA appears to have GL << AL ( due to CH > TH mutational pressure ) and CL >> TL ( due to CL > TL mutational pressure ) in the third base of triplet codon in L strand genes . Therefore they seem to have predominant C > T mutational pressure without strand bias . D . rerio ( zebrafish ) , X . laevis ( frog ) , and M . musculus ( mouse ) present GL << AL ( due to CH > TH mutational pressure ) , but similar number of CL and TL . Therefore , mtDNA of these sequences is thought to have CH > TH , with strand bias . The existence of TL > CL is not clear . Finally , mtDNA of H . sapiens , P . troglodytes ( Chimpanzee ) , and G . domesticus ( Chicken ) shows clear CH > TH and TL > CL as mentioned in the main manuscript . Interestingly , TL > CL seems to be slightly stronger in the mitochondria of chicken than that of human ( or chimp ) . We suggest there would be some differences in the mechanism of mtDNA replication across the evolution tree . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01510 . 7554/eLife . 02935 . 016Figure 4—figure supplement 2 . Correlation of triplet codon frequencies between from observed and from simulated evolutions under the mtDNA somatic mutational signature . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 016 To quantify whether the somatic mutational signature we have defined can fully explain the trinucleotide frequency of human mtDNA , we performed evolutionary simulations . First , we simulated the evolution of a random DNA sequence under the mutational signature described here . By mutational pressure alone , the random sequence starts losing certain hypermutable trinucleotides until eventually reaching a stationary sequence composition . The actual sequence composition of the human mitochondrial genome strongly resembles this stationary distribution ( Pearson's r = 0 . 83; p < 0 . 0001; Figure 4—figure supplement 2 ) . In a second simulation , a random sequence encoding the exact amino acid sequence of the reference mitochondrial genome was evolved by synonymous mutations under the observed mtDNA signature until reaching a stationary sequence composition ( mutation–selection equilibrium ) . These simulations also eventually approximate the observed human mitochondrial genome ( Pearson's r = 0 . 96 , p < 0 . 0001; Figure 4B ) . These analyses strongly suggest that the mitochondrial mutation signature observed in cancer cells closely reflects the mutation signature active in the germline , which has continuously shaped the mitochondrial genome during human evolution . Next , we assessed the functional impact of somatic mtDNA mutations . Of the 1907 substitutions , 1153 ( 60 . 5% ) were in the 13 protein-coding genes . These include 63 nonsense , 4 stop-lost , 878 missense , and 208 silent substitutions ( Supplementary file 2 ) . In addition , out of 251 indels we observed , 110 occurred within protein-coding genes ( Supplementary file 2 ) . Of the missense substitutions , 245 ( 27 . 9% ) were recurrent , affecting 107 distinct mtDNA sites . Although this very high level of mutation clustering could , at first sight , be interpreted as evidence for positive selection , we found that silent substitutions were also frequently recurrent ( 28 recurrent variants in 13 mtDNA sites ) , with no substantial difference in recurrence rates between silent and missense mutations ( p = 0 . 19; Figure 5—figure supplement 1 ) . We believe this recurrence to be the consequence of a high mtDNA mutation rate with restricted mutational signature ( CH > TH and TL > CL ) . Independently recurring mutations in human germline mtDNA are well described across human evolution ( Levin et al . , 2013 ) . The ratio of somatic missense to silent substitutions ( Rms:s ) is apparently higher ( 4 . 2 , 878/208 ) than that observed for cancer-associated somatic mutations in nuclear DNA ( generally around 2:1 to 3:1 across tumor types ) ( Greenman et al . , 2007; Nik-Zainal et al . , 2012a ) . At face value , this again could be interpreted as evidence for positive selection . However , as described above , the somatic mtDNA mutational signature shows extreme strand asymmetry and the same mutational signature has been operative in the germline over evolutionary time . Thus , the dominant mutational signature has already acted on potentially synonymous sites in the mitochondrial genome ( Figure 4A ) , meaning that any new somatic changes are much less likely to be silent . In keeping with this , a dN/dS ratio ( See ‘Materials and Methods’ ) calculated taking into account both the mutational signature and the mtDNA codon usage revealed that missense mutations accumulate at a frequency very close to that expected under neutrality ( dN/dS = 1 . 21; 95% confidence interval , 1 . 015–1 . 434; p = 0 . 031 ) . This indicates that despite the apparent high ratio of missense to silent mutations , the vast majority of mtDNA mutations are passengers with no convincing evidence suggesting the existence of driver mitochondrial DNA mutations . Additional gene-by-gene analysis further revealed that no single gene had a higher than expected rate of missense or nonsense mutations ( Supplementary file 4 ) . For nonsense substitutions and frameshift indels , we observe a somewhat different picture . Taking into account the mutation signature and amino acid composition of the mitochondrial genome , the overall ratio of nonsense mutations to silent mutations is exactly that expected by chance ( dNonsense/dS = 1 . 004; 95% confidence interval , 0 . 699–1 . 443; p = 0 . 98 ) . However , while missense and silent substitutions exhibited equivalent variant allele fractions ( average VAFs; 40 . 1% and 40 . 9% , respectively; p = 0 . 8 ) , nonsense substitutions presented significantly lower VAFs ( average 26 . 4%; p = 6x10−5 ) , as did frameshift indels ( average 25 . 0%; p = 2 × 10−3; Figure 5A ) . Taken together , these data suggest that nonsense mutations occur at the expected rate given the underlying mutational process . However , while silent and missense substitutions frequently achieve high allele fractions in tumor cells due to the effects of random drift , there are significantly greater constraints on mitochondrial genomes carrying protein-inactivating mutations . The inference here is that cancer cells carrying such deleterious mutations at or near homoplasmy are at a selective disadvantage and hence do not contribute to clonal expansions , underlining the importance of functional mitochondria to cancer cells . The extent of such disadvantage may vary according to tumor type: for example colorectal cancers show less negative selection compared to breast cancers ( p = 0 . 028; Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 02935 . 017Figure 5 . Selection and mutational process for mtDNA somatic substitutions . ( A ) Truncating mutations ( nonsense substitutions and frame-shifting ( FS ) coding indels ) present significantly lower VAF . ( B ) Change of VAF of mtDNA somatic mutation between primary and metastatic ( or late ) cancer tissues . ( C ) Mutational signature for mtDNA across various tumor types . None of the three highlighted mechanisms or nuclear DNA double-strand breaks repair mechanism ( BRCA ) match with the mtDNA mutational signature . * Only substitutions in protein-coding genes considered . ( D ) A proposed model of mtDNA mutational process . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01710 . 7554/eLife . 02935 . 018Figure 5—figure supplement 1 . Number of recurrent substitutions between silent and missense substitutions . 100 sites were randomly selected from silent substitutions ( at third base of triplet codon ) and missense substitutions ( at first and second base of triplet codon ) . No significant difference was observed among these three groups . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01810 . 7554/eLife . 02935 . 019Figure 5—figure supplement 2 . Comparison of VAF of protein-truncating mutations ( nonsense substitution and indels ) across tumor types . Four tumor types with more than 10 protein-truncating mutations are shown . Fisher's exact were applied between breast and other tissue types . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 01910 . 7554/eLife . 02935 . 020Figure 5—figure supplement 3 . Negligible impacts of external mutagens ( UV and tobacco smoking ) to the somatic mtDNA mutations . No evidence of UV and tobacco smoking was identified even in melanoma and lung cancers , respectively . ( Left ) We compared the proportion of C > T ( and G > A ) substitutions in the CpC ( GpG ) context ( mutational signature for UV [Alexandrov et al . , 2013] ) between melanomas and breast cancers ( controls ) . Because UV shows trivial impact to the nuclear DNA somatic mutations of breast cancers ( Alexandrov et al . , 2013 ) , the vast majority of mtDNA C > T substitutions in the CpC context from breast cancers were not generated by UV . ( Right ) We compared the proportion of C > A ( G > T ) substitutions between lung and breast ( control ) cancers . C > A ( G > T ) substitutions are dominantly generated by tobacco smoking . Like UV , the impact of tobacco smoking to the somatic mutations of breast cancers is trivial ( Alexandrov et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02935 . 020 We found 171 mtDNA substitutions in mitochondrial tRNA sequences , which are very similar to the expected number ( 168 . 2 , p = 0 . 82 ) from the mutational signature . Interestingly , none of the substitutions was located in the trinucleotide anticodon site of the tRNA ( expected number = 7 . 6 , p = 0 . 006 ) . This suggests that mutations in tRNA anticodons confer a similar selective disadvantage as protein-truncating mutations , presumably because such mutations would lead to systematic erroneous aminoacylation of nascent proteins during translation of the relevant codon . Next , we assessed whether any specific somatic mutations showed evidence of positive selection . Out of the 1907 somatic substitutions , 16 ( 0 . 8% ) overlapped with known disease-associated mtDNA mutations , such as mutations frequently detected in MELAS ( Mitochondrial Encephalomyopathy , lactic acidosis , and stroke-like episodes ) and LHON ( Leber hereditary optic neuropathy ) ( Supplementary file 2 ) . In addition , ten mutations within mitochondrial protein-coding , tRNA and rRNA genes showed significantly higher recurrent rate than expected from background mutational signature ( Supplementary file 5 ) . However , it remains unclear whether this high recurrence reflects positive selection , because any factors not included in our background model of the mutational process , such as local mutation hotspots , could also explain a mild excess of mutations at a given nucleotide . We investigated whether somatic mtDNA mutations are more likely to become homoplasmic later in tumor evolution by assessing paired cancer samples , either primary and metastasis ( breast , colorectal , and prostate ) or primary and relapse ( myeloma ) ( Figure 5B and Supplementary file 1 ) . As mentioned earlier , 73 late ( metastasis or relapse ) cancer samples were sequenced in addition to the primary tissues . Among the mtDNA mutations identified in either of the paired cancer samples , a number of different patterns were observed . There were mutations at high VAF in the primary not found in the metastasis ( n = 49 ) ; mutations in the metastasis not found in the primary ( n = 49 ) ; and shared mutations ( n = 71 ) at high or low VAF , sometimes with evidence for drift ( VAF difference >0 . 2 ) between the two samples ( n = 25 ) . These data , particularly the mutations found in the metastasis only , suggest that mitochondrial mutations can occur throughout the time course of tumor evolution , and still drift to homoplasmy with appreciable frequency , as suggested previously ( Coller et al . , 2001 ) . To assess the plausibility of this conclusion , we modeled the dynamics of mtDNA mutations based on a few simplifying assumptions ( See ‘Materials and Methods’ , Evolutionary dynamics of neutral mitochondrial mutations ) . We find that the expected number of neutral mitochondrial mutations drifting to homoplasmy increases linearly with mutation rate and number of cell divisions . Based on a mutation rate of 10−7/base-pair/generation ( Coller et al . , 2001; Hudson and Chinnery , 2006 ) , this leads to an average ∼1 homoplasmic mutation for every 1000 cell generations . We also explored whether the mutational forces that are so critical to shaping the nuclear genome during tumor evolution could affect the mitochondrial genome . In cancers associated with exogenous mutagens , such as tobacco-associated lung cancer and ultraviolet light-associated melanomas , we found no evidence of the mutational signatures characteristic of these carcinogens among the mtDNA mutations ( Figure 5C , Figure 5—figure supplement 3 ) . Moreover , BRCA1 and BRCA2 mutations showed no evident influence on mitochondrial genomes in breast cancer ( Figure 5C ) , in contrast to their effects on nuclear genomes exhibiting an even distribution of mutations across all trinucleotide contexts ( Nik-Zainal et al . , 2012a; Alexandrov et al . , 2013 ) . Taken together , it appears that the primary mtDNA mutational process is endogenous to mitochondria and is very different to those operating in nuclear DNA . It is surprising that the endogenous mutational process has far greater impact than any external forces , as the physicochemical interactions of ultraviolet light or the chemicals in cigarette smoke with DNA should be similar in both genomes . The simulations described above suggest the major explanation to be that the endogenous mutation rate is several orders of magnitude greater than that expected for exogenous carcinogens , thus swamping any signal .
In theory , there are two potential sources of the mtDNA variants we observed in cancer tissues: ( 1 ) somatically acquired , or de novo , mutations accumulated during the cancer clone's lineage of cell divisions from the fertilized egg or ( 2 ) low-level heteroplasmic mtDNA present in the oocyte ( therefore maternally inherited ) amplified in cancer but lost from normal tissue by random drift ( He et al . , 2010; Freyer et al . , 2012; Payne et al . , 2013 ) . We believe the majority of the variants we find are genuinely acquired somatically . First , of the 45 pairs of somatic mutations phased together on the same copy of the mtDNA genome , at least 33 ( 73 . 3% ) showed a clear sub-clonal relationship and therefore their occurrence is separated in time , or apparently somatic . Secondly , 63 . 4% of our substitutions were not previously reported as germline polymorphisms . This is a much higher rate than reported for equivalent analyses on heteroplasmic variants in non-cancer samples ( 8/37; 21 . 6% ) ( Li et al . , 2010 ) , although methodological differences may somewhat contribute to this apparent difference ( Goto et al . , 2011; Avital et al . , 2012 ) . Thirdly , if the variants were due to inherited low-level heteroplasmy , we would not expect to see such variation across tissue types , since all tissue types derive from the fertilized egg . It is difficult to distinguish whether the variants we observe occur before or after the initiating driver mutations that herald tumorigenesis , but our analysis of paired samples does suggest that they can occur both early and late . Given the homogeneity of the mutational signature across tumor types and its inferred resemblance to the germline mtDNA mutational process , we would hypothesize that new mutations occur at a fairly constant and high rate per mitochondrial genome replication throughout all cell divisions . On the basis of the mutational signature observed here , somatic substitutions are unlikely to be attributable to reactive oxygen species ( ROS ) , as previous reports have suggested ( Polyak et al . , 1998; Larman et al . , 2012 ) . Guanine oxidation by ROS predominantly causes G:C > T:A transversion ( Thilly , 2003; Delaney et al . , 2012 ) , which constitute only 4 . 0% of the mutations in our data ( Figure 5C ) . Instead , we propose three replication-coupled mechanisms that can explain the strand asymmetric CH > TH and TL > CL mutational signature and define a model of the mtDNA mutational process ( Figure 5D ) . First , the parent H strand , displaced and single-stranded during mtDNA replication ( Holt and Reyes , 2012 ) , could be more prone to cytosine deamination ( generating CH > TH ) and/or adenine deamination ( Lindahl , 1993; Saccone et al . , 1999; Faith and Pollock , 2003 ) ( generating TL > CL ) . Secondly , endogenous mtDNA polymerase ( POLG ) replication errors ( Zheng et al . , 2006 ) ( which show the pattern of C > T and A > G substitutions ) could be preferentially generated on the leading strand ( Pavlov et al . , 2002 ) . Thirdly , there may be differences between the efficiency of repair between the leading and lagging strand ( Pavlov et al . , 2003 ) . Further , the mutation pattern reported here is consistent with the hypothesized bidirectional initiation of mtDNA genome replication ( Yasukawa et al . , 2005 , 2006; Holt and Reyes , 2012 ) . It appears that most of the mtDNA missense mutations we observe become fixed in tumor progenitor cells without distinct physiological advantage . All the statistical testing performed in this study—variant allele fraction comparison across different categories of somatic mutations , number of recurrent mutations , and dN/dS ratio—suggest that mtDNA somatic substitutions accumulate largely neutrally . This is not different from previous observations in nuclear genomes: of the thousands of somatic mutations found in a cancer genome , many fewer than a hundred are believed to confer a selective advantage to the cancer cell ( Stratton et al . , 2009 ) . In contrast , protein-truncating mutations showed evidence of negative selection , at the level of constraints on the allele fraction achieved . The implication of this is that the inactivating mutations occur at an appreciable rate , but the fraction of mitochondrial genomes per cell carrying these variants cannot increase beyond a certain limit without impairing the selective fitness of that cell . Having a sizable number of mitochondria with fully intact proteome remains critical to the fitness of a cancer cell .
All the sequences were generated by Illumina platforms ( either Genome Analyzer or HiSeq 2000 ) . With respect to TCGA data , we downloaded aligned bam files through UCSC CGHub ( http://cghub . ucsc . edu ) . Sequencing reads were aligned on the human reference genome build 37 ( GRCh37 ) and human reference mtDNA sequence ( revised Cambridge reference sequence , rCRS [Andrews et al . , 1999] ) , mainly by BWA alignment tool . Samtools ( Li and Durbin , 2009 ) and Varscan2 ( Koboldt et al . , 2012 ) were used for manipulating sequence reads and for calling somatic mutations , respectively . Sequence data have been deposited in the European Genome-phenome Archive ( EGA; https://www . ebi . ac . uk/ega/home ; study accession # EGAS00001000968; dataset accession numbers EGAD00001001014 for primary samples and EGAD00001001015 for metastatic samples ) . Sample accession numbers are available in the Supplementary file 6 . Most of the currently available whole-exome capture kits , including Agilent Technologies SureSelect Human All Exon 50 Mb ( Agilent Technologies Inc . , Santa Clara , CA ) used mostly in this study , do not target mtDNA genes ( Falk et al . , 2012 ) . However , because of the abundance of mtDNA in human cells ( 100–100 , 000 copies per cell ) , it is expected that a number of mtDNA fragments could be off-target captured . We checked whether the amount of off-target mtDNA reads was sufficient for mtDNA variant detection . Whole-exome sequencing ( normal samples ) generated by CGP ( n = 855 ) , WUGSC ( Washington University Genome Sequencing Center; n = 140 ) , and BCM ( Baylor College of Medicine; n = 85 ) contained ∼100 off-target mtDNA reads per 1M autosomal reads ( Figure 1—figure supplement 4 ) . We concluded that these could be sufficient for the downstream analyses , because ordinary 10 Gb whole-exome data would provide ∼60× read depth for mtDNA here . However , whole-exome data sequenced by BI ( Broad Institute; n = 436 ) included far less , ∼3 off-target mtDNA per 1 M autosomal reads , which would show ∼2× mtDNA read depth per 10 Gb exome sequencing ( Figure 1—figure supplement 4 ) . It may be due to ‘improved’ exome-capture protocols by BI to increase the DNA-capture efficiency and on-target rate ( Fisher et al . , 2011 ) . Therefore , we did not include whole-exome data sequenced from BI for further analysis . 139 samples were sequenced by both whole-genome and whole-exome sequencing . From these , we compared the amount of off-target mtDNA reads from whole-exome sequencing with that of whole-genome sequencing . It showed clear positive linear correlation ( Figure 1—figure supplement 1 ) . Given the abundance of mtDNA in the cancer cells , 1–214× coverage cancer whole nuclear genome sequencing provides extensive coverage of mtDNA ( average read depth = 7901 . 0×; Supplementary file 2 ) enabling accurate identification of somatic mutations , even if heteroplasmic . Whole-exome sequencing data were also included because off-target reads provided sufficient coverage ( average read depth = 92 . 1× ) to analyze mtDNA mutations . This high coverage of mtDNA , especially from whole-genome sequencing , permitted us to identify heteroplasmic variants ( our detection threshold was 3%; see ‘Variant calling’ for more details ) . However , because sample swaps and/or DNA cross-contaminations would definitely generate false-positive somatic variants , we filtered out suspicious DNA samples as described below . 1 . Major sample swapsA subset of tumor and normal sequencing pairs , of which the nuclear genotypes were not matching with each other , were removed from further analyses . We randomly selected 320 common single-nucleotide polymorphism sites on the 22 human autosomes , of which the minor allele frequency is ∼50% ( 45–55% ) according to The 1000 Genomes Project ( 1000 Genomes Project Consortium et al . , 2010 ) . Of the 320 sites , homozygous positions in normal tissues ( which showed >90% variant allele fraction ( VAF ) with bases Q score >20 ) were compared with the corresponding genotypes in the counterpart cancer . Sample pairs were removed if the genotype mismatch rate was greater than 0 . 1 ( Nhet+NwtNhom+Nhet+Nwt; Nhet , number of heterozygote positions; Nhom , number of homozygote positions; Nwt , number of wild-type positions ) ( Figure 1—figure supplement 5A ) . We note 0 is expected for the rate when genotyping is perfect and sample pairs are from the same individual . By contrast , 0 . 5 is expected when samples were from different individuals . 2 . Minor cross-contaminationWe estimated DNA cross-contamination levels with the VAF of autosomal homozygous SNPs genotyped from the common ( population minor allele frequency ∼50% ) SNP sites . Theoretically , if there is no sequencing ( and mapping ) error , all the homozygote SNP sites in pure samples should present 100% VAFs . However , when samples are contaminated , corresponding VAFs are reduced because the contaminant has only an ∼25% of chance of having homozygote SNPs on the same site . Therefore , minor contamination levels ( C ) of each cancer sequencing data were estimated as below:C=2×∑ ( RCwt ) −Ne∑ ( RDhom ) −Ne , where RDhom is sequencing read depth , RCwt is read-count of wild-type alleles , and Ne is number of sequencing errors on each autosomal homozygote SNP site . For high accuracy , we only counted base with sufficient quality score ( Q > 20 ) . In order to estimate Ne , we assumed a conservative rate ( sequencing error rate = 0 . 001 ) . We considered sites covered by at least 10 reads and 90% VAF ( Figure 1—figure supplement 5B ) . 95% confidence intervals of cross-contamination levels were calculated using binomial distribution . In order to clear somatic variants , here we made the very conservative assumption that somatic variants present in excess of 5-times of the 95% upper limit of C levels were true somatic rather than false-positives by low-level of cross-contamination . 3 . Germline polymorphisms and back mutationsWe further checked samples for contamination using known mtDNA polymorphisms . Because human mtDNA is small ( 16 , 569 bp ) and extensively explored previously , most of germline mtDNA polymorphisms are already known . For example , 97 . 7% of the 39 , 036 inherited substitutions were known polymorphisms in the mtDB database ( Ingman and Gyllensten , 2006 ) . Therefore , when a tumor sample is contaminated by other samples , many somatic-like mtDNA substitutions by contaminants are likely to be overlapped with known mtDNA polymorphisms . At the same time , low-level contamination would generate excessive back mutations , which appeared to reverse germline common polymorphisms into wild-type alleles . Taken together , both the number of somatic substitutions known in mtDB and number of back mutations can be good indicators for mtDNA cross-contamination . Therefore , we filtered out tumor tissues with ≥3 known potentially somatic mutations or with ≥2 back mutations from the further analyses ( Figure 1—figure supplement 5A and B ) . We extracted mtDNA reads using Samtools ( Li and Durbin , 2009 ) . We used VarScan2 ( Koboldt et al . , 2012 ) for initial variant calling with a few options ( --strand-filter 1 ( mismatches should be reported by both forward and reverse reads ) , --min-var-freq 0 . 03 ( minimum VAF 3% ) , --min-avg-qual 20 ( minimum base quality 20 ) , --min-coverage 3 and --min-reads2 2 ) . With respect to the --strand-filter , it generally removes variant when >90% of mismatches are reported from either of the H or the L mtDNA strand . However , where only reads with a specific orientation are could be aligned dominantly ( i . e . in both extreme region of mitochondrial reference genome; only L strand reads could be aligned on the 5′ extreme of mtDNA ) , we compared strand bias between ‘perfect matches’ ( # perfect matches from L strand reads / total # perfect matches ) and mismatches ( # mismatches from L strand reads / total # mismatches ) . If the difference between those two bias <0 . 1 , the mutations were rescued . Of the 1907 mutations , 54 ( 2 . 8% ) were rescued accordingly . Putative somatic variants called by VarScan2 were further filtered using criteria shown below . At least 4 unique reads supporting variants and all variant reads at least 20 phred scale sequencing quality score ( Q 20 = 1% sequencing error rate ) and at least 3% variant allele fractions ( VAFs ) . A . Regardless of in WGS and in WES , the ≥4 mismatches and the ≥3% VAF criteria must be satisfied simultaneously . B . However , in WGS , the minimum number of reads ( n = 4 ) criterion is not essential , because the ≥3% VAF criterion is much more stringent ( 3% VAF request at least 240 mismatches ( >>4 ) given mtDNA coverage is ∼8000 for WGS ) . C . In WES , the ≥3% VAF criterion is relatively less important than in WGS , because the ≥4 mismatches criterion is more stringent . For example , 4 mismatches in 90x ( WXS average ) coverage region ( VAF = 4 . 4% ) automatically fulfill the ≥3% VAF criterion . For less covered regions ( i . e . <40x coverage; n = 285 out of total 1907 substitutions ) , the VAF criterion becomes less important , because 4 mismatches would generate ≥10% VAF , much higher than the minimum threshold ( i . e . 3% ) . As results , we are missing lower heteroplasmic variants ( i . e . variants with 3–10% heteroplasmic levels ) from low coverage samples ( mostly by WXS ) . The lower sensitivity of WXS is also confirmed in our validation study ( see “Validation of somatic variants” below ) . There is no minimum threshold for total coverage ( # perfect matches + # mismatches ) . To increase sensitivity for detecting mutations , we rescued mutations with 3 unique variant reads ( with at least 20 phred scale sequencing quality score ) when VAFs is ≥ 20% . Of 1907 somatic substitutions , 32 ( 1 . 7% ) were rescued accordingly . All somatic variants presenting with VAFs lower than our very conservative threshold for minor cross-contamination ( 5-times 95% upper limit of contamination levels for each tumor sample , see above “Minor cross-contamination of DNA samples” ) were removed . When we could not estimate cross-contamination levels because of low sequencing depth of coverage ( for nuclear genome ) , a conservative criterion ( 10% contamination level threshold ) was explicitly used . Substitutions were further visually inspected using IGV ( Thorvaldsdottir et al . , 2013 ) . Thirteen frequent false-positive variants ( shown below ) by misalignment due to extensive level of homopolymers in rCRS and due to sequencing error in the reference mtDNA genome ( 3107N , see Mitomap ( http://www . mitomap . org/bin/view . pl/MITOMAP/CambridgeReanalysis ) for more information ) were explicitly removed:1 . Misalignment due to ACCCCCCCTCCCCC ( rCRS 302-315 ) A302C , C309T , C311T , C312T , C313T , G316C2 . Misalignment due to GCACACACACACC ( rCRS 513-525 ) C514A , A515G , A523C , C524G3 . Misalignment due to 3107N in rCRS ( ACNTT , rCRS 3105-3109 ) C3106A , T3109C , C3110A We compared our variant calls with common inherited mtDNA polymorphisms deposited in the mtDB database as of 24th July 2013 ( Ingman and Gyllensten , 2006 ) . Gene annotation of somatic variants was done using custom script based on human mtDNA gene information ( Ruiz-Pesini et al . , 2007 ) . To validate the sensitivity and specificity of variant calling in this study , 19 tumor and normal pairs ( which were originally whole-genome sequenced ) were whole-exome sequenced and mtDNA variants were assessed independently . Among the 28 somatic substitutions originally detected from the 19 tumor–normal whole-genome sequencing pairs , 20 ( 71 . 4% ) were called as somatic ( Figure 1—figure supplement 3 ) . In addition , 5 ( 17 . 9% ) presented evidence of variant reads in the validation set , although it was filtered out because of its low read depth of coverage in exome sequencings ( showed 2–5 variant reads ) . Moreover , because 3 remaining sites were not sufficiently covered in the validation set to call somatic variants , these could not be evidence of the inaccuracy of whole-genome sequencing data , therefore not considered in the accuracy validation . Taken together , all the 25 somatic substitutions by whole-genome sequencing were highly likely to be true positives , therefore we concluded it provided ∼100% accuracy in the mtDNA somatic substitution assessment . Actually , the high accuracy of whole-genome sequencing is very likely and what we expect , because it provides extensive coverage of mtDNA ( average read depth >7 , 500× ) , ∼3% heteroplasmic variants would present >200 variant reads . By contrast , the validation set ( whole-exome sequencing ) is called 21 somatic substitutions . Of these , 20 were common with whole-genome sequencing , and one was incorrectly called as somatic though it was actually germline substitutions in the whole-genome sequencing data . In addition , as mentioned above , the validation set missed 8 somatic substitutions called by whole-genome sequencing . Six out of eight undercalls ( 75% ) were low heteroplasmic substitutions in whole-genome sequencing , ranging from 3 . 36% to 8 . 68% . Based on these data , we suggest 71 . 4% sensitivity ( 20/28 ) and 95 . 2% specificity ( 20/21 ) for exome-sequencing in detecting upto 3% heteroplasmic somatic mtDNA substitutions in cancer . We further checked the correlation of heteroplasmy level between the 20 mtDNA somatic mutations called both whole-genome and whole-exome sequencing . It showed great linear relationship ( R2 = 0 . 97 , Figure 1—figure supplement 2 ) , further suggesting whole-exome sequencing data are appropriate for accurate detection of mtDNA somatic mutations . We phased 72 somatic substitution pairs , which arose in a single cancer sample and which located sufficiently close ( from 10 bp to ∼500 bp ) , therefore both sites could be sequenced by same sequence fragments ( Supplementary file 3 and Figure 2—figure supplement 1 ) . We classified them as ‘different strand’ , ‘co-clonal’ , and ‘sub-clonal’ using criteria as follows: Different strand: the two somatic substitutions are obligate on different strands . Reads that report wild-type1 ( wt ) -substitution2 ( subs ) and subs1-wt2 , but subs1-subs2 , are observed . Co-clonal: reads reporting wt1-wt2 and subs1-subs2 are only observed . Sub-clonal: One substitution is sub-clonal to the other , but the two are definitely phased . Reads subs1-subs2 and either subs1-wt2 or wt1-subs2 are observed . To understand the relationship between tumor types and number of mtDNA mutations , Poisson regression and ANOVA were applied to our dataset using R software ( http://www . r-project . org ) . Fit1<-glm ( Nsub∼CovT+CovN , family=poisson ( ) ) Fit2<-glm ( Nsub∼CovT+CovN+t , family=poisson ( ) ) anova ( Fit1 , Fit2 , test=“Chisq” ) , where Nsub is number of mtDNA substitutions of each sample , CovT and CovN are coverage of tumor and normal mtDNA , respectively , ( if Cov is >200 , we replaced it by 200 ) , t is tumor types . Poisson regression was applied to our breast cancer dataset . Fit1<-glm ( Nsub∼CovT+CovN+a , family=poisson ( ) ) where Nsub is number of mtDNA substitutions of each sample , CovT and CovN are coverage of tumor and normal mtDNA , respectively , ( if Cov is >200 , we replaced it by 200 ) , a is age at diagnosis . p-value in estimation of a was shown in the manuscript . Different mutational processes generate different combinations of mutation types , termed ‘signatures’ ( Nik-Zainal et al . , 2012a ) . For example , ultraviolet ( UV ) light and tobacco smoking ( polycyclic aromatic hydrocarbons ) frequently generate C > T transitions and G > T transversions on non-transcribed ( coding ) strands in melanoma and lung cancers , respectively ( Pleasance et al . , 2010a , 2010b ) . To understand the mutational processes influencing cancer mtDNA , we correlated the 1907 mtDNA substitutions with 21 cancer specific mutational signatures in the nuclear DNA recently identified ( Alexandrov et al . , 2013 ) . However , none of the signature could explain the highly unique mtDNA substitutions . Mutational signature and strand bias were assessed as described in our previous reports ( Alexandrov et al . , 2013 ) . Briefly , the immediate 5′ and 3′ sequence context was extracted from rCRS . Substitution rate for each trinucleotide context was calculated with the number of substitution normalized by the frequency of the trinucleotide context observed in the rCRS , in the L and H strand , respectively . For analyses of substitutions falling in the mtDNA genes ( 13 protein-coding and 22 tRNA genes ) , transcribed/non-transcribed strand was also considered for comparison . In order to prove the strand bias is not transcription but replication-coupled , we checked strand biases of polymorphisms in the 12 L strand protein-coding genes , 1 H strand protein-coding gene ( MT-ND6 ) , and/or 22 tRNAs ( Figure 3—figure supplement 1 ) . For this specific purpose , we did not consider the sequence context ( immediate 5′ and 3′ bases ) because it over-classifies mutations ( i . e . the number of mutation classes ( n = 96 ) is larger than that of mutations ) . In other words , 12 classes of substitutions ( six classes of possible base substitutions ( C > A , C > G , C > T , T > A , T > C , T > G ) × two strands ( L and H strands ) ) were considered . Substitution rates are ratio between observed and expected numbers ( H0 = same mutation rate for all substitution classes ) for each substitution class . In order to understand which model ( replicative or transcriptional strand ) is appropriate to explain the strand-bias , Chi-square tests were used between the number of observed mutations for each class and expected ones under the background signature . We counted the codon frequencies in 13 mtDNA protein-coding genes . Because 12 L strand protein-coding genes and 1 H strand gene ( MT-ND6 ) are under opposite mutational pressure ( T > C and G > A for L strand genes; A > G and C > T for MT-ND6 ) , we separated L and H strand genes for this analysis . T > C skew and G > A skew were calculated as shown below , to understand the TL > CL and CH > TH ( equivalent to GL > AL ) substitutions during the evolution of human mtDNA:T>Cskew=NC−NTNC+NT and G>Askew=NA−NGNA+NG , where NA , NC , NG , and NT are number of A , C , G , and T base in the 3rd position of triplet codons in mtDNA genes , respectively . For the assessment of mtDNA codon usage of other animal species , we analyzed the mtDNA sequence of Caenorhabditis elegans ( accession# NC_001328 ) , Drosophila melanogaster ( accession# NC_001709 ) , D . rerio ( accession# NC_002333 ) , Xenopus laevis ( accession# NC_001573 ) , Mus musculus ( accession# EU450583 ) , Gallus domesticus ( accession # NC_235570 ) , and Pan troglodytes ( NC_001643 ) . We considered only L strand mtDNA genes in the cross-species analysis . To compare the number of recurrent substitutions between silent and missense substitutions , we randomly selected 100 substitutions each from 198 silent substitutions in the third base of triplet codons , 440 missense substitutions in the first base of triplet codons , and 405 missense substitutions in the second base of triplet codons . We counted the number of recurrent substitutions in each group . This was iterated 300 times independently . ANOVA testing was applied to determine the difference between the three groups ( Figure 5—figure supplement 1 ) . To estimate dN/dS values for missense mutations ( wmis ) , we used an adaptation of the method described previously ( Greenman et al . , 2006 ) . Briefly , the rate of mutations is modeled as a Poisson process , with a rate given by a product of the mutation rate and the impact of selection . To obtain accurate estimates of dN/dS , we used two separate models , one using 12 single-nucleotide substitution rates and a more complex one accounting for any context dependence effect by 1-nucleotide upstream and downstream using 192 substitution rates . For example in the 12-rate model , the expected number of A > C mutations ( λA>C ) would be modeled as follows:λsyn , A>C=rA>C*Lsyn , A>Cλmis , A>C=rA>C*wmis*Lmis , A>Cwhere Lsyn , A>C and Lmis , A>C are the number of sites that can suffer a synonymous and missense A > C mutation , respectively , which are calculated for any particular sequence . The likelihood of observing the number of missense A > C mutations ( Nmis , A>C ) given the expected λmis , A>C is then calculated as:Lik=Poisson ( Nmis , A>C | rA>C , wmis ) and the likelihood of the entire model is the product of all individual likelihoods . Wmis is fixed to be equal in all 12 ( or 192 ) equations describing each substitution type , and a hill-climbing algorithm is used to find the maximum likelihood estimates for all rate and selection parameters . Likelihood Ratio Tests are then used to test deviations from neutrality ( wmis = 1 ) . The dN/dS ratio reported in the main text corresponds to the full context dependent model with 192 substitution rates . This method allows quantifying the strength of selection avoiding the confounding effect of gene length , sequence composition , different rates of each substitution type , and context-dependent mutagenesis . Along with the 1907 somatic mtDNA substitutions , we identified 109 and 142 somatic short insertions and deletions , respectively , from the 1675 cancer mtDNA sequences using Varscan2 ( Supplementary file 2 ) . We model the evolutionary dynamics of mitochondrial mutations under random drift and derive a simple equation for the expected number of homoplasmic mutations . There exist multiple levels at which mitochondrial mutations evolve: within mitochondria , in the cytoplasm , and on the cellular level ( Rand , 2011 ) . In this study , we focus on the dynamics in a single cell , which represents the founder of the last clonal expansion in the tumor cell population . The cellular dynamics during a clonal expansion is difficult to describe analytically , but it is important to realize that mutations of a clonal expansion preserves the allele frequencies of neutral variants and that mutations that occur after the expansion are unlikely to contribute to measurable allele frequencies , as the population becomes large . We model the evolutionary dynamics of mitochondrial mutations in the cytoplasm of a single cell by a Wright–Fisher process ( Wright , 1931 ) , in which the number of mitochondria in a subsequent generation is a binomial sample of the mitochondria in the previous generation . The number of mitochondria M is kept fixed . The marginal allele frequency X of a single site has two absorbing boundaries , X = 0 and X = M ( homoplasmy ) , and the probability of fixation of an allele at frequency X by neutral drift is ρ = X/M ( Wright , 1931 ) . Note that this process leads , on the population level , to a dichotomization of heteroplasmic variants to either go extinct or become homoplasmic and fixate in a cell . Mutations on any of L ( = 16 , 569 nt ) sites in the mitochondrial genome are assumed to occur at a uniform rate μ per nucleotide per cell division , which is of order 10−7 , based on a human inter-generational comparison ( Coller et al . , 2001 ) . Hence the rate of neutral evolution is simply μLM/M = μL ( Kimura , 1984 ) . Lastly , the expected time to fixation in the Wright–Fisher process is t = 2M . Putting these things together , the expected number of mutant alleles N in a cell initially without any mitochondrial mutations after T generation isE[N]=μ L ( T−2M ) This equation predicts a linear accumulation of neutral mutations over time , with a delay imposed by number of mitochondrial copies . A similar behavior has been reported using numerical simulations ( Coller et al . , 2001 ) . When also considering heteroplasmic mutations , the expected number of alterations may be slightly higher . To check whether our model yields the correct behavior , we use the following numbers: the observed order of magnitude of mitochondrial mutations per patient was N = 1 . The sequencing coverage on the mitochondrial genome indicates that there were of order M = 100 mitochondrial genome copies present per cancer cell . The expected number of mutations per cell division is μL = 1 . 6 × 10−3 , it therefore requires around 1000 cell generations T to accumulate on average one homoplasmic mutation . This number of generations appears realistic for regenerating tissues . As expected , epithelial cancers had among the highest observed number of mitochondrial mutations , while hematopoietic cancers typically had lower numbers . Statistical testing was performed using R software . All p-values were calculated by two-tailed testing . Figures were generated using R and Microsoft Excel software . | The DNA in a cell's nucleus must be copied faithfully , and divided equally , when a cell divides to produce two new cells . Mistakes—or mutations—are sometimes made during the copying process , and mutations can also be introduced by exposing DNA to damaging agents known as mutagens , such as UV light or cigarette smoke . These mutations are then maintained in all of the descendants of the cell . Most of these mutations have no impact on the cell's characteristics ( ‘passenger mutations’ ) . However , ‘driver mutations’ that allow cells to divide uncontrollably and spread to other body sites can lead to cancer . Mitochondria are cellular compartments that are responsible for generating the energy a cell needs to survive and are also responsible for initiating programmed cell death . Mitochondria contain their own DNA—entirely separate from that in the nucleus of the cell—that encodes the proteins most essential for energy production . Mitochondrial DNA molecules are frequently exposed to damaging molecules called reactive oxygen species that are produced by the mitochondria . Therefore , these reactive oxygen species have been thought to be one of the most important causes of mitochondrial DNA mutations . In addition , because cancer cells produce energy differently to normal cells , mutations in the mitochondrial DNA that change the ability of the mitochondria to produce energy have been conventionally thought to help normal cells to become cancerous . However , conclusive evidence for a link between cancer and mitochondrial DNA mutations is lacking . Ju et al . examined the mitochondrial DNA sequences taken from 1675 cancer biopsies from over thirty different types of cancer and compared these to normal tissue from the same patients . This revealed 1907 mutations in the mitochondrial DNA taken from the cancer cells . The pattern of the mutations suggests that the majority of the mutations are not introduced from reactive oxygen species , but from the errors the mitochondria themselves make in the process of duplicating their DNA when a cell divides . Unexpectedly , known mutagens , such as cigarette smoke or UV light , had a negligible effect on mitochondrial DNA mutations . Contrary to conventional wisdom , Ju et al . found no evidence that the mitochondrial DNA mutations help cancer to develop or spread . Instead , like passenger mutations found in the DNA in the cell nucleus , most mitochondrial genome mutations have no discernible effect . However , Ju et al . revealed that DNA mutations that damage normal mitochondrial activity are less likely to be maintained in cancer cells . Presumably , mitochondria containing these proteins produce less energy , and so a cell containing too many of these mutations will find it harder to survive . This shows that having enough correctly functioning mitochondria is essential for even cancer cells to thrive . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology",
"genetics",
"and",
"genomics"
] | 2014 | Origins and functional consequences of somatic mitochondrial DNA mutations in human cancer |
Amyloid precursor protein ( APP ) is enriched at the synapse , but its synaptic function is still poorly understood . We previously showed that GABAergic short-term plasticity is impaired in App knock-out ( App-/- ) animals , but the precise mechanism by which APP regulates GABAergic synaptic transmission has remained elusive . Using electrophysiological , biochemical , moleculobiological , and pharmacological analysis , here we show that APP can physically interact with KCC2 , a neuron-specific K+-Cl- cotransporter that is essential for Cl- homeostasis and fast GABAergic inhibition . APP deficiency results in significant reductions in both total and membrane KCC2 levels , leading to a depolarizing shift in the GABA reversal potential ( EGABA ) . Simultaneous measurement of presynaptic action potentials and inhibitory postsynaptic currents ( IPSCs ) in hippocampal neurons reveals impaired unitary IPSC amplitudes attributable to a reduction in α1 subunit levels of GABAAR . Importantly , restoration of normal KCC2 expression and function in App-/- mice rescues EGABA , GABAAR α1 levels and GABAAR mediated phasic inhibition . We show that APP functions to limit tyrosine-phosphorylation and ubiquitination and thus subsequent degradation of KCC2 , providing a mechanism by which APP influences KCC2 abundance . Together , these experiments elucidate a novel molecular pathway in which APP regulates , via protein-protein interaction with KCC2 , GABAAR mediated inhibition in the hippocampus .
APP is a type I single pass transmembrane protein highly expressed in the central nervous system ( CNS ) which is processed by α- , β- , and γ-secretases ( Hardy and Selkoe , 2002 ) to generate beta amyloid ( Aβ ) peptide fragments that comprise the amyloid plaques found in Alzheimer’s disease ( AD ) patients post-mortem . The deleterious impact of Aβ has been studied extensively , with the recent discovery of the potent toxicity of Aβ oligomers on neuronal and synaptic activities ( Hsieh et al . , 2006; Kim et al . , 2006; Wei et al . , 2010 ) . Despite advances in our understanding of how Aβ generation drives AD progression , the physiologic role of APP has proven more difficult to elucidate ( De Strooper and Annaert , 2000 ) . APP has been detected in vesicular fractions of dendrites and axons ( Schubert et al . , 1991 ) , suggesting a role for APP in synaptic activity . Moreover , we have shown that APP is required for normal adult neurogenesis ( Wang et al . , 2014a ) and formation of the neuromuscular junction ( Wang et al . , 2009; Yang et al . , 2007 ) . Many lines of evidence have suggested that APP function can impact the electrophysiological properties of neurons ( Kamenetz et al . , 2003; Klevanski et al . , 2015; Palop and Mucke , 2010; Priller et al . , 2006; Selkoe , 2002; Wang et al . , 2014a; Yang et al . , 2009 ) . Paired-pulse inhibition of GABAergic IPSCs is significantly reduced in App-/- hippocampal slices ( Seabrook et al . , 1999 ) . However , the molecular mechanism underlying this phenotype is not fully understood . Given that GABAergic function is commonly disturbed in many neuronal disorders including AD ( Braat and Kooy , 2015a; Verret et al . , 2012 ) , identifying mechanisms whereby APP regulates GABAergic signaling and synaptic inhibition may provide a link between the endogenous function of APP and the etiology of AD . The K+-Cl- cotransporter , KCC2 , is broadly expressed in neuronal membrane of the adult CNS ( Kaila et al . , 2014 ) . KCC2 functions in setting the proper intracellular Cl- concentration ( [Cl-]i ) by transporting Cl- against the concentration gradient ( Ben-Ari , 2002; Ben-Ari et al . , 2012 ) . KCC2 maintains low [Cl-]i in mature neurons , which is essential for maintaining proper postsynaptic inhibition mediated by GABAA receptors ( GABAARs ) and glycine receptors ( GlyRs ) ( Braat and Kooy , 2015a; Kaila et al . , 2014; Rivera et al . , 1999 ) . Impaired KCC2 activity and subsequent increases in [Cl-]i occur in several neurological disorders ( Boulenguez et al . , 2010; Coull et al . , 2003; Tang et al . , 2016 ) , leading to a depolarizing action of GABAAR mediated currents due to a positive shift of EGABA , and over-excitation of neuronal network activity ( Ben-Ari , 2002; Ben-Ari et al . , 2012; Blaesse et al . , 2009 ) . Recent experiments on KCC2 processing suggest that the intrinsic ion transport rate , cell surface stability , and trafficking of plasmalemmal KCC2 are rapidly and reversibly modulated by the ( de ) phosphorylation of critical serine and tyrosine residues at the C-terminus of this protein ( Lee et al . , 2010 , 2007 ) . Increased tyrosine phosphorylation of KCC2 leads to enhanced degradation of KCC2 and thus reduced KCC2 protein levels ( Lee et al . , 2011 , 2010 ) . Therefore , abundance of KCC2 protein and the EGABA of a neuron can be regulated by phosphorylation and degradation of KCC2 . However , which signaling pathways regulate these phosphorylation events and KCC2 degradation are largely unknown . In the current study , we aimed to elucidate the synaptic mechanisms underlying APP regulation of hippocampal GABAergic inhibition . We show that APP mediates GABAergic inhibition via a direct protein-protein interaction with KCC2 , which stabilizes KCC2 on the cell membrane . APP deficiency causes a loss of acting force holding KCC2 on site , resulting in increased KCC2 degradation via both tyrosine-phosphorylation and ubiquitination , decreased KCC2 levels and , sequentially , a depolarizing shift of EGABA and impairment in GABAAR activities .
In order to test the idea that loss of APP changes the cellular ionic balance of Cl- , thereby altering GABAergic transmission , we used gramicidin perforated patch clamp measurements in acute hippocampal brain slices ( Figure 1A ) that allows maintenance of an intact intracellular chloride concentration to estimate the IV-relationship of GABAergic currents ( Chavas and Marty , 2003 ) . We observed a positive shift of EGABA in App-/- hippocampal CA1 compared to WT controls ( WT , −83 ± 6 mV; App-/- , −64 ± 5 mV; p=0 . 02 ) ( Figure 1B–D ) , accompanying with identical resting membrane potentials ( WT , −61 ± 1 . 4 mV; App-/- , −59 ± 1 . 5 mV; p=0 . 2 ) . We repeated perforated patch recordings in hippocampal cultures and observed a similar depolarizing shift of EGABA in App-/- neuronal cultures ( WT , −82 . 5 ± 4 mV; App-/- , −67 ± 3 . 8 mV; p=0 . 02 ) ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 20142 . 003Figure 1 . GABA reversal potential shifts toward depolarization in hippocampus of App-/- mice . ( A ) Slice preparations were made from the hippocampus and neurons were recorded in CA1 ( inset ) . Pipettes were positioned to clamp neurons ( Rec ) while puffing molecules onto the cell ( Puff ) . Scale bar: 10 µm . ( B ) Sample traces showing perforated patch recording of EGABA in WT and App-/- hippocampal slices . Currents were recorded at the indicated holding potentials shown to the left of each trace . ( C ) Graph shows I–V plots for the traces shown in B . ( D ) Quantification of EGABA shows a significant depolarizing shift in App-/- mice compared to WT controls ( WT , n = 16 cells from five mice; App-/- , n = 20 cells from five mice ) . *p<0 . 05; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 00310 . 7554/eLife . 20142 . 004Figure 1—figure supplement 1 . GABA reversal potential shifts toward depolarization in App-/- hippocampal cultures . ( A ) Sample traces showing perforated patch recordings of EGABA in WT and App-/- mice hippocampal cultures . Currents were recorded at the indicated holding potentials shown to the left of each trace . ( B ) Graph shows I–V plots for the traces shown in A . ( C ) Quantification of EGABA shows a significant depolarizing shift in App-/- mice compared to WT ( WT , n = 14 cells from three independent cultures , 2–3 mice/per culture; App-/- , n = 12 cells from three independent cultures , 2–3 mice/per culture ) . *p<0 . 05; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 004 Loss of APP impairs GABAergic synaptic transmission ( Seabrook et al . , 1999; Yang et al . , 2009 ) . Given that we see a change in EGABA in hippocampal neurons , we next asked whether loss of APP alters hippocampal IPSCs and GABAAR expression . To do so , we performed dual whole-cell recordings in primary hippocampal cultures . We simultaneously evaluated electrically evoked presynaptic action potentials of a GABAergic interneuron labeled with GAD67+/GFP , while patching a neighboring glutamatergic neuron and recording post-synaptic unitary IPSCs ( uIPSCs ) at a holding potential of −60 mV ( Figure 2A ) . The mean uIPSC amplitude , evoked by a single presynaptic action potential in the GABAergic neuron , was significantly reduced in App-/- hippocampal cultures compared to WT control cultures ( WT , 178 ± 42 pA; App-/- , 80 ± 19 pA; p=0 . 03 ) ( Figure 2B–C ) . Similarly , puffing the GABAAR agonist isoguvacine significantly reduced the amplitude of inhibitory currents in App-/- neurons compared to WT neurons ( WT , 245 ± 26 pA; App-/- , 146 ± 24 pA; p=0 . 01 ) ( Figure 2F–G ) . In contrast , the paired-pulse ratio ( PPR ) of uIPSCs in response to two consecutive action potentials in GABAergic neurons , at an inter-spike interval of 100 ms and 150 ms to assess synaptic release probability ( Zucker and Regehr , 2002 ) , was similar between WT and App-/- ( Figure 2D–E ) ( 100 ms; WT , 0 . 47 ± 0 . 04; App-/- , 0 . 56 ± 0 . 03; p=0 . 2; 150 ms; WT , 0 . 51 ± 0 . 06; App-/- , 0 . 55 ± 0 . 05; p=0 . 75 ) . Consistent with impaired post-synaptic function accounting for these changes in APP mutants , we did not see a decrease in miniature IPSC ( mIPSC ) frequency , a measure of presynaptic efficacy ( Fiszman et al . , 2005; Goswami et al . , 2012 ) , when we compared App-/- and WT hippocampal slices ( Frequency; WT , 2 . 9 ± 0 . 8 Hz; App-/- 2 . 6 ± 0 . 7 Hz; p=0 . 8 ) ( Figure 2—figure supplement 1C–D ) . Meanwhile mIPSC amplitude remained unchanged as well ( Amplitude; WT , 14 . 8 ± 0 . 4 pA; App-/- , 13 . 9 ± 0 . 8 pA; p=0 . 7 ) ( Figure 2—figure supplement 1C–D ) . In line with similar frequencies of mIPSCs in WT and App-/- hippocampus , we observed identical numbers of GABAergic neurons in the hippocampus of App-/- and WT mice ( Figure 2—figure supplement 1A–B , Figure 2—source data 1 ) . Finally , we looked at the abundance of GABAARs in the hippocampus of APP mutant and WT animals . App-/- hippocampal lysates display reduced immunoreactivity of the α1 subunit , mediating fast inhibition , of the GABAAR but not the other GABAAR subunits , ( Figure 2H–I , Figure 2—source data 1 ) . Together , these data suggest that loss of APP may cause postsynaptic deficits in GABAAR mediated GABAergic inhibition . 10 . 7554/eLife . 20142 . 005Figure 2 . Decreased inhibitory IPSC amplitude and GABAAR α1 protein levels in App-/- mice . ( A ) Images of 12–14 DIV hippocampal cultures of Gad67+/GFP mice . In the upper panel , under fluorescent illumination , GABAergic neurons appear green , and were stimulated ( Sti ) while recording from nearby glutamatergic cells ( Rec ) and puffing solutions ( puff , lower panel DIC ) . Scale bar: 10 µm . ( B ) Sample traces showing postsynaptic uIPSC responding to presynaptic action potentials induced by short depolarizing voltage pulse injection ( 2 ms ) to GABAergic neurons . ( WT , black line; App-/- , red line ) . ( C ) Quantification of uIPSC amplitude shows a significant decrease in App-/- mice . ( WT , n = 18 cells from five mice; App-/- , n = 17 cells from five mice ) . ( D ) Sample traces showing uIPSC recordings responding to injection of paired pulses to presynaptic GABAergic neurons ( 100 ms interpulse interval , left trace and 150 ms interpulse interval , right trace ) . ( E ) Quantification of paired pulse ratio ( PPR ) of uIPSCs with 100 ms and 150 ms interpulse intervals shows no significant difference between genotypes . ( F ) Sample traces showing evoked inhibitory currents responses to puffing 100 µM isoguvacine . ( G ) Quantification of isoguvacine-evoked inhibitory current amplitudes shows a significant decrease in App-/- mice . ( WT , n = 14 cells from three mice; App-/- , n = 14 cells from three mice ) . ( H ) Representative immunoblots of hippocampal extracts from WT and App-/- littermates . ( I ) Quantification of the immunoblots reveals a significant decrease of GABAAR α1 , but not other GABAAR subunits , levels in App-/- mice . Representative immunoblots of western blotting were from single experiment using three pairs hippocampal lysates , two repeats . *p<0 . 05; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 00510 . 7554/eLife . 20142 . 006Figure 2—source data 1 . Contains source data for Figure 2 and all accompanying Figure 2—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 00610 . 7554/eLife . 20142 . 007Figure 2—figure supplement 1 . Similar numbers of GABAergic interneurons and mIPSC in WT and App-/- hippocampus . ( A ) Images of hippocampal sections from App-/--Gad67+/GFP and WT-Gad67+/GFP littermates . Scale bar: 400 µm; Enlarge: 20 µm . ( B ) Quantification shows no difference in the number of GABAergic interneuron between WT and App-/- hippocampus . Each value represents the mean±SEM of at least three sections per genotype , two mice/genotype . ( C ) Sample traces showing mIPSCs from App-/- and WT littermates . ( D ) Quantification of the amplitude and frequency of mIPSCs shows no significant difference between WT and App-/- mice ( WT , n = 16 cells from three mice; App-/- , n = 15 cells from three mice ) . Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 00710 . 7554/eLife . 20142 . 008Figure 2—figure supplement 2 . Identical mEPSC and GluRs levels in between WT and App-/- hippocampus . ( A ) Sample traces showing mEPSCs from App-/- and WT littermates . ( B ) Quantification of the amplitude and frequency of mEPSCs shows no significant difference between WT and App-/- mice ( WT , n = 7 cells from three mice; App-/- , n = 9 cells from three mice ) . ( C ) Representative immunoblots of hippocampal extracts from WT and App-/- littermates . ( D ) Quantification of the immunoblots reveals identical GluR1 and GluR2 levels in between WT and App-/- hippocampus . Representative immunoblots were obtained using three pairs of hippocampal lysates of two independent experiments . Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 008 The electrical response of GABAARs depends on the Cl- equilibrium potential . The best characterized effectors of the Cl- distribution in the CNS are KCC2 and the Na-K-2Cl cotransporter ( NKCC1 ) ( Ben-Ari , 2002; Kaila et al . , 2014; Lee et al . , 2011; Rivera et al . , 1999 ) . Since neuronal [Cl-]i is mainly determined by the opposing activities of the Cl- extruding transporter , KCC2 , and the Cl- importing NKCC1 ( Blaesse et al . , 2009 ) , we compared hippocampal KCC2 and NKCC1 levels in WT and App-/- mice . We found that KCC2 , but not NKCC1 , protein levels were significantly reduced in App-/- hippocampus compared to WT littermate controls ( Figure 3A–B , Figure 3—source data 1 ) . KCC2 is mainly localized to the cell surface ( Gauvain et al . , 2011 ) . To determine whether plasma membrane localization of KCC2 is decreased in APP mutants , we used surface biotinylation in hippocampal tissue ( Figure 3A ) or HEK293 cells transfected with KCC2 alone or KCC2 and hAPP695 ( Figure 3C ) . We observed significantly reduced plasma membrane KCC2 levels in both App-/- hippocampus and HEK293 cells without co-transfection of hAPP695 ( Figure 3B and D , Figure 3—source data 1 ) . To examine KCC2 levels in intact tissue , we performed immunohistochemical staining in slice with anti-KCC2 antibody . We observed a selective loss of KCC2 immunoreactivity in App-/- mice in the CA1 region of the hippocampus ( Figure 3E ) . To determine if this is due to altered transcription of KCC2 in APP mutants , we compared hippocampal KCC2 mRNA levels in App-/- and WT controls . KCC2 mRNA abundance was similar in App-/- and WT hippocampus ( Figure 3F , Figure 3—source data 1 ) . Together , these results indicate that APP is required to maintain normal KCC2 protein levels in the hippocampus via a post-transcriptional mechanism . 10 . 7554/eLife . 20142 . 009Figure 3 . Full length APP is required for maintaining normal total and surface KCC2 protein levels . ( A ) Representative immunoblots of hippocampal extracts from WT and App-/- littermates . ( B ) Quantification of the immunoblots reveals a significant decrease in surface and total KCC2 protein levels in App-/- mice , while NKCC1 expression levels remain unchanged . Representative immunoblots of western blotting were from single experiment using three pairs of hippocampal lysates , three independent experiments ( C ) Representative immunoblots of HEK293 cells transfected with KCC2 and hAPP695 . ( D ) Quantification shows surface and total KCC2 protein levels are elevated when HEK293 cells are transfected with both constructs . ( E ) DAB staining for KCC2 in hippocampal sections from App-/- and WT littermates shows decreased KCC2 levels in CA1 of hippocampus in App-/- mice . Scale bar: 400 µm . ( F ) qRT-PCR displays similar KCC2 mRNA levels in WT and App-/- hippocampus . APP mRNA serves as a positive control . Each value represents the mean±SEM of at least three samples per genotype . ( G ) Representative immunoblots of transfected HEK293 cells . ( H ) Quantification of the immunoblots reveals an increase in KCC2 levels when we co-transfect hAPP695 , but not when co-transfect with C99 or with sAPPβ . ( I ) Representative immunoblots of hippocampal extracts from sAppβKi and WT littermates . ( J ) Quantification of the immunoblots reveals a significant decrease of KCC2 protein levels in sAppβKi mice . Representative immunoblots were from single experiment using three pairs of HEK293 cells/ or hippocampal lysates , two repeats . *p<0 . 05; **p<0 . 01; **p<0 . 001; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 00910 . 7554/eLife . 20142 . 010Figure 3—source data 1 . Contains source data for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 010 To determine whether APP could stabilize KCC2 protein , we co-transfected HEK293 cells with KCC2 and hAPP695 cDNAs and assayed subsequent KCC2 protein levels . Cells co-transfected with both constructs displayed higher levels of KCC2 protein compared to cells transfected with only KCC2 ( Figure 3G ) . APP is proteolytically processed into multiple extracellular and intracellular fragments . We tested these fragments individually to determine which fragment conferred KCC2 stability . We transfected HEK293 cells with KCC2+C99 , KCC2+sAPPβ , or KCC2+hAPP695 . Only full length APP but not the intracellular or extracellular fragments of APP prevented KCC2 degradation ( Figure 3H , Figure 3—source data 1 ) . Furthermore , using a sAppβ knock in mouse model ( sAppβKi ) , we observed similarly reduced KCC2 levels in sAppβKi hippocampus compared to WT mice ( Figure 3I–J , Figure 3—source data 1 ) . Together , our results indicate that full length APP is required to maintain normal levels of KCC2 protein . To test whether APP regulates KCC2 through direct protein-protein interaction , we performed immunoprecipitation of KCC2 in HEK293 cells co-transfected with constructs encoding KCC2 and hAPP695 . First , Coomassie blue staining was performed in transfected HEK293 cells showing a distribution of all proteins , including APP , pulled down by anti-KCC2 antibody ( Figure 4—figure supplement 1 ) . Next , APP was detected in anti-KCC2 immunoprecipitates ( IP ) . Similarly , immunoprecipitation of APP co-precipitated KCC2 ( Figure 4A ) . However , immunoprecipitation of the GABAAR α1 subunit co-transfected with hAPP695 , we did not observe a direct physical interaction between APP and GABAAR α1 subunit ( Figure 4B ) . Immunoprecipitation of KCC2 using hippocampal lysates co-precipitated APP as well ( Figure 4C ) . Further , the APP-KCC2 interaction result was strengthened by proximity ligation assay ( PLA ) which would sensitively detect a very close surface interaction between proteins ( Söderberg et al . , 2006 ) ( Figure 4D ) . 10 . 7554/eLife . 20142 . 011Figure 4 . APP interacts with KCC2 to limit KCC2 tyrosine phosphorylation and degradation . ( A ) KCC2 interacts with APP in HEK293 cells by co-IP with both KCC2 and APP antibodies . Rabbit IgG ( IP: IgG ) was used as a negative control . ( B ) GABAAR α1 subunit does not coimmunoprecipitate with APP . ( C ) KCC2 interacts with APP by co-IP in hippocampal lysates . ( D ) KCC2 interacts with APP by proximity ligation assay ( PLA ) . HEK293 cell was transfected with APP and KCC2 , APP alone or KCC2 alone . Red: PLA signal indicates an existence of APP-KCC2 interaction; Blue: DAPI . Scale bar: 20 µm . ( E ) Hippocampal extracts from WT and App-/- littermates were immunoprecipitated with an anti-KCC2 antibody ( IP: KCC2 ) and probed with 4G10 antibody ( anti-Phospho-tyrosine ) . Tyrosine phosphorylation of KCC2 is increased in App-/- mice . Total KCC2 levels are also decreased in App-/- mice . ( Extracts were from three mice per genotype ) . ( F ) Representative immunoblots of KCC2 transfected HEK293 cells incubated with PP2 or vehicle . ( G ) Quantification of the immunoblots reveals a significant increase in KCC2 protein levels in PP2 treated HEK293 cells . Representative immunoblots of western blotting were from single experiment using three pairs of HEK293 lysates , two repeats . *p<0 . 05; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 01110 . 7554/eLife . 20142 . 012Figure 4—source data 1 . Contains source data for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 01210 . 7554/eLife . 20142 . 013Figure 4—figure supplement 1 . Coomassie-stained SDS-PAGE gel showing KCC2 binding proteins after immunoprecipitated with an anti-KCC2 antibody ( IP: KCC2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 013 KCC2 is regulated post-translationally by tyrosine phosphorylation ( Lee et al . , 2010; Wake et al . , 2007 ) . Increased tyrosine phosphorylation of KCC2 enhances lysozomal degradation and thus decreases KCC2 levels ( Lee et al . , 2011 , 2010 ) . We hypothesized that APP may normally interact with KCC2 to prevent tyrosine phosphorylation and subsequent degradation . To test this hypothesis , we examined whether tyrosine phosphorylated KCC2 levels increased in the absence of APP . Hippocampal tissue from App-/- and WT mice was lysed and immunoprecipitated with anti-KCC2 antibody ( Lee et al . , 2010 ) . Precipitates were blotted for tyrosine phosphorylated proteins using the anti-P-Tyr antibody , 4G10 . We observed a robust increase in tyrosine phosphorylated KCC2 levels in APP mutants compared to littermate controls ( Figure 4E ) . Moreover , total KCC2 protein levels remained decreased in App-/- ( Figure 4E ) ( Relative IOD: WT , 544 . 86; App-/- , 412 . 56 . hippocampal lysates of three mice per genotype ) indicating an increase in tyrosine phosphorylation of KCC2 accompanies the reduction in total KCC2 levels . To further demonstrate that KCC2 instability is due , at least in part , to phosphorylation by a tyrosine kinase in the absence of APP , we tested whether blockade of tyrosine kinases by PP2 , a potent inhibitor of Src-family tyrosine kinases ( Bi et al . , 2000; Lee et al . , 2007 ) , enhances KCC2 protein levels . Treatment of HEK293 cells transfected with KCC2 with PP2 at a concentration of 20 µM for 30 min significantly increased KCC2 levels compared to vehicle controls ( Figure 4F–G , Figure 4—source data 1 ) . Together , these results suggest that APP-KCC2 interaction functions , at least in part , to limit tyrosine phosphorylation of KCC2 , loss of APP leads to excessive tyrosine phosphorylation and premature degradation of KCC2 , resulting in reduced KCC2 protein levels in APP mutants . Next , we tested whether restoring KCC2 function in APP mutant mice would rescue changes in EGABA in hippocampal neurons . Chloride extrusion enhancers , CLP257 and CLP290 , have been recently shown to restore Cl- extrusion capacity only in neurons with reduced KCC2 activity ( Gagnon et al . , 2013 ) . We incubated hippocampal slices from APP mutant mice with CLP257 at 100 µM for 2 hr , and compared EGABA in slices treated with CLP257 or vehicle . Activation of KCC2 in App-/- hippocampus resulted in a hyperpolarizing shift of EGABA ( CLP257 , −66 ± 3 mV; Control , −54 ± 2 mV; p=0 . 03 ) ( Figure 5A–B , Figure 5—source data 1 ) . We next tested whether restoration of Cl- extrusion by chronic treatment with CLP290 ( Gagnon et al . , 2013 ) rescues KCC2 expression levels . CLP290 or vehicle was intraperitoneally injected once daily for seven consecutive days in App-/- mice , and hippocampal KCC2 levels were analyzed by western blotting . We observed a significant increase in KCC2 levels in CLP290 treated animals compared to vehicle treated App-/- mice ( Figure 5C–D , Figure 5—source data 1 ) . 10 . 7554/eLife . 20142 . 014Figure 5 . Restoring KCC2 expression and function in App-/- mice rescues hippocampal EGABA , GABAAR expression and IPSC amplitude . ( A ) Sample traces showing perforated patch recording of EGABA in CLP257 or vehicle treated App-/- hippocampal slices . Currents were recorded at the indicated holding potentials shown to the left of each trace . ( B ) Quantification of EGABA shows a significant hyperpolarizing shift in App-/- in CLP257 treated slices compared to vehicle controls ( CLP257 , n = 5 cells from three mice; Control , n = 5 cells from three mice ) . ( C ) Representative immunoblots of hippocampal extracts from App-/- treated with CLP290 or vehicle . ( D ) Quantification of the immunoblots reveals increased KCC2 protein levels in CLP290 treated App-/- mice . ( E ) Representative immunoblots of HEK293 cells transfected with KCC2 incubated with CLP257 or vehicle . ( F ) Quantification of the immunoblots reveals a significant increase in surface KCC2 protein levels when treated with CLP257 , while total KCC2 levels remain unchanged , suggesting CLP257 might regulate KCC2 activity through blocking the latter’s turnover . ( G ) Sample traces of evoked inhibitory currents in response to puffs of 100 µM isoguvacine . ( H ) Quantification of isoguvacine-evoked inhibitory current amplitudes shows a significant increase in CLP257 treated App-/- hippocampus compared to vehicle controls ( CLP257 , n = 15 cells from three mice; Control , n = 15 cells from three mice ) , but no difference between CLP257 treated WT hippocampus and vehicle treated controls ( CLP257 , n = 9 cells from three mice; Control , n = 11 cells from three mice ) , suggesting that restoring KCC2 function rescues GABAAR mediated responses . ( I ) Representative immunoblots of hippocampal extracts from App-/- treated with CLP290 or vehicle . ( J ) Quantification of the immunoblots reveals increased GABAAR α1 protein levels in CLP290 treated App-/- mice . Representative immunoblots of western blotting were from single experiment using three pairs of HEK293 cells/ or hippocampal lysates of two independent experiments . *p<0 . 05; **p<0 . 01; ***p<0 . 001; Student’s t-test; quantified data of H , two-way ANOVA with post hoc tests . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 01410 . 7554/eLife . 20142 . 015Figure 5—source data 1 . Contains source data for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 015 It has been proposed that CLPs might increase KCC2 levels by reducing membrane KCC2 turnover ( Gagnon et al . , 2013 ) . We then examined , in 293 cells transfected with KCC2 , if treatment with CLP257 would indeed increase levels of membrane KCC2 , and observed significantly enhanced surface KCC2 levels in CLP257 treated cells compared to vehicle control ( Figure 5E–F , Figure 5—source data 1 ) . These experiments could not distinguish whether CLP restored KCC2 levels by increasing insertion or decreasing removal of KCC2 . Instead , the study suggested a rescue of KCC2 surface levels and function by CLPs in App-/- mice . We observed decreases in GABAAR α1 subunit protein levels and smaller evoked IPSC amplitudes in APP mutants ( Figure 2 ) . We then tested whether potentiating KCC2 function would rescue GABAAR mediated inhibitory currents in APP mutant hippocampal slices . Evoked whole-cell inhibitory currents , induced by puffing the GABAAR agonist isoguvacine , were recorded after incubation with CLP257 or vehicle for 2 hr , respectively . Evoked inhibitory current amplitudes were significantly larger in CLP257 treated App-/- slices compared to vehicle controls ( App-/-: CLP257 , 410 ± 34 pA; Vehicle , 278 ± 31 pA; p=0 . 008 ) ( Figure 5G–H ) . Importantly , CLP257 treated WT slices did not differ significantly compared to vehicle controls ( WT: CLP257 , 612 ± 99 pA; Vehicle , 510 ± 109 pA; p=0 . 5 ) ( Figure 5G–H ) . Additionally , chronic treatment with CLP290 rescued GABAAR α1 subunit protein levels as well ( Figure 5I–J , Figure 5—source data 1 ) . These results support a model in which APP loss of function results in reduced KCC2 levels and decreased KCC2 function , which causes the attenuation of GABAAR α1 expression and GABAAR mediated inhibitory current amplitudes , leading to a decrease of inhibitory tone in the hippocampus of APP mutants . To test whether GABAAR α1 level changes are dependent on the presence of KCC2 , we transfected HEK293 cells with GABAAR α1 or β2 subunits with and without co-transfection of KCC2 . Interestingly , we only observed significantly reduced GABAAR α1 , but not β2 , subunit levels in the absence of KCC2 ( Figure 6A–B , Figure 6—source data 1 ) , suggesting a selective role of KCC2 on GABAAR α1 expression . 10 . 7554/eLife . 20142 . 016Figure 6 . GABAAR α1 subunit is regulated by KCC2 expression through affecting intracellular Cl- . ( A ) Representative immunoblots of GABAAR α1+KCC2 , GABAAR α1+pcDNA3 , GABAAR β2+KCC2 , GABAAR β2+pcDNA3 , transfected HEK293 cells blotted with anti-GABAAR antibodies . ( B ) Quantification of the immunoblots reveals a significant increase in GABAAR α1 levels in HEK293 cells co-transfected with GABAAR α1 and KCC2 , indicating KCC2 positively regulates expression of GABAAR α1 . ( C ) GABAAR-α1 does not co-immunoprecipitate with KCC2 , suggesting that there was no protein-protein interaction between KCC2 and GABAAR α1 . ( D ) Representative immunoblots of hippocampal extracts from WT and App-/- littermates . ( E ) Quantification of the immunoblots reveals identical PLIC-1 levels in between WT and App-/- hippocampus . ( F ) Representative immunoblots of WT hippocampal slices incubated with a KCC2 inhibitor , VU02450551 , or vehicle . ( G ) Representative immunoblots of App-/- hippocampal slices incubated with a KCC2 inhibitor , VU02450551 , or vehicle . ( H ) Quantification of the immunoblots reveals a significant decrease only in GABAAR α1 levels of hippocampal tissue treated with VU02450551 . ( I ) Representative immunoblots of hippocampal extracts from P7 and P35 WT mice . ( J ) Quantification of the immunoblots reveals significantly increased KCC2 and GABAAR α1 levels in P35 compared to P7 WT mice . Representative immunoblots of western blotting were from single experiment using three pairs of HEK293 cells/ or hippocampal lysates , two repeats . *p<0 . 05; **p<0 . 01; ***p<0 . 001; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 01610 . 7554/eLife . 20142 . 017Figure 6—source data 1 . Contains source data for Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 01710 . 7554/eLife . 20142 . 018Figure 6—figure supplement 1 . Immunofluorescent staining of hippocampal KCC2 in P14 and P35 WT mice shows increased KCC2 levels in P35 WT mice . Scale bar: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 018 The mechanisms underlying the actions of KCC2 on GABAergic synapses are currently unclear ( Chudotvorova et al . , 2005 ) . We next conducted experiments to elucidate how KCC2 regulates GABAAR α1 expression and thus GABAAR mediated inhibition . During immunoprecipitation of the GABAAR α1 subunit co-transfected with KCC2 , we did not observe a direct physical interaction between KCC2 and GABAAR α1 subunit ( Figure 6C ) , indicating that KCC2 regulation of GABAAR α1 expression was unlikely through a direct protein-protein interaction . In a previous report the ubiquitin-like protein PLIC-1 , which regulates the membrane trafficking of GABAAR , was shown to directly interact with GABAARs and promote their accumulation at the cell surface ( Saliba et al . , 2008 ) . We next examined if KCC2 regulates GABAAR α1 levels through affecting PLIC-1 expression , however , Western blotting result showed identical PLIC-1 levels in hippocampus of WT and App-/- mice ( Figure 6D–E , Figure 6—source data 1 ) . It has been shown that expression of GABAAR subunits is regulated by KCC2 expression through affecting intracellular Cl- gradient ( Succol et al . , 2012 ) . We thus proposed that KCC2-mediated Cl- extrusion might underlie KCC2 regulation of GABAAR α1 protein levels . If this was the case , treatment of WT hippocampus with a KCC2 inhibitor , VU02450551 , would result in a reduction of α1 subunit of GABAAR protein levels as seen in App-/- hippocampus . Hippocampal GABAAR α1 , several other GABAAR subunits and GluRs levels were analyzed by western blotting after incubation with VU02450551 or vehicle for 2 hr in WT hippocampus slices . We observed a significant decrease in GABAAR α1 levels , but not that of other GABAARs and GluRs subunits in VU02450551 treated WT hippocampus compared to vehicle control ( Figure 6F and H , Figure 6—source data 1 ) . Moreover , incubation of App-/- hippocampus with KCC2 inhibitor , VU02450551 , further reduced levels of GABAAR α1 subunit but left GluR1 levels unchanged ( Figure 6G and H , Figure 6—source data 1 ) . These experiments suggested that blockade of Cl- extrusion by VU02450551 selectively affected levels of GABAAR α1 , but not other GABAAR subunits tested in the present study . It was well known that the expression of KCC2 and GABAergic inhibition parallel neuronal maturation and the emergence of low intracellular Cl- ( Ben-Ari et al . , 2012; Blaesse et al . , 2009; Rivera et al . , 1999 ) . KCC2 exports Cl- and is weakly expressed at birth and upregulated as the brain matures ( Plotkin et al . , 1997; Rivera et al . , 1999 ) . To further evaluate the notion that rescue of GABAAR α1 expression and GABAAR mediated inhibition in App-/- hippocampus by restoring KCC2 levels involves alteration in intracellular Cl- , we compared hippocampal KCC2 and GABAAR α1 levels by western blotting in between postnatal day 7 ( P7 ) ( with high intracellular Cl- concentration ) and P35 ( with low intracellular Cl- concentration ) WT mice , and observed significantly increased GABAAR and KCC2 levels in P35 compared to P7 hippocampus ( Figure 6I–J , Figure 6—source data 1 ) . Similar to western blotting , immunofluorescent staining in P14 and P35 hippocampus displayed higher KCC2 levels in P35 compared to P14 mice ( Figure 6—figure supplement 1 ) . Data thus far suggested that rescue of GABAAR α1 levels by restoring KCC2 in App-/- hippocampus involved , at least in part , an alteration of KCC2 mediated Cl- extrusion . Having shown that one way for APP to regulate KCC2 levels was to limit tyrosine phosphorylation and sequential degradation of KCC2 , we speculated that blockade of tyrosine phosphorylation of KCC2 by mutating Y903A and Y1087A ( mKCC2 ) which are two tyrosine phosphorylating sites of KCC2 may restore KCC2 levels in the absence of APP . However , mKCC2 levels were significantly lower without hAPP695 than with ( Figure 7A–B , Figure 7—source data 1 ) . Meanwhile , mKCC2 levels are identical in PP2 treated and non-treated HEK293 cells ( Figure 7C–D , Figure 7—source data 1 ) confirming that mKCC2 did not undergo tyrosine phosphorylation , which met desired effect . Moreover , both IP and PLA examinations displayed direct protein-protein interaction between APP and mKCC2 ( Figure 7E–F ) , suggesting that mutation of KCC2 at tyrosine phosphorylating sites did not disturb APP-KCC2 interaction . We then hypothesized that APP deficiency might cause a loss of acting force to hold mKCC2 on site leading to an increase in mKCC2 internalization , resulting in enhanced mKCC2 degradation via an unknown pathway rather than tyrosine phosphorylation . Interestingly , similar to co-transfection of mKCC2 with hAPP695 constructs , HEK293 cells co-transfected with mKCC2 and cylindromatosis ( CYLD , a deubiquitinating enzyme ) displayed higher levels of mKCC2 protein levels compared to cells transfected with only mKCC2 ( Figure 7G–H , Figure 7—source data 1 ) , implicating that mKCC2 might undergo ubiquitination and subsequent degradation in the absence of APP . 10 . 7554/eLife . 20142 . 019Figure 7 . APP interacts with mKCC2 to limit mKCC2 ubiquitination and degradation . ( A ) Representative immunoblots of HEK293 cells transfected with mKCC2 and hAPP695 . ( B ) KCC2 protein levels are elevated when HEK293 cells are transfected with both constructs . ( C ) Representative immunoblots of mKCC2 transfected HEK293 cells incubated with PP2 or vehicle . ( D ) Quantification of the immunoblots reveals identical KCC2 protein levels in between PP2 treated HEK293 cells and vehicle control . ( E ) mKCC2 interacts with APP . Rabbit IgG ( IP: IgG ) was used as a negative control . ( F ) PLA shows that mKCC2 interacts with as well . HEK293 cell were transfected with both APP and mKCC2 , APP alone , or mKCC2 alone . Red: PLA signal indicates an existence of APP-mKCC2 interaction; Blue: DAPI . Scale bar: 20 µm . ( G ) Representative immunoblots of transfected HEK293 cells . ( H ) Quantification of the immunoblots reveals a significantly increased KCC2 protein levels in HEK293 cell transfected with hAPP695 and CYLD , respectively . Representative immunoblots of western blotting were from single experiment using three pairs of lysates , two repeats . **p<0 . 01; ***p<0 . 001; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 01910 . 7554/eLife . 20142 . 020Figure 7—source data 1 . Contains source data for Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 020 Next , treating mKCC2 transfected HEK293 cells ( in the presence or absence of APP co-transfection ) with the ubiquitination inhibitor MG132 , we observed that MG132 increased levels of mKCC2 only in the absence ( Figure 8A left–B , Figure 8—source data 1 ) but not in the presence ( Figure 8A right–B , Figure 8—source data 1 ) of hAPP695 , supporting that APP functions to limit mKCC2 ubiquitination . Moreover , using MG132 to block ubiquitination in HEK293 cells transfected with KCC2 , we confirmed that ubiquitination occurred in KCC2 as well in the absence of APP ( Figure 8C–D , Figure 8—source data 1 ) . 10 . 7554/eLife . 20142 . 021Figure 8 . APP deficiency causes an increase in KCC2 ubiquitination leading to a reduction of KCC2 levels . ( A ) Representative immunoblots of HEK293 cells transfected with mKCC2 alone , mKCC2 and hAPP695 , incubated with MG132 or vehicle . ( B ) Quantification of the immunoblots reveals a significantly increased mKCC2 levels in HEK293 cell transfected with mKCC2 alone and incubated with MG132 . ( C ) Representative immunoblots of HEK293 cells transfected with KCC2 alone ( incubated with MG132 ) ; KCC2 and hAPP695 . ( D ) Quantification of the immunoblots reveals a significantly increased KCC2 protein levels in KCC2 alone with MG132 incubation , which is similar to co-transfection of hAPP695 and KCC2 , compared to KCC2 alone incubated with vehicle . ( E ) Representative immunoblots of supernatant before and after IP pull down of KCC2 in WT and App-/- mice . ( F ) Hippocampal extracts from WT and App-/- littermates immunoprecipitated with an anti-KCC2 antibody ( IP: KCC2 ) and probed with ubiquitin antibody . Ubiquitinated KCC2 is obviously increased in App-/- compared to WT mice ( Each extract was from three mice per genotype , repeated twice ) . ( G ) Representative immunoblots of App-/- hippocampus incubated with MG132 or vehicle . ( H ) Quantification of the immunoblots reveals a significantly increased KCC2 protein levels in MG132 treated App-/- hippocampus implicating KCC2 underwent ubiquitination in the absence of APP . ( I ) Representative immunoblots of surface KCC2 levels of HEK293 cells and App-/- hippocampus incubated with MG132 or vehicle . ( J ) Quantification of the immunoblots reveals a significantly increased surface KCC2 levels in MG132 treated compared to vehicle control . Representative immunoblots of western blotting were from single experiment using three pairs of HEK293 cells/ or hippocampal lysates , two repeats . *p<0 . 05; ***p<0 . 001; Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 02110 . 7554/eLife . 20142 . 022Figure 8—source data 1 . Contains source data for Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 022 We then examined whether KCC2 ubiquitination contributed to reduction of KCC2 levels in App-/- hippocampus . Hippocampal tissue from App-/- and WT mice was lysed and immunoprecipitated with anti-KCC2 antibody . Precipitates were blotted for ubiquitin/ubiquitinated proteins using the anti-ubiquitin antibody . We observed a robust increase in ubiquitinated KCC2 levels in App-/- hippocampus compared to littermate controls ( Figure 8E–F ) . Furthermore , incubating hippocampus containing App-/- slices with MG132 significantly increased KCC2 levels ( Figure 8G–H , Figure 8—source data 1 ) . Moreover , surface KCC2 levels could also be increased by MG132 blockade of KCC2 ubiquitination in HEK293 cells and App-/- hippocampus ( Figure 8I–J , Figure 8—source data 1 ) . Data thus far indicated that KCC2 levels could be regulated post-translationally by ubiquitination as well as tyrosine phosphorylation in the absence of APP . APP-KCC2 interaction played a crucial role in suppressing KCC2 internalization and subsequent degradation via both tyrosine phosphorylation and ubiquitination . APP deficiency resulted in a loss of acting force to hold membrane KCC2 , leading to enhanced KCC2 degradation in the hippocampus . GABA has not only phasic but also tonic mode of action ( tonic GABA current ) and the latter involves extrasynaptic GABAARs and/or reactive astrocytic GABA ( Curia et al . , 2009; Jo et al . , 2014 ) , and tonic GABA action has been implicated in AD mouse models ( Jo et al . , 2014; Wu et al . , 2014; Yarishkin et al . , 2015 ) . We have previously shown that tonic GABA currents were impaired in the dentate gyrus ( DG ) of App-/- mice attributing to abnormal genesis and survival of DG newborn neurons ( Wang et al . , 2014b ) . We thus evaluated whether abnormal tonic GABA currents also occurred in APP deficient CA1 and , observed significantly reduced amplitude of tonic GABA current in the CA1 of App-/- mice compared to WT controls ( WT , 19 . 1 ± 2 . 7 pA; App-/- , 9 . 1 ± 0 . 7 pA; p=0 . 002 ) ( Figure 9—source data 1 ) . Interestingly , restoring KCC2 by CLP257 incubation of App-/- hippocampal slice did not rescue impaired tonic GABA current ( App-/- , 9 . 1 ± 0 . 7 pA; App-/-+CLP257 , 11 . 9 ± 1 . 6 pA; p=0 . 1 ) ( Figure 9—source data 1 ) , suggesting that aberrant KCC2 levels and function might underlie impaired phasic but not tonic GABA current in the CA1 of App-/- mice . 10 . 7554/eLife . 20142 . 023Figure 9 . Reduced tonic GABA current in CA1 of App-/- hippocampus . ( A ) Sample traces of tonic GABA current in WT , App-/- and App-/- incubated with 100 μM CLP257 . ( B ) Quantification of tonic GABA current amplitude shows a significant decrease in App-/- mice and CLP does not rescue tonic GABA current in App-/- hippocampus ( WT , n = 10 cells from five mice; App-/- , n = 10 cells from five mice; App-/-+CLP257 , n = 9 cells from five mice ) . *p<0 . 05; **p<0 . 01; one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 02310 . 7554/eLife . 20142 . 024Figure 9—source data 1 . Contains source data for Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 024
In this study , we demonstrate that APP regulation of KCC2 expression and function is required to maintain appropriate GABAergic signaling capacity in the hippocampus as summarized in Figure 10 . This is supported by several lines of evidence: ( 1 ) APP deficiency results in a depolarizing shift of EGABA along with reduced amplitude of uIPSCs and tonic GABA current , ( 2 ) APP deficiency results in a significant reduction in the abundance of hippocampal KCC2 , but not NKCC1 , and GABAAR α1 subunit expression , ( 3 ) Phasic GABAergic signaling phenotypes can be rescued by CLPs enhancing KCC2 expression and function , suggesting that impaired regulation of KCC2 by APP is responsible for these fast inhibition deficits in APP mutants , while altered tonic GABA current might involve different mechanisms ( Jo et al . , 2014; Wu et al . , 2014 ) . Together , these data indicate that regulation of KCC2 by APP plays a critical role in the maintenance of normal hippocampal GABAergic inhibition . Identical mIPSC amplitude in between WT and App-/- hippocampus indicates that reduction in GABAAR α1 expression in App-/- does not impair basal inhibitory activity that requires activation of a portion of GABAAR , which is commonly observed when correlating altered receptors with miniature activities ( Caraiscos et al . , 2004; Chudotvorova et al . , 2005 ) . 10 . 7554/eLife . 20142 . 025Figure 10 . A working model shows the mechanism underlying APP regulation of GABAAR mediated inhibition . KCC2 transports Cl- outside the neuron maintaining low [Cl-]i in mature neurons and therefore insures an Cl- influx upon GABA binding to the GABAARs . APP and KCC2 physically interacts to limit tyrosine phosphorylation , ubiquitination and sequential degradation of KCC2 to maintain abundant membrane KCC2 levels and hyperpolarizing EGABA in WT hippocampal neuron . APP deficiency leads to a loss of limit on tyrosine phosphorylation and ubiquitination of KCC2 and thus reduces KCC2 protein levels , resulting in depolarizing shift of EGABA , reduction of GABAAR α1 and GABA mediated inhibition , which can be rescued by restoration of KCC2 expression and function . DOI: http://dx . doi . org/10 . 7554/eLife . 20142 . 025 In previous studies , we have shown that impaired GABAergic short-term plasticity evaluated by abnormal PPR of IPSCs is due , in part , to presynaptic alterations of L-type calcium channels ( LTCCs ) ( Yang et al . , 2009 ) . However , we did not see changes in the PPR of uIPSCs in the present study . This may be due to the fact that IPSCs were induced by field stimulation in previous study , which activates multiple presynaptic GABAergic neurons , and corresponding LTCC currents . By contrast , in the present study , uIPSCs were evoked by a single presynaptic action potential in a single GABAergic neuron such that the involvement of presynaptic LTCCs is minimized . This system allowed us to focus on postsynaptic KCC2 dysfunction in APP mutant hippocampal neurons . KCC2 is the principle mechanism used by neurons to lower internal [Cl-] which is required for adequate inhibitory synaptic transmission upon activation of GABAARs and GlyRs ( Braat and Kooy , 2015a; Kaila et al . , 2014; Rivera et al . , 1999 ) . However , little is known about what regulates KCC2 expression and function . In an attempt to understand why KCC2 levels are diminished in APP mutants , we discovered a physical interaction between APP and KCC2 . This interaction protects KCC2 from protein phosphorylation and ubiquitination which target the protein for degradation . These results suggest that APP is a novel protein binding-partner for KCC2 . APP holds KCC2 on membrane through protein-protein interaction to maintain appropriate KCC2 levels , internal Cl- and GABAAR mediated inhibition . Though increased activation of glutamatergic receptors has been shown to downregulate KCC2 expression and GABAAR mediated current ( Lee et al . , 2011 ) , neither amplitude , frequency of mEPSCs ( Amplitude; WT , 13 . 3 ± 0 . 4 pA; App-/- , 12 . 6 ± 0 . 8 pA; p=0 . 5; Frequency; WT , 0 . 2 ± 0 . 05 Hz; App-/- 0 . 4 ± 0 . 2 Hz; p=0 . 3 ) nor GluRs protein levels changed in hippocampus of APP mutants compared to WT controls ( Figure 2—figure supplement 2A–D , Figure 2—source data 1 ) . The results suggest that excitatory synaptic activity mediated by AMPA receptors may not be affected by loss of APP at the age tested . Changes in KCC2 levels that occur in APP mutants can have wide ranging effects on the status of GABAergic signaling . Modulation of KCC2 protein levels has been reported to affect GABAAR density and GABAergic inhibition ( Chudotvorova et al . , 2005; Succol et al . , 2012 ) . Therefore , we postulated that altered KCC2 function in APP mutants accounts for changes in GABAergic signaling . To test this , we pharmacologically enhanced KCC2 function in App-/- mice and observed a rescue in both amplitudes of isoguvacine induced currents and GABAAR α1 expression levels , indicating a rescue of GABAergic functional deficits by restoring KCC2 function . Moreover , we show that blockade of KCC2 mediated Cl- extrusion in WT hippocampus resulted in a reduction of GABAAR α1 levels as seen in App-/- mice and , GABAAR α1 levels are positively correlated with KCC2 expression and function during early stage of development . We thus reasoned that increase in KCC2 mediated Cl- extrusion might be responsible for upregulation of surface GABAAR α1 levels . Though the exact mechanism underlying how restoring KCC2 rescued surface GABAAR α1 levels remains unclear , given that APP interacts directly with KCC2 , and increasing KCC2 expression and function rescued GABAergic signaling phenotypes in App-/- hippocampus , we propose that KCC2 is the primary defect caused by APP deletion to alter GABAergic signaling , while changes in GABAAR α1 levels and GABAAR mediated inhibition are secondary to altered KCC2 expression and function . Our results indicate that full length APP is required to maintain normal levels of KCC2 protein . Together with a recent study of the AppΔCT15-DM mice , a mouse model lacking the last 15 amino acids of APP as well as the APP homologue , APLP2 , which demonstrated impairments of synaptic plasticity and hippocampal-dependent behavior ( Klevanski et al . , 2015 ) , it is likely that APP proteolysis may be a mechanism for turning off normal APP functions which needs to be taken into account when exploring the pathophysiology of APP with respect to KCC2 regulation . GABAergic signaling has been shown to be disrupted in many diseases of the nervous system ( Braat and Kooy , 2015a ) , including fragile X syndrome and autism spectrum disorder ( Braat and Kooy , 2015b; Han et al . , 2014 ) . Changes in KCC2 function may account for many GABAergic signaling deficits , as it has been shown to be involved in the pathogenesis of epilepsy , schizophrenia , autism and aging brain ( Ben-Ari et al . , 2012; Ferando et al . , 2016; Tang et al . , 2016; Tao et al . , 2012 ) . Though the GABAergic system appears to be resistant to Aβ toxicity ( Selkoe , 2002 ) , recent research has demonstrated dysfunctional GABAergic inhibition in AD ( Verret et al . , 2012 ) . In the present study , we have described a cellular mechanism by which loss of APP function downregulates KCC2 to impair GABAergic signaling . Given the results presented here in APP mutant hippocampal neurons , if APP function is compromised in the context of progressing AD , this may lead to the gradual loss of GABAergic signaling capacity in the hippocampus and associated changes in memory function . Together with a most recent study revealing a critical role of KCC2 in information storage in aging brain ( Ferando et al . , 2016 ) , the present study suggests that KCC2 might serve as a therapeutic target to improve GABAergic inhibition to potentially restore hippocampal function and improve memory in AD .
All experiments and animal housing were in accordance with procedures approved by the Ethics Committee for animal research at South China Normal University , according to the Guidelines for Animal Care established by the National Institute of Health . Both female and male mice were used in this study . App-/- and wild type ( WT ) mice ( Zheng et al . , 1995 ) were generated by crossing App+/- males and females obtained from Model Animal Research Center of Nanjing University ( Nanjing , China ) . App-/- mice were crossed with Gad67+/GFP mice ( Tamamaki et al . , 2003 ) ( gift from Dr . Yuqiang Ding at Tongji University School of Medicine , Shanghai , China ) when necessary to generate App-/--Gad67+/GFP and WT-Gad67+/GFP controls . Soluble APP ectodomain beta ( sAppβ ) knock in mice ( Li et al . , 2010 ) , full length human APP ( hAPP695 ) , APP C-terminal fragments ( C99 ) and sAPPβ constructs were from Hui Zheng lab at Baylor College of Medicine . The KCC2 construct was gift from John A Payne ( School of Medicine , University of California ) . The GABAAR α1 construct was gift from Gangyi Wu ( South China Normal University , Guangzhou , China ) . CYLD construct was gift from Shaocong Sun ( University of Texas MD Anderson Cancer Center , Texas , USA ) . GABAAR β2 construct was from Yong Zhang ( Peking University ) . Cultured hippocampal neurons were prepared as described ( Yang et al . , 2009 ) with some modifications . Briefly , we dissected hippocampus from App-/--Gad67+/GFP and WT-Gad67+/GFP littermates of P0/1 pups , trypsinized the tissue at 37°C for 30 min , and then gently triturated in culture medium containing 10% heat-inactivated fetal bovine serum using fire-polished pipettes , before centrifuging at 1000 rpm for 10 min . Cells were resuspended in Neurobasal Medium supplemented with B27 and L-glutamine ( Invitrogen , Carlsbad , CA ) , and plated at 1–2 × 104 cells/ml onto 13 mm poly-L-lysine coated coverslips . One day after plating , 4 M cytosine-D-arabinofuranoside ( Ara-C ) ( Sigma ) was added to prevent astrocyte proliferation . Cultures of 12–14 DIV were used for patch clamp recordings . Mice brains were dissected and immediately frozen in liquid nitrogen . Total RNA was extracted using Trizol reagent . RNA pellets were resuspended in diethylpyrocarbonate-treated water and RNA concentration was measured using a Nanodrop2000c spectrometer ( Thermo Scientific ) . RNA was DNase treated ( DNase I , amplification grade , Invitrogen ) and then reverse-transcribed using the Superscript III First-Strand Synthesis System ( Invitrogen ) . Relative quantification of gene expression was performed using Taq Manprobes and an ABI Prism 7000 Sequence Detection System . Platinum Quantitative PCR Super Mix-UDG w/ROX ( Invitrogen ) was used with the following primers and probes: App F-primer , ( 5’–3’ ) CCAAGAGGTCTACCCTGAACTGC; App R-primer , ( 5’–3’ ) AGGCAACGGTAAGGAATCACG; Actβ F-primer ( 5’–3’ ) GTGACGTTGACATCCGTAAAGA; Actβ R-primer ( 5’–3’ ) GTAACAGTCCGCCTAGAAGCAC; Slc12a5 F-primer ( 5’–3’ ) GGGCAGAGAGTACGATGGC; Slc12a5 R-primer ( 5’–3’ ) TGGGGTAGGTTGGTGTAGTTG . Assay efficiencies were experimentally determined using a five-point dilution series of cDNA spanning a 100-fold range in concentration . 0 . 025 µg cDNA template was used per reaction . Statistical analysis was performed on 2-∆∆Ct values . Mice were anesthetized with intraperitoneal injection of 20% urethane ( 0 . 01 ml/g ) and perfused transcardially with 0 . 9% physiological saline and 4% paraformaldehyde ( PFA ) in phosphate buffer ( 0 . 01 M PBS , pH 7 . 4 ) . Following perfusion , the brains were dissected and post-fixed overnight in 4% PFA followed by a sucrose series ( 10% , 20% and 30% ) solution at 4°C for cryoprotection . Serial coronal/sagittal sections of tissue were obtained using Lycra frozen slicer 1800 with a thickness of 30 μm . Free-floating sections were rinsed in PBS three times for 5 min and processed for antigen retrieval by boiling in 10 mM citrate buffer ( pH 6 . 0 ) for 5 min . The sections were incubated in 3% H2O2 in PBS for 15 min at room temperature ( RT ) to quench endogenous peroxidases , washed three times for 5 min and subsequently incubated in 0 . 3% Triton X-100 in PBS for 2 hr . To block nonspecific binding , sections were incubated in 3% bovine serum albumin ( BSA ) for 30 min . Sections were then incubated overnight with primary antibody against KCC2 ( Neuromab , 75–013; Millipore , 07–432 ) at 4°C . After three washes with PBS , secondary antibody ( CWS ) application was performed at RT for 2 hr followed by three additional washes with PBS for 10 min . The secondary antigen was visualized with 3 , 3-diaminobenzidine tetrahydrochloride ( DAB , Sola ) . The reaction was terminated by rinsing in PBS and then sections were mounted on gelatin-subbed slides , dehydrated in ascending alcohol concentrations , cleared through xylenes , and covered with DPX resin . KCC2 immunofluorescent staining was conducted as described ( Yang et al . , 2009 ) . Mouse brains were homogenized using SDS lysis buffer ( 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM EDTA pH 8 . 0 , 1% SDS ) containing complete protease inhibitor cocktail ( Roche ) . After 1 min of homogenization , the cellular debris was removed by centrifugation at 14000 rpm for 10 min at 4°C and supernatant was collected and denatured for 20 min at 75°C . Tissue lysates were subjected to SDS-PAGE ( Bio-Rad ) and transferred to nitrocellulose membranes . Membranes were blocked for 0 . 5 hr using 5% non-fat dry milk in TBS containing 0 . 5% Tween-20 ( TBST ) , then incubated with specific primary antibodies against KCC2 ( Neuromab , 75–013; Millipore , 07–432; Santa Cruz , sc-19419 ) , NKCC1 ( Abcam , ab191289 ) , APP ( Abcam , ab15272 ) , 4G10 ( Millipore , 16–452 ) , PLIC-1 ( ABGENT , AP2176c ) , GABAAR subunits: α1 ( Millipore , 06–868 ) , α5 ( Abcam , ab175195 ) , β2 ( Abcam , ab156000 ) , δ ( Abcam , ab110014 ) , γ2 ( Abcam , ab87328 ) , GluR1 subunit ( Abcam , ab31232 ) and GluR2 subunit ( Abcam , ab206293 ) of AMPA receptors , and ubiquitin ( Abcam , ab7254 ) . Anti γ-tubulin antibody ( Sigma , T6557 ) was used as a loading control . Antibodies were diluted at 1:1000 to 1:2000 at use . After three washes with TBST , secondary antibody ( CWS ) application was performed at RT for 1 hr using 5% milk in TBST followed by three additional washes with TBST . Bands were visualized using Immobilon Western ECL system and analyzed with Gel Pro Analysis software . For surface KCC2 detection , biotinylation of cell surface proteins was performed as described previously ( Lee et al . , 2010 ) . Briefly , HEK293 cells transfected with KCC2 constructs or cultured neurons were washed two times with PBS containing 0 . 5 mM MgCl2 and 1 mM CaCl2 ( PBS-CM ) and then incubated with 2 ml of PBS-CM containing 1 mg/ml Sulfo-NHS-SS-Biotin ( Pierce ) for 30 min at 4°C . After labeling , the biotin reaction was quenched by washing three times with ice-cold PBS-CM containing 50 mM glycine and 0 . 1% BSA . Cells were then lyzed with 1 ml lysis buffer and the lysate was collected . Protein concentration was determined using Micro BCA protein assay kit ( Thermo Scientific ) . After correction for protein concentration , 40 μl of 1:1 slurry of UltraLink NeutrAvidin ( Pierce ) was added to the lysates to pull down the biotin-labeled surface proteins for 2 hr at 4°C . The NeutrAvidin beads were washed and bound materials were analyzed by SDS-PAGE . To obtain a KCC2 plasmid in which the tyrosine phosphorylating sites , Y903A and Y1087A , were mutated ( mKCC2 ) , the KCC2 open reading frame ( ORF ) was amplified from the pCAGIG vector , incorporate Y903A and Y1087A mutation using the stratagene QuickChange method as described ( Lee et al . , 2010 ) . For mutation of Y903 the following primers were used: ( sense ) TCCGGATCCACTCGAGATGCTCAAC; ( antisense ) AATGTCTTCTCGAAGGTG TATGCTG . For mutation of Y1087 the following primers were used: ( sense ) TACACCTTCGAGAAGACATTGG; ( antisense ) : ATTTACGTAG CGGCCGCTCAGGAGT AGATGGATATCTGCAGAATTCCATGAAGTTTTCATCTCC . Commercially-available HEK293 cells , original transformated as reported ( Graham et al . , 1977 ) , were co-transfected with KCC2 and GABAAR α1 or β2; KCC2 and APP/ or different APP fragments . Whenever necessary , KCC2 was replaced by mKCC2 . Transfected HEK293 cells were washed with ice-cold PBS , centrifuged at 5000 rpm for 10 min at 4°C , and the precipitate was collected and homogenized using SDS lysis buffer , centrifuged at 14 , 000 rpm for 10 min at 4°C and supernatant and denatured for 20 min at 75°C . Supernatants were analyzed by western blotting . HEK293 cells were grown in DMEM ( high glucose with L-glutamine ) containing 10% FBS . For immunoprecipitation , HEK293 cells were plated onto 10 cm culture dishes , and grown to 70% confluency . Cell transfections were performed with lipofectamine 2000 ( Invitrogen ) , following the manufacturer’s recommended protocol . 24–48 hr after transfection , cells were washed once with cold PBS then homogenized with 1 ml cold lysis buffer ( 50 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1% NP40 , and protease inhibitor ) in 2 ml tubes . To remove cell debris , lysates were centrifuged for 20 min at 14 , 000 rpm at 4°C . Supernatants were incubated with 2 μl of KCC2 ( Millipore , 07–432 ) , APPc ( Wang et al . , 2007 ) , or GABAAR α1 ( Millipore , 06–868 ) , antibodies at 4°C overnight . Protein A agarose beads at 30 μl ( Millipore ) was added to each sample and they were incubated at 4°C for 2–3 hr . To control for nonspecific binding , protein lysates incubated with rabbit or mouse IgG and beads were processed in parallel . Subsequently , beads were washed three times with 500 μl of cold lysis buffer . Immunoprecipitates were eluted from the beads by adding 30 μl of SDS-PAGE sample buffer and heating for 20 min at 75°C . The eluates ( 15 μl ) were analyzed by western blotting . For hippocampal lysates immunoprecipitation , we followed the HEK293 cells protocol above . Coomassie blue staining was conducted as described ( Savas et al . , 2015 ) . HEK293 cells were grown in DMEM containing 10% FBS . For PLA ( Söderberg et al . , 2006 ) , HEK293 cells were plated onto 10 cm culture dishes , and grown to 70% confluency . On the day of experiment , the wells were washed with 1xPBS and then added with 4% PFA and incubated for 10 min at RT without agitation . Then the cells were washed with PBS for 3 times , 5 min each , with agitation and treated with 0 . 5% Triton X-100 in PBS ( PBST ) for 10 min without agitation . Wash the cells with TBST for 3 times , 5 min each , with agitation . Block with Duolink II Blocking Solution ( 1x ) for 1 hr at 37°C and then incubate with specific primary antibodies against KCC2 ( Neuromab , 75–013 ) or APP ( Abcam , ab15272 ) . After wash with 1x Duolink II Wash Buffer A , cells were incubated with two PLA probes ( Duolink II anti-Mouse MINUS and Duolink II anti-Rabbit PLUS ) 1:5 in Antibody Diluent . Wash the slides in 1x Wash Buffer A for 2 times , 5 min each , with agitation at RT . Incubate the slides with Ligation-Ligase solution for 30 min at 37°C . Wash the slides in 1x Wash Buffer A with for 2 times , 2 min each . Then incubate the slides with Amplification-Polymerase solution for 100 min at 37°C . Wash the slides for 2 times , 10 min each , in 1x Duolink II Wash Buffer B and incubate with DAPI for 15 min at 37°C . Data was analyzed with a fluorescence microscope . KCC2 enhancers , CLP257 and CLP290 ( Gagnon et al . , 2013 ) were gifts from Yves De Koninck ( Institut universitaire en santé mentale de Québec , Qc , Canada ) . CLP257 incubation: CLP257 was dissolved in 200 µl DMSO as 100 mM stock solution and diluted using aCSF to 100 µM . For slice recording , hippocampal slices ( 350 μm thick ) of App-/- mice were cut using a vibratome ( LeicaVT1000S ) as described above for whole-cell recording in acute hippocampal slices . Slices were incubated for 2 hr in aCSF containing 100 μM CLP257 or DMSO and babbled with 95% O2 and 5% CO2 . For surface and total KCC2 protein levels evaluation , HEK293 cells , which was transfected with KCC2 were incubated for 2 hr in 100 μM CLP257 or DMSO . CLP290 intraperitoneal injection: CLP290 was dissolved in HPCD and intraperitoneally administered once daily at a dose of 100 mg/kg for one week . HPCD injection group was used as vehicle control . Unless otherwise stated , data are displayed as mean±SEM . Student’s t-test was used for statistical analysis in two-group comparison . For comparison among multiple groups , one-way ANOVA or two-way ANOVA with post hoc tests were used . Statistical significance was set at p<0 . 05 . | Alzheimer’s disease is the most common form of dementia . One of the hallmarks of the disease is the formation of sticky protein clumps called amyloid plaques in the brain . These plaques are formed from specific fragments of a protein called APP . The intact form of APP is essential for synapses ( the junctions across which neurons transmit signals ) to form and work correctly . The hippocampus is one of the first brain regions to be affected in Alzheimer’s disease and is important for forming memories and emotions . In the hippocampus , GABAA receptors at synapses normally tightly regulate synaptic signaling by reducing the ability of the receiving neuron to respond , but this inhibition is disrupted in Alzheimer’s disease . Studies suggest that APP can affect how GABAA receptors transmit signals , but it is not known how it does so . One possibility is that APP regulates a protein called KCC2 that helps to maintain the inhibitory effect of GABAA receptors . To investigate this , Chen et al . genetically modified mice to lack the gene that produces APP . These mice had a lower level of KCC2 in their hippocampus than normal mice , and their GABAA receptors were less able to inhibit synaptic signaling . Further experiments demonstrated that APP physically interacts with KCC2 and maintains normal levels of the protein by preventing it from being chemically modified and degraded . Chen et al . also showed that treating mice that lack APP with specific compounds can restore KCC2 activity and return the behavior of synaptic GABAA receptors to normal . Future studies in mice ( and eventually people ) that exhibit symptoms of Alzheimer's disease will help to determine whether KCC2 is important in the development of the disease . If so , modifying the levels of the KCC2 protein in the brain could potentially help to slow down memory loss in Alzheimer’s disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | APP modulates KCC2 expression and function in hippocampal GABAergic inhibition |
In vertebrate development , the sequential and rhythmic segmentation of the body axis is regulated by a “segmentation clock” . This clock is comprised of a population of coordinated oscillating cells that together produce rhythmic gene expression patterns in the embryo . Whether individual cells autonomously maintain oscillations , or whether oscillations depend on signals from neighboring cells is unknown . Using a transgenic zebrafish reporter line for the cyclic transcription factor Her1 , we recorded single tailbud cells in vitro . We demonstrate that individual cells can behave as autonomous cellular oscillators . We described the observed variability in cell behavior using a theory of generic oscillators with correlated noise . Single cells have longer periods and lower precision than the tissue , highlighting the role of collective processes in the segmentation clock . Our work reveals a population of cells from the zebrafish segmentation clock that behave as self-sustained , autonomous oscillators with distinctive noisy dynamics .
Populations of coordinated oscillators occur in a variety of biological systems . Examples include the rhythmic flashing of fireflies , the spiral aggregation of microbes , and the daily oscillation of circadian clocks in nearly all organisms . Communication between the individual oscillators can influence whether oscillations are maintained , i . e . their persistence , as well as their period and their precision . Without examining the properties of an individual in isolation from its neighbors , a state that we define as autonomous , it is challenging to assign the relative contribution of individual and collective processes to the observed rhythmic behavior of the population . During vertebrate embryogenesis , coordinated genetic oscillations occur in a population of cells in the posterior-most tissues of the body axis , the tailbud and presomitic mesoderm ( PSM ) . These oscillations generate a rhythmic spatial pattern . This “segmentation clock” is thought to subdivide the embryonic body into morphological segments , called somites , which arise rhythmically and sequentially from the PSM . Persistent oscillating gene expression within the tailbud and PSM corresponds to segment formation in chick , mouse , and zebrafish ( Palmeirim et al . , 1997; Dequéant et al . , 2006; Krol et al . , 2011 ) . Looking across biological systems , persistent and coherent rhythms in a population can be the product of synchronized cell-autonomous oscillators , or alternatively can be the outcome of population-level coupling of otherwise non-oscillatory cells . The autonomy of circadian clock neurons was demonstrated by recording daily oscillations in firing rate and gene expression from single cells for several cycles in the absence of their neighbors ( Welsh et al . , 1995; Webb et al . , 2009 ) . In contrast , some microbial systems have been shown to produce oscillations only when at critical densities that allowed cell-to-cell communication , otherwise the isolated cells were not rhythmic ( Gregor et al . , 2010; Danino et al . , 2010 ) . Therefore , to test for autonomy of cellular oscillators in the segmentation clock , it is imperative to determine whether individual cells can oscillate in the absence of signals from their neighbors . Historically , the term autonomy has appeared many times in the segmentation clock literature , starting with the observation that gene expression in explanted PSM can oscillate in the absence of neighboring tissues ( Palmeirim et al . , 1997; 1998; Maroto et al . , 2005 ) . This means the PSM is autonomous at the tissue level . The question of whether individual segmentation clock cells are able to oscillate autonomously , that is , when fully separated from the tissue , has been debated for decades . Early theoretical arguments explored this possibility ( Cooke and Zeeman , 1976 ) , as well as scenarios where coupling led to oscillations ( Meinhardt , 1986 ) . The possibility for an auto-regulatory negative feedback loop arising from the transcription and translation of members of the Hes/Her gene family would be consistent with a cell-autonomous mechanism ( Hirata et al . , 2002; Lewis , 2003; Monk , 2003 ) , and oscillations in this gene family have been observed across vertebrate species ( Krol et al . , 2011 ) . However , the discovery of oscillations in the Delta-Notch system in all vertebrates and in many genes of the Wnt and FGF intercellular signaling pathways in mouse and chick , raises the possibility that communication between cells may play a critical role in the generation and/or maintenance of the oscillations ( Dequéant et al . , 2006; Krol et al . , 2011; Goldbeter and Pourquié , 2008 ) . Two pioneering studies have attempted to address cellular autonomy in the chick and mouse segmentation clocks . In the first study , cells isolated from chick PSM , then cultured in suspension and fixed at subsequent time intervals , showed changes in cyclic gene expression ( Maroto et al . , 2005 ) . Due to the unavoidable uncertainty in reconstructing a time series from static snapshots of different cells , the authors of this study were not able to distinguish between noisy autonomous oscillators and stochastic patterns of gene expression , and highlighted the need for real-time reporters to investigate the autonomy of PSM cells . In the second study , the first real-time reporter of the segmentation clock , a luciferase reporter of Hes1 expression in mouse , allowed individual mouse PSM cells to be observed in vitro ( Masamizu et al . , 2006 ) . Three cells were reported , showing at most 4 expression pulses with variable duration and amplitude , which appeared to damp out . This study concluded that PSM cells may be “unstable” oscillators , and highlighted the role of intercellular coupling for maintenance of oscillations . Reflecting this , the authors modeled the cells as excitable systems dependent on noise or signals from their neighbors for pulse generation . Thus , the degree of autonomy of completely isolated cells from the segmentation clock in mouse and chick remains unclear . Working from the segmentation phenotypes of mutant zebrafish and the identity of the mutated genes , cases were originally made both for and against cell-autonomous oscillators in the zebrafish segmentation clock ( Jiang et al . , 2000; Holley et al . , 2002 ) . More recent results bring support to the idea of autonomous oscillators that are synchronized to each other . Treatment of zebrafish embryos with a γ-secretase inhibitor targeting the Notch intercellular domain ( DAPT ) leads to a loss of spatial coherence in oscillating gene expression ( Riedel-Kruse et al . , 2007 ) . Additionally , imaging of individual cells in mutant embryos with reduced Notch signaling show that oscillations persist under these conditions ( Delaune et al . , 2012 ) , though this does not rule out the possibility of Notch or other signaling factors playing a role in promoting oscillations . We recently developed an in vitro primary culture system to image gene expression in individual tailbud and PSM cells ( Webb et al . , 2014 ) , allowing the possibility of autonomous oscillations to be tested directly . In this paper we measure the intrinsic properties of single zebrafish cells isolated from the segmentation clock in the tailbud and show that they behave as autonomous genetic oscillators in vitro , in the absence of cell-cell or tissue-level coupling . We observe a striking variability in the cells’ dynamics and find that a long-timescale noise in a theoretical description of the autonomous oscillator can account for this . We then ask how the behavior of these cell-autonomous oscillators compares to oscillations at the level of the intact embryo . It is thought that collective processes at cellular and tissue level influence the period of segmentation in zebrafish . Theoretical analysis of the collective behavior of many cellular oscillators with time-delayed coupling shows that this process can set a collective period in a synchronized population ( Herrgen et al . , 2010 ) . This predicts that cells isolated from the tailbud will have a different period than the population . In addition , the precision of oscillation in a synchronized population can be higher than that of component oscillators ( Herzog et al . , 2004; Garcia-Ojalvo et al . , 2004 ) , and this scenario has been proposed for the segmentation clock ( Masamizu et al . , 2006 ) . However , an alternative case of synchronized oscillators with the same individual precision as the population has also been considered for the segmentation clock ( Horikawa et al . , 2006 ) . Knowledge of the period and precision of cells isolated from the segmentation clock should enable tests of these ideas . Using our autonomous oscillator data we characterize the period and precision of individual cells; we find that individual cells have a longer period and are less precise than the population in the tissue . Together , these results have implications for the pace-making circuit and the collective organization of the segmentation clock .
To test whether single cells behave as autonomous oscillators requires a cyclic gene expression reporter and a primary cell culture system . The cyclic bHLH transcription factor Her1 has been proposed to act within a core negative feedback loop that drives oscillations ( Holley et al . , 2000; Oates and Ho , 2002; Schröter et al . , 2012 ) and has previously been used to follow segmentation clock dynamics in the embryo ( Delaune et al . , 2012; Soroldoni et al . , 2014 ) . We used our transgenic zebrafish line Looping , which expresses a Her1-VenusYFP ( Her1-YFP ) fusion reporter driven from the regulatory elements of the her1 locus with accurate temporal and spatial dynamics in the embryo ( Soroldoni et al . , 2014 ) . Although cells are thought to slow their oscillations as they leave the tailbud and differentiate in the PSM , progenitor cells in the tailbud are thought to maintain a regular rhythm throughout development ( Delaune et al . , 2012; Giudicelli et al . , 2007; Morelli et al . , 2009; Uriu et al . , 2009; Ay et al . , 2014; Shih et al . , 2015 ) . In search of these cells , we first explanted intact tailbuds from 8-somite stage embryos homozygous for the Looping transgene ( n=3 ) ( Figure 1A; Figure 1—figure supplement 1 ) into culture and recorded the Her1-YFP signal within a central region of the tissue . After local background subtraction , we generated time series of average intensities of the regions of interest , estimated period from inter-peak intervals and measured amplitudes for each cycle ( Figure 1—figure supplement 2 ) . We observed persistent Her1-YFP oscillations ( Figure 1—figure supplement 3 ) that did not slow down ( period 42 . 5 ± 11 . 4 min ( mean ± SD ) at 26°C ) ( Figure 1—source data 1 ) . 10 . 7554/eLife . 08438 . 003Figure 1 . Zebrafish segmentation clock cells oscillate autonomously in culture . ( A ) Confocal section through the tailbud of a Looping zebrafish embryo in dorsal view where the dotted line indicates the anterior limit of tissue removed . Nuclei are shown in red and YFP expression in green . Scale bar = 50 μm . Kupffer’s vesicle ( Kv ) , notochord ( Nc ) , presomitic mesoderm ( PSM ) , tailbud ( Tb ) . ( B ) A representative 40x transmitted light field with dispersed low-density Looping tailbud-derived cells . Individual cells highlighted with black arrowheads; green arrowhead shows cell with green time series in ( D ) . Scale bar = 10 μm . Pie chart: More than half of the in vitro population of Looping tailbud cells ( n = 321 out of 547 cells combined from 4 independent culture replicates as described in Materials and methods ) expresses the Her1-YFP reporter . Some expressing cells are disqualified because they move out of the field of view ( 4% ) , touch other cells by colliding in the field of view ( 12% ) or following division ( 2% ) for a total of 14% , or do not survive until the end of the 10-hr recording ( 7% ) . ( C ) Montage of timelapse images ( transmitted light , top; YFP , bottom ) of a single tailbud cell ( green arrowhead in panel B ) over 10-hr recording . Scale bar = 10 μm . ( D ) YFP signal intensity ( arbitrary units ) measured by tracking a regions of interest over 3 single tailbud cells ( green trace follows cell marked by green arrowhead in B , gray traces are two additional cells from culture ) . Plotted in 2-min intervals . ( E ) Plot of Her1-YFP ( black ) and H2A-mCherry ( red ) signal intensity over time measured together from a representative cell . ( EI ) . Nuclear YFP signal accumulates and degrades over time , as shown in the overlay of H2A-mcherry signal ( red channel ) and Her1-YFP signal ( green channel ) during troughs ( 297 , 372 ) and peaks ( 342 , 402 ) in Her1 expression . mCherry signal in the nucleus is relatively constant . Plotted in 2-min intervals . ( F ) Plot of YFP intensity ( a . u . ) over time in a fully isolated tailbud cell within a single well of a 96-well plate . Plotted in 2-min intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00310 . 7554/eLife . 08438 . 004Figure 1—source data 1 . Summary table of segmentation clock tissue and cellular oscillatory properties . Summary of statistics of time series traces recorded and analyzed in vitro in tailbud explants or tailbud cells . Peaks were identified , and the period/amplitude of cycles was determined as described in Materials and methods . A maximum period is defined in the method at 140 min , approximately twice the mean . Serum only cells were from the same cell suspension as those that were then treated with Fgf8b for the experiments 280711 and 250112 . Pooled data from N = 2 independent cultures , for a total of n = 52 serum only cells . Pooled data from N = 4 independent cultures , for a total of n = 149 Fgf8b treated cells . To culture fully isolated cells , a cell suspension was serially diluted in wells within a 96-well plate , producing wells with a single tailbud cell . These were also treated with Fgf8b . N = 5 independent 96-well plate experiments , with a total of n = 10 fully isolated cells in these experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00410 . 7554/eLife . 08438 . 005Figure 1—source data 2 . Summary table of low-density segmentation clock cell experiments . Description of in vitro cultured tailbud cell population treated with Fgf8b ( n = 547 ) , using multiple donor embryos in each of 4 independent experimental replicates ( N = 4 ) , carried out on separate days . Across the 29 fields recorded , we observed cell divisions in both YFP-negative ( 30 , 5% of total cells ) and YFP-positive cells ( 13 , 2% of total cells ) . We found a range in the number of cell divisions from 0 to 5 cells per field , with an average of 1 . 5 ( ±1 SD ) divisions per field . The categories of disqualification list the first event in a recording that led to disqualification . For example , four divisions in YFP-positive cells occurred after the cell had been disqualified for another reason ( movement in and out of field , touching another cell ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00510 . 7554/eLife . 08438 . 006Figure 1—source data 3 . Time series data from low-density segmentation clock cells . XLS file containing all time series data for each of the 147 low-density segmentation clock cells in the presence of Fgfb . The file contains 4 work-sheets corresponding to each of the 4 independent replicates and to the plots in Figure 1—figure supplement 5 . In each sheet , each cell is described by 3 neighboring columns: average fluorescence , local background , and background subtracted signal . Cells are also listed by their field of view in the original microscopy files . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00610 . 7554/eLife . 08438 . 007Figure 1—figure supplement 1 . Her1-YFP-expressing cells in the zebrafish tailbud . A confocal section through the tailbud of a Her1-YFP and Histone 2A-mCherry expressing 8-somite stage Looping zebrafish embryo in both lateral and dorsal orientations . The approximate location of the segmentation clock cells removed by surgery to generate the tailbud explants and single cell cultures is shown with the dashed line . Nuclei are shown in red and YFP in green . Scale bar = 50 μm . Kupffer’s vesicle ( Kv ) , notochord ( Nc ) , presomitic mesoderm ( PSM ) , tailbud ( Tb ) , neural progenitors ( Np ) , yolk cell ( yolk ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00710 . 7554/eLife . 08438 . 008Figure 1—figure supplement 2 . Peak finding in time series to estimate period and amplitude . ( A ) An example of peak finding from a representative low-density cell trace , showing the steps used to find and estimate inter-peak intervals and amplitude . Top: raw data ( red line ) and background ( grey line ) . Middle: peak finding ( red triangles ) and filtering ( blue triangles ) from the smoothed signal ( thin blue line ) of the background subtracted trace ( red line ) . Bottom: resulting peaks ( blue dots ) and troughs ( green dots ) . ( B ) Definition of period T as inter-peak interval ( orange double arrow ) , and amplitude A as the average of peak heights relative to the trough ( green double arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00810 . 7554/eLife . 08438 . 009Figure 1—figure supplement 3 . Persistent oscillations in explanted tailbud . ( A ) Montage of brightfield and corresponding YFP images from representative explanted tailbud over ~7 hr recording . Brightfield image is a single z-plane , while YFP signal is shown as an average projection of the entire stack . Oscillations in the tailbud were measured using a region of interest ( ROI; black circle ) to extract average YFP intensity values over time . We placed this region on the most central “tailbud” area , as we observed “PSM”-like protrusions ( black arrowheads ) emerging from the explant . These areas , as expected for PSM , showed brighter and increasing YFP expression over time , which then switched off . ( B ) Intensity over time plot for the ROI shown in A . The tailbud area shows 9 cycles over the recording with no slowing of the period . Peak finding was performed as in Figure 1—figure supplement 2: background subtracted raw data ( thick red line ) was smoothed ( thin blue line ) and peaks were identified ( red down triangles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 00910 . 7554/eLife . 08438 . 010Figure 1—figure supplement 4 . Time series of low-density segmentation clock cells in serum-only culture . ( A ) Individual tailbud cells from experiment 280711 in the presence of serum . Black time trace is the raw data after background subtraction , the background level is the black line , the smoothed curve ( red ) was used for peak counting and identification of peaks and troughs ( blue circles ) . Details of smoothing and peak finding are given in Materials and methods and Figure 1—figure supplement 2 . ( B ) As above for experiment 250112 . Corresponding cells grown in the presence of serum + Fgf8b from the same tailbud cell suspensions in this figure are found in Figure 1—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01010 . 7554/eLife . 08438 . 011Figure 1—figure supplement 5 . Time series of low-density segmentation clock cells . Traces from each independent low-density culture replicate ( serum + Fgf8b ) are shown in separate panels ( #070312: yellow , #012512: green , #251012: red , #280711: blue ) . Each raw trace ( black ) was smoothed ( red ) and peaks and troughs were identified ( blue circles ) . This is the complete low-density cell data set from which representative examples shown in Figure 1B–D are chosen . Details of smoothing and peak finding are given in Materials and methods and Figure 1—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01110 . 7554/eLife . 08438 . 012Figure 1—figure supplement 6 . Characterization of Ntla and Tbx16 antibodies . ( A , D ) Graphical representation of the full-length Ntla and Tbx16 proteins with exons depicted in different colors . Blue arrows show the predicted T-box domain from amino acid 35 to 212 , and 31 to 213 , for Ntla and Tbx16 respectively . The Ntla antibody ( clone D18-4 , IgG1 ) was generated using a peptide from amino acid 1 to 261 , while the corresponding sequence for the Tbx16 antibody ( clone C24-1 , IgG2a ) spans the region from amino acid 232 to 405 ( bracketed in red in both cases ) . ( B , E ) Representative examples showing ntla mRNA and protein expression patterns in wild type and in ntla mutant embryos at 90% epiboly . Immunolabeling using the Ntla antibody was followed by in situ hybridization using a ntla riboprobe . The same procedure was used to characterize the Tbx16 antibody except that wild type embryos are compared to tbx16 morpholino-injected embryos and a tbx16 riboprobe was used . Scale bar = 150 μm . ( C , F ) Immunolabeling of wild type embryos injected at 1-cell-stage with capped mRNAs ( Ntla-T2A-mKate2CAAX , Tbx6-T2AmKate2CAAX , Tbx6l-T2A-mKate2CAAX or Tbx16-T2A-mKate2CAAX ) , fixed at 4 hr post fertilization and imaged at the animal pole where the endogenous genes are not expressed . Both Ntla ( C ) and Tbx16 ( F ) antibodies bind only in embryos injected with ntla and tbx16 mRNA respectively , demonstrating antibody specificity . Scale bar = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01210 . 7554/eLife . 08438 . 013Figure 1—figure supplement 7 . Expression of tailbud markers in vivo and in low-density cultures of segmentation clock cells . ( A ) z-stack projection showing the expression patterns of Ntla ( green ) and Tbx16 ( red ) protein in a 12-somite stage wild type embryo detected using immunohistochemistry with monoclonal antibodies D18-4 ( IgG1 ) to Ntla and C24-1 ( IgG2a ) to Tbx16 . Scale bar = 120 μm ( B-C ) Close up view of the boxed area in ( A ) showing a single confocal section at dorsal ( B ) and ventral ( C ) locations in the tailbud . Scale bar = 60 μm . ( D ) Representative panels showing expression of Ntla ( green ) and Tbx16 ( red ) in single cells within low-density tailbud cultures ( serum + Fgf8b ) after 5 hr in vitro . Nuclei are labeled with DAPI ( blue ) . Cells single-positive for Ntla and Tbx16 are visible , as are cells co-expressing both proteins . Scale bar = 20 μm . ( E ) Quantification of nuclear fluorescence intensity of experiment in ( D ) showing populations of Ntla-positive , Tbx16-positive , and Ntla/Tbx16 co-expressing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01310 . 7554/eLife . 08438 . 014Figure 1—figure supplement 8 . Analysis of low-density segmentation clock cell cultures . ( A ) Histogram of the number of peaks observed in each cell in the presence of serum + Fgf8b ( blue ) compared to those from cells in the presence of serum alone ( orange ) . ( B ) Histogram of periods from all measured cycles for low-density cells in the presence of serum + Fgf8b ( blue ) compared to those from cells in the presence of serum alone ( orange ) . ( C ) Histogram of amplitudes from all measured cycles for low-density serum + Fgf8b data set . See Figure 1—source data 1 for statistics . Average amplitude ( D ) and period ( E ) plotted vs . time intervals for all cycles in the four different serum + Fgf8b low-density experiments ( rows ) . Data is grouped in bins of 100 min , error bars show variance . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01410 . 7554/eLife . 08438 . 015Figure 1—figure supplement 9 . Time series of fully isolated segmentation clock cells . ( Top row ) An example 40x transmitted light field of a single , isolated tailbud cell in a 96-well plate well . Scale bar = 20 μm . The cell’s corresponding fluorescence time series at the right shows raw YFP intensity over time ( black ) , smoothed signal ( red ) , and automatically detected peaks ( green arrowheads ) . Cells were obtained from N = 5 independent replicates . From 43 YFP-expressing cells , 33 were disqualified due to death , division or movement from imaging field , leaving n = 10 for analysis . ( Bottom panels ) As above , each plot shows the background-subtracted average YFP intensity levels from a single cell over time ( black ) , smoothed signal ( red ) , and peaks ( green arrowheads ) for the remaining 9 fully isolated segmentation clock cells . Peak finding was first performed as in Figure supplement 1–2 . Due to the higher noise levels in the fully isolated segmentation clock cells , we modified the parameters of the algorithm to be less stringent , and we introduced an additional step of curation to remove the detected peaks that had very low amplitude and were considered spurious . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 015 We next used our primary culture protocol ( Webb et al . , 2014 ) to generate low-density cultures of tailbud cells from multiple Looping tailbuds , which were recorded for 10 hr . In identical culture conditions to the explanted tailbuds , most individual cells that expressed Her1-YFP displayed a few pulses that appeared to damp out early in the recording ( N = 2 independent cultures , n = 52 total cells , median cycles = 2 ) ( Figure 1—figure supplement 4; Figure 1—source data 1 ) . Fgf is expressed at elevated levels in the tailbud and is proposed to play roles in maintaining an oscillatory progenitor state in the embryo ( Dubrulle et al . , 2001; Sawada et al . , 2001; Ishimatsu et al . , 2010; Niwa et al . , 2007 ) . In contrast to the serum-only treatment , when Fgf8b was added to part of the same cell suspension within a divided imaging dish , a persistent rhythmic behavior was observed ( Figure 1—figure supplement 5 ) . The number of cycles increased dramatically , spanning the recording interval and without altering the period ( N = 4 independent cultures , n = 547 total cells; median cycles = 5; Figure 1—source data 1 ) . An imaged field contained 15 to 20 cells ( Figure 1B , black arrowheads ) and more than 50% of these cells ( 321/547 ) had YFP signal ( Figure 1B , Figure 1—source data 2 ) . Typically , cells remained rounded and displayed blebs , as expected from early zebrafish progenitors in vivo and in vitro ( Diz-Muñoz et al . , 2010; Maître et al . , 2012; Ruprecht et al . , 2015 ) . Cells from the same embryos cultured in parallel maintained tailbud marker expression ( Figure 1—figure supplement 6 , Figure 1—figure supplement 7 ) , suggesting that under these conditions the cells retain a tailbud progenitor phenotype ( Martin and Kimelman , 2010; 2012 ) . For the remainder of the manuscript we focus on the Fgf8b-treated cells as a model to understand the properties of segmentation clock oscillators . From the 321 cells that were YFP-positive at the beginning of the recording , we first excluded from analysis any cell that moved out of the imaging field , died , or touched another cell either through movement or division at any point in the recording . From the remaining 189 autonomous cells , we observed rhythmic expression ( two or more pulses of expression ) in 147 , which we term the low-density data set ( Figure 1B–D; Figure 1—figure supplement 5; Figure 1—source data 2; Figure 1—source data 3 ) . An illustrative imaged field is shown in Video 1 . Although some cells continued to oscillate after division , analysis of these rare events was complicated by the tendency of the daughters to strongly adhere to each other , and is beyond the scope of this study . We found a distribution of periods ( 78 . 8 ± 15 . 3 min [mean ± SD from 442 cycles] ) and amplitudes ( Figure 1—figure supplement 8A , B ) . Importantly , we observed no systematic slowing in successive oscillations in our data suggesting that the cultures were stationary over this interval ( Figure 1—figure supplement 8E ) . 10 . 7554/eLife . 08438 . 016Video 1 . Low-density segmentation clock cells oscillate in vitro . Field of view containing cell in Figure 1B–C , highlighted by the red arrow . This field contains 18 cells in total , with 9 expressing YFP . We observed 5 cell divisions , the highest number in any experiment , including one non-YFP cell , 3 YFP-positive cells , which are excluded from analysis because of division , and one YFP-positive cell disqualified due to contact with another cell in the field prior to the division . The remaining 5 YFP-positive cells , including the highlighted cell , are part of the low-density data set . Total duration = 10 hr; Time interval = 2 min; field size = 410 x 410 μm; Scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 016 To rule out the possibility that oscillations in the YFP signal were influenced by focal drift , we recorded from individual cells co-expressing Her1-YFP and the nuclear marker H2A-mCherry . We found that the Her1-YFP signal co-localized with the nuclear mCherry signal , which did not oscillate ( n = 8 cells , Figure 1E , EI ) indicating that focal drift does not contribute strongly to changes in YFP intensity . Combined , these results show that cells from the zebrafish tailbud do not need cell-cell contact to maintain oscillations . Nevertheless , diffusible factors could be released rhythmically in these cultures and maintain oscillations . To test the ability of a fully isolated cell to oscillate , we used serial dilution to obtain and image single tailbud cells isolated in individual culture chambers . We found that these fully isolated cells can also sustain oscillations ( N = 5 , n = 10 , period = 62 ± 21 min ( mean ± SD ) , median cycles = 5 ) ( Figure 1F; Figure 1—figure supplement 9; Figure 1—source data 1; Video 2 ) . 10 . 7554/eLife . 08438 . 017Video 2 . Isolated segmentation clock cells oscillate in vitro . Field of view corresponding to cell in top row of Figure 1—figure supplement 9 . Total duration = 6 . 2 hr; Time interval = 2 . 14 min , field size = 205 x 205 μm , Scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 017 Together , these data reveal the existence of a population of autonomously oscillating cells from the zebrafish segmentation clock . Mechanisms of oscillation in zebrafish based on reaction-diffusion processes ( Meinhardt and Gierer , 2000 ) , and which rely on diffusion across the tissue and do not contain cell-autonomous oscillators , are therefore not supported by our results . In addition , in nearly all cases the period of individual cells is longer than that of the tissue , indicating a role for tissue-level processes in setting the period of segmentation . The oscillatory signal we observe from individual cells is reporting the state of a pace-making circuit component ( Schröter et al . , 2012 ) . A remarkable feature of these oscillatory signals is their variability between cells in the population ( Figure 1—figure supplement 5 ) . We observe a spectrum of behaviors in the low-density data set including cells that start or stop oscillating during the experiment , cells that start and then stop , and stuttering rhythms where cycles are missed ( Figure 2A ) . Plotting amplitude and period of consecutive cycles from the whole low-density data set indicate that amplitude displays a slow variation revealed by correlations ( Figure 2B ) , while period does not show any appreciable correlation at this timescale ( Figure 2C; Figure 2—figure supplement 1A–C ) . We did not find a correlation between the period and amplitude of each cycle ( Figure 2—figure supplement 1D ) . 10 . 7554/eLife . 08438 . 018Figure 2 . Dispersed low-density cells show a variety of behaviors compatible with slow amplitude fluctuations . ( A ) Representative background-subtracted traces displaying different oscillatory behaviors: persistent oscillations , oscillations that initiate , stop , or start and stop within the recording time of 600 min . ( B ) Amplitude correlations in successive cycles from the Fgf-treated low-density data set ( 645 cycles measured from 147 cells ) are shown as red squares . Blue crosses show correlations from the same data set with pairs of peaks drawn at random from the same list . Amplitude values are normalized to the mean of the data set . ( C ) Period correlations in successive cycles from the Fgf-treated low-density data set are shown as red squares . Blue crosses show correlations from the same data set with period values drawn at random from the same list . Period is normalized to the mean of the data set . ( D ) Left: Scheme defining amplitude and period and corresponding limit cycle illustrating fluctuations in μ and ω , which are parameters controlling amplitude and frequency , respectively . Middle: equation of the generic Stuart-Landau oscillator model , which describes the time evolution of phase θ and amplitude r of the oscillator . Right: illustration of the Hopf bifurcation showing how the limit cycle ( blue circle ) collapses and becomes a fixed point ( blue dot ) as μ changes from positive to negative values . ( E ) Simulated traces generated with the Stuart-Landau model with colored noise in parameter μ and white noise in the oscillator variables , showing behaviors corresponding to those observed in the data , compare to panel A . ( F ) Amplitude correlations in successive cycles from the simulated oscillator are shown in red squares . Blue crosses show correlations from the same data set with pairs of peaks drawn at random . ( G ) Period correlations in successive cycles from the simulated oscillator shown in red squares . Blue crosses show correlations from the same data set with inter-peak-intervals from pairs of peaks drawn at random . ( H ) Heat plot of the fraction of time spent oscillating as measured by number of peaks occurring over time given the median period observed in the synthetic data , as the variance and correlation time of colored noise fluctuations in μ vary . Oscillating fraction of time for the Fgf-treated low-density data set ( Figure 1—figure supplement 5; Figure 1—source data 3 ) would be found in the shaded region of the heat plot . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01810 . 7554/eLife . 08438 . 019Figure 2—figure supplement 1 . Statistics of amplitude and period correlations . ( A ) Definition of averages Ai* ( blue dot ) of a quantity A measured in consecutive cycles ( red square ) , and the distance Di ( green line ) to the identity ( grey line ) , employed in panels ( B ) and ( C ) . ( B ) Histograms of normalized average consecutive amplitudes Ai* ( top ) and the distances Di to the identity line ( bottom ) , computed for all cycles in the low-density data set ( Figure 2B ) . ( C ) Histograms of normalized average consecutive periods Ti* ( top ) and the distances Di to the identity line ( bottom ) , computed for all cycles in the low-density data set ( Figure 2C ) . ( D-I ) Correlation of amplitude and period . Amplitude and period values are normalized to the mean of the data set for each cycle . ( D ) Fgf-treated low-density data set: 645 cycles measured from 147 cells . ( E-I ) Simulated oscillators with μ=1 , b=1 , σz2=0 . 486 , and τμ=476 min . ( E ) q=0 , min-1 , ( F ) q=−0 . 001 , ω=2π/78 min-1 , ( G ) q=0 . 001 , ω=2π/78 . 3 min-1 , ( H ) q=−0 . 005 , ω=2π/96 . 7 min-1 , ( I ) q=0 . 005 , ω=2π/65 . 5 min-1 . Since non-isochronicity affects the instantaneous frequency for q≠0 , we adjust the value of the autonomous frequency ω to keep the mean period of oscillation at 78 min . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 01910 . 7554/eLife . 08438 . 020Figure 2—figure supplement 2 . Numerical simulations . ( A ) Numerical simulation of equation ( S30 ) , see Supplementary ile 1 , with μ = 1 , b = 1 , ω = 2π/78 min-1 , q = 0 , . Signal x ( t ) oscillates with a constant amplitude and a period T = 78 min . There is no additive noise and the phase θ ( t ) increases monotonically with time , θ ( t ) ~ ω t , so when wrapped in the interval [−π , π] the phase is periodic with T = 78 min . ( B-D ) Numerical simulations of equation ( S30 ) with μ = 1 , b = 1 , ω = 2π/78 min-1 , q = 0 , , σμ2=6 . 84 and τμ=476 min . Each case B-D shows a different dynamical state that depends on the trajectory of μ ( t ) . In all cases , top panel shows signal x ( t ) oscillatory behavior showing amplitude fluctuations . Second panel is μ ( t ) =μ+ξμ ( t ) . Third panel is amplitude of the signal r ( t ) =x ( t ) 2+y ( t ) 2 . When μ ( t ) < 0 , the system crosses the Hopf bifurcation and amplitude drops to zero . Fourth panel shows the additive white noise in variable x , ξx ( t ) . Bottom panel shows phase θ ( t ) increases monotonically in time θ ( t ) ~ ω t , when wrapped in the interval [−π , π] the phase is periodic with T = 78 min . Phase is not defined when r = 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 02010 . 7554/eLife . 08438 . 021Figure 2—figure supplement 3 . Both additive noise and color noise are necessary for the theory to describe the observed fluctuations . ( A ) Amplitude is not affected by additive noise . Data points show the median peak amplitude for 1000 stochastic simulations ( S30 ) , see Supplementary file 1 , for σμ2=0 . Error bars are the 68% confidence interval . Median of the amplitude remains constant as we increase σz2 , while amplitude fluctuations increase . ( B ) Median period is not affected by additive noise . Data points show the median value of the period for 1000 stochastic simulations ( S30 ) for σμ2=0 . Error bars are the 68% confidence interval . The median period from numerical simulations does not change when the variance of the additive noise increases . Dashed line joins data points . ( C ) Additive noise does not affect oscillating time fraction . Since the mean period is constant and additive noise does not modulate the amplitude , the oscillating time fraction remains constant . Data points show the median value of the oscillating time fraction for 1000 stochastic simulations ( S30 ) for σμ2=0 . Error bars are the 68% confidence interval . The dashed line joins data points . ( D ) Amplitude correlations of x ( t ) in successive cycles for fixed σz2 = 0 . 486 and σμ2=0 . Red squares are data points , blue crosses show results of a randomized list . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 021 We adopted a theoretical approach to better understand these observations . We chose a generic Stuart-Landau ( SL ) model that describes the phase θ and amplitude r of an oscillator in the vicinity of a supercritical Hopf bifurcation ( Figure 2D ) ( Strogatz , 1994 ) . In this description , the amplitude of oscillations grows with the square of the distance to the bifurcation . Existing genetic regulatory network models possess supercritical Hopf bifurcations ( Verdugo and Rand , 2008a; 2008b; Wu and Eshete , 2011 ) , though the topology and parameter values of the pace-making circuit remain unclear ( Schröter et al . , 2012; Uriu et al . , 2009; Lewis , 2003; Trofka et al . , 2012 ) . The SL description allows us to examine the effects of noise strength and correlation time on frequency and amplitude , neatly separated and in combination ( Supplementary file 1 ) . Amplitude fluctuations observed in the data occur on a timescale that is similar to that of the oscillator period , and could be the result of global changes in the cell state . Slow fluctuations in gene expression and signaling systems have been reported in a variety of systems ( Süel et al . , 2006; Sigal et al . , 2006; Huang , 2009; Chang et al . , 2008; Albeck et al . , 2013; Aoki et al . , 2013 ) . We introduced slow fluctuations in the parameter μ that controls the amplitude of oscillators in the theory . Motivated by the lack of correlation between period and amplitude in the data ( Figure 2—figure supplement 1D–I ) , we set the coupling between these processes to zero . Slow amplitude fluctuations can drive the oscillators in and out of the oscillatory state ( Figure 2E; Figure 2—figure supplement 2 ) , and introduce correlations in the amplitude of consecutive cycles that are comparable to the experimental data ( Figure 2F ) . Interestingly , the trend to higher amplitude variance at higher amplitude values , and the existence of a low occurrence of high relative amplitude cycles are not captured by the theory . To describe period fluctuations and their weak correlation observed in the data we introduced an additive white noise in the variables of the oscillator ( Figure 2C , G ) ( Supplementary file 1 ) . Since oscillators move in and out of the oscillatory state , a key observable in both model and data is the fraction of the total time that a cell oscillates . We performed simulations over a range of values of correlation time and variance in μ and found a region in parameter space that corresponds to the behavior of isolated cells ( Figure 2H ) . In contrast to these effects , changing the variance of the additive white noise did not affect the distribution of amplitudes , the correlations of amplitudes , or the fraction of the oscillating time ( Figure 2—figure supplement 3 ) . While these results cannot rule out potential cell-type differences in the population , the theory is consistent with a population of cells having the same oscillatory mechanism , captured during different time windows of their dynamics . Importantly , it provides for the first time an observation of the longer timescales of noise in the autonomous oscillations of cells from the segmentation clock , regardless of its source . We next compared the precision of the most reliable of the cellular oscillators to the precision of collective oscillations in the intact embryo . From the 147 cells in the low-density data set , we selected those cells with persistent oscillatory behavior , defined as cells exhibiting sequential peaks over 80% of the recording time ( n = 54 cells; Figure 3—source data 1 ) . We used a wavelet transform to extract the phase of oscillatory signals ( Figure 3A ) . We then evaluated precision by means of the quality factor Q , defined as the ratio of the decay time of the autocorrelation function and the period of oscillations ( Morelli and Jülicher , 2007 ) . All time series were sampled using time windows of equivalent lengths ( 6 . 5 cycles ) ( Figure 3—figure supplement 1 ) ; this procedure and the observed distribution of Q values are described in Supplementary file 1 . We found that persistent cells had a mean period of 78 . 3 min ( Figure 3B , inset ) . This is consistent with the period inferred from the inter-peak intervals measured from the entire low-density data set . We calculated a median quality factor QP ~4 from the persistent cells’ oscillations ( Figure 3B ) . Our analysis excluded dividing cells , and as cell division is thought to introduce phase noise in the time series ( Delaune et al . , 2012 ) , the precision of dividing cells would be lower . 10 . 7554/eLife . 08438 . 022Figure 3 . Precision of persistent segmentation clock oscillators ( A ) Quality factor workflow for time series analysis for an example persistent oscillator . Sub-panel 1: Background-subtracted intensity over time trace from a single tailbud cell ( black ) with phase ( gray ) . Sub-panel 2: Wavelet transform of the intensity trace with cosine ( light blue ) of the phase information ( gray ) . Sub-panel 3: Autocorrelation of the phase trace and fit ( green ) of the decay ( for details see Supplementary file 1 ) . The period of the autocorrelation divided by its correlation time is the quality factor plotted in B for each cell ( blue ) . ( B ) Distribution of quality factors QP for persistent segmentation clock oscillators ( blue; range 1–28 , median 4 . 6 ± 5 . 8 ) compared to quality factors QE for the oscillating tailbud tissue in the embryo ( red; range 1–117 , median 10 ± 21 ) . To compare between time series of different lengths we used sampling windows to calculate the quality factors , see theoretical supplement for details . Median values are indicated by dotted lines . Inset: Distribution of periods in single tailbud cells . ( C ) Estimation of tissue-level quality factor determined by measuring from an ROI placed over posterior PSM tissue in whole embryo timelapse of a single Looping embryo ( Soroldoni et al . , 2014 ) . The intensity trace ( black ) and cosine ( light blue ) correspond to the average signal in the ROI over time . The period of the fit of the autocorrelation ( green ) divided by its correlation time is the quality factor plotted in B ( red ) . ( D ) Distribution of quality factors for persistent segmentation clock oscillators ( blue ) replotted from B compared with the distribution of quality factors for circadian fibroblasts ( orange; range 1–149 , median 20 ± 27 ) . Median values are indicated by dotted lines . Inset: Distribution of periods in circadian fibroblasts . ( E ) Precision decreases with increasing additive noise . Top panel , quality factor Q vs . variance σz2 of the additive noise , from numerical simulations ( S30 ) . Dots are the median value and error bars display the 68% confidence interval for 1000 stochastic simulations . Black line and shaded region indicates the median and the 68% confidence interval of persistent cells’ oscillations . Bottom panel , p-value of a two-sample Kolmogorov–Smirnov test vs . variance σz2 . We test whether the persistent cells oscillations and the quality factors obtained from simulations come from the same distribution . In the absence of amplitude fluctuations σμ2 = 0 , for σz2 = 0 . 486 we have Q = 4 . 6 and a p-value of 0 . 78 . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 02210 . 7554/eLife . 08438 . 023Figure 3—source data 1 . Precision and period calculation for persistent segmentation clock oscillators . Each set of panels shows , successively , the background-subtracted average YFP intensity levels over time from a single persistently oscillating cell in black; the cosine of the phase calculated from the wavelet transformation in blue; and the autocorrelation function in green . The dashed green curve shows the analytical fit of the autocorrelation . Both period and quality factor can be calculated from this procedure ( see Supplementary file 1 ) . This is the complete persistent cell data set , a sub-set of the low-density set , from which the plots of period andquality factor QP in Figure 3B and D are generated . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 02310 . 7554/eLife . 08438 . 024Figure 3—source data 2 . Precision and period calculation for the tissue-level segmentation clock in the zebrafish embryo . As for data set supplement 3–1 , each set of panels shows , successively , the background-subtracted average YFP intensity levels from a region of posterior PSM tissue in a Looping embryo in black; the cosine of the phase calculated from the wavelet transformation in blue; and the autocorrelation function in green . The dashed green curve shows the analytical fit of the autocorrelation . Both period and quality factor can be calculated from this procedure . The original intensity versus time data comes from Soroldoni et al . ( 2014 ) . This is the complete dataset from time-lapse data of 24 embryos from which the plot of quality factor QEmbryo in Figure 3B is generated . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 02410 . 7554/eLife . 08438 . 025Figure 3—source data 3 . Precision and period calculation for persistent circadian clock oscillators . As for data set supplement 3–1 , each set of panels shows , successively , the background-subtracted intensity levels from a single persistently oscillating Per2-Lucifcerase-expressing fibroblast over time in black; the cosine of the phase calculated from the wavelet transformation in blue; and the autocorrelation function in green . The dashed green curve shows the analytical fit of the autocorrelation . Both period and quality factor can be calculated from this procedure . The original intensity versus time data comes from Leise et al . ( 2012 ) . This is the complete fibroblast dataset from which the plot of quality factor QF in Figure 3D is generated . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 02510 . 7554/eLife . 08438 . 026Figure 3—figure supplement 1 . Quality factor value depends on length of time series . Time series length is defined in terms of the number of cycles . The plot shows the quality factor from stochastic simulations for two parameter sets A and B that display QA = 4 and QB = 10 when the number of cycles used to compute Q is 4 . The quality factor Q decreases with increasing number of cycles in the time series in both cases , but QA remains consistently smaller than QB . For all the different values analysed , the distributions have a p-value < 10–10 using the Kolmogorov-Smirnov test ( asterisks ) . Thus , while the quality factor may depend on the length of the time series , we can use it to compare different datasets as long as we compare time series with the same number of cycles . For the comparison in this work , we use 6 . 5 cycles . DOI: http://dx . doi . org/10 . 7554/eLife . 08438 . 026 We compared the precision of persistent oscillators in vitro to the precision of the tailbud during an equivalent developmental time window using tissue-level oscillation data from embryos in Soroldoni et al . ( 2014 ) . We calculated a median quality factor QE ~10 in the embryo ( Figure 3B , C; Figure 3—source data 2 ) . This value indicates that the precision of the tissue level oscillations in vivo is a factor of 3 times higher than the typical persistent segmentation clock cell in vitro . This suggests that in the embryo , coupling may increase the precision of the individual cells by a similar amount . Given that there are many cells with lower Q factor than the median , including those that were too noisy to be included in the persistent oscillator set , this increase in precision should be considered as a lower bound . To place the precision of persistent cells in context with another autonomous cellular oscillator , we analyzed the precision of the circadian clock using time series from single mammalian fibroblasts expressing a Period2-Luciferase reporter ( Leise et al . , 2012 ) and found these cells oscillated with a median quality factor QF ~20 ( Figure 3D; Figure 3—source data 3 ) . Thus , cells isolated from the zebrafish segmentation clock are less precise than the mouse circadian clock in single cells . We investigated the precision of simulated oscillators , comparing it to the experimental data . Noise in the parameter μ , which controls the amplitude and produced the heterogeneity of oscillator behavior discussed above , did not result in the observed range of quality factors ( Supplementary file 1 ) . In contrast , introducing white noise in the variables of the SL oscillator described the precision of the data ( Figure 2D ) . This choice of noise was motivated by a lack of correlation in subsequent periods ( Figure 2C , G ) . The experimentally observed precision of persistent cells from the segmentation clock was located within a restricted range of variance in this noise ( Figure 3E ) . Together these data demonstrate a role for collective processes in increasing the precision of the tissue-level segmentation clock above that of the individual isolated cells . Our findings improve understanding of the segmentation clock at both single cell and tissue level . Single cell oscillators isolated from the zebrafish segmentation clock are autonomous . Nevertheless , tissue-level contributions aid in establishing the period and increasing the precision of segmentation . In zebrafish , inhibiting the Delta-Notch pathway that mediates synchronization between neighboring oscillators results in slower segmentation ( Herrgen et al . , 2010 ) . This change in period has been interpreted as the result of losing the collective effects of coupling with delays ( Herrgen et al . , 2010; Morelli et al . , 2009 ) . An untested prediction of this scenario is that an individual isolated tailbud cell should slow when removed from coupling within the tissue . Our experiments allowed us to compare the period of the low-density cells to that of the explanted Looping tailbud cultured under the same conditions . We first noted that the period of the oscillations measured in tailbud explants ( 42 . 5 ± 11 . 4 min , mean ± SD ) ( Figure 1—figure supplement 3 ) was longer than the segmentation period in intact embryos of 27 ± 1 min ( ± SD ) at 26°C ( Schröter et al . , 2008 ) , a slowing of approximately 1 . 5-fold over the intact embryo . A comparable increase in period was not reported with explanted mouse tailbuds ( Masamizu et al . , 2006; Lauschke et al . , 2012 ) . However , a general developmental slowing of explanted zebrafish tissue has been previously reported ( Langenberg et al . , 2003 ) . The reason for this is unknown , and likely involves chemical or mechanical differences in culture compared to the embryo . Recent studies of embryonic cell shape and migration have successfully utilized in vitro assays over minutes to tens-of-minutes time scales ( Diz-Muñoz et al . , 2010; Maître et al . , 2012; Ruprecht et al . , 2015 ) , but the dynamics of longer-term zebrafish primary cell culture remains relatively unexplored ( Westerfield , 2000 ) . The differentiation of several lineages in primary cell culture appears to be slowed compared to the embryo , although this has not prevented the identification of relevant molecular regulatory mechanisms in this context ( Xu et al . , 2013; Huang et al . , 2012 ) . The zebrafish segmentation clock can maintain stable oscillatory output over a three-fold change in frequency due to temperature differences ( Schröter et al . , 2008 ) , suggesting that its mechanism is robust to alterations in global growth conditions . Nevertheless , until the mechanism of this general slowing in vitro and its influence on the molecular and cellular processes within the segmentation clock are understood , we must remain circumspect in our interpretations . The period inferred from the low-density data set was approximately two-fold longer than that of explanted tailbuds using the same time series analysis ( Figure 1—source data 1 ) . This supports the expectation that coupling with time delays between segmentation clock cells in the zebrafish leads to a decrease in the period of the synchronized population ( Herrgen et al . , 2010; Morelli et al . , 2009 ) . However , the magnitude of the difference is larger than anticipated from the segmentation period of intact embryos with reduced Delta-Notch signaling , where the effect of delayed coupling was estimated at 20% ( Herrgen et al . , 2010 ) . This difference may be due to additional , as yet unknown coupling pathways in the tissue , and/or to the existence of signals in the tissue that alter the base period with which the cell can tick , and which are diluted by the low-density culture . Our own observations with a range of Fgf8b concentrations indicate that it does not affect period ( Figure 1—figure supplement 8D , and data not shown ) , suggesting that Fgf signaling is unlikely to be responsible . In this assay , rather than being instructive for the period of oscillations ( Ishimatsu et al . , 2010 ) , Fgf appears to be permissive for the oscillatory state ( Dubrulle et al . , 2001 ) . In summary , our results demonstrate that individual cells have a longer period than the period in the tissue . Thus , they provide independent support for a role for collective processes in determining the period of the tissue-level segmentation clock . A striking finding of our studies was the heterogeneity of oscillations across the population of cells . We propose that this heterogeneity can be described as the trajectories of self-sustained oscillators in the vicinity of a Hopf bifurcation . In this scenario , excursions across the bifurcation stop and start cycling behavior , but the underlying oscillatory mechanism remains . Fgf signalling appears to push the oscillators above this bifurcation , and the characteristic longer-timescales in the amplitude noise that we have observed may come from the inherent dynamics of the Erk network downstream of the Fgf receptor ( Aoki et al . , 2013; Albeck et al . , 2013 ) . One feature of oscillators in the vicinity of a bifurcation is that they may be more readily synchronized by coupling ( Gonze et al . , 2005; Barral et al . , 2010 ) . Individual cells from the mouse segmentation clock are also noisy , but the contribution of cell-cell signaling to maintenance of oscillations , either by contact or through diffusible factors , remains unclear ( Masamizu et al . , 2006 ) . It is possible that in the mouse intercellular signaling may be required for sustaining oscillations , as has been observed for Delta-Notch signaling in neural progenitors ( Imayoshi et al . , 2013 ) . Striking the optimal balance between autonomous and collectively-maintained oscillations could be a tune-able evolutionary strategy in developing multi-cellular systems . The relationship between cell intrinsic circuits , local cell communication and tissue-level patterns will be important for understanding development , as well as engineering strategies for synthetic cellular systems ( Sprinzak et al . , 2010; Matsuda et al . , 2015 ) .
The Looping zebrafish line expresses a fusion of the Her1 protein and YFP driven by the endogenous her1 regulatory elements contained in a BAC transgene ( Soroldoni et al . , 2014 ) . For tailbud explant cultures , intact posterior tissue including PSM and tailbud was dissected from transgenic zebrafish embryos between 5 and 8 somite stage . The ectodermal tissue layer was removed from the explant and discarded . Tailbud pieces were dissected away from the remaining PSM by making a lateral cut across the explant , just posterior to the base of the notochord and Kupffer’s vesicle ( Figure 1A ) . Explants were then transferred to fibronectin-coated glass bottom 35-mm petri dishes ( MatTek , Ashland , Massachusetts ) and maintained in a small volume of L15 medium ( Sigma , St . Louis , Missouri ) with 10% fetal bovine serum ( Invitrogen , Waltham , Massachusetts ) during imaging . Cultures of tailbud cells were generated for in vitro imaging as described in ( Webb et al . , 2014 ) . Briefly , tailbud explants were generated as described above . For each independent replicate , multiple tailbud pieces ( each containing ~1000 cells ) were pooled and incubated in trypsin/EDTA ( Sigma ) for 20 min at room temperature . To quench trypsin , tailbud cells were dispersed into a small volume of L15 medium ( Sigma ) with 10% fetal bovine serum ( Invitrogen ) by pipetting , then plated onto a fibronectin-coated glass bottom 35-mm petri dish ( MatTek ) and allowed to adhere for 20 min Additional L15 medium containing 10% serum ± mouse Fgf8b ( 75 ng/uL; R&D Systems , Minneapolis , Minnesota ) was added to the culture prior to imaging . For the 96-well plate format , serial dilutions of the cell suspension to a final volume of 2 microliters were plated into individual wells . Again , additional L15 medium with 10% serum and Fgf8b was added to each well prior to imaging . Using this strategy , about 50% of plated wells have a single cell . Imaging was performed as described in Webb et al . ( 2014 ) . Briefly , transmitted light , YFP and mCherry images were acquired using an EM-CCD camera ( Andor xIOn 888 , Northern Ireland ) fixed to an inverted widefield microscope with a 40x lens ( Axiovert 200M; NeoFluor 40x , NA 0 . 75 , Zeiss , Germany ) . The temperature of the imaging dish was maintained at 26°C using a Warner heating chamber ( Harvard Apparatus , Cambridge , Massachusetts ) . Using iQ2 software , we acquired multiple fields within the dish over a 2-min interval and a 10-hr recording time . In each recorded field , we counted the total number of cells that were YFP+ out of all viable cells ( n = 321 out of 547 , 59% ) . We first disqualified from analysis YFP+ cells that moved out of the field ( 4% ) , came into contact with another cell or divided ( 14% ) , or were not viable at the end of the 10-hr recording time ( 7% ) . The remaining cells ( 34% ) were tracked in transmitted light images using a region of interest ( ROI ) tool in Fiji ( ImageJ , NIH ) . Average and maximum intensities across the ROI were then measured and interpolated across images with a customized plug-in ( ROI interpolator [Soroldoni et al . , 2014 ) ] ) . Before peak detection , any low frequency trends in the baseline of these raw intensity traces were removed by subtracting a local background obtained by measuring signal next to each cell ( Figure 1—figure supplement 2A ) . For each background subtracted cell trace , we detected peaks using a custom Matlab script that uses the findpeaks function . The algorithm first smoothens the time series and then detects local maxima ( Figure 1—figure supplement 2A ) . Local minima are then found between pairs of maxima using the function min restricted to the time interval between successive peaks . Peaks are subsequently discarded if they are smaller than 0 . 1 times the dynamic range of the time series , or if they are too close ( less than 40 min ) to the beginning or end of the time series , see the example trace in Figure 1—figure supplement 2A . Period is estimated as the time interval between consecutive peaks ( Figure 1—figure supplement 2B ) . Period estimates are discarded if their value is larger that 140 min , interpreting these events as elapsed time between disconnected peaks . Amplitude is defined as the average of the difference between a peak's height and its two adjacent minima , to take into account signal drifting during one cycle . Monoclonal antibodies were generated to the Ntla protein , the zebrafish T/Brachyury homolog , and to Tbx16 , the product of the spadetail locus . 8 μg of Ntla ( amino acids 1–261 ) or Tbx16 peptide ( amino acids 232–405 ) fused to GST was injected into Balb/c mice; sera were screened via ELISA . Each antiserum with a positive signal was further tested for tissue-specific binding in 15-somite stage wild-type and mutant or morpholino-injected embryos . Hybridoma cell lines were produced from one mouse; antibodies were purified from the supernatants . The antibody with highest signal-to-noise ratio was used for experiments ( Ntla , clone D18-4 , IgG1; Tbx16 , clone C24-1 , IgG2a ) . The validation of these antibodies is shown in Figure 1—figure supplement 6 . Cell dispersals from the same cell suspension used for time-lapse imaging ( L15 medium , plus 10% serum and Fgf8b ) were cultured separately on Conconavalin-A ( Sigma ) coated glass bottom dishes ( MatTek ) and maintained in the incubator at 28°C . These cells were fixed in 4% paraformaldehyde ( Sigma ) after 5 hr in culture overnight at 4°C . Prior to staining , cells were washed 4 × 5 min in PBS at room temperature and then blocked for 1 hr at room temperature in a PBS solution containing 1% BSA and 0 . 1% Triton ( Sigma ) . For staining , cells were incubated with monoclonal antibodies for Ntla ( D18 , IgG1 ) and Tbx16 ( C24 , IgG2a ) ( 1:5000 ) overnight at 4°C . Primary antibody was removed and the cultures were washed 3 × 20 min in PSM at room temperature . Cultures were incubated in secondary antibodies GFP-anti-IgG1 and Cy5-anti-IgG2a ( 1:500 , Molecular Probes , Eugene , Oregon ) for 2 hr at room temperature , prior to DAPI staining and final washes ( Figure 1—figure supplement 7 ) . These data were generated and published in Leise et al . ( 2012 ) and kindly donated for analysis . We use wavelet transforms to generate a phase time series from the raw traces . We compute the autocorrelation function of this phase and fit it to obtain an estimate of the period T and correlation time tc , which allows us to compute the quality factor Q = tc/T ( Supplementary file 1 ) . The observed behavior can be described using a Stuart-Landau model , which captures the generic behavior of an oscillator close to a supercritical Hopf bifurcation that gives rise to sustained , limit cycle oscillations . We introduce slow fluctuations in the parameter that controls the distance to the bifurcation to capture amplitude fluctuations , and white noise in the variables of the oscillator to capture frequency fluctuations ( Supplementary file 1 ) . | The timing and pattern of gene activity in cells can be very important . For example , precise gene activity patterns in 24-hour circadian clocks help to set daily cycles of rest and activity in organisms . In such scenarios , cells often communicate with each other to coordinate the activity of their genes . To fully understand how the behavior of the population emerges , scientists must first understand the gene activity patterns in individual cells . Rhythmic gene activity is essential for the spinal column to form in fish and other vertebrate embryos . A group of cells that switch genes on/off in a coordinated pattern act like a clock to regulate the timing of the various steps in the process of backbone formation . However , it is not clear if each cell is able to maintain a rhythm of gene expression on their own , or whether they rely on messages from neighboring cells to achieve it . Now , Webb et al . use time-lapse videos of individual cells isolated from the tail of zebrafish embryos to show that each cell can maintain a pattern of rhythmic activity in a gene called Her1 . In the experiments , individual cells were removed from zebrafish and placed under a microscope to record and track the activity of Her1 over time using fluorescent proteins . These experiments show that each cell is able to maintain a rhythmic pattern of Her1 expression on its own . Webb et al . then compared the Her1 activity patterns in individual cells with the Her1 patterns present in a larger piece of zebrafish tissue . The experiments showed that the rhythms in the individual cells are slower and less precise in their timing than in the tissue . This suggests that groups of cells must work together to create the synchronized rhythms of gene expression with the right precision and timing needed for the spinal column to be patterned correctly . In the future , further experiment with these cells will allow researchers to investigate the genetic basis of the rhythms in single cells , and find out how individual cells work together with their neighbors to allow tissues to work properly . | [
"Abstract",
"Introduction",
"Results",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"physics",
"of",
"living",
"systems",
"computational",
"and",
"systems",
"biology"
] | 2016 | Persistence, period and precision of autonomous cellular oscillators from the zebrafish segmentation clock |
Sphingolipids are important structural components of cell membranes and prominent signaling molecules controlling cell growth , differentiation , and apoptosis . Sphingolipids are particularly abundant in the brain , and defects in sphingolipid degradation are associated with several human neurodegenerative diseases . However , molecular mechanisms governing sphingolipid metabolism remain unclear . Here , we report that sphingolipid degradation is under transcriptional control of SIRT1 , a highly conserved mammalian NAD+-dependent protein deacetylase , in mouse embryonic stem cells ( mESCs ) . Deletion of SIRT1 results in accumulation of sphingomyelin in mESCs , primarily due to reduction of SMPDL3B , a GPI-anchored plasma membrane bound sphingomyelin phosphodiesterase . Mechanistically , SIRT1 regulates transcription of Smpdl3b through c-Myc . Functionally , SIRT1 deficiency-induced accumulation of sphingomyelin increases membrane fluidity and impairs neural differentiation in vitro and in vivo . Our findings discover a key regulatory mechanism for sphingolipid homeostasis and neural differentiation , further imply that pharmacological manipulation of SIRT1-mediated sphingomyelin degradation might be beneficial for treatment of human neurological diseases .
First isolated from brain extract in the 1880s , sphingolipid is a class of natural lipids containing a backbone of sphingoid base sphingosine ( Chen et al . , 2010; Merrill et al . , 2007; Pralhada Rao et al . , 2013 ) . Sphingolipids not only are the structural components of cell membranes , but also act as signaling molecules to control various cellular events , including signal transduction , cell growth , differentiation , and apoptosis ( Hannun and Obeid , 2008; van Meer et al . , 2008 ) . Particularly , sphingolipids are enriched in microdomains/lipid rafts , a liquid-ordered phase in plasma membrane , where sphingomyelin , glycosphingolipids , and cholesterol form unique platforms for many different proteins that are important for nutrient transport , organelle contact , membrane trafficking , and homotypic fusion ( Brown and London , 1998 ) . The homeostasis of sphingolipids is maintained by a highly coordinated metabolic network that links together various pathways with ceramide as a central node ( Figure 1—figure supplement 1A ) . Firstly , ceramide can be de novo synthesized in endoplasmic reticulum ( ER ) , starting from the condensation of serine and fatty acyl-CoA by serine palmitoyltransferase ( SPT ) . Ceramide is then transported to the Golgi complex and further converted into more complex forms of sphingolipids , such as glycosphingolipids or sphingomyelin . Secondly , ceramide can be regenerated by hydrolysis of complex sphingolipids in the Golgi complex and plasma membrane , which is catalyzed by a class of specific hydrolases and phosphodiesterases . For example , regeneration of ceramide from sphingomyelin can be mediated by plasma-membrane-bound sphingomyelin phosphodiesterases ( SMPDs ) , including SMPDL3B , a GPI-anchored lipid raft SMPD ( Heinz et al . , 2015 ) . By degrading sphingomyelin , SMPDL3B regulates plasma membrane fluidity , which in turn impacts TLR-mediated innate immunity in macrophages ( Heinz et al . , 2015 ) and modulates insulin receptor signaling in podocytes ( Mitrofanova et al . , 2019 ) . Sphingolipid metabolism is highly sensitive to environmental/nutritional perturbations . Sphingolipid biosynthesis and accumulation can be induced by high-fat diet ( HFD ) feeding in multiple tissues in mice ( Choi and Snider , 2015 ) . Sphingolipids are also accumulated during aging ( Giusto et al . , 1992; Lightle et al . , 2000 ) , and caloric restriction , a dietary regimen known to extend life span and delay a number of age-associated diseases , decreases sphingolipid accumulation by reducing the activity of SPT ( Tacconi et al . , 1991 ) . Dietary serine restriction also alters sphingolipid diversity and synthesis to constrain tumor growth ( Muthusamy et al . , 2020 ) . Disruption of sphingolipid homeostasis has been involved in the pathogenesis of a number of human diseases , such as Niemann-Pick disease and neurodegenerative Alzheimer’s , Parkinson’s , and Huntington’s diseases . These human diseases are generally the outcomes of defect in enzymes that degrade the sphingolipids ( Brice and Cowart , 2011; Alvarez-Vasquez et al . , 2005 ) . Such degradation defect leads to accumulation of sphingolipids , which in turn dramatically alters cellular membrane structure and signaling , thereby triggering various diseases . Consequently , manipulation of one enzyme or metabolite in sphingolipid degradation pathways may result in unexpected changes in cellular metabolic programs and related cellular functions ( Brice and Cowart , 2011; Alvarez-Vasquez et al . , 2005 ) . However , molecular mechanisms that regulate the expression of these sphingolipid degrading enzymes remain unclear . In the present study , we investigated the role of SIRT1 in regulation of sphingolipid degradation in mouse embryonic stem cells ( mESCs ) and mouse embryos . SIRT1 is an NAD+-dependent protein deacetylase critical for multiple cellular processes , including metabolism , inflammation , stress response , and stem cell functions ( Houtkooper et al . , 2012; Tang et al . , 2014; Han et al . , 2008 ) . SIRT1 is also a key regulator of animal development ( McBurney et al . , 2003; Cheng et al . , 2003; Wang et al . , 2008 ) and is particularly important in the central nervous system . For instance , SIRT1 has a major influence on hypothalamic function , and cell-type specific SIRT1 mutations result in defects in systemic energy metabolism , circadian rhythm , and the lifespan of the animal ( Dietrich et al . , 2010; Ramadori et al . , 2010; Ramadori et al . , 2011; Satoh et al . , 2013 ) . SIRT1 also modulates dendritic and axonal growth ( Hisahara et al . , 2008; Michán et al . , 2010 ) , and regulates synaptic plasticity and memory formation in adult brain ( Gao et al . , 2010 ) . Moreover , SIRT1 has been shown to ameliorate neurodegenerative phenotypes in animal models of Alzheimer’s , Parkinson’s , and Huntington’s disease ( reviewed in Herskovits and Guarente , 2014 ) . However , despite multiple mechanisms proposed for these critical roles of SIRT1 in the brain , how SIRT1 regulates neural development and functions remains unclear . Through global metabolomics and cellular metabolic characterizations , we discovered that SIRT1 deficiency in mESCs results in abnormal accumulation of sphingolipids , primarily due to reduced degradation of these lipids . We further found that this abnormal accumulation of sphingolipids does not affect the maintenance of pluripotent mESCs , but delays their neural differentiation during in vitro neural differentiation and in vivo mouse embryogenesis . Moreover , we provide evidence that SIRT1 regulates sphingolipid metabolism through deacetylation of c-Myc transcription factor , which promotes the expression of SMPDL3B and subsequent sphingomyelin degradation in mESCs . Together , our study identifies the SIRT1-c-Myc axis as an important regulatory mechanism for cellular sphingolipid metabolism and neural differentiation .
During a large-scale unbiased metabolomic analysis of WT and SIRT1 KO mESCs cultured in a serum-containing M10 medium , we discovered that SIRT1 KO mESCs display altered lipid metabolism , particularly metabolites involved in sphingolipid metabolism ( Figure 1A and Supplementary file 1 ) , in addition to previously reported metabolic defects in methionine metabolism ( Tang et al . , 2017 ) . Specifically , SIRT1 KO mESCs had a dramatic accumulation of sphingomyelin in both complete medium and a methionine restricted medium ( Figure 1B , Figure 1—figure supplement 1B , and Supplementary file 1 ) . Moreover , deletion of SIRT1 in mel1 human ESCs by CRISPR/Cas9-mediated gene editing technology ( Figure 1—figure supplement 1C ) also led to accumulation of several types of sphingomyelin regardless of medium methionine contents ( Supplementary file 2 ) , indicating that SIRT1 regulates sphingolipid metabolism in ESCs independently of cellular methionine metabolism . To confirm that deletion of SIRT1 indeed increases sphingomyelin contents in ESCs , we loaded WT and SIRT1 KO mESCs cultured in serum-free ESGRO medium with a green-fluorescent dye labeled sphingomyelin , BODIPY FL-C5-sphingomyelin , for 30 min at 4°C , then chased at 37°C for 30 min . Both WT and SIRT1 KO mESCs were labeled with this green-fluorescent sphingomyelin ( Figure 1C ) . However , SIRT1 KO mESCs had marked accumulation of BODIPY FL-C5-sphingomyelin inside cells , presumably in ER and Golgi , compared to WT mESCs . Quantitative FACS analysis showed that SIRT1 KO mESCs have about a 50% increase in cellular levels of BODIPY FL-C5-sphingomyelin compared to WT mESCs ( Figure 1D ) . An enzyme-coupled colorimetric assay further revealed a ~60% increase of endogenous sphingomyelin in SIRT1 KO mESCs ( Figure 1E ) . Interestingly , the accumulation of sphingomyelin was specific to mESCs , as SIRT1 KO MEFs had a comparable staining intensity of BODIPY FL-C5-sphingomyelin as WT MEFs ( Figure 1—figure supplement 1D ) . Additionally , SIRT1 KO mESCs had a similar staining intensity of BODIPY FL-C5-Ceramide compared to WT mESCs ( Figure 1—figure supplement 1E ) , suggesting that accumulation of sphingolipids in SIRT1 KO mESCs is specific to sphingomyelin . SIRT1 KO mESCs in above analyses were previously generated in R1 mES cell line using traditional gene targeting technology ( McBurney et al . , 2003 ) . To further confirm our observation that SIRT1 deficiency in mESCs induces accumulation of sphingomyelin , we deleted Sirt1 gene in another widely used full pluripotent mES cell line , E14 cells ( Wakayama et al . , 1999 ) , by CRISPR/Cas9-mediated gene editing technology ( Figure 1F ) . Consistent with observations in SIRT1 KO mESCs , these SIRT1 KO E14 mESC clones also had an enhanced staining of BODIPY FL-C5-sphingomyelin when analyzed by confocal fluorescence imaging ( Figure 1G ) and by quantitative FACS analysis ( Figure 1H ) . Taken together , our results indicate that deletion of SIRT1 in ESCs results in accumulation of sphingomyelin in independent ES cell lines . Cellular levels of sphingomyelin are regulated by a tight balance between their synthesis and breakdown , which are mediated by activities of sphingomyelin synthases ( SGMSs ) and sphingomyelin phosphodiesterases ( SMPDs ) , respectively ( Figure 1—figure supplement 1A ) . Many of these enzymes were highly expressed in mESCs ( Figure 2—figure supplement 1A and B ) . To better understand how SIRT1 deficiency in mESCs leads to sphingomyelin accumulation , we surveyed the expression levels of these enzymes in WT and SIRT1 KO mESCs . SIRT1 KO mESCs had significantly reduced expression of a sphingomyelin synthase Sgms2 ( Figure 2A and Figure 2—figure supplement 1B ) , and a dramatic reduction of a sphingomyelin phosphodiesterase SMPDL3B , one of the most highly expressed SMPDs in mESCs ( Figure 2—figure supplement 1A ) , in both ESGRO and M10 media ( Figure 2A–C ) . Since sphingomyelin was accumulated in SIRT1 KO mESCs , we focused on the reduction of Smpdl3b . As shown in Figure 2D , the reduced expression of SMPDL3B in SIRT1 KO mESCs was coupled with a decreased rate to clear away preloaded BODIPY FL-C5-sphingomyelin in a time-lapse video analysis , suggesting that reduction of SMPDL3B-mediated sphingomyelin degradation may be responsible for accumulation of sphingomyelin observed in SIRT1 KO mESCs . To test this possibility , we manipulated the levels of SMPDL3B in WT and SIRT1 KO mESCs and analyzed their impacts on cellular levels of sphingomyelin . Stable overexpression of SMPDL3B significantly reduced accumulation of BODIPY FL-C5-sphingomyelin in SIRT1 KO but not WT mESCs when cells were cultured in serum-containing M10 medium ( Figure 3A–C ) . In serum-free ESGRO medium , overexpression of SMPDL3B reduced accumulation of BODIPY FL-C5-sphingomyelin and endogenous sphingomyelin in both WT and SIRT1 KO mESCs ( Figure 3—figure supplement 1A–C ) . Moreover , the ability of SMPDL3B to reduce cellular levels of sphingomyelin is dependent on its enzymatic activity , as a catalytic inactive mutant of this enzyme , SMPDL3B H135A ( Heinz et al . , 2015; Mitrofanova et al . , 2019 ) , failed to decrease the levels of BODIPY FL-C5-sphingomyelin and endogenous sphingomyelin in WT and SIRT1 KO mESCs ( Figure 3—figure supplement 1D , Figure 3D–F ) . These results indicate that SMPDL3B is capable of removing sphingomyelin in mESCs , particularly in SIRT1 KO mESCs . Conversely , shRNA-mediated stable knockdown of SMPDL3B ( Figure 3G ) enhanced the accumulation of BODIPY FL-C5-sphingomyelin ( Figure 3H and I ) and endogenous sphingomyelin ( Figure 3J ) in WT mESCs but not further in SIRT1 KO mESCs , indicating that accumulation of sphingomyelin observed in SIRT1 KO mESCs is primarily due to reduced expression of SMPDL3B . As an NAD+-dependent protein deacetylase that deacetylates histones , transcription factors , cofactors , as well as splicing factors , SIRT1 has been shown to modulate gene expression at multiple levels . We confirmed that SIRT1 indeed regulates the expression of Smpdl3b and sphingomyelin degradation through its catalytic activity , as a SIRT1 catalytic inactive mutant ( H355Y , HY ) failed to rescue defective Smpdl3b expression and reduce BODIPY FL-C5-sphingomyelin accumulation in SIRT1 KO mESCs compared to WT SIRT1 protein ( Figure 4 ) . When interrogated each step along the expression of Smpdl3b gene in SIRT1 KO mESCs , we found that qPCR primers designed to amplify different segments of mature Smpdl3b mRNA all detected a reduced abundance of the full-length mature Smpdl3b mRNA upon SIRT1 deletion in mESCs ( Figure 5—figure supplement 1A and B ) . Moreover , the abundance of Smpdl3b mRNA was reduced in both nuclear and cytosolic fractions in SIRT1 KO mESCs ( Figure 5—figure supplement 1C ) . Northern blotting analysis using random probes generated from the full-length Smpdl3b cDNA further showed that deletion of SIRT1 in mESCs reduces the abundance of the full-length Smpdl3b mRNA without detectable accumulation of other minor isoforms ( Figure 5—figure supplement 1D ) . Finally , RNA-seq analysis of total Ribo-minus RNA ( total RNA after depletion of ribosomal RNAs ) revealed that the abundance of RNA species from both exonic and intronic regions of Smpdl3b gene were reduced in SIRT1 KO mESCs ( Figure 5—figure supplement 1E and Supplementary file 3 ) , and no defective splicing of Smpdl3b pre-mRNA was detected in these cells ( not shown ) . All these observations strongly suggest that the reduction of SMDPL3B expression in SIRT1 KO mESCs is due to defective transcription of Smpdl3b gene . In support of this notion , SIRT1 KO mESCs had a drastic depletion of Pol II near the TSS of Smpdl3b gene , along with decreased deposition of an activation mark H3K4me3 yet increased deposition of a repression mark H3K27me3 ( Figure 5A ) , indicative of a strong attenuation of transcriptional activation of Smpdl3b gene in SIRT1 KO mESCs . Sequence analysis of the TSS region revealed multiple potential transcription factors ( TFs ) that may target Smpdl3b gene , including two known SIRT1 deacetylation substrates , c-Myc and N-Myc ( Tang et al . , 2017; Menssen et al . , 2012; Figure 5B ) . To determine the promoter region ( s ) and associated TF ( s ) that are responsible for the transcription suppression of Smpdl3b gene in SIRT1 KO mESCs , we designed small gRNAs ( sgRNAs ) to target different potential TF loci along the Smpdl3b promoter ( Figure 5—figure supplement 2A , top ) , then analyzed their impacts on the expression of Smpdl3b after transfecting into WT and SIRT1 KO mESCs generated from a mouse ES cell line stably expressing a dox-inducible dCas9 and BirA-V5 ( dCas9 mESCs , Figure 5—figure supplement 2B; Liu et al . , 2017 ) . It has been demonstrated that transfected sgRNAs in these cells are able to guide the deactivated Cas9 ( dCas9 ) to bind to their targeting loci without further cleavage , resulting in altered transcription of the target gene ( Liu et al . , 2017 ) . Compared to control Gal4 sgRNA and other sgRNAs , a sgRNA targeting a locus near 528 bp downstream of the TSS of Smpdl3b gene rescued the defective expression of Smpdl3b mRNA in SIRT1 KO dCas9mESCs ( Figure 5C ) . Further bioinformatic analysis showed that this locus is overlapped with previously mapped binding regions of two TFs , c-Myc and EZH2 ( Figure 5—figure supplement 2A , bottom ) . c-Myc is a known SIRT1 deacetylation substrate , and deacetylation of c-Myc by SIRT1 has been reported to increase its stability and activity ( Menssen et al . , 2012 ) . We have previously shown that c-Myc is hyperacetylated but unstable in SIRT1 KO mESCs , which reduces its binding to target promoters thereby decreasing their transcription ( Tang et al . , 2017 ) . The association of c-Myc protein to the promoter of Smpdl3b gene was indeed significantly reduced in SIRT1 KO mESCs by a ChIP-qPCR assay ( Figure 5A , c-Myc ) . Moreover , inhibition of c-Myc activity by 10058-F4 ( Yin et al . , 2003; Figure 5D ) or knocking down c-Myc with siRNAs ( Figure 5E ) significantly reduced the mRNA abundance of Smpdl3b in WT mESCs but not or to a less extend in SIRT1 KO mESCs , indicating that c-Myc is a key transcription factor in SIRT1-mediated regulation of Smpdl3b . Furthermore , SIRT1 promoted the transcription of Smpdl3b in part through deacetylating c-Myc , as a c-Myc mutant with its major acetylation site mutated to R to mimic deacetylated c-Myc ( K323R , KR ) , partially rescued the expression of Smpdl3b in SIRT1 KO mESCs compared to empty vector ( V ) , WT c-Myc , and a c-Myc mutant with its major acetylation site mutated Q to mimic acetylated c-Myc ( K323Q , KQ ) ( Figure 5—figure supplement 2C and Figure 5F ) . Finally , a Smpdl3b promoter luciferase reporter containing a mutant c-Myc binding site ( E-box ) displayed a dramatically reduced activity in mESCs compared to a WT Smpdl3b promoter luciferase reporter ( Figure 5G , pGL3-Smpdl3b E-box Mut vs pGL3-Smpdl3b WT in WT mESCs ) . Additionally , this mutant luciferase reporter had a comparable low activity in SIRT1 KO mESCs vs WT mESCs , further suggesting that the differential expression levels of Smpdl3b in WT and SIRT1 KO mESCs is mediated by c-Myc . EZH2 , an H3K27me3 methyltransferase and the functional enzymatic component of the Polycomb Repressive Complex 2 ( PRC2 ) , is also a deacetylation substrate of SIRT1 ( Wan et al . , 2015 ) . Deacetylation of K348 of EZH2 by SIRT1 has been reported to reduce its stability and activity ( Wan et al . , 2015 ) , which is consistent with our current observation that deletion of SIRT1 in mESCs significantly increased the occupancy of EZH2 and H3K27me3 on the promoter of Smpdl3b gene ( Figure 5A , EZH2 , H3K27me3 ) . However , in contrast to c-Myc , neither inhibition of EZH2 activity by its inhibitor EPZ6438 ( Figure 5—figure supplement 2D ) nor knockdown of EZH2 by siRNAs ( Figure 5—figure supplement 2E–G ) significantly affected the expression of Smpdl3b in mESCs , particularly in SIRT1 KO mESCs , indicating that blocking EZH2-catalyzed H3K27 trimethylation alone is not sufficient to rescue SIRT1 deficiency-induced transcriptional suppression of Smpdl3b in mESCs . Collectively , our data demonstrate that SIRT1 activates the expression of Smpdl3b in mESCs primarily through deacetylation of c-Myc . SMPDL3B-catalyzed sphingomyelin degradation has been shown to reduce plasma membrane fluidity ( Heinz et al . , 2015 ) . As expected from their reduced expression of SMPDL3B , SIRT1 KO mESCs had a reduced fraction of ordered structures ( thereby increased membrane fluidity ) compared to WT mESCs when probed with Di-4-ANEPPDHQ , an electrical potential sensitive fluorescent dye for detection of microdomains and ( dis ) ordered membrane in live cells ( Figure 6A , vehicle ) . Treatment with methyl-β-cyclodextrin ( MβCD ) , a cholesterol-extracting agent , further decreased the membrane order , particularly in SIRT1 KO mESCs ( Figure 6A , MβCD ) . In line with this observation , pathways involved in cell surface receptor signaling pathway and intracellular signal transduction were among the most significantly disrupted Gene Ontology ( GO ) Biological Process ( GO BP ) in SIRT1 KO mESCs when compared with WT mESCs in our Ribo-minus RNA-seq dataset ( Figure 6—figure supplement 1 ) . Sphingolipids are bioactive lipids important for stem cell survival and differentiation ( Bieberich , 2008; Wang et al . , 2018 ) . Since we previously observed that SIRT1 KO mESCs have a compromised pluripotency ( Tang et al . , 2017 ) , we investigated whether sphingomyelin accumulation in SIRT1 KO mESCs is responsible for their reduced pluripotency . Compared to WT mESCs , SIRT1 KO mESCs had a reduced staining intensity of Alkaline Phosphatase ( AP ) , a marker of undifferentiated ESCs ( Figure 6B ) , along with decreased expression of OCT3/4 and Nanog , two pluripotent stem cell markers , and increased expression of Nestin , a neuroectodermal stem cell ( NSC ) marker ( Figure 6C ) . However , sphingomyelin treatment did not consistently affect the AP staining intensity ( Figure 6B ) nor induced any significantly changes on the expression of pluripotency markers in either WT or SIRT1 KO mESCs ( Figure 6C , OCT3/4 and Nanog ) . In contrast , sphingomyelin dose-dependently increased the expression of Nestin , a NSC marker that was induced in SIRT1 KO mESCs , but not in WT mESCs ( Figure 6C , Nestin ) . Moreover , overexpression of WT but not H135A mutant SMPDL3B reduced the expression of Nestin in SIRT1 KO mESCs ( Figure 6D ) , suggesting that SMPDL3B deficiency-resulted sphingomyelin accumulation may interfere neural differentiation in SIRT1 KO mESCs instead of their pluripotency . Consistent with this observation , during a 4-week in vitro neural differentiation of mESCs ( Figure 7A; Ying et al . , 2003; Abranches et al . , 2009 ) , the mRNA levels of Sirt1 and Smpdl3b were significantly reduced in WT E14 mESCs , along with dramatic decrease of Nanog and Oct4 and massive induction of several NSC and neural differentiation factors , such as Sox3 , Nestin , Notch3 , and Tau ( Figure 7B , WT ) . However , both the reduction of pluripotency markers and the induction of NSC/neural differentiation factors were significantly blunted when SIRT1 was deleted in E14 mESCs ( Figure 7B , KO ) , indicating that SIRT1 deficiency impairs in vitro neural differentiation of these cells . Further cellular and morphological analyses by immunofluorescence staining of progenitor and neuronal markers showed that progenitors and neurons differentiated from WT E14 mESCs have high expression of marker proteins and typical mature neuronal morphology , including elongated axons and dendrites , after 4-week differentiation ( Figure 7C , WT ) . Progenitors and neurons differentiated from SIRT1 KO E14 mESCs , on the other hand , had low levels of these markers and lacked typical neuronal morphology ( Figure 7C , KO ) . Additionally , SIRT1 KO mESCs also displayed defective neural differentiation after in vitro differentiation ( Figure 7D ) , indicating that SIRT1 deficiency in mESCs impairs neural differentiation in vitro in a cell line independent manner . To validate that defective neural differentiation of SIRT1 KO mESCs is related to their accumulation of sphingomyelin , we analyzed whether adding back SMPDL3B in these cells will rescue their neural differentiation defects . Morphologically , putting back SMPDL3B into SIRT1 KO mESCs increased neurons with elongated axons and dendrites ( Figure 8A and B , SIRT1 KO SMPDL3B ) , which was associated with the increased fraction of cells positive of several neural markers when analyzed by FACS analysis ( Figure 8C , SIRT1 KO SMPDL3B ) . Moreover , the expression of several progenitor and neuronal markers was also increased by adding back SMPDL3B in SIRT1 KO mESCs when analyzed by immunofluorescence staining ( Figure 8D , SIRT1 KO SMPDL3B ) or by qPCR ( Figure 8E , SIRT1 KO SMPDL3B ) . Overexpression of SMPDL3B in WT mESCs , however , disrupted the expression of progenitor markers and morphology of neurons ( Figures 8B , C and D , SIRT1 WT SMPDL3B ) , suggesting that a balanced sphingomyelin degradation is required to maintain normal neural differentiation . Conversely , in vitro neural differentiation of WT and SIRT1 KO mESCs with or without stable knockdown of SMPDL3B revealed that reduction of this enzyme disrupts neural differentiation in both WT and SIRT1 KO mESCs ( Figure 8F ) , indicative the importance of this enzyme in normal neural differentiation . Finally , putting back WT SIRT1 protein into SIRT1 KO mESCs rescued expression of progenitor marker SOX1 and NSC marker Nestin as well as neuronal morphology after in vitro neural differentiation ( Figure 8G , SIRT1 KO-WT ) . In contrast , putting back a catalytic inactive SIRT1 mutant failed to restore marker protein expression and/or neuronal morphology ( Figure 8G , SIRT1 KO-HY ) , confirming that the neural differentiation defects observed in SIRT1 KO mESCs are primarily due to a lack of SIRT1 deacetylase activity . Taken together , our observations indicate that SIRT1-mediated transcription of Smpdl3b and sphingolipid degradation influence neural differentiation in vitro . To assess the importance of sphingolipid metabolism in SIRT1-regulated neural differentiation in vivo , we investigated whether embryonic SIRT1 deficiency is associated with altered sphingomyelin accumulation and neural differentiation in mice . Consistent with our previous observations ( Tang et al . , 2017 ) , systemic deletion of SIRT1 in C57BL/6J mice leads to intrauterine growth retardation when dams were fed on a regular chow diet ( containing 4% fat ) ( Figure 9A , chow ) . The mRNA levels of Smpdl3b were significantly reduced in the brain of SIRT1 KO E18 . 5 embryos ( Figure 9B ) . However , these embryos did not display any detectable defects in brain sphingomyelin levels , nor in expression of a number of neural progenitor and neuron markers ( Figure 9C and D , chow ) despite reported developmental defects in other systems ( McBurney et al . , 2003; Cheng et al . , 2003; Wang et al . , 2008 ) . Since HFD feeding has been shown to induce sphingolipid biosynthesis and turnover of sphingolipids in multiple tissues ( Choi and Snider , 2015 ) , we tested whether maternal HFD feeding could induce sphingomyelin accumulation and disrupt neural development in SIRT1 KO embryos . Intriguingly , maternal feeding of an HFD diet containing 36% fat for 4–8 weeks before breeding significantly reduced intrauterine growth of embryos , particularly on SIRT1 KO embryos , at E18 . 5 ( Figure 9A , HFD ) . Maternal HFD feeding also elevated sphingomyelin contents in all tested regions of brain in SIRT1 KO but not WT E18 . 5 embryos ( Figure 9C , HFD ) . These maternal HFD feeding-induced gross and metabolic alterations were associated with reduced expression of many intermediate progenitor and mature neuron markers ( Figure 9D , HFD ) without significant changes on early stage neuroepithelial cell markers and oligodendrocyte markers ( not shown ) , suggesting that SIRT1 deficiency-induced sphingomyelin accumulation specifically delays neuron maturation in mouse embryos .
Highly enriched in the nervous system , sphingolipids are important for the development and maintenance of the functional integrity of the nervous system ( van Echten-Deckert and Herget , 1758; Olsen and Færgeman , 2017 ) . Perturbations of the sphingolipid metabolism has been shown to rearrange the plasma membrane , resulting in development of various human diseases , particularly neurological diseases ( Piccinini et al . , 2010 ) . However , despite these diverse biological functions , the transcriptional regulation of sphingolipid metabolism is largely unknown . In the present study , we show that cellular sphingomyelin degradation is under transcriptional control of SIRT1 , an important cellular metabolic sensor . We provide evidence that the SIRT1-Myc axis is vital for transcriptional activation of SMPDL3B , a major GPI-anchored plasma-membrane-bound sphingomyelin phosphodiesterase in mESCs . This transcriptional regulation directly impacts cellular levels of sphingomyelin and membrane fluidity , and is important in regulation of neural differentiation in response to developmental signals ( Figure 9E ) . Our findings therefore identify a unique genetic regulatory pathway for sphingolipid homeostasis . Given the high sensitivity of SIRT1 to nutritional and environmental perturbations ( e . g . activation upon caloric restriction and repression after HFD feeding or during aging Cantó and Auwerx , 2009; Imai , 2009 ) , our study further suggests that SIRT1-Myc-regulatetd sphingolipid degradation may be an important element in mediating reported environmental influence on sphingolipid metabolism ( Choi and Snider , 2015; Giusto et al . , 1992; Lightle et al . , 2000; Tacconi et al . , 1991 ) . It will be of great interest to test this possibility in future studies . As the most conserved mammalian NAD+-dependent protein deacetylase , SIRT1 has a number of important functions in the brain , including regulation of late stage of neural development and protection against a number of neurodegenerative diseases ( Herskovits and Guarente , 2014 ) . In particular , SIRT1 has been shown to modulate the neural and glial specification of neural precursors ( Prozorovski et al . , 2008; Kang et al . , 2010 ) and repress low glucose induced proliferation and neurogenesis of neural stem and progenitor cells ( NSCs ) in vitro ( Fusco et al . , 2016 ) . Our observations in the present study demonstrate that SIRT1 is also a key metabolic regulator for the differentiation of neural progenitors/NSCs from ESCs . Our results show that SIRT1 is highly expressed in mESCs cells ( Figure 7B ) , where it functions to promote c-Myc-mediated transcriptional activation of SMPDL3B and sphingomyelin degradation ( Figures 1–4 ) . This action of SIRT1 appears to have minimal impacts on the pluripotency of mESCs ( Figure 6B and C ) , but instead is important for maintenance of a proper membrane fluidity for normal neural differentiation in response to nutritional/developmental cues ( Figure 9E ) . Thus , by interacting with different protein factors , SIRT1 is important for neural differentiation and development at multiple stages . How impaired degradation of sphingomyelin might influence the differentiation of ESCs remains unclear . Our observations that SMPDL3B deficiency in SIRT1 KO mESCs is associated with increase of membrane fluidity ( Figure 6A ) and that pathways involved in cell surface receptor signaling pathway are one of the most significantly downregulated biological processes in SIRT1 KO mESCs ( Figure 6—figure supplement 1 ) suggest that impaired signaling transduction may underlie the impaired neural differentiation of these cells . This idea is consistent with the notion that sphingomyelin is important for formation of microdomains/lipid rafts on the plasma membrane for organization of many signaling proteins ( Brown and London , 1998 ) . Future studies are needed to directly test whether the plasma membrane/microdomain association of signaling proteins involved in neural differentiation ( e . g . insulin and/or bFGF signaling pathways ) is disrupted and whether the phosphorylation of downstream signal transduction factors is reduced in SIRT1 KO mESCs upon induction of neural differentiation . Our findings further suggest that an appropriate content of sphingomyelin , thereby a suitable degree of membrane fluidity , is required to maintain a proper signaling transduction for normal neural differentiation , as too much sphingomyelin resulted from direct sphingomyelin supplementation ( Figure 6C ) or SMPDL3B knockdown ( Figure 8F ) and too little sphingomyelin resulted from SMPDL3B overexpression ( Figure 8B and D ) all impair the expression of neural markers and differentiation . Our observations that Smpdl3b promoter is targeted by both c-Myc and EZH2 , consequently with ‘co-localization’ of antagonistic epigenetic marks H3K4me3 and H3K27me3 at the same locus near its TSS ( Figure 4A and S5A ) , suggest that Smpdl3b in mESCs might be a bivalent promoter-associated gene ( Bernstein et al . , 2006; Sanz et al . , 2008; Vastenhouw and Schier , 2012 ) . The bivalent chromatin domains in ESCs often mark lineage regulatory genes , and it has been proposed that bivalent domains might repress lineage control genes by H3K27me3 during pluripotency while keeping them poised for activation upon differentiation with H3K4me3 ( Vastenhouw and Schier , 2012 ) . Interestingly , the bivalent chromatin domain on Smpdl3b appears to keep it active in mESCs while poising it to be repressed upon differentiation . Moreover , while long-recognized as a transcription repressor through deacetylation of histones , SIRT1 plays an active role in remodeling this bivalent domain by stabilizing c-Myc while restricting EZH2-induced H3K27me3 in mESCs ( Figure 9E ) . Our present study reveals that this SIRT1/c-Myc/EZH2-regulated bivalent domain remodeling enables swift membrane remodeling in response to developmental signals , allowing more efficient and synchronous neural differentiation . Premature disruption of this chromatin remodeling complex in mESCs alters membrane fluidity ( Figure 5A ) , which may in turn affect developmental signaling transduction ( e . g . insulin , bFGF ) and impair neural differentiation . Future studies will be needed to elucidate the general role of this bivalent chromatin switch in regulation of pluripotency vs lineage genes , as well as in process of somatic cell reprogramming and/or transformation . The transcriptional regulation of sphingolipid metabolism and neural differentiation by the SIRT1-Myc axis in mESCs revealed in our present study is very intriguing , as we have previously reported that the same regulatory axis is key for methionine metabolism and maintenance of pluripotency in mESCs ( Tang et al . , 2017 ) . While our data show that accumulation of sphingomyelin in SIRT1 KO mESCs is independently of methionine metabolism ( Figure 1—figure supplement 1B and Supplementary file 1 ) , additional studies are needed to assess how the SIRT1-Myc regulatory axis coordinates diverse metabolic processes to shape stem cell fates in response to different environmental signals . It is also worth noting that SIRT1 KO mESCs have additional lipid metabolic defects , including depletion of monoacyglycerols , accumulation of plasmalogens , acetylcholine , and monohydroxy fatty acids , and altered phospholipids , regardless of medium methionine concentrations ( Figure 1—figure supplement 1B and Supplementary file 1 ) . It will be of importance to evaluate the contribution of these lipid metabolic defects to the observed hypersensitivity of SIRT1 KO embryos to maternal HFD feeding-induced intrauterine growth retardation ( Figure 9A ) in future studies . Our study has a few important implications . Firstly , the previously uncharacterized transcriptional regulation of SIRT1 on sphingomyelin degradation directly links cellular levels of sphingomyelin and membrane fluidity with cellular energy status , and provides a possible molecular mechanism for the beneficial impacts of SIRT1 small molecule activators and/or NAD+-boosting dietary supplements on human neurodegenerative diseases . Secondly , given the prevalence of obesity and metabolic syndrome in the reproductive population in modern society , the hypersensitivity of SIRT1 KO embryos to maternal HFD feeding-induced intrauterine growth retardation and neurodevelopmental defects suggests that pharmacological activation of SIRT1 by small molecule activators and/or NAD+-boosting dietary supplements might be able to attenuate maternal obesity-associated neonatal complications and defective childhood neurodevelopment ( Iessa and Bérard , 2015; Helle and Priest , 2020; Tong and Kalish , 2021 ) . In summary , our study uncovers a SIRT1-Myc-mediated transcriptional regulation of sphingomyelin degradation that modulates neural differentiation of ESCs . This finding highlights the importance of SIRT1 and its regulation in mESC differentiation and embryonic development , and may have important implications in potential therapeutic strategies again human neurodegenerative diseases and/or maternal obesity-induced adverse developmental outcomes .
Please see the appendix . WT ad SIRT1 KO mESCs generated in R1 mESC line have been reported previously ( Tang et al . , 2014; McBurney et al . , 2003 ) . They were a gift from Dr . Michael McBurney at Ottawa Hospital Research Institute . ES-E14TG2a ( E14 ) mESC line was purchased from ATCC . WT and SIRT1 KO E14 mESCs were generated by CRISPR/Cas9 mediated gene editing technology using lentivirus carrying either all-in-one empty vector pCRISPR-CG01 vector or pCRISPR-CG01 containing different sgRNAs targeting mouse Sirt1 gene ( GeneCopoeia ) . Stable single colonies were picked up and screened with immunoblotting assay using anti-SIRT1 antibodies . Three independent WT and SIRT1 KO E14 mESCs were used for the experiments to minimize the potential off-target effects of each individual line . mESCs stably transfected with pEF1α-FB-dCas9-puro ( Addgene #100547 ) and pEF1α-BirA-V5-neo ( Addgene #100548 ) vectors ( dCas9 mESCs ) are described previously ( Liu et al . , 2017 ) . WT and SIRT1 KO dCas9 mESCs were generated by a similar strategy as WT and SIRT1 KO E14 mESCs . Different sgRNA sequences targeting promoter region of Smpdl3b ( Supplementary file 4 ) were cloned into plasmid pSLQ1651-sgRNA ( F + E ) -sgGal4 ( Addgene #100549 ) and then were packed into lentivirus . The WT and SIRT1 KO dCas9 mESCs were infected with those lentiviruses to deliver sgRNA into cells , which then guided the dCas9 to bind on the promotor region of Smpdl3b and interfere its expression . WT and SIRT1 KO mESCs with stable Smpdl3b knockdown were generated by infecting WT and SIRT1 KO mESCs with lentivirus containing vector pLKO . 1 or constructs expressing shRNAs against Smpdl3b ( B11 , B12 , and C1 ) ( Sigma ) . WT and SIRT1 KO mESCs with stable overexpression of SMPDL3B were generated with lentivirus carrying vector pLenti-III-ef1α ( Addgene ) or constructs expressing the full-length SMPDL3B protein . The SMPDL3B H135A mutant was constructed with QuikChange II XL Site-Directed Mutagenesis Kit ( Agilent Technologies , 200522–5 ) against pLenti-III-ef1α-Smpdl3b by using primers described in Supplementary file 4 . The expression of SMPDL3B in these cells was analyzed by immunoblotting or immunofluorescence staining . WT and SIRT1 KO mESCs with stable overexpression of WT or a catalytic inactive H355Y mutant ( HY ) were generated using lentivirus carrying empty vector pLenti-III-ef1α ( Addgene ) or constructs expressing the full-length WT or HY SIRT1 proteins . To knockdown the Ezh2 and C-Myc gene expression , WT and Sirt1 KO mESCs were transfected with siRNA against mouse Ezh2 ( ThermoFisher , 4390771-s65775; siNeg: 4390843 ) and c-Myc ( Santa Cruz sc-29227; siNeg: siRNA-A sc-37007 ) , with Lipofectamine RNAiMAX ( ThermoFisher Scientific ) . The knockdown of genes expression was evaluated by Quantitative real-time PCR 48 hr after transfection . All mouse stem cells were maintained on gelatin-coated plates in the ESGRO Complete Clonal Grade Medium ( Millipore ) , and then cultured in the M10 medium ( High-glucose DMEM , 10% ES cell FBS , 2 mM L-glutamine , 1 mM sodium pyruvate , 0 . 1 mM nonessential amino acids , 10 μM 2-mercaptoethanol , and 500 units/ml leukocyte inhibitory factor ) for some experiments . Mel1 hESCs was a gift from Dr . Andrew Elefanty and Edouard Stanley at the University of Queensland , Australia . WT and SIRT1 KO mel1 hESCs were generated by CRISPR/Cas9 mediated gene editing technology using a plasmid containing Cas9 , gRNA ( AGAGATGGCTGGAATTGTCC ( - strand ) ) and a GFP indicator . The GFP positive cells were purified by flow cytometry , and were either grown on 10 cm dishes with a serial dilution , or in the 96-well plates at a density of 1 cell/well . Single colonies were picked up and subjected to immunofluorescence assay with anti-SIRT1 antibodies . The cell colonies without SIRT1 staining were sequenced to confirm the mutation . Three independent SIRT1 KO mel1 lines were used for the experiments to minimize the potential off target effects of each individual line . All cell lines were not authenticated at our end . All original and genetically modified ESCs are routinely checked ( at least every 6 months ) by the NIEHS Quality Assurance Laboratory for contamination of mycoplasma and other microbes by prolonged culture followed with qPCR-based assays , and they were all free of mycoplasma in our study . Whole body SIRT1 knockout , heterozygote and their age-matched littermate WT mice on the C57BL/6J background have been reported before ( Tang et al . , 2014 ) . They were housed in individualized ventilated cages ( Techniplast , Exton , PA ) with a combination of autoclaved nesting material ( Nestlet , Ancare Corp . , Bellmore , NY and Crink-l’Nest , The Andersons , Inc , Maumee , OH ) and housed on hardwood bedding ( Sani-chips , PJ Murphy , Montville , NJ ) . Mice were maintained on a 12:12 hr light:dark cycle at 22 ± 0 . 5°C and relative humidity of 40% to 60% . Mice were provided ad libitum autoclaved rodent diet ( NIH31 , Harlan Laboratories , Madison , WI ) and deionized water treated by reverse osmosis . Mice were negative for mouse hepatitis virus , Sendai virus , pneumonia virus of mice , mouse parvovirus 1 and 2 , epizootic diarrhea of infant mice , mouse norovirus , Mycoplasma pulmonis , Helicobacter spp . , and endo- and ectoparasites upon receipt and no pathogens were detected in sentinel mice during this study . Mice were randomly assigned to experimental groups after they were allowed to acclimate for at least one week prior to experiments . WT and SIRT1 KO mESCs were cultured in the complete M10 medium containing 200 μM methionine or methionine restricted M10 medium containing 6 μM methionine for 6 hr ( n = 5 biological replicates ) . WT and SIRT1 KO mel1 hESCs were cultured in serum-free TeSR-E8 medium containing 116 μM methionine or a methionine restricted medium containing 6 μM methionine for 6 hr on Matrigel ( n = 4 biological replicates ) . Cells were then harvested and profiled by metabolomics analysis as previously described ( Tang et al . , 2017 ) . Specifically , about 100 μl of packed cell pellet per sample were submitted to Metabolon , Inc ( Durham , NC , USA ) , where the relative amounts of small molecular metabolites were determined using four platforms of Ultrahigh-Performance Liquid Chromatography-Tandem Mass Spectroscopy ( UPLC-MS/MS ) as previously described ( Evans et al . , 2014 ) . All methods utilized a Waters ACQUITY UPLC and a Thermo Scientific Q-Exactive high-resolution/accurate mass spectrometer interfaced with a heated electrospray ionization ( HESI-II ) source and Orbitrap mass analyzer operated at 35 , 000 mass resolution . Raw data collected from above four analyses were managed by the Metabolon Laboratory Information Management System ( LIMS ) , extracted , peak-identified and QC processed using Metabolon’s hardware and software . The hardware and software foundations for these informatics components were the LAN backbone , and a database server running Oracle 10 . 2 . 0 . 1 Enterprise Edition . The final relative abundance of metabolites in each sample was normalized by the respective total protein concentration . To confirmed the alteration of sphingolipids in SIRT1 KO mESCs , WT and SIRT1 KO mESCs cultured in serum-free ESGRO medium or serum-containing M10 medium were incubated with 5 µM BODIPY FL C5-Sphingomyelin ( N- ( 4 , 4-Difluoro-5 , 7-Dimethyl-4-Bora-3a , 4a-Diaza-s-Indacene-3-Pentanoyl ) Sphingosyl Phosphocholine ) ( Invitrogen , D3522 ) or 5 µM BODIPY FL C5-Ceramide ( N- ( 4 , 4-Difluoro-5 , 7-Dimethyl-4-Bora-3a , 4a-Diaza-s-Indacene-3-Pentanoyl ) Sphingosine ) ( Invitrogen , D3521 ) together with 5 µM delipidated BSA ( as a delivery carrier ) in Hanks' buffered salt solution containing 10 mM HEPES ( HBSS/HEPES buffer pH 7 . 4 ) at 4°C for 30 min to load SM/ceramide . They were then incubated in medium without BODIPY FL C5-Sphingolipid at 37°C for additional 30 min . The intensity of cellular BODIPY FL C5-Sphingomyelin/Ceramide were analyzed by Zeiss LSM 780 UV confocal microscope and by Flow cytometry analysis ( Abs 505 nm and Em 511 nm ) . To analyze the degradation of sphingomyelins , WT and SIRT1 KO mESCs were pre-load with BODIPY FL C5-Sphingomyelin at 4°C for 30 min . They were then incubated in medium without BODIPY FL C5-Sphingomyelin at 37°C , and the dynamics of loaded BODIPY FL C5-Sphingomyelins in cells were followed using Zeiss LSM 780 UV confocal microscope for additional 12 hr at 37°C . The relative contents of endogenous sphingomyelins in WT and SIRT1 KO mESCs were analyzed using a commercially available Sphingomyelin Assay Kit ( Abcam ab133118 ) per manufacturer’s instruction . Total RNAs were isolated from mESCs or mice tissues using Qiagen RNeasy mini-kit ( 74104 ) . The nuclear and cytoplasmic RNA were separated and enriched by Fisher BioReagents SurePrep Nuclear or Cytoplasmic RNA Purification Kit ( Fisher Scientific , BP280550 ) . The cDNA was synthesized with ABI High-Capacity cDNA Reverse Transcription Kits ( 4374967 ) and further analyzed with qPCR using iQ SYBR Green Supermix ( Biorad ) . Three biological replications are performed for each experiment and raw data are normalized to the expression level of Rplp0 mRNA levels . The primers used in RT-PCR are listed in ( Supplementary file 4 ) . Immunofluorescence analysis mESCs grown on 0 . 1% gelatin coated coverslips were washed with PBS and fixed with 4% paraformaldehyde ( PFA ) in PBS ( pH 7 . 4 ) solution for 20 min at room temperature . They were then incubated with 1% glycine/PBS for 10 min , and cell membrane was permeabilized with 0 . 3% Triton X-100 in 1% glycine/PBS for 10 min . Cells were further blocked with 1% BSA and 0 . 05% Tween 20 in PBS for 30 min , incubated with primary antibodies ( Key Resources Table ) diluted with the blocking solution for overnight at 4°C , then the secondary antibodies Alexa Fluor 488 , 594 and 633 ( for flow cytometry sorting ) ( Invitrogen , A-11008 , A-11032 , A-21052 ) at 1:1000 in PBS for 1 hr at room temperature . Cells were counterstained for Nuclei with DRAQ5 Fluorescent Probe Solution ( Thermal Fisher , 62251 ) or directly mounted on glass slides with VECTASHIELD Antifade Mounting Media ( VECTOR LABORATORY ) which contains DAPI . The images of stained cells are acquired by Zeiss LSM 780 UV confocal microscope . The probe for Northern Blot hybridization is generated by using North2South Biotin Random Prime Labeling Kit ( ThermoFisher , 17075 ) . A 100 ng DNA product , which was synthesized from PCR reaction by using primer pair ‘5’- CACCGCTAGCGCCACCatgacgctgctcgggtggctgata-3’ and 5’- CACCGCGGCCGCtaacacctccagtacgtgcaggct-3’’ and the cDNA synthesized from total RNA isolated from mESCs , was used as a template to yield biotin-labeled single strand DNA probe that covers the full-length Smpdl3b mRNA sequence . Forty µg of total RNA isolated from mESC were separated with agarose electrophoresis with RNA Gel Loading Dye ( 2X ) ( ThermoFisher , R0641 ) and NorthernMax 10X Running Buffer ( Ambion , AM8671 ) . The separated total RNA samples were further transblotted to positively charged nylon transfer membrane ( Cat . 77016 ) by using S and S TurboBlotter Rapid Downward Transfer System ( DAIGGER Scientific ) and SSC buffer ( ThermoFisher , AM9763 ) . The RNA samples transferred to membrane were further crosslinked by using Stratalinker UV Crosslinker ( Model 1800 ) immediately upon completion of transblotting . The hybridization was performed by using North2South Chemiluminescent Hybridization and Detection Kit ( ThermoFisher , 17097 ) and the signaling of positive hybridization on membrane was detected with Chemiluminescent Nucleic Acid Detection Module ( ThermoFisher , 89880 ) . All procedures were performed by strictly following protocols for each kit provided by manufacturers . Cells were washed once with PBS , and were then lysed and scraped with 2 x SDS loading buffer without bromophenol blue . Samples were boiled for 10 min , and quantified . Equal amount of protein lysates was loaded and resolved on SDS-PAGE gel and transferred onto an PVDF membrane ( Millipore ) . Blots were blocked with 5% BSA for 1 hr , incubated with primary antibodies at 4°C overnight , incubated with secondary antibodies for 2 hr , and detected by Odyssey ( LI-Cor inc ) . To determine the association of RNA polymerase II ( Pol II ) , c-Myc , EZH2 , and histone marks H3K4me3 and H3K27me3 on mouse Smpdl3b locus in WT and SIRT1 KO mESCs , cells were fixed , harvested , and sonicated . The resulting sonicated chromatin was processed for immunoprecipitation with respective antibodies ( Key Resources Table ) as previously described ( Shimbo et al . , 2013 ) . Total RNA was extracted from WT and SIRT1 KO mESCs cultured in ESGRO medium in triplicates . All RNA-seq libraries were prepared with the TruSeq Stranded/Ribo kit ( Illumina , San Diego , CA ) and sequenced using the pair-end 76 bp protocol at about 520 million reads per library using the NovaSeq platform ( Illumina ) per the manufacturer’s protocol . Adaptor sequences were removed by Trim Galore ( v0 . 4 . 4 ) . Then reads were aligned to mouse genome version GRCm38/mm10 using STAR ( v2 . 5 . 3a ) ( Dobin et al . , 2013 ) with Gencode vM18 annotation . Gene expression values were quantified using RSEM ( v1 . 2 . 28 ) ( Li and Dewey , 2011 ) and differences in gene expression between experimental conditions were estimated using R package DEseq2 ( Love et al . , 2014 ) with input reads count from FeatureCounts ( Liao et al . , 2014 ) in Subread . GO Biological Process enrichment analysis was performed on 2541 significantly downregulated genes ( q < 0 . 01 ) in SIRT1 KO mESCs in g:Profiler ( https://biit . cs . ut . ee/gprofiler/gost ) . Gene set enrichment analysis ( GSEA ) ( v4 . 1 . 0 ) was implemented against all gene ontology ( GO ) gene sets in Molecular Signatures database ( MsigDB v7 . 2 ) with 10000 permutations ( min size 15 , max size 500 , FDR q < 0 . 25 ) . Potential transcription factors ( TFs ) on the promoter of Smpdl3b gene were predicted using ‘Match’ from geneXplain ( genexplain . com ) , in which the association scores , including ‘Core Motif Similarity’ and ‘Weight Matrix Similarity’ were calculated ( Chen et al . , 2008 ) . A higher score implies a higher chance of the gene being the target of this TF . To test the possible roles of transcriptional factor c-Myc and EZH2 in regulation of Smpdl3b expression in mESCs , WT and SIRT1 KO mESCs were treated with c-Myc Inhibitor CAS 403811-55-2–Calbiochem ( 10058-F4 ) ( Millipore Sigma , 475956 ) at 10 μM or EZh2 inhibitor Tazemetostat ( EPZ-6438 ) ( MCE , HY-13803 ) at indicated concentrations for 48 hr . WT and SIRT1 KO mESCs were also transfected with siRNA against mouse c-Myc ( Santa Cruz , sc-29227 ) or EZH2 ( ThermoFisher , 4390771 ) to knockdown their expression respectively . Cells were collected for 48 hr after transfection for qPCR analysis . To further determine the influence of the acetylation status of c-Myc on expression of Smpdl3b , mouse c-Myc protein was first cloned into the pHAGE-EF1α-HA-Puro vector ( Zheng et al . , 2016 ) . The major acetylation site of c-Myc protein , K323 , was then mutated to either R to mimic deacetylated c-Myc ( K323R ) or Q to mimic acetylated c-Myc ( K323Q ) using QuickChange II Site-Directed Mutagenesis Kit ( Agilent Technologies , 200522–5 ) against pHAGE-EF1α-HA-Puro-c-Myc . pHAGE-EF1α-HA-Puro vector , WT , K323R , and K323Q c-Myc constructs were transfected into WT and SIRT1 KO mESCs using Lipofectamine 3000 Reagent ( Invitrogen , L3000001 ) . The overexpression of WT and mutant c-Myc protein in cells were confirmed by immunofluorescence staining of transfected cells with anti-c-Myc antibody ( Abcam ) . To test the importance of the enzymatic activity of SMPDL3B in regulation of sphingomyelin degradation and neural differentiation , mouse WT SMPDL3B protein was first cloned into the pLenti-III-EF1α vector ( Zhang et al . , 2011 ) . An active site H135 was then mutated to A using QuickChange II Site-Directed Mutagenesis Kit ( Agilent Technologies , 200522–5 ) against pLenti-III-EF1α-Smpdl3b . The sequences of cloning and mutagenesis primers were listed in Supplementary file 4 . To directly analyze the transcriptional regulation of Smpdl3b expression by c-Myc/SIRT1 , firefly luciferase reporters driven by a 3 . 1 kb mouse Smpdl3b promoter fragment ( amplified by 5’- tcttacgcgtgctagcccgggctcgagACTCATCCAAAGGACCCAGGTT-3’ and 5’- tttatgtttttggcgtcttCCATGGGGCAGCAGGCACACATG-3’ ) containing either wild type ( WT ) or a mutant c-Myc binding site ( E-box ) were cloned into pGL3 basic vector . The E-box mutant was constructed with QuikChange II XL Site-Directed Mutagenesis Kit ( Agilent Technologies , 200522–5 ) using primers 5’- cgcgggttcccaccttgtggccagaagatcttctgggcagaactactcgtttggc-3’ and 5’- gccaaacgagtagttctgcccagaagatcttctggccacaaggtgggaacccgcg-3’ . The WT or mutant plasmids were then transfected into WT and SIRT1 KO mESCs together with the control pRL-TK plasmid ( Renilla Luciferase , Promega ) . Cells were cultured for 48 hr and the luciferase activity was measured using the Dual-Luciferase Reporter Assay System ( Promega , E1751 ) . The final firefly luciferase activity was normalized to the co-expressed renilla luciferase activity . WT and SIRT1 KO mESCs were seeded on glass cover slips and cultured in ESGRO medium overnight . They were then preincubated for 1 hr with or without 2 . 5 mM MβCD ( Sigma-Aldrich , C4555 ) , and stained with 5 μM di-4-ANEPPDHQ ( ThermoFisher , D36802 ) for 30 min . Coverslips were mounted using ProLongGold ( Invitrogen ) , images were acquired on a Zeiss LSM780 confocal microscope . The fluorescent dye was excited at 488 nm and images from 30 individual colonies in each group were acquired at 560 nm for emission from ordered phase and 620 nm for emission from disordered phase . The images were further analyzed ImageJ according to a method described previously ( Owen et al . , 2012 ) . In vitro neural differentiation of WT and SIRT1 KO mESCs were performed essentially as described ( Ying et al . , 2003; Abranches et al . , 2009 ) . Specifically , WT and SIRT1 KO mESCs maintained in ESGRO Complete PLUS Clonal Grade Medium ( Millipore SF001-500P ) were gently dissociated with 0 . 05% Trypsin and plated onto 0 . 1% gelatin coated cell culture dish at a density of 1 × 104 cells / cm2 with RHB-A medium ( Clontech TaKaba Cellartis , Y40001 ) . Medium was changed every 2 days and cultured for 4 days . Cells were then dissociated with 0 . 05% Trypsin again and plated into cell culture dish coated with 1 µg/ml Laminin ( Sigma , L2020 ) at a density of 2 × 104 cells/cm2 in RHB-A medium supplemented with 5 ng/ml murine bFGF ( Sigma , SRP4038-50UG ) . Medium was changed again every 2 days for the next 3 weeks . Cell morphology was monitored during the differentiation . The differentiated neural cells were maintained in RHB-A: Neurobasal ( ThermoFisher , 10888022 ) : B27 Supplement ( ThermoFisher , 17504044 ) ( 1:1:0 . 02 ) medium to for a better survival . To quantify the factions of differentiated cells at different stages during in vitro neural differentiation , differentiated cells were harvested with trypsin and fixed with 4% PFA . After fixation , cells are washed with PBS and immunofluorescence stained with different neural markers , and analyzed by FACS . The alkaline phosphatase staining assay was performed using the Alkaline Phosphatase staining kit II as per manufacturer's instructions ( Stemgent , Cambridge , MA; cat . no . 00–0055 ) . To investigate the effects of maternal HFD feeding on embryonic development of control and SIRT1 KO mice , 6- to 8-week-old SIRT1 heterozygous ( Sirt1+/- ) female mice were fed with either control chow diet ( NIH-31 contains 4% fat ) or a HFD ( D12492 contains 36% fat ) for 4–8 weeks . They were then bred with age matched Sirt1+/- ± mice fed with chow diet . Early next morning , females with the mating plug ( E0 . 5 ) were separated from the males into a new cage and put back on the HFD . Embryos from E14 . 5 and E18 . 5 were then collected and analyzed . The total feeding time on the HFD is up to 11 weeks . Embryos were collected from at least four dams ( pregnant females ) for each time point , this sample size was estimated using Chi square based on 100% penetrance of body weight reduction of SIRT1 KO embryos and a 95% power . All animal procedures were reviewed and approved by National Institute of Environmental Health Sciences Animal Care and Use Committee , under an Animal Study Proposal number 2017–0008 STL . All animals were housed , cared for , and used in compliance with the Guide for the Care and Use of Laboratory Animals and housed and used in an Association for the Assessment and Accreditation of Laboratory Animal Care , International ( AAALAC ) Program . Values are expressed as mean ± standard error of mean ( SEM ) from at least three independent experiments or biological replicates , unless otherwise indicated in the figure legend . Significant differences between the means were analyzed by the two-tailed , unpaired , Student’s t-test , and differences were considered significant at *p<0 . 05 using Microsoft Office Excel ( Version 16 . 16 . 27 ) . No methods were used to determine whether the data met assumptions of the statistical approach ( e . g . test for normal distribution ) . Bioinformatic analyses of RNA-seq data are detailed in RNA-seq analysis section . | All cells in the brain start life as stem cells which are yet to have a defined role in the body . A wide range of molecules and chemical signals guide stem cells towards a neuronal fate , including a group of molecules called sphingolipids . These molecules sit in the membrane surrounding the cell and play a pivotal role in a number of processes which help keep the neuronal cell healthy . Various enzymes work together to break down sphingolipids and remove them from the membrane . Defects in these enzymes can result in excess levels of sphingolipids , which can lead to neurodegenerative diseases , such as Alzheimer’s , Parkinson’s and Huntington’s disease . But how these enzymes are used and controlled during neuronal development is still somewhat of a mystery . To help answer this question , Fan et al . studied an enzyme called SIRT1 which has been shown to alleviate symptoms in animal models of neurodegenerative diseases . Stem cells were extracted from a mouse embryo lacking the gene for SIRT1 and cultured in the laboratory . These faulty cells were found to have superfluous amounts of sphingolipids , which made their membranes more fluid and reduced their ability to develop into neuronal cells . Further investigation revealed that SIRT1 regulates the degradation of sphingolipids by promoting the production of another enzyme called SMPDL3B . Fan et al . also found that when female mice were fed a high-fat diet , this caused sphingolipids to accumulate in their embryos which lacked the gene for SIRT1; this , in turn , impaired the neural development of their offspring . These findings suggest that targeting SIRT1 may offer new strategies for treating neurological diseases . The discovery that embryos deficient in SIRT1 are sensitive to high-fat diets implies that activating this enzyme might attenuate some of the neonatal complications associated with maternal obesity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"cell",
"biology"
] | 2021 | SIRT1 regulates sphingolipid metabolism and neural differentiation of mouse embryonic stem cells through c-Myc-SMPDL3B |
Centromeres vary greatly in size and sequence composition , ranging from ‘point’ centromeres with a single cenH3-containing nucleosome to ‘regional’ centromeres embedded in tandemly repeated sequences to holocentromeres that extend along the length of entire chromosomes . Point centromeres are defined by sequence , whereas regional and holocentromeres are epigenetically defined by the location of cenH3-containing nucleosomes . In this study , we show that Caenorhabditis elegans holocentromeres are organized as dispersed but discretely localized point centromeres , each forming a single cenH3-containing nucleosome . These centromeric sites co-localize with kinetochore components , and their occupancy is dependent on the cenH3 loading machinery . These sites coincide with non-specific binding sites for multiple transcription factors ( ‘HOT’ sites ) , which become occupied when cenH3 is lost . Our results show that the point centromere is the basic unit of holocentric organization in support of the classical polycentric model for holocentromeres , and provide a mechanistic basis for understanding how centromeric chromatin might be maintained .
The centromere is a defining feature of eukaryotic chromosomes and is essential for the segregation of chromosomes during cell division , as it organizes the proteinaceous kinetochore for attachment to the spindle apparatus at mitosis . Centromeres are universally marked by the variant histone cenH3 ( also called CENP-A in many organisms ) that replaces canonical histone H3 in centromeric nucleosomes , and most commonly localize to a single position along the chromosome ( Malik and Henikoff , 2009 ) . However , the DNA on which centromeric nucleosomes assemble is not conserved and varies greatly in size and composition . It ranges from genetically defined point centromeres that assemble a single cenH3-containing nucleosome to epigenetically defined regional centromeres of several kb or Mb of tandemly repeated DNA to holocentromeres that extend along the length of entire chromosomes . With the exception of budding yeast point centromeres , where there is a 1:1 relationship between a single cenH3 nucleosome and the functional centromere , the precise organization of centromeric chromatin has remained elusive . One of the main issues standing in the way of uncovering the distribution of centromeric nucleosomes is the fact that most regional centromeres are localized to homogeneous tandemly repetitive regions of the genome , making it difficult to map individual nucleosomes . C . elegans is amenable to address this question because the genome is repeat-poor , making it possible to precisely map centromeric regions . Classical cytogenetic observations have demonstrated that C . elegans chromosomes are holocentric , whereby mitotic spindle fibers attach along the length of chromosomes and pull them to the poles as straight bars rather than from a single position that defines the more familiar monocentric chromosomes ( Schrader , 1935; Albertson and Thomson , 1982 ) . Two models have been put forward for how holocentric chromosomes might be organized ( Schrader , 1947 ) . The ‘diffuse centromere’ model predicts that the centromere is truly distributed along the length of the chromosomes , and that spindle fiber attachments form randomly . The ‘polycentromere’ model predicts that there are a number of discrete sites dispersed along the chromosomes , creating a holocentric appearance when observed at cytological resolution ( Figure 1A ) . 10 . 7554/eLife . 02025 . 003Figure 1 . Genome-wide distribution of cenH3 . ( A ) Classic holocentromere models proposed by Schrader ( Schrader , 1947 ) : diffuse and polycentric holocentromeres . The diffuse model predicts full centromere coverage of the chromosomes . The polycentric model predicts discrete centromeric sites that together give the appearance of holocenticity . ( B ) Genome browser view of 525 kb on Chr I for cenH3 X-ChIP-chip ( Gassmann et al . , 2012 ) , cenH3 N-ChIP-seq ( this study ) , H3 . 3 N-ChIP-chip ( Ooi et al . , 2010 ) and H3K9me3 X-ChIP-chip ( Liu et al . , 2011 ) , showing the close correspondance of cenH3 N-ChIP and X-ChIP signals for domains , but not peaks , and the positive correlation of cenH3 signal with H3K9me3 signal and the negative correlation of cenH3 signal with H3 . 3 signal . Log2 ratios of IP and input are shown to enable comparison between microarray and sequencing data . CenH3 peaks are marked by asterisks . ( C ) Genome browser view of cenH3 N-ChIP-seq ( this study ) and cenH3 X-ChIP-chip ( Gassmann et al . , 2012 ) at two representative cenH3 peaks marked in ( B ) with red boxes . ( D ) Number and genomic distribution of cenH3 peaks called per chromosome . See also Figure 1—source data 1 . ( E ) Heatmaps and average plots of input and cenH3 N-ChIP signal within a 2-kb window around all 707 cenH3 peaks . Each line of the heatmaps represents an individual cenH3 site . The heatmaps are sorted from high to low cenH3 signal . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 00310 . 7554/eLife . 02025 . 004Figure 1—source data 1 . Genomic coordinates of cenH3 peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 00410 . 7554/eLife . 02025 . 005Figure 1—figure supplement 1 . Solubilization of chromatin . Agarose gel image of DNA isolated from soluble and insoluble chromatin after 2 min MNase digestion and needle extraction . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 00510 . 7554/eLife . 02025 . 006Figure 1—figure supplement 2 . Comparison of cenH3 to H3K9me3 and H3 . 3 on a genome-wide scale . ( A ) Genome browser view of cenH3 N-ChIP signal on a genome-wide scale . Log2 ratios of ChIP and input are shown . ( B ) Correlation density plots of cenH3 N-ChIP ( this study ) and cenH3 X-ChIP ( Gassmann et al . , 2012 ) with H3K9me3 X-ChIP ( Liu et al . , 2011 ) ( left ) and H3 . 3 N-ChIP ( Ooi et al . , 2010 ) ( center ) . Correlation density plot of cenH3 N-ChIP with cenH3 X-ChIP ( right ) . Average log2 ratios of ChIP over input in 10 kb-windows are plotted and used to calculate regression lines and correlation coefficients ( r ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 00610 . 7554/eLife . 02025 . 007Figure 1—figure supplement 3 . Analysis of cenH3 distribution and occupancy on a genome-wide scale . ( A ) Distance between cenH3 peaks ranked by size for the entire genome ( left ) and for individual chromosomes ( right ) . ( B ) Gene density at cenH3 sites . ( C ) Average cenH3 ChIP signal in the N-ChIP data ( this study ) and the X-ChIP data ( Gassmann et al . , 2012 ) within a 2-kb window around all 707 cenH3 sites . Average occupancy at domains defined in Gassmann et al . ( Gassmann et al . , 2012 ) is indicated by dashed lines for both data sets . ( D ) Box plots of fragment counts at cenH3 peaks , cenH3 domains , and regions between cenH3 domains for cenH3 N-ChIP data , illustrating that the highest signal genome-wide is found at cenH3 peaks . Differences of cenH3 ChIP and input samples are shown . The average cenH3 signal at each peak , domain , and region between domains was calculated , and the average values were plotted . N = 707 ( peaks ) , 2865 ( domains ) , 2868 ( regions between domains ) . For cenH3 peaks , a 200-bp window around the summit of the peaks was used . Domains were defined in Gassmann et al . ( Gassmann et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 007 Consistent with either model , C . elegans cenH3 ( also called HCP-3 ) localizes to a characteristic band along the length of the chromosome at mitosis ( Buchwitz et al . , 1999 ) . A previous study mapped C . elegans cenH3 using a microarray-based approach and found that it occupies ∼2900 broad domains that account for about half of the genome , but that there was only enough cenH3 to cover 4% of the genome , suggesting that cenH3 nucleosomes assemble at random positions within the domains ( Gassmann et al . , 2012 ) . These findings thus seemingly supported the diffuse centromere model . However , mitotic microtubules must attach to discrete sites for chromosome segregation , and the number of microtubules attached during C . elegans mitosis has been estimated to about 100 for all six chromosomes combined ( O’Toole et al . , 2003 ) . This left open the question of the relationship between these diffuse domains and discrete microtubule attachment sites . To identify potential kinetochore attachment sites in C . elegans , we profiled cenH3 nucleosomes with single base-pair resolution . While we observed domains of low occupancy similar to those described in the earlier study , we also discovered discrete sites of much higher cenH3 occupancy that are distributed independently of the domains . Depletion of the machinery needed for incorporation of cenH3 nucleosomes resulted in reduced occupancy of cenH3 at these sites . As an independent indicator of centromeric localization , we also profiled the inner kinetochore protein CENP-C ( also called HCP-4 in C . elegans ) . We found that cenH3 sites coincide with high CENP-C signal , indicating that they serve as attachment sites for the kinetochore , consistent with a polycentric organization of the chromosome . Individual sites resemble budding yeast point centromeres and coincide with transcription factor hotspots , which become occupied by transcription factors when cenH3 is lost , providing a clue as to how kinetochore sites might be selected and maintained .
To precisely localize cenH3-containing nucleosomes and identify potential kinetochore attachment sites , we digested chromatin from mixed-stage embryos with micrococcal nuclease ( MNase ) and solubilized the majority of the chromatin by cavitation , a method adapted from Jin and Felsenfeld , 2007 ( Figure 1—figure supplement 1 ) . We subsequently performed native chromatin immunoprecipitation ( ChIP ) of cenH3 from the soluble chromatin , followed by paired-end sequencing ( N-ChIP-seq ) , resulting in single base-pair resolution maps of cenH3-associated DNA fragments . As expected from a previous ChIP-microarray map of C . elegans cenH3 using formaldehyde crosslinking ( X-ChIP-chip ) ( Gassmann et al . , 2012 ) , we found that cenH3 is broadly distributed throughout the genome . CenH3 is enriched towards the arms relative to the centers of the autosomes , while the distribution on the X chromosome is relatively even ( Figure 1—figure supplement 2A ) . In C . elegans , chromosome arms tend to be enriched for repeats and are associated with marks of heterochromatin ( The C . elegans Sequencing Consortium , 1998; Liu et al . , 2011 ) . Indeed , the distribution of cenH3 is positively correlated with the distribution of trimethylation of lysine 9 on histone H3 ( H3K9me3 ) , a mark of transcriptionally silent regions , in both our ChIP-seq and the previously published ChIP–chip data ( Figure 1—figure supplement 2B , left panels . r = 0 . 44 , p<2 . 2 × 10−16 for correlation with cenH3 X-Chip and r = 0 . 31 , p<2 . 2 × 10−16 for correlation with N-ChIP ) ( Gu and Fire , 2010; Liu et al . , 2011 ) . This is consistent with previous findings that have associated H3K9 methylation with the acquisition of cenH3 in fission yeast ( Folco et al . , 2008; Kagansky et al . , 2009 ) . We therefore wondered if the enrichment of cenH3 is associated with lower nucleosome turnover . The replication-independent variant histone H3 . 3 is incorporated into chromatin when nucleosomes are replaced and serves as a measure of replication-independent nucleosome turnover ( Ahmad and Henikoff , 2002; Mito et al . , 2005; Goldberg et al . , 2010; Ooi et al . , 2010 ) . Consistent with the hypothesis that cenH3 is associated with lower nucleosome turnover , we found that the distributions of H3 . 3 and cenH3 are negatively correlated in both our ChIP-seq and the previously published ChIP–chip data ( Figure 1—figure supplement 2B , center panels . r = −0 . 65 , p<2 . 2 × 10−16 for correlation with cenH3 X-Chip and r = −0 . 21 , p<2 . 2 × 10−16 for correlation with N-ChIP ) . Replication-independent nucleosome turnover is mainly driven by transcription ( Deal et al . , 2010; Teves and Henikoff , 2011 ) , consistent with the previously described anti-correlation between cenH3 and RNA polymerase II ( Gassmann et al . , 2012 ) . CenH3 has been found to be localized to 10–12 kb wide domains that occupy about half of the genome ( Figure 1B , first track ) ( Gassmann et al . , 2012 ) . Despite the use of a different methodology ( low-salt native chromatin preparation and MNase digestion instead of formaldehyde fixation and sonication , a different antibody and ChIP-seq instead of ChIP–chip ) , our data showed a very similar domain pattern ( Figure 1B , second track , Figure 1—figure supplement 2B , right panel . Correlation with cenH3 X-ChIP r = 0 . 67 , p<2 . 2 × 10−16 ) . The previously published data ( Gassmann et al . , 2012 ) pointed to an anti-correlation of cenH3 occupancy with transcription in the germline and in the early embryo . Because there is little RNA Polymerase II activity in the early embryo , and transcriptional profiles are not directly transmitted during the maternal–zygotic transition , the authors proposed that it is the memory of germline transcription transmitted to the embryo that excludes cenH3 incorporation . We found that the domains were both negatively correlated with previously published H3 . 3 data ( Figure 1B , third track ) and positively correlated with previously published H3K9me3 patterns ( Figure 1B , fourth track ) ( Ooi et al . , 2010; Liu et al . , 2011 ) . H3 . 3 is abundantly incorporated into chromatin in the germline and throughout early embryonic development ( Ooi et al . , 2010 ) , whereas H3K9me3 is associated with transcriptionally silent chromatin where nucleosome turnover and H3 . 3 incorporation are low ( Ahmad and Henikoff , 2002; Mito et al . , 2005; Gu and Fire , 2010; Ooi et al . , 2010; Liu et al . , 2011 ) . This suggests that it is nucleosome turnover that excludes the deposition of cenH3 and shapes the domain-like distribution of cenH3 across the genome , which can happen even in absence of transcription in the early embryo . The levels of cenH3 within a cell only allow for the occupancy of 4% of the genome , and each cenH3 domain can therefore only contain a limited number of cenH3 nucleosomes per cell ( Gassmann et al . , 2012 ) . The large-scale correspondence of our native ChIP-seq data to the previously published crosslinked ChIP–chip data provides independent confirmation of this interpretation . However , we wondered whether there might also be preferred sites of centromeric nucleosome positioning within the cenH3 domains of C . elegans , as predicted by the polycentromere model . These sites would appear as sites of high cenH3 occupancy in a population average . We indeed found that cenH3 was highly enriched at discrete , dispersed loci ( Figure 1B , C ) . As these loci appeared as very well-defined peaks , we removed background by subtracting the input signal and considered sites with 30 or more normalized counts ( equivalent to the mean plus 7 times the standard deviation of the genome-wide signal ) in at least one of two biological replicates as positive . We identified about 100 cenH3 peaks on each chromosome ( 707 peaks total; Figure 1D ) , with the distance between peaks ranging from 290 bp to 1 . 9 Mb ( median 83 kb; Figure 1—figure supplement 3A , Figure 1—source data 1 ) . We averaged the signal around all 707 cenH3 peaks , represented by a single centered peak in the cenH3 ChIP data that is highly enriched compared to input ( Figure 1E ) . The peaks were enriched in gene-poor regions of the genome ( Figure 1—figure supplement 3B ) . To our surprise , 607 out of 707 of these peaks resided outside of the domains described previously ( Gassmann et al . , 2012 ) and corresponded to sites of only slight local cenH3 enrichment in the X-ChIP data ( Figure 1—figure supplement 3C ) . In the N-ChIP data , cenH3 occupancy was much higher at these peaks compared to the domains ( Figure 1—figure supplement 3D ) . We hypothesized that these sites are preferred for the deposition of centromeric nucleosomes and serve as potential kinetochore attachment sites . Previous studies in other organisms observed that centromeric regions were sensitive to MNase or resulted in patterns inconsistent with the presence of canonical nucleosomes ( Polizzi and Clarke , 1991; Takahashi et al . , 1992; Dalal et al . , 2007; Krassovsky et al . , 2012 ) . We found that the cenH3 peaks were sensitive to MNase and disappeared with progressive MNase digestion ( Figure 2A , B ) , even at MNase conditions where nucleosome arrays remain intact and that would be considered underdigested by most standards ( Figure 2—figure supplement 1 , left panel ) . In contrast , the chromatin features around the cenH3 peaks were remarkably unaffected ( Figure 2A , Figure 2—figure supplement 2 ) . As a control , we compared the cenH3 peaks to the +1 nucleosomes at transcription start sites ( Chen et al . , 2013 ) . Occupancy of these well-positioned nucleosomes also decreased with progressing MNase digestion , but to a lesser extent ( Figure 2C ) . We quantified MNase sensitivity at cenH3 peaks , at the nucleosomes immediately flanking the cenH3 peaks , and at the +1 nucleosomes by dividing the occupancy of these features at each time point by the occupancy at the first time point . This analysis revealed that cenH3 nucleosomes were more sensitive to MNase than both flanking and +1 nucleosomes ( Figure 2D ) . These findings suggest that the sites of high cenH3 occupancy contain nucleosomes with similar properties as centromeric nucleosomes in other organisms . 10 . 7554/eLife . 02025 . 008Figure 2 . CenH3 peaks are especially MNase sensitive . ( A ) Genome browser view of input chromatin and cenH3 ChIP signal within a 25-kb window surrounding two representative cenH3 peaks . Tracks for occupancy after 1 min , 2 min , 5 min and 10 min of MNase digestion are shown . ( B ) Average input signal ( left ) and cenH3 ChIP signal ( right ) within a 1-kb window around all 707 cenH3 sites after the indicated MNase digestion intervals . The dashed red line indicates the midpoint of the cenH3 nucleosome and the dashed black lines indicate the midpoints of the flanking nucleosomes . ( C ) Average input signal within a 1-kb window around 7043 transcriptional start sites ( TSS ) after the indicated MNase digestion intervals . The dashed green line indicates the midpoint of the +1 nucleosome . TSS were defined by Chen et al . ( Chen et al . , 2013 ) . ( D ) MNase sensitivity plot for the cenH3 nucleosome and the flanking nucleosomes shown in ( B ) and the +1 nucleosome shown in ( C ) . The occupancy of these nucleosomes at each MNase digestion time point was divided by the occupancy at the first time point . N = 707 ( cenH3 nuc ) , 1414 ( flanking nucs ) , 7043 ( +1 nuc ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 00810 . 7554/eLife . 02025 . 009Figure 2—figure supplement 1 . Progress of MNase digestion . Agarose gel images of input DNA after MNase digestion of chromatin from wildtype and knl-2 ( RNAi ) embryos and wildtype adults . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 00910 . 7554/eLife . 02025 . 010Figure 2—figure supplement 2 . CenH3 sites are preferentially digested by MNase . Genome browser view of input chromatin at two representative cenH3 sites after 1 min , 2 min , 5 min , and 10 min of MNase digestion , illustrating that the MNase-sensitivity of cenH3 nucleosomes is visible in the input chromatin . CenH3 sites are marked in red . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 010 Incorporation of cenH3 into chromatin depends on the kinetochore protein KNL-2 ( Maddox et al . , 2007 ) . To test if the signal at the cenH3 peaks results from KNL-2-dependent incorporation of cenH3 , we analyzed the chromatin upon KNL-2 knockdown . KNL-2 depletion by RNAi led to an embryonic lethal phenotype with 99 ± 1% penetrance ( n = 8 ) . We confirmed by microscopy that this was caused by chromosome segregation defects and found by immunofluorescence that the cenH3 signal became undetectable in embryos . These results were consistent with published findings ( Maddox et al . , 2007 ) and suggested that our depletion of KNL-2 successfully reduced the presence of cenH3 and thus the functionality of centromeres . ChIP experiments revealed that cenH3 occupancy at the cenH3 sites was much reduced in knl-2 ( RNAi ) embryos compared to wildtype ( Figure 3 ) for similar levels of MNase digestion ( Figure 2—figure supplement 1 , center panel ) . This effect extended genome wide ( Figure 3—figure supplement 1A ) and was not caused by changes in the input chromatin , as the overall occupancy and positioning of most canonical nucleosomes and other DNA binding factors remained unchanged in knl-2 ( RNAi ) embryos ( Figure 3—figure supplement 1B ) , and the input chromatin showed the same correlation with wildtype chromatin as between wildtype replicates ( R = 0 . 971 and 0 . 969 for wildtype vs knl-2 ( RNAi ) and R = 0 . 973 for wildtype vs wildtype; comparison of normalized fragment counts in 10-bp bins , N = 7633808 ) . 10 . 7554/eLife . 02025 . 011Figure 3 . CenH3 peaks depend on cenH3 loading . ( A ) Genome browser view of cenH3 ChIP in wildtype and knl-2 ( RNAi ) embryos , with two enlarged cenH3 peaks . KNL-2 is required for cenH3 loading onto chromatin . Differences between ChIP and input are shown . CenH3 peaks are marked by asterisks . ( B ) Average cenH3 ChIP signal within a 2-kb window around all 707 cenH3 sites in wildtype and knl-2 ( RNAi ) embryos . Differences between ChIP and input are plotted . ( C ) Heatmap of difference in cenH3 ChIP signal between wildtype and knl-2 ( RNAi ) . Each line of the heatmap represents an individual cenH3 site . The heatmap is sorted by decreasing difference between wildtype and knl-2 ( RNAi ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 01110 . 7554/eLife . 02025 . 012Figure 3—figure supplement 1 . Comparison of wildtype and knl-2 ( RNAi ) chromatin . ( A ) Box plots of fragment counts at cenH3 peaks , cenH3 domains , and regions between cenH3 domains for cenH3 ChIP in wildtype and knl-2 ( RNAi ) embryos . Differences of cenH3 ChIP and input samples are shown . The average cenH3 signal at each peak , domain , and region between domains was calculated , and the average values were plotted . N = 707 ( peaks ) , 2865 ( domains ) , 2868 ( regions between domains ) . For cenH3 peaks , a 200-bp window around the summit of the peaks was used . Domains were defined in Gassmann et al . ( Gassmann et al . , 2012 ) . ( B ) Genome browser view of cenH3 ChIP from wildtype and knl-2 ( RNAi ) embryos . Log2 ratios of ChIP and input are plotted to visualize domains ( top ) , and normalized counts are plotted for an enlarged region to show that the differences between the two samples are dominated by the ChIP , while input samples are highly similar . CenH3 peaks are marked by asterisks . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 012 It is conceivable that with a partial depletion of the factor required for cenH3-assembly into chromatin , relatively high cenH3 occupancy is maintained at the sites that are functional due to perdurance of the protein . Indeed , cenH3 is still locally enriched at cenH3 sites in the knockdown , albeit at much reduced levels , while the broad domains of weak enrichment are lost ( Figure 3A , Figure 3—figure supplement 1 ) . These results indicate that the signal at the identified cenH3 sites indeed depends on the incorporation of cenH3 into centromeric nucleosomes by the cenH3-specific assembly machinery . CenH3 can be incorporated at low levels into nucleosomes away from the centromeres ( Camahort et al . , 2009; Lefrancois et al . , 2009; Lopes da Rosa et al . , 2011; Krassovsky et al . , 2012; Lefrancois et al . , 2013; Lacoste et al . , 2014 ) , and so the presence of cenH3 itself is thus not a sufficient measure for the presence of a centromere . To test if the cenH3 peaks indeed correspond to centromeric sites , we compared it to the kinetochore . Previous findings from our lab have suggested that in budding yeast the chromatin fraction that remains insoluble under native conditions after MNase digestion is strongly enriched for kinetochore complexes ( Krassovsky et al . , 2012 ) . As there is a one-to-one relationship between the centromere and the kinetochore in budding yeast , and the kinetochore components are conserved between eukaryotes , it can be inferred that kinetochore-bound chromatin remains mostly insoluble under these conditions . We therefore analyzed the distribution of the MNase fragments associated with the chromatin fraction that remained insoluble after MNase digestion and needle extraction and found peaks corresponding to each cenH3 peak ( Figure 4A , B ) . This suggested the presence of insoluble complexes , potentially kinetochores , at every cenH3 site in at least part of the cell population analyzed . The insoluble chromatin signal is reduced in knl-2 ( RNAi ) embryos ( Figure 4—figure supplement 1A , B ) , supporting the interpretation that these peaks correspond to kinetochores . Interestingly , the peaks in the insoluble chromatin were more resistant to MNase than cenH3 nucleosomes in the soluble fraction , and the insoluble chromatin signal persisted even after 10 min of MNase digestion ( Figure 4A , B ) , suggesting that the proteins that render cenH3 chromatin insoluble during extraction help to protect it from nuclease digestion . 10 . 7554/eLife . 02025 . 013Figure 4 . CenH3 sites are bound by the kinetochore . ( A ) Genome browser view of cenH3 ChIP ( native , 2 min MNase ) , insoluble chromatin ( native , 2 min and 10 min MNase ) , and CENP-C ChIP ( formaldehyde-crosslinked , 2 min MNase ) signal at two representative cenH3 sites . ( B and C ) Heatmaps and average plots of insoluble chromatin signal after 2-min and 10-min MNase digestion ( B ) and CENP-C ChIP signal ( C ) within a 2-kb window around all 707 cenH3 sites . Each line of the heatmaps represents an individual cenH3 site . Heatmaps are sorted by decreasing signal . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 01310 . 7554/eLife . 02025 . 014Figure 4—figure supplement 1 . Insoluble chromatin at centromeric sites in knl-2 ( RNAi ) embryos . ( A ) Average insoluble chromatin signal after 10 min MNase digestion within a 2-kb window around all 707 cenH3 sites in wild-type and knl-2 ( RNAi ) embryos . ( B ) Heatmap of the difference in insoluble chromatin signal between wildtype and knl-2 ( RNAi ) . Each line of the heatmap represents an individual cenH3 site . The heatmap is sorted by decreasing difference between wildtype and knl-2 ( RNAi ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 01410 . 7554/eLife . 02025 . 015Figure 4—figure supplement 2 . Quantification of kinetochore occupancy . ( A ) Box plots of fragment counts at cenH3 peaks , cenH3 domains , and regions between domains for insoluble chromatin ( native ) , illustrating that the highest signal genome-wide is found at cenH3 peaks . ( B ) Box plots of fragment counts at cenH3 peaks , cenH3 domains , and regions between domains for CENP-C ChIP ( formaldehyde-crosslinked ) , illustrating that the highest signal genome-wide is found at cenH3 peaks . Differences of CENP-C ChIP and input samples are shown . For ( A ) and ( B ) , the average cenH3 signal at each peak , domain , and region between domains was calculated , and the average values were plotted . N = 707 ( peaks ) , 2865 ( domains ) , 2868 ( regions between domains ) . For cenH3 peaks , a 200-bp window around the center of the high occupancy sites was used . Domains were defined in Gassmann et al . ( Gassmann et al . , 2012 ) . ( C ) Normalized number of insoluble chromatin peaks coinciding with cenH3 peaks , cenH3 domains and regions between domains . ( D ) Normalized number of CENP-C peaks coinciding with cenH3 peaks , cenH3 domains and regions between domains . For ( C ) and ( D ) , the number of insoluble chromatin peaks and CENP-C peaks overlapping with each group was divided by its genome coverage ( 141 , 400 bp for peaks , 42 , 722 , 880 bp for domains , 46 , 119 , 910 bp for rest of the genome ) and then multiplied by 10 Mb . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 015 To test directly if the cenH3 peaks coincide with kinetochore attachment sites , we performed ChIP on the inner kinetochore protein CENP-C . This protein binds cenH3 and is required for the assembly of the kinetochore complex that links centromeric chromatin to the microtubule ( Moore and Roth , 2001; Oegema et al . , 2001; Cheeseman et al . , 2004; Carroll et al . , 2010; Kato et al . , 2013 ) . In fact , CENP-C can organize the entire functional kinetochore in the absence of cenH3 ( Gascoigne et al . , 2011; Przewloka et al . , 2011; Hori et al . , 2013 ) . Although CENP-C remains with centromeric DNA throughout the cell cycle in other organisms , in C . elegans CENP-C localizes to chromatin only during mitosis , but not interphase , when it is in the cytoplasm ( Moore and Roth , 2001 ) . As a consequence , only a fraction of the cells analyzed contains CENP-C on chromatin , which limits the dynamic range of ChIP signal that is achievable . Because the kinetochore complex is highly insoluble , no DNA was recovered in native ChIP of CENP-C ( data not shown ) . We therefore profiled CENP-C using MNase followed by formaldehyde crosslinking and solubilization with SDS . We found that CENP-C is enriched at cenH3 sites ( Figure 4A , C ) . Neither the signal in the insoluble fraction nor the signal for CENP-C was enriched over the previously identified cenH3 domains compared to the rest of the genome ( Figure 4—figure supplement 2A , B ) . Despite the presence of non-centromeric enrichment in the insoluble chromatin fraction and the lower dynamic range of the CENP-C ChIP data , we called peaks in these two data sets and compared them to the cenH3 peak calls . 460 of the 2060 insoluble chromatin peaks and 163 of the 347 insoluble chromatin peaks coincided with cenH3 peaks . In contrast , only 174 insoluble chromatin peaks and 26 CENP-C peaks fell within the domains , in both cases fewer sites than expected by chance ( p<0 . 001 ) . Normalized to the genome coverage of domains and peaks , this amounts to an 800-fold enrichment of insoluble chromatin peaks and an almost 2000-fold enrichment of CENP-C peaks at cenH3 peaks compared to cenH3 domains ( Figure 4—figure supplement 2C , D ) . These results indicate that the cenH3 peaks identified in this study act as the preferred sites of kinetochore formation . The precise co-localization of cenH3 , CENP-C and insoluble chromatin peaks confirm that these sites correspond to centromeres . Moreover , the number of sites lies in the same order of magnitude as the number of microtubules observed during mitosis , thus providing a mechanistically reasonable alternative to the conundrum of how domains that cover half of the genome can organize a relatively small number of microtubule attachment sites . Centromeric nucleosomes in other organisms protect only 80–120 bp of wrapped DNA , compared to 147 bp for canonical nucleosomes ( Dalal et al . , 2007; Krassovsky et al . , 2012; Hasson et al . , 2013; Zhang et al . , 2013 ) , probably due to the reduced wrapping of DNA around them ( Henikoff and Furuyama , 2012 ) . To examine the size and positioning of the nucleosomes at the 707 centromeric sites , we divided the fragments in the input and cenH3 ChIP samples into size classes of fragments >140 bp representing nucleosomes , and ≤140 bp representing sub-nucleosome-sized particles . In the input sample , we found that two well-positioned nucleosomes flank the cenH3 peaks ( Figure 5A ) . These nucleosomes are also visible in modENCODE data for mononucleosomes prepared under native conditions , but not upon formaldehyde-crosslinking , presumably because they become crosslinked to the centromere ( Figure 5—figure supplement 1 ) . The native input sample also revealed the presence of sub-nucleosome-size fragments over the center of the cenH3 sites , while few of these fragments were found in the flanking regions ( Figure 5A ) . In the cenH3 ChIP sample , only relatively few nucleosome-size fragments were recovered , while the majority of the signal came from fragments <140 bp ( Figure 5B ) . The insoluble chromatin showed a similar pattern , indicating that the particles bound to DNA at cenH3 sites are similar in the insoluble and in the soluble chromatin fractions ( Figure 5C ) . This analysis showed that centromeric sites consist of two well-positioned nucleosomes flanking a single cenH3 nucleosome that wraps less DNA than is wrapped by a canonical nucleosome . 10 . 7554/eLife . 02025 . 016Figure 5 . CenH3 nucleosomes protect small DNA fragments . ( A , B , C ) Normalized fragment counts in input ( A ) , cenH3 ChIP ( B ) and insoluble chromatin ( C ) samples at centromeric sites . MNase fragments were divided into nucleosomal ( 141–500 bp ) and small ( 21–140 bp ) size classes . Average signals within a 1-kb window around all 707 cenH3 sites are plotted . Dashed lines mark the centers of the flanking nucleosomes in ( A ) ( black lines ) or the centromeric nucleosome in ( B ) ( blue line ) and in ( C ) ( magenta line ) . ( D ) Cartoon illustrating how MNase fragment size distributions shown in ( E and F ) were determined . Fragments that cross the center of the cenH3 nucleosome or of the flanking nucleosomes were counted . ( E and F ) MNase fragment size distribution after 2 min ( E ) and 10 min ( F ) of MNase digestion . Input fragments at flanking nucleosomes ( black ) and cenH3 ChIP fragments ( blue ) and insoluble chromatin fragments ( magenta ) at centromeric nucleosomes are shown . Cartoons of the protected particles are shown below each panel . ( G and H ) Comparison of worm holocentromere and budding yeast point centromere . ( G ) C . elegans holocentromere . Centromere model and cenH3 ChIP over input ratio ( all size classes; left y-axis ) and nucleosomal signal from input ( 141–500 bp; right y-axis ) . Average signals within a 1-kb window around all 707 cenH3 sites are shown . ( H ) Budding yeast point centromere . Centromere model and data from Krassovsky et al . ( Krassovsky et al . , 2012 ) , cenH3 ChIP over input ratio ( left y-axis ) and input signal ( right y-axis ) from all 16 centromeres . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 01610 . 7554/eLife . 02025 . 017Figure 5—figure supplement 1 . Comparison of input signal at centromeric sites in native and formaldehyde-crosslinked samples . Nucleosome-size fragments at centromeric sites from native and formaldehyde-crosslinked input samples ( top , MNase-seq , this study ) , and mononucleosome-size fragments from native and formaldehyde-crosslinked samples from the modENCODE database ( http://intermine . modencode . org/ ) , derived from microarray data ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 01710 . 7554/eLife . 02025 . 018Figure 5—figure supplement 2 . Fragment size distribution for different degrees of MNase digestions . ( A ) MNase fragment size distribution of fragments crossing the dyad of the nucleosomes flanking the cenH3 peaks ( top ) or of all fragments genome wide ( bottom ) . Fragment size distributions for the four indicated MNase digestion time points are shown . ( B ) MNase fragment size distribution of fragments within cenH3 domains ( top ) and regions between domains ( bottom ) for input and cenH3 ChIP after 2 min of MNase digestion . Domains were defined in Gassmann et al . ( Gassmann et al . , 2012 ) . The MNase fragment size distribution of fragments crossing the center of the cenH3 nucleosome at cenH3 sites is shown in light blue for comparison , where the gray area marks the peak size interval . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 01810 . 7554/eLife . 02025 . 019Figure 5—figure supplement 3 . Particle size estimation from overdigested insoluble chromatin . Half-height width of the average signal of insoluble chromatin at cenH3 peaks after 10 min MNase . Average signal within a 1-kb window around all 707 cenH3 sites is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 019 To estimate the size of the DNA associated with these nucleosomes , we plotted the size-distribution of fragments that cross the center of each particle ( Figure 5D ) . In the input sample after 2 min of MNase digestion , the fragments that cross the dyad of the flanking nucleosomes show a distribution that peaks at 166 bp , consistent with canonical nucleosomes ( Figure 5E , black line ) . The peak lies at 166 bp rather than the 147 bp minimal protected fragment size for mononucleosomes because of the relatively light MNase digestion required to prevent the loss of cenH3 peaks ( Figure 2B ) . Progressive MNase digestion reduced the size of the DNA fragments protected by these particles in a manner expected for nucleosomes , analogous to the size pattern observed for bulk chromatin ( Figure 5—figure supplement 2A ) . A second peak representing dinucleosomes is also visible at about 330 bp ( Figure 5E , black line ) . The fragments that cross the center of the cenH3 nucleosomes in the cenH3 ChIP sample after 2 min of MNase digestion show a strikingly different distribution that is left-shifted and indicate that the nucleosomes that occupy these sites protect only about 60–120 bp ( Figure 5E , blue line ) . The fragments in the insoluble chromatin fraction show a very similar distribution , supporting this size estimate ( Figure 5E , magenta line ) . The dinucleosome peak in the cenH3 ChIP and insoluble chromatin , representing the centromeric nucleosome and one flanking canonical nucleosome , is equally left-shifted to about 260 bp ( Figure 5E , blue and magenta lines ) . These shorter dinucleosome fragments , protected by two neighboring nucleosomes on the same molecule of DNA , can only be explained by the presence of a smaller centromeric nucleosome , because the inferred size estimates are not affected by possible MNase encroachment on the centromeric nucleosome . After 10 min of MNase digestion , the peak of the distribution of MNase fragments that cross the center of the cenH3 sites in the insoluble fractions lies at 66 bp ( Figure 5F , magenta line ) , indicating that the minimal protected size of these particles lies in the 60–100 bp range . Moreover , the width at half-height of the average 10 min-digested insoluble chromatin peak that aligns with the cenH3 peaks is 82 bp ( Figure 5—figure supplement 3 ) . The inferred fragment size protected by the centromeric nucleosome is thus 60–100 bp . These results confirm the findings in other organisms that cenH3 nucleosomes at centromeric sites in C . elegans wrap less DNA than that wrapped by canonical nucleosomes . CenH3 ChIP also revealed mild enrichment over broad domains . Fragment size analysis revealed that the majority of cenH3 nucleosomes in these regions of the genome protect about 135–155 bp of DNA , and that this size distribution is similar in regions between domains ( Figure 5—figure supplement 2B ) . This level of protection is consistent with the findings in other organisms that cenH3 can incorporate into canonical-type nucleosomes away from centromeres , in some cases as cenH3-H3 . 3 heterotypic nucleosomes ( Camahort et al . , 2009; Krassovsky et al . , 2012; Lacoste et al . , 2014 ) . This further suggests that the cenH3 domains may be distributed independently of the centromere . Taken together , fragment size analysis thus revealed that each centromeric site consists of a single cenH3-containing nucleosome that is flanked by two well-positioned canonical nucleosomes ( Figure 5G ) . This chromatin landscape is reminiscent of the budding yeast centromere , where a single cenH3 nucleosome assembles on a genetically defined sequence ( Furuyama and Biggins , 2007; Henikoff and Henikoff , 2012 ) . This sequence is flanked by binding sites for centromere-specific protein complexes ( Cbf1 and Cbf3 ) that in turn position two flanking canonical nucleosomes ( Figure 5H ) ( Densmore et al . , 1991; Krassovsky et al . , 2012 ) . The stable binding of Cbf1 and Cbf3 prevent the centromeric DNA from being occupied by canonical nucleosomes ( Kent et al . , 2011; Krassovsky et al . , 2012 ) . In C . elegans , the flanking nucleosomes are positioned closer together than in yeast , presumably because there are no sequence-specific DNA-binding proteins between the centromeric and the flanking nucleosomes . Thus , the dispersed centromeric sites in C . elegans holocentromeres consist of a cenH3 nucleosome that is associated with 60–100 bp of DNA flanked by two well-positioned nucleosomes , a chromatin pattern with striking similarities to budding yeast point centromeres . Budding yeast point centromeres assemble on a genetically defined sequence ( Clarke and Carbon , 1980 ) . To determine whether the C . elegans centromeric sites have common sequence properties , we searched for motifs using MEME ( Bailey et al . , 2009 ) . We found a 15-nt GA-repeat-rich motif that was common to 297 of the 80-bp cores of the centromeric sites ( Figure 6A ) . A weaker , but similar GA-rich motif was common to all 707 centromeric sites ( Figure 6—figure supplement 1A ) . The 15-nt motif also matched more than 60 , 000 sites in the genome that are not associated with high cenH3 signal and so is not sufficient to determine cenH3 occupancy . 10 . 7554/eLife . 02025 . 020Figure 6 . Centromeres coincide with transcription factor hotspots . ( A ) MEME motif for the 80-bp cores of centromeric sites ( left ) and the 80-bp cores of high occupancy target ( HOT ) sites ( right ) . ( B ) Heatmaps and average plots of cenH3 ChIP , insoluble chromatin and CENP-C ChIP signal within a 2-kb window around HOT sites , illustrating that HOT sites are highly occupied by cenH3 , insoluble chromatin and CENP-C . Each line of the heatmaps represents an individual HOT site . Heatmaps are sorted by decreasing signal . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 02010 . 7554/eLife . 02025 . 021Figure 6—figure supplement 1 . CenH3 site motif and characterization of HOT sites . ( A ) MEME motif for 80-bp cores of all centromeric sites . ( B ) Venn diagram illustrating the overlap between cenH3 sites and HOT sites . Hypergeometric p=8 . 6 × 10−161 . ( C ) Average signal of nucleosome-size input MNase fragments ( 141–500 bp ) within a 2-kb window around all 248 HOT sites . ( C ) Average cenH3 ChIP signal within a 2-kb window around all 248 HOT sites in wildtype and knl-2 ( RNAi ) embryos . Differences between ChIP and input are plotted . ( D ) Heatmap of difference in cenH3 ChIP signal between wildtype and knl-2 ( RNAi ) at all 248 HOT sites . Each line of the heatmap represents an individual HOT site . The heatmap is sorted by decreasing difference between wildtype and knl-2 ( RNAi ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 021 GAGA-rich sequences are well-characterized targets for the GAGA factor ( GAF/Trl ) in D . melanogaster ( van Steensel et al . , 2003 ) . However , we could not identify a GAF/Trl homologue in the C . elegans genome . Instead , the motif we identified is almost identical to the motif associated with C . elegans high occupancy target ( HOT ) sites ( Figure 6A ) . These sites were uncovered by the modENCODE consortium through binding-site analysis of 22 transcription factors and operationally defined as sites that are bound by ≥15 transcription factors ( Gerstein et al . , 2010; Niu et al . , 2011 ) . The sequences at these sites do not contain DNA motifs of known C . elegans transcription factors and are therefore expected to bind transcription factors with low affinity . We found that 117 out of 248 HOT sites coincided with cenH3 sites ( Figure 6—figure supplement 1B; hypergeometric p=8 . 6 × 10−161 ) . Although this degree of overlap is striking , the actual overlap of cenH3 sites and transcription factor hotspots is likely to be much larger , given the fact that that only 22 transcription factors have been used for the definition of HOT sites , but that there are 934 predicted transcription factors encoded in the C . elegans genome ( Reece-Hoyes et al . , 2005; Gerstein et al . , 2010 ) . We also found that HOT sites show a high signal for cenH3 ChIP , insoluble chromatin , CENP-C ChIP ( Figure 6B ) and well-positioned flanking nucleosomes ( Figure 6—figure supplement 1C ) . Moreover , the cenH3 ChIP signal is reduced in KNL-2-depleted animals ( Figure 6—figure supplement 1D , E ) . These data suggest that HOT sites and centromeric sites share a similar chromatin landscape and are targeted by both cenH3 nucleosomes and transcription factors . Embryonic cells contain both cenH3 and transcription factors , thus complicating the analysis of the chromatin landscape at centromeric sites . To probe the chromatin landscape in cenH3-depleted cells , we analyzed our previously published data for affinity-purified adult muscle cells , where cenH3 protein is below detection levels ( Figure 7A ) , and cenH3 mRNA is significantly depleted ( Bayesian t-test; q = 0 . 0036 ) ( Steiner et al . , 2012 ) . These samples were MNase-digested >10 min to enrich for mononucleosomes ( Figure 2—figure supplement 1 , right panel ) . Centromeric nucleosomes are unstable under these MNase conditions , and associated fragments are expected to be depleted ( Figure 2 ) . Despite differences between chromatin preparations from embryos for native ChIP input and whole nuclear DNA extraction from adults for MNase-seq , the overall nucleosome landscapes obtained were very similar ( Figure 7—figure supplement 1A ) . In total adult samples , which contain nuclei from dividing germline and embryonic cells , we found a depletion of signal at centromeric sites and well-positioned nucleosomes flanking the sites , reminiscent of the chromatin landscape in embryos ( Figure 7B , black line ) . In muscle nuclei , the flanking nucleosomes were also present ( Figure 7B , red line ) , and protected a very similar size range of fragments as in the total nuclei sample ( Figure 7C , left panel ) . However , the centromeric sites in the muscle nuclei sample were occupied by MNase-stable particles ( Figure 7B , red line ) that protect sub-nucleosome-size fragments ( Figure 7C , right panel ) and likely represent non-nucleosomal DNA-binding proteins . 10 . 7554/eLife . 02025 . 022Figure 7 . Transcription factor occupancy upon loss of cenH3 as cells exit the cell cycle . ( A ) CenH3 in adult germline , intestine and muscle . Staining of a worm section with anti-cenH3 , anti-NPP-9 ( nuclear pores; staining control ) and DAPI are shown . Germ cells in diakinesis are marked with asterisks , muscle cells with arrowheads in the merge . Scale bar is 3 µm . ( B ) Average MNase-seq signal within a 2-kb window around centromeric sites for adult total and adult muscle nuclei , illustrating that centromeric sites remain occupied in cenH3-depleted cells . MNase digestion >10 min . ( C ) MNase fragment size distribution at flanking nucleosomes and centromeric sites for total adult and adult muscle nuclei , illustrating that at least a sub-population of the particles occupying the centromeric sites are not canonical nucleosomes . ( D ) Heatmap and average plot of HLH-1 ChIP signal from adults within a 2-kb window around cenH3 sites . HLH-1 is a HOT site transcription factor . ( E ) Heatmaps and average plots of ChIP signal for another nine of the transcription factors used to define HOT sites within a 2-kb window around cenH3 sites , data from Gerstein et al . ( Gerstein et al . , 2010 ) . All transcription factors in ( E ) were profiled in the third larval instar except ALR-1 ( second larval instar ) and PHA-4 ( adults ) . Differences between ChIP and input are plotted in ( D and E ) . Each line of the heatmaps represents an individual cenH3 site . Heatmaps are sorted by decreasing signal . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 02210 . 7554/eLife . 02025 . 023Figure 7—figure supplement 1 . Comparison of chromatin from embryos and adults . ( A ) Genome browser view of MNase-seq input signal from embryonic nuclei ( native ) , total adult nuclei ( dimethylformamide fixed ) , and adult muscle nuclei ( dimethylformamide fixed ) . ( B ) Overlap of NPP-9::mCherry used for purifying muscle nuclei to obtain chromatin profiles ( Steiner et al . , 2012 ) and HLH-1::GFP , a muscle-specific HOT site transcription factor ( Gerstein et al . , 2010 ) in a live worm . Gut granules show auto-fluorescence in the green channel . Muscle nuclei are marked with arrowheads in the merge . DOI: http://dx . doi . org/10 . 7554/eLife . 02025 . 023 To test directly whether centromeric sites become occupied by transcription factors upon depletion of cenH3 , we profiled the muscle-specific transcription factor HLH-1 in young adults by X-ChIP-seq . HLH-1 is the C . elegans myoD homologue and is required for proper myogenesis ( Krause , 1995 ) . It is exclusively expressed in the same cells that have been used for the muscle chromatin profiling ( Figure 7—figure supplement 1B ) . We found that HLH-1 is enriched at the majority of centromeric sites ( Figure 7D ) . We also analyzed the previously published transcription factor ChIP-seq datasets for binding at centromeric sites ( Gerstein et al . , 2010 ) . We found that all tested transcription factors profiled in larval instar 2 or later stages , when few somatic cells are still dividing , are enriched at centromeric sites ( Figure 7E ) . These results show that if cenH3 is depleted , centromeric sites are not occupied by canonical nucleosomes , but are bound by sub-nucleosome-size particles at least some of which are known transcription factors . As centromere function is needed only in dividing cells , this possibility is consistent with the observation that some HOT sites have post-mitotic functions as enhancers ( Kvon et al . , 2012; Chen et al . , 2013 ) .
Holocentricity is a common mode of chromosome organization , having evolved from monocentricity at least 13 times , including organisms as diverse as nematodes , moths , and sedges ( Melters et al . , 2012 ) . Based on cytological observations two very different models for holocentricity were proposed more than 60 years ago: diffuse centromeres and polycentromeres ( Schrader , 1947 ) . We use high resolution mapping of centromeric nucleosomes to demonstrate that the polycentromere model is correct for C . elegans , and that C . elegans holocentromeres consist of about 100 centromeric sites on each chromosome . The number of discrete centromeric sites is in excess over the observed number of microtubule attachments , which has been estimated to be ∼100 for all six chromosomes by electron tomography ( O’Toole et al . , 2003 ) . It is possible that only about 15% of the centromeric sites identified in this study are attached in each mitotic cell , which seems reasonable given that we analyzed a diverse cell population . The availability of multiple centromeric sites might reflect redundancy to assure faithful segregation in every cell cycle , although it is also possible that some sites are rendered inaccessible for cenH3 nucleosomes in some cell lineages due to changes in expression profiles during development . Polycentromere and diffuse centromere mechanisms might not be mutually exclusive , insofar as a large fraction of cenH3 is incorporated into broad domains that cover half of the genome . It has previously been reported that these domains anti-correlate with transcription ( Gassmann et al . , 2012 ) , and we show that they also anti-correlate with H3 . 3 and correlate with H3K9me3 . These correlations all indicate that cenH3 is preferentially located in regions of low nucleosome turnover , and the intrinsic instability of cenH3 nucleosomes may contribute to losing it from ‘open’ chromatin ( Conde e Silva , 2007 ) . The domains are in part shaped by transcription in the germline and are present in the early embryo , however , there is no significant RNA Pol II-dependent transcription during the first two rounds of embryonic cell division . In contrast , H3 . 3 is deposited in the germline and has a well-established role in the inheritance of chromatin states ( Ooi et al . , 2006; Ooi et al . , 2010; Jullien et al . , 2012 ) . Specifically , H3 . 3 is retained both in mature sperm and oocytes , suggesting that it transmits epigenetic information through both the maternal and the paternal germline . The maintenance of H3 . 3 in sperm might explain how the domain pattern is established on paternal chromatin upon fertilization , even though cenH3 is not maintained in mature sperm ( Gassmann et al . , 2012 ) . However , because these domains do not align with the kinetochore , they probably do not have a direct centromere function , although they might serve as cenH3 ‘reservoirs’ , in parallel with the suggestion that Drosophila transcription factor hotspots might serve as transcription factor reservoirs ( Moorman et al . , 2006 ) . We have found that individual centromeric sites resemble budding yeast point centromeres: a single cenH3 nucleosome flanked by two well-positioned nucleosomes . Point centromeres are genetically defined in budding yeast , and satellite repeats help position centromeric nucleosomes at regional centromeres in many species . However , neo-centromeres can form on sequences that are not normally linked to centromeres ( Marshall et al . , 2008; Shang et al . , 2013 ) . These observations suggest that centromeric nucleosomes are inherited in a sequence-independent way , so that it might seem surprising that a distinct sequence motif is associated with C . elegans centromeric sites . However , given that the motif is short and its abundance in the genome by far exceeds the number of centromeric sites , there is no evidence that it is a direct target for cenH3-nucleosome loading . Rather , the DNA at these sites might disfavor the formation of canonical nucleosomes , allowing centromeric nucleosomes to form in these ‘gaps’ in the chromatin landscape . A similar model for worm holocentromeres had been proposed by Gu and Fire ( Gu and Fire , 2010 ) based on finding ∼120-bp ‘holes’ in the nucleosome landscape that could fit small nucleosomes the size of those previously shown for Drosophila cenH3 ( Dalal et al . , 2007 ) . In C . elegans , virtually any piece of DNA injected into the gonad will concatamerize , acquire a centromere and segregate with varying efficiency ( Stinchcomb et al . , 1985; Mello et al . , 1991; Yuen et al . , 2011 ) . Opportunistic assembly of centromeric nucleosomes at sites of accessible DNA predicts that cenH3 will be loaded onto any fragment of DNA that contains accessible stretches . The fact that new extrachromosomal arrays initially partition passively and only acquire segregation competence after a few cell cycles indicates that centromeric competence can be acquired epigenetically in C . elegans ( Yuen et al . , 2011 ) , consistent with an opportunistic gap-filling model . We found that centromeric sites coincide with HOT sites , which are occupied by many transcription factors without having high binding affinity for any of them . When cells exit the cell cycle and cenH3 is no longer expressed , ‘holes’ in the chromatin landscape might open up and allow HOT site transcription factors to bind by mass action . Although many HOT site transcription factors are cell type-specific , their non-specific binding to holes vacated by cenH3 nucleosomes would result in high HOT site transcription factor occupancy in multiple cell-types . Indeed , we found that cenH3 is lost in adult muscle cells , but that centromeric sites remain occupied , in part by the muscle-specific transcription factor HLH-1 and presumably also by other HOT site transcription factors . Virtually all transcription factors profiled in postembryonic tissues ( larval instar 2 and later stages ) were found at many , if not all centromeric sites . Replacement of cenH3 nucleosomes by transcription factors at HOT sites upon exit from the cell cycle may then result in their reported enhancer activity ( Kvon et al . , 2012; Chen et al . , 2013 ) . CenH3 protein has been shown to turn over completely during the mitotic cell cycle , to disappear during the pachytene stage of meiotic prophase , to reappear when nuclei progress into diplotene and to be absent from mature sperm ( Gassmann et al . , 2012 ) . These observations imply that the centromeric sites need to be marked during certain stages of the cell cycle in order to be repopulated at later stages . The coincidence of cenH3 and HOT sites raises the possibility that low-affinity binding of transcription factors by mass action prevents encroachment of nucleosomes and thus , helps to maintain holocentric sites over the course of development . Our results resolve the long-standing question whether holocentromeres are polycentric or diffuse . We show that C . elegans holocentromeres are organized as dispersed point centromeres , consistent with the polycentromere model . Our discovery of the coincidence of centromeric sites with transcription factor hotspots points to a possible mechanism for centromeric site selection and maintenance .
We used the standard wild-type strain N2 and OP64 grown at 20°C . Synchronized populations were cultured on peptone-rich plates seeded with E . coli strain NA22 . To deplete KNL-2 by RNAi , synchronized populations were grown on NA22 until fourth larval instar ( L4 ) , washed in M9 buffer and transferred to bacteria expressing dsRNA that targets knl-2 for 24 hr . Embryos were harvested from adults by sodium hypochlorite treatment . Worms were decapitated with a razor blade in M9 buffer , freeze cracked on dry ice and fixed 2 min in methanol and 4 min in acetone at −20°C . Samples were incubated with anti-cenH3 ( rabbit ) ( Buchwitz et al . , 1999 ) and anti-NPP-9 ( mouse ) ( Sheth et al . , 2010 ) antibodies overnight at 4°C and with Cy3 donkey anti-rabbit and DyLight 488 donkey anti-mouse antibodies ( Jackson ImmunoResearch ) for 1 hr at 37°C . Washes were carried out with phosphate buffered saline containing 1% Tween-20 ( PBS-T ) throughout . Samples were incubated in PBS-T containing 0 . 01 mg/ml 4′ , 6-diamidino-2-phenylindole ( DAPI ) before being mounted . Images were acquired using a Nikon Eclipse 90i microscope ( 60x lens ) . N2 embryos were treated in 0 . 1U/ml chitinase ( Sigma ) for 30–60 min and washed with buffer A ( 15 mM Tris–HCl pH7 . 5 , 2 mM MgCl2 , 340 mM sucrose , 0 . 2 mM spermine , 0 . 5 mM spermidine , 0 . 5 mM phenylmethanesulfonate [PMSF] ) . Nuclei were isolated using a glass Dounce homogenizer with 15 strokes each of the loose- and tight-fitting inserts in buffer A supplemented with 0 . 1% Trition X-100 and 0 . 25% NP-40 substitute . The homogenate was diluted five times with buffer A , the debris were removed by spinning at 100×g for 2 min and nuclei were pelleted by spinning at 1000×g for 10 min . Nuclei were transferred to 1 ml 10 mM Tris pH7 . 5 , 2 mM MgCl2 , 0 . 5 mM PMSF and pre-warmed for 5 min at 37°C . CaCl2 to a final concentration of 2 mM , and 0 . 1 units of micrococcal nuclease ( MNase; Sigma–Aldrich ) was added . After 1 , 2 , 5 or 10 min the reaction was stopped by the addition of ethylenediaminetetraacetic acid ( EDTA ) to a final concentration of 30 mM . A light MNase digestion corresponding to 2 min in this time-course experiment was used for all other experiments unless otherwise noted . Chromatin was solubilized by cavitation using needle extraction ( 4 times 20 gauge , 4 times 26 gauge ) , a protocol modified from Jin and Felsenfeld , 2007 . Soluble chromatin was collected by spinning at 1000×g for 5 min and the supernatant was pooled with additional chromatin solubilized by incubating the pellet in 10 mM Tris pH7 . 5 , 10 mM EDTA , 0 . 1% Trition X-100 , 0 . 5 mM PMSF for 4 hr at 4°C . The remaining pellet was retained as the insoluble chromatin fraction . Soluble chromatin fractions were combined , NaCl adjusted to 100 mM , debris removed by spinning 4 times at maximum speed for 5 min and pre-cleared by incubation with Dynabeads protein A ( Invitrogen ) . From this input fraction , cenH3 was isolated by incubation with 4 µl anti-cenH3 antibody overnight and protein A dynabeads for 2 hr . Beads were washed three times in 10 mM Tris pH7 . 5 , 100 mM NaCl , 10 mM EDTA , 0 . 1% Trition X-100 , 0 . 5 mM PMSF and twice in 10 mM Tris pH7 . 5 , 100 mM NaCl , 10 mM EDTA , 0 . 5 mM PMSF . Chromatin was treated with RNase and Proteinase K , and DNA was isolated with phenol:chloroform and precipitated with ethanol in the presence of glycogen . For CENP-C ChIP , nuclei were prepared and MNase treated as for native ChIP , except that MNase incubation was done in HM2 ( 50 mM HEPES pH7 . 4 , 2 mM MgCl2 , 0 . 5 mM PMSF ) for 2 min . MNase was inactivated by addition of EGTA to 5 mM and nuclei were washed once in HM2 . Chromatin was crosslinked in 1% formaldehyde for 10 min . Crosslinking was quenched by adding glycine to 125 mM for 10 min . Nuclei were washed with HM2 and lysed in 50 mM Tris pH7 . 5 , 10 mM EDTA , 1% SDS , 0 . 5 mM PMSF by vortexing for 2 min . The lysate was diluted to 20 mM Tris pH7 . 5 , 150 mM NaCl , 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 0 . 5 mM PMSF and sonicated with a Sonic Dismembrator Model 500 ( Fisher Scientific ) for 40s at 30% amplitude . Debris removal , pre-clearing and CENP-C ChIP were done as for native ChIP , with the antibody from Moore and Roth , 2001 . Beads were washed twice with 20 mM Tris pH7 . 5 , 150 mM NaCl , 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , once each in 20 mM Tris pH7 . 5 , 500 mM NaCl , 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA and 20 mM Tris pH7 . 5 , 250 mM LiCl , 1% sodium deoxycholate , 1% NP-40 substitute , 2 mM EDTA and once in TE . Crosslinks were reversed overnight at 65°C , chromatin was treated with RNase and Proteinase K , and DNA was isolated with phenol:chloroform and precipitated with ethanol in the presence of glycogen . For HLH-1 ChIP , OP64 worms were washed in PBS , ground under liquid nitrogen and resuspended in PBS containing 1x Complete Protease Inhibitor Cocktail ( Roche ) . Proteins were crosslinked with 1% formaldehyde for 15 min , the reaction was quenched with 125 mM glycine for 10 min , the volume increased to 50 ml , and chromatin pelleted by spinning at 2000×g for 10 min . The pellet was washed again in PBS , resuspended in 50 mM Tris pH7 . 5 , 10 mM EDTA , 1% SDS containing 1x Complete Protease Inhibitor Cocktail ( Roche ) , incubated 10 min at room temperature , diluted to 20 mM Tris pH7 . 5 , 150 mM NaCl , 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA containing 1x Complete Protease Inhibitor Cocktail ( Roche ) and sonicated with a Sonic Dismembrator Model 500 ( Fisher Scientific ) for 4 min at 30% amplitude . Debris was removed by spinning twice at maximum speed for 5 min . The extract was incubated with an anti-FLAG M2 antibody ( Sigma ) overnight at 4°C . Protein G beads pre-blocked with BSA and yeast tRNA were added for 4h . Beads were washed and DNA isolated as described for CENP-C ChIP above . Libraries were prepared using a modified Illumina paired-end library protocol as described in Henikoff et al . , 2011 . Cluster generation , followed by 25 rounds of paired-end sequencing in an Illumina Hi-Seq 2000 , was performed by the FHCRC Genomics Shared Resource . After processing and base-calling by Illumina software , paired-end reads were mapped to the C . elegans genome release WS220 using Novoalign ( http://www . novocraft . com ) with default parameters , except that each multiple hit was mapped to one site chosen at random ( Novoalign parameter -r Random ) . The number of inserts aligned to each 10-bp interval of the genome was counted , and the interval counts were normalized by dividing by the total number of counts for all intervals , and then scaled by multiplying by the number of bases in the genome . We considered fragments >140 bp to represent nucleosomes , the in silico equivalent of excising a gel slice around the ∼150-bp size range from an MNase-digested chromatin ladder and extracting the DNA for single-end sequencing . As we use a modified paired-end sequencing protocol to include all fragments >25 bp ( Henikoff et al . , 2011 ) , we can accomplish the size cut more precisely by mapping only the reads in the nucleosome size range . Simple repeat regions were downloaded from www . wormbase . org and excluded from all analyses . To call peaks , given the discrete nature of the sites of high cenH3 signal , we set a threshold and considered all the features with higher counts as peaks . For cenH3 peak calling , input counts were subtracted from cenH3 ChIP counts for two biological replicates ( 2 min MNase ) , and peaks that exceeded 30 counts in at least one of the biological replicates were considered positive . For CENP-C peak calling , input counts were subtracted from CENP-C ChIP counts , and peaks that exceeded 20 counts were considered positive . For insoluble chromatin peak calling , peaks that exceeded 100 counts were considered positive . To normalize against input , we used log2-ratios only to compare to previously published array data . Otherwise , we consider input reads a separate “blank” experiment that we subtract from the ChIP counts . | During cell division , the chromosomes in the original cell must be replicated and these ‘sister chromosomes’ must then be divided equally between the two new daughter cells . At first , the sister chromosomes are held together near a region called the centromere , which is important because the microtubules that pull the sister chromosomes apart attach themselves to the centromere . In many cases , the centromere is a small region near the middle of the chromosomes , which produces a classic X shape . However , in some organisms centromeres span the entire length of the chromosomes . There are at least 13 plant and animal lineages with such holocentromeres . Inside the nucleus of cells , DNA is wrapped around molecules called histones . There are five major families of histones , and histones belonging to one of these families—the H3 histones—are replaced by cenH3 variant histones at both conventional centromeres and holocentromeres . There are many unanswered questions about holocentromeres . In particular , do holocentromeres truly extend along the full length of the chromosomes , or are they found at a large number of specific sites ? Now Steiner and Henikoff have studied the distribution of cenH3 in the genome of the worm C . elegans to investigate holocentromeres in greater detail . These experiments showed that the holocentromere in C . elegans is actually made of about 700 individual centromeric sites distributed along the length of the chromosomes . Each of these sites contains just one nucleosome that contains cenH3 , and these sites are likely to be the sites that microtubules attach to during cell division . Surprisingly , the same sites can also act as so-called ‘HOT–sites’: these sites are bound by many proteins that are involved in regulating the process by which genes are expressed as proteins , which suggests a link between centromeres and these regulatory proteins . The work of Steiner and Henikoff describes how centromeric nucleosomes are distributed across the genome , but why and how cenH3 ends up at these particular 700 sites remains an open question . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2014 | Holocentromeres are dispersed point centromeres localized at transcription factor hotspots |
Cancer clone evolution takes place within tissue ecosystem habitats . But , how exactly tumors arise from a few malignant cells within an intact epithelium is a central , yet unanswered question . This is mainly due to the inaccessibility of this process to longitudinal imaging together with a lack of systems that model the progression of a fraction of transformed cells within a tissue . Here , we developed a new methodology based on primary mouse mammary epithelial acini , where oncogenes can be switched on in single cells within an otherwise normal epithelial cell layer . We combine this stochastic breast tumor induction model with inverted light-sheet imaging to study single-cell behavior for up to four days and analyze cell fates utilizing a newly developed image-data analysis workflow . The power of this integrated approach is illustrated by us finding that small local clusters of transformed cells form tumors while isolated transformed cells do not .
Organoid cultures grown from cell-lines or primary cells have been successfully employed to study molecular mechanisms during different stages of tumorigenesis ( Clevers , 2016; Havas et al . , 2017; Simian and Bissell , 2017 ) . However , they are derived from primary material that usually allows oncogenic activation in all cells of the tissue and therefore cannot reproduce the localized expansion at a defined part of the tissue that is seen in the patient situation . To study tumor initiation , models need to be established wherein only a few tumorigenic cells expand in the context of its immediate non-tumor microenvironment ( Tabassum and Polyak , 2015 ) and the interplay of these populations can be visualized . Long-term imaging of primary organoids has been achieved via light-sheet microscopy ( Drost et al . , 2015; Serra et al . , 2019; Verissimo et al . , 2016 ) and these efforts have benefited from the lower phototoxicity provided by SPIM imaging . However , regarding fast dividing cells in tumor organoids , tracking single-cell dynamics necessitates high resolution imaging which in turn limits the time frame in which organoids can be imaged without phototoxic effects ( Held et al . , 2018 ) . Conversely , imaging primary organoids for longer time periods requires an offset of temporal and cellular resolution that eventually cannot allow single-cell fate tracking ( Dekkers et al . , 2016 ) . Two models have been used to understand tumor progression and heterogeneity ( Plaks et al . , 2015 ) ; the ( i ) hierarchical model refers to tumor propagating cells as cancer stem cells ( CSCs ) ( Kreso and Dick , 2014 ) and the ( ii ) stochastic model states that every cell within a tumor is equally likely to be the cell of origin and facilitate tumor initiation and progression ( Vogelstein et al . , 1988; Williams et al . , 2007 ) . These models for tumor propagation can be reconciled when considering that the population of oncogene driven cells have the capacity to interconvert between differentiated- and stem-like states ( Chaffer and Weinberg , 2015; Gupta et al . , 2011; Plaks et al . , 2015 ) , a flexibility that is supported by the microenvironment ( McGovern et al . , 2009; Quail et al . , 2012 ) . These concepts , however , have not been probed in the context of tumor initiation upon stochastic oncogene activation in an established cell-layer . Here we present a novel model of breast tumorigenesis where only single cells express oncogenes within healthy mammary acini and are able to establish localized outgrowth . We thereby overcome the above-mentioned limitations of studying tumorigenesis events in the face of tissue-wide transformation . Furthermore , we perform long-term imaging of these stochastic breast tumor acini for the first time at a temporal resolution that allows us to follow single-cell fates . We also integrate this approach with an image analysis pipeline capable of segmenting cells in their dynamic progression towards tumorigenesis , so they can be tracked individually over time .
For modeling tumorigenesis in breast tissue , we use an inducible mouse model of breast cancer ( TetO-MYC/TetO-Neu/MMTV-rtTA transgenic mice ) ( Fry et al . , 2016; Moody et al . , 2002; Podsypanina et al . , 2008 ) that has been shown to recapitulate hallmarks of human breast disease ( Havas et al . , 2017; Jechlinger et al . , 2009; Figure 1a ) . In this tractable transgenic mouse model , the activity of two potent oncogenes – MYC and Neu ( the rodent homolog for the human ERBB2 gene used in this mouse model ) – are under the control of a Tet-O promoter that can only be activated in the presence of both the rtTA ( reverse tetracycline-controlled transactivator ) protein and the doxycycline compound . This three-part inducible system allows for temporal control of oncogenic expression by regulating doxycycline in the medium or animal diet and spatial control by the MMTV promoter that confines expression of the rtTA protein to the cells of the mammary lineage ( Bockamp et al . , 2002 ) . We modify this tissue wide tumorigenesis model ( tri-transgenic ( T ) model; Figure 1a , upper panels ) to generate a stochastic system by retaining only the oncogenic constructs , making them TetO-MYC/TetO-Neu mice ( bi-transgenic ( B ) model; Figure 1a , lower panels ) . The rtTA inducer gene is then lentivirally delivered to single cells within luminal epithelial acini derived from the bi-transgenic ( B ) mouse glands , granting only this subset of cells the competence to express transforming oncogenes and thereby preventing tissue wide transformation . Hence , this novel ( B ) model should allow acini to present with two cell populations –tumorigenic and non-tumorigenic– and could be used to study the dynamics of early tumorigenesis in an otherwise normal epithelium ( Figure 1a , lower right panels ) . Primary mammary epithelial cells derived from transgenic mice were seeded in 3D matrigel as single cells to be grown for 48 hr to form small acini consisting of around 15 cells that arrange around a small lumen . A small number of single cells in these acini were then transduced with lentiviral particles ( Figure 1a , middle panel ) . In the tissue-wide tumorigenesis model ( T ) , acini were transduced with the reporter virus ( pLv-pGK-H2B-GFP ) that marks a subset of cells with H2B-GFP , while tissue wide rtTA expression is driven in all cells ( Figure 1a , left panels ) . To achieve stochastic tumorigenesis , bi-transgenic ( B ) acini were transduced with the inducer-reporter virus ( pLv-pGK-rtTA-p2A-H2B-GFP ) that expresses rtTA and reporter H2B-GFP in only these single cells within the normal epithelium ( Figure 1a , right panels ) . Then , doxycycline was supplemented in the medium to induce tumorigenic growth in rtTA expressing cells . Immunofluorescent staining of 3D matrigel cultures , for both sets of doxycycline-induced transduced acini , was used to validate transgene specific protein expression of the MYC oncogene ( human ) in only the transduced cells of ( B ) acini as opposed to all cells of ( T ) acini ( Figure 1b , anti-human MYC specific antibody ) . We FACS sorted H2B-GFP marked cells from both systems respectively for qPCR analysis of transgene specific MYC and Neu mRNA expression . This way we normalized doxycycline dosage in both systems to obtain the same mRNA expression levels of oncogenes using 800 ng/ml doxycycline in the tissue wide ( T ) system and 600 ng/ml doxycycline in the stochastic ( B ) system ( Figure 1—figure supplement 1 ) . To verify oncogene induction patterns and to understand the impact on epithelial morphology in the acini , 3D matrigel cultures of both doxycycline-induced ( T ) and ( B ) transduced acini were analyzed by immunofluorescent staining for polarity marker distribution . Lentivirus-transduced structures from the tissue wide induction system ( T ) presented with filled lumens that contained both infected ( GFP positive ) as well as non-infected oncogene-expressing cells . These structures exhibited overall disrupted epithelial polarity and some random microlumen , as reported previously ( Jechlinger et al . , 2009; Figure 2a ) . For the stochastic tumor induction system ( B ) , we observed 2 phenotypes after 96 hr of oncogene expression; some of the ( B ) acini showed exclusively H2B-GFP marked cells in hyperplastic areas that also displayed double cell rim morphology and disturbed epithelial polarity ( Figures 2a and 96 hours ON DOX , left panels ) , while the other ( B ) acini did not show expansion of the few H2B-GFP positive cells ( Figures 2a and 96 hours ON DOX , right panels ) . Intrigued by these 2 distinct phenotypes , we analyzed transduced ( H2B-GFP expression ) ( B ) acini for their oncogenic MYC expression ( Figure 2b ) . We verified that multi layered regions in acini contained mainly H2B-GFP positive cells that expectedly stained for MYC ( Figure 2b , middle panels ) . For normal appearing regions , this analysis showed that the inability of single H2B-GFP positive , transduced cells to expand over 96 hr was not due to the lack of MYC expression ( Figure 2b , lower panels ) . To follow up these observations in more detail over time , we bred the nuclear reporter H2B-mCherry into the T and B mice to mark all the cells in the acinus for inverted light-sheet microscopy ( Luxendo InVi SPIM , Figure 3—figure supplement 1 ) . The InVI SPIM was adjusted for non-phototoxic , long-term imaging ( up to 4 days , every 10 min with 1 µm z-spacing ) . The T/H2B-mCherry acini transduced with reporter virus proliferated swiftly upon doxycycline addition showing expansion of both the marked and unmarked cells ( Figure 3a ) ; a sturdy tumor phenotype developed , manifested by multi-cell-layered rims and pronounced proliferation-associated-apoptosis in all acini ( Figure 3—Video 1 ) . In contrast , B/H2B-mCherry acini transduced with inducer-reporter virus , displayed phenotypic variation upon induction of oncogenes in the transduced cells . Some acini showed fast clonal expansion of oncogene-expressing cells that form multilayer clusters in the acinus rim . This proliferative phenotype seems to stem from several transduced cells in vicinity to each other at the start of time-lapse imaging ( Figure 3b , upper panel; Figure 3—Video 2 ) . Again , more sparsely infected acini , did not sustain proliferation of the oncogene-expressing cells ( Figure 3b , lower panel; Figure 3—Video 3 ) . Taken together , the phenotypes for ( B ) and ( T ) acini are consistent with the respective behavior observed with the 96 hr end point immunofluorescent analysis ( Figure 2a and b ) , verifying these prior observations and excluding phototoxic effects due to longitudinal SPIM imaging . To analyze the dual-color light-sheet movies ( H2B-mCherry—all cells in the acinus , H2B-GFP—transduced cells within the acinus ) on a single-cell level , we developed a big image data compatible image analysis pipeline that allows efficient visualization of longitudinal big image data , nuclear segmentation and single-cell tracking in 3D ( Figure 3c , Figure 3—figure supplement 3 and Figure 4—figure supplement 2 ) . The Big Data Tools plugin ( Big Data Processor ( Tischer , Norlin , and Pepperkok ) ) for Fiji ( Schindelin et al . , 2012 ) was used to visualize the 2D image stacks recorded over time . It uses lazy loading to stream the image files , which can be up to a few terabytes in size per acinus over 3–4 days of imaging . It also allows for easy cropping , binning , chromatic shift correction between channels and file format conversion to Imaris compatible formats . Once the raw data have been pre-processed , the CATS plug-in from Fiji ( Tischer and Pepperkok ) was used for segmenting the cell nuclei in the green channel images . The CATS plug-in uses machine learning algorithms to predict the probabilities of all pixels in the image and classify them into pre-defined classes . Three classes — background , nucleus , boundary — were trained manually on the CATS tool and the segmented images were exported to the commercially available Imaris software ( Bitplane AG , 2020 , Software available at http://bitplane . com ) for 3D rendering and tracking . The nucleus class channel from the CATS tool was used for surface rendering and the surfaces in all time points were tracked over the course of the movie . Cell tracking allowed us to follow the clonal evolution for each transduced cell in the acinus over 3 days . Single transduced cells within one acinus show a difference in proliferation and cell fate as indicated in representative tracks ( Figure 4a ) . To better understand the parameters that positively affect a transduced cell in the stochastic tumorigenesis model to start proliferating and establishing a tumor within a normal epithelium , we extracted the center of mass coordinates of all cells in the acinus at the start of the imaging and represented them as dots in a 3D space: depicted as magenta for normal cells and green for transduced cells in Figure 4b ( upper right panel ) . Equipped with the tumor formation success of each stochastically transduced acinus , we hypothesized that tumors originate from groups of independently transduced oncogene-expressing cells in closer vicinity within the acinus . To explore this in detail , we defined clusters of cells in the stochastic model as a group of oncogene-expressing cells that are closer to each other than to other oncogene-expressing cells of the same acinus . Notably , these clusters contain both oncogene-expressing cells as well as normal cells of the acinus . The cluster volume was described as the sphere located at the center of mass of a cluster with a diameter equal to the distance between the two farthest oncogene-expressing cells of the cluster . Next , we extracted 9 features that represent reasonable factors that could influence tumor establishment in the epithelium such as size of the acinus ( represented by number of cells in the acinus ) , density of cells in the acinus , infection efficiency ( represented by number of infected cells in the acinus ) , number of cells in a defined cluster volume , number of independently transduced cells in a cluster volume , distances between all cells in a cluster volume , distances between transduced cells in a cluster volume , fraction of transduced cells in the cluster volume and number of cell-cell contacts between transduced cells in a cluster volume ( Figure 4b , left panel ) . We analyzed 20 acini and all the transduced cells in these acini ( n = 150 ) at the start of the imaging to ascertain the effect of the immediate non-transformed cell microenvironment on tumor cell proliferation . Since clusters of oncogene-expressing cells could be tracked over time , each cluster was associated with either a tumor outcome , or not . To identify which features were linked to this outcome , we chose a model selection approach , based on information theory . We fitted a logistic regression model for all possible linear combinations of features and selected the best model based on the Akaike information criterion ( with correction for small sample sizes ) ( Calcagno and Mazancourt , 2010 ) . In this model only one feature , the ‘number of transduced cells in the cluster’ , can predict whether a cluster can form a tumor . Each additional transduced cell in a cluster increases the odds of this cluster forming a tumor by 9 ( Figure 4b , lower right panel ) . Since other models within a few information criterion units of the best one can be considered equally good at predicting tumor formation , we also looked at features included in models within 3 information criterion units from the best one . Relative importance of features can be estimated by the sum of the information criterion values of the models in which each feature appears ( Buckland et al . , 1997 ) . Computing relative importance across the selected models also revealed ‘number of transduced cells in the cluster’ as the most important feature , followed by ‘number of cells in a cluster’ and ‘number of contacts between transduced cells in a cluster’ . However , the only feature with significant contribution to tumor formation is ‘number of transduced cells in the cluster’ . It is important to note that this result does not rule out contribution from other factors which more targeted experiments may be able to reveal . We also analyzed the effect of whole acinar features at the beginning and end of the imaging period to try and detect acinus-specific advantages , like acinus size , cell density and cell proliferative rate . Acini with transduced cell clusters that displayed a tumor outcome were compared to acini where the cell clusters did not result in tumor outcomes ( Figure 4—figure supplement 1a ) . Acini with tumors only differed significantly from those without tumor on the number and proliferation rate of transduced cells in line with the observation that clusters of transduced cells are the primary indicators of tumor outcome . To account for this and possibly other acinus-specific effects , we also evaluated a logistic regression model including the three most important features identified above plus a random effect for acinus of origin and interaction terms for the starting number of normal cells in the acinus . In this model again , the number of oncogene-expressing cells in the cluster appeared as the main contributing factor to tumorigenesis with an odds ratio of 6 . 7 ( Figure 4—figure supplement 1b ) . That tumors are mainly initiated by clusters of transduced cells is further illustrated by representative acini shown in Figure 4c; proliferating cells that establish tumor foci at the end of the long-term imaging cluster together at the start of the time lapse imaging and the non-proliferative cells are more sparsely located within the acinus . In line , we found examples of similar sized acini bearing the same number of transduced cells that showed differential cluster expansion depending on the ‘numbers of transduced cells within a cluster’ ( Figure 4—figure supplement 2; comparison a to b ) . Consequently , and again depending on the number of cells within the specific cluster , we observed clusters behaving differently within the same organoid; the organoids shown in Figure 4a , Figure 4c ( top and middle panels ) and Figure 4—figure supplement 2a , display clusters of single cells that do not expand over time , whereas clusters containing several cells show massive expansion ( beyond normal expected cells division rates due to growth of the acinus ) .
Our results indicate that a proximity-controlled interaction or signaling network between different transformed cells might guide tumor outgrowth in a normal epithelium ( Greaves and Maley , 2012; Reid et al . , 2010 ) . This might be due to the repressive effect that an intact polarized tissue layer exerts on single early stage cancerous cells ( Lee and Vasioukhin , 2008 ) . According to our modeling attempts the simple parameter of ‘number of transduced cells within a cluster’ is enough to impinge on overall epithelial integrity and most likely already accounts for sub-features like ‘distances between transduced cells in a cluster volume’ or ‘number of cell-cell contacts between transduced cells in a cluster volume’ , which did not pan out as predictive factors on their own . These observations , however , do not exclude the importance of such traits , which should be explored in more detail for the different situations we report ( proliferative versus non proliferative clusters ) . In addition , the increased probability for tumor outgrowth upon proximity effects of closely located transformed cells , might also be rooted in paracrine effects ( e . g . via microRNAs [Kosaka et al . , 2012] in the cluster , or hydrogen ion dynamics/pH [Reshkin et al . , 2014] ) . The questions outlined above can now be assessed employing the presented 3D system and longitudinal imaging pipeline together with appropriate fluorescent-tagged reporter proteins . Studies on loss of important polarity proteins have highlighted their function as non-canonical tumor suppressors in breast tumorigenesis ( McCaffrey et al . , 2012; Partanen et al . , 2012 ) ; however , other reports ( Macara and McCaffrey , 2013; Xue et al . , 2013 ) cannot confirm these observations which were all obtained from tissue wide transformed model systems , leaving conclusions open for alternative studies . Similarly , it will interesting to trace the effect of prominent tumor factors in breast cancer , such as SCRIBL/MYC ( Zhan et al . , 2008 ) or the Hippo pathway ( Calses et al . , 2019 ) on overall tissue integrity and promotion of tumor growth in a none tissue wide expression setting . Clearly , to better interrogate the effect of oncogenic drivers on epithelial polarity in more detail , to understand deficiencies in cell-cell interactions towards tumorigenesis and to settle conflicting reports as mentioned above , there is a need for a more detailed analysis , employing a model system that does not show modification of all cells in the tissue . The interaction of tumor cells with the immediate microenvironment has been subject of extensive studies with regards to immune cells ( Binnewies et al . , 2018 ) and other tumor associated cell types ( Tabassum and Polyak , 2015 ) ; however , the interaction with the normal neighboring cells has not been explored in real time using an organotypic mammalian model system . Rather , questions of cell competition in heterogenous tissues have mainly been addressed in either 2D culture systems or in the Drosophila wing ( Tamori and Deng , 2011 ) . Alternatively , organ-specific mammalian cell types that self-organize in a manner similar to in vivo can now be studied in specialized 3D culture conditions in vitro ( Lancaster and Knoblich , 2014 ) . Furthermore , these cultures can be used to model disease and serve as an alternative system for drug testing that better recapitulate effects as compared to conventional 2D cell culture ( Clevers , 2016; Simian and Bissell , 2017 ) . Our adaptation of such culture systems , to mark and at the same time enable single cells to express regulatable oncogenes , gave rise to the first stochastic breast tumor model using mammary luminal epithelial acini . Together with a newly established imaging and data processing pipeline , we have developed an integrated approach that allows us to follow cell fates and to interrogate cell-cell interactions of a tumor cell with the normal epithelium . We report that cells within a polarized , differentiated epithelial cell layer are able to establish tumorigenic outgrowth upon stochastic activation of strong oncogenic drivers , supporting the concept of interconversion between differentiated- and stem-like states ( Chaffer and Weinberg , 2015; Gupta et al . , 2011; Plaks et al . , 2015 ) for tumor initiating cells . However , not all cells expressing the very powerful combination of MYC and Neu oncogenes can expand , demonstrating an inhibitory effect that needs to stem from the normal epithelial microenvironment in the presented system . The ability to distinguish marked tumor cells from the normal epithelium will now allow us to perform single-cell RNA sequencing analysis on select sorted cells . This will help delineate the signaling networks within the immediate tumor microenvironment and further clarify the molecular basis of these distinct processes . The amenability of 3D multicellular systems to interference with small molecule inhibitors , viral shRNA vectors and genomic editing has the potential to further our understanding of the mechanisms important during tumor initiation . In the future , this will also permit drug screens on the heterogeneous populations of normal and malignant cells , as presented in the described stochastic breast tumor model . The ability to follow tumor regression and establishment of minimal residual disease within a tissue like acinus structure will help distinguish between overall cytotoxic drugs and tumor cell specific inhibitory drugs . Taken together , we strongly believe that our integration of a true stochastic tumor model with the ability to image single-cell fates will successfully bridge the gap between genetically modified model systems and the clinical situation , helping gain novel insights on breast cancer .
The mouse strains TetO-MYC/MMTV rtTA ( D'Cruz et al . , 2001 ) and TetO-Neu/MMTV rtTA ( Moody et al . , 2002 ) , that have been previously described , were bred in order to establish the tri-transgenic strain TetO-MYC/TetO-Neu/MMTV rtTA ( T ) or bi-transgenic strain TetO-MYC/TetO-Neu ( B ) . Reporter H2B-mCherry was crossed into the B and T lines using a R26-H2B-mCherry line ( Abe and Fujimori , 2013 ) ( RIKEN , CDB0239K ) . All ten mammary glands were harvested ( from virgin female mice between 8–10 weeks old ) , digested and singularized for establishing acinar cultures . All mice used in this study were housed according to the guidelines of the Federation of European Laboratory Animal Science Associations ( FELASA ) . Rational for the use of these oncogenes: ERBB2 is overexpressed in ~20% of breast cancers ( Arteaga et al . , 2012 ) , MYC in 15–50% of human breast cancer ( Blancato et al . , 2004 ) . The combination of MYC and ERBB2 is found in highly aggressive human breast cancer ( Al-Kuraya et al . , 2004 ) :and - in fact- Neu ( the rodent homolog for the human ERBB2 gene used in this mouse model ) and MYC strongly accelerate tumor onset In the combined transgenic animals ( average 45 days ) as compared to single transgenic animals ( MYC 155 days , Neu 99 days ) , In all cases tumors regress rapidly to non-palpable state following oncogene silencing . The lentivirus design is based on pWPXL backbone , which was a gift from Didier Trono ( Addgene #12257 ) . The coding region from the original plasmid was excised using ClaI and NdeI in order to insert a new multiple cloning site ( MCS ) . The pGK promoter was PCR amplified from pLVPT-GDNF-rtTR-KRAB-2SM2 , which was a gift from Patrick Aebischer and Didier Trono ( Addgene #11647 ) and cloned using XhoI and EcoRI restriction sites . For the plasmid pLenti-rtTA-GFP the synthetic region rtTA-p2A-H2B-GFP was cloned downstream of the pGK promoter using EcoRI and NheI sites . The plasmid pLenti-Null-GFP is derived from the pLenti-rtTA-GFP by removing the rtTA sequence , using the restriction sites EcoRI and BamHI , and retaining H2B-GFP in the coding region . For production of lentivirus particles , we seeded 1 . 6 × 107 HEK-293T cells ( Lenti-X - Clontech Cat . # 632180 ) in 500 cm2 square dishes ( Corning Cat . # 431110 ) . After 24 hr , the cells were supplemented with medium containing 25 µM of chloroquine diphosphate ( Sigma-Aldrich Cat . # C6628 ) . After a 5 hr incubation , using 360 µg of polyethyleneimine ( 4 µg for each µg of plasmid ) , we transfect the cells with a mixture of endotoxin free plasmids: 20 µg pCMV-VSV-G ( Addgene #8454 ) ; 30 µg psPAX2 ( Addgene #12260 ) ; 40 µg transfer plasmids pLenti-rtTA-GFP or pLenti-Null-GFP . We harvested the medium after 48 hr , 72 hr and 96 hr after transfection . Concentration of the lentivirus from the collected medium was performed using an ultracentrifuge ( Beckman Sw32 rotor ) at 25 , 000 rpm for 2 hr at 4°C . The lentivirus pellet was resuspended in 1000 µl of HBBS buffer , aliquoted and stored at −80°C . The lentivirus titer was measured using FACS analysis as described by Kutner and colleagues ( Kutner et al . , 2009 ) . Mammary glands harvested from mice ( see above ) , were digested in order to prepare a single-cell solution . For this , the tissue was divided in four loosely capped 50 ml falcon , each supplemented with 5 ml serum-free media ( DMEM/F12 supplemented with 25 mM HEPES and 1% Pen Strep ( 100 U/ml Penicillin; 100 µg/ml Streptomycin; ThermoFisher Cat . # 15140122 ) ) and 750 U of Collagenase Type 3 ( Worthington Biochemical Corp Cat . # LS004183 ) , 20 µg of Liberase ( Roche Cat . # 5401020001 ) and incubated overnight at 37°C and 5%CO2 . The glands were then mechanically disrupted using a 5 ml pipette , and washed in PBS before being pelleted at 1000 rpm for 5 min . The cell pellet was resuspended in 5 ml of 0 . 25% Trypsin-EDTA and incubated for 45 min at 37°C and 5%CO2 . The enzymatic reaction was then neutralized using 40 ml of serum supplemented medium ( DMEM/F12 with 25 mM HEPES , 1% Pen Strep and 10% FBS Tetracycline Free certified ( Biowest Cat . # S181T ) . The cells were pelleted again , resuspended in Mammary Epithelial Cell Basal Medium ( PromoCell Cat . # C-21210 ) and seeded in collagen coated plates ( Corning Cat . # 354400 ) overnight at 37°C and 5%CO2 . This allows for epithelial cells to adhere to the surface of the plates while the other cell types float on top in the medium and can be easily removed by vacuum suction . The epithelial cells were detached from the collagen coated plates by incubating them with 0 . 25% Trypsin-EDTA for 5–7 min at 37°C and 5%CO2 , following inactivation with serum supplemented media . The single-cell solution was pelleted , resuspended in MEBM and counted . We mixed 50 , 000 cells with 90 µl of Matrigel Matrix basement Membrane growth factor reduced phenol red free ( Corning Cat . # 356231 ) , and seeded this mixture into a 12 well plate ( Corning Cat . # 3336 ) and incubated it for 30–40 min until the matrigel solidified . The gels were supplemented with 1 . 5 ml MEBM and allowed to grow at 37°C and 5%CO2 . For transduction , after 3 days of growth , the gels were mechanically disrupted and placed in a 15 ml falcon . Two disrupted gels were placed in one 15 ml falcon with 2 ml of MEBM supplemented with 25U of Collagenase type I and 5 µg of Liberase . Following incubation in this solution for 2 hr at 37°C and 5%CO2 , when the matrigel was totally digested , the acini were washed 3 times with 15 ml of serum supplemented media and once with 15 ml of serum free media , and pelleted at 1000 rpm for 5 min . We then supplemented the acinus pellet ( from two original gels ) in 10 µl of MEBM and added 6 × 105 lentivirus particles to the solution . We then mixed this solution with 90 µl matrigel and plated it in 35 mm dishes ( Greiner Bio-One Cat . # 627160 ) and placed in incubator for 30–40 min until the matrigel solidified . The gels were supplemented with 3 ml MEBM and incubated for 2 days at 37°C and 5%CO2 in order to allow for acinus recovery and lentiviral gene expression . For induction of oncogenes in the cells of the acini , doxycycline ( Sigma Cat . # D9891 ) was supplemented in the media . 800 ng/ml of doxycycline was used to induce T acini and 600 ng/ml was used for B acini . qPCR analysis was used to standardize the doxycycline dosage for B acini ( see below ) . The qPCR technique was performed following the MIQE guidelines , where the total RNA was isolated from the mammary gland acini using RNA PureLink Mini Kit ( ThermoFisher Cat . # 12183018A ) and 2 . 5 ug was reverse transcribed to cDNA using SuperScript VILO cDNA Synthesis Kit ( ThermoFisher Cat . # 11754050 ) . Using Primer3 software we designed specific primers for DNA intercalating fluorescent dye approach for the transgenes Neu ( Forward: CGTTTTGTGGTCATCCAGAACG and Reverse: CTTCAGCGTCTACCAGGTCACC ) and c- MYC ( Forward: GCGACTCTGAGGAGGAACAAGA and Reverse: CCAGCAGAAGGTGATCCAGACT ) . As endogenous controls , mCherry ( Forward: GAGGCTGAAGCTGAAGGAC and Reverse: GATGGTGTAGTCCTCGTTGTG ) and Pum1 ( Forward: AATGTGTGGCCGGATCTTGT and Reverse: CCCACAGTGCCTTATACACCA ) were used . Primer efficiency was verified and established between 95% and 105% Each sample was analyzed in duplicate and non-template controls were used in each qPCR run . Analyses were carried out using a StepOne device ( Applied Biosystems , USA ) . Analysis of relative gene expression data was performed according to the 2−ΔΔCq method and the results were expressed as fold change of ΔΔCq values obtained from the reference T800 acini ( Figure 1—figure supplement 1 ) . Matrigel cultures were grown as described above and plated on Nunc Lab-Tek II ( Thermo Cat . # 155382 ) chambers . At pre-defined timepoints , the gels were fixed using 4% PFA for 2–3 min , following 3 washes with PBS . The gels were blocked with 10% goat serum for 2 hr at room temperature , followed by incubation with primary antibodies was done overnight at 4°C . The remaining immunofluorescence staining was performed as per standard protocol for c-MYC ( Cell Signaling Technologies , Cat . # D84C12 , 1:900 ) , alpha6-integrin ( Millipore Cat . # MAB1378 , dilution 1:80 ) and ZO1 ( Life Technologies Cat . # 61–7300 , dilution 1:500 ) . The nuclei were counter stained with 1:1000 DAPI ( ThermoFisher Cat . # 62248 , 1 mg/ml , dilution 1:1000 ) and mounted in anti-fading mounting medium ( VECTASHIELD Mounting Medium with DAPI ( Vecto Cat . # H1500-10 ) ) . Please note that the c-MYC antibody ( Cell Signaling Technologies , Cat . # D84C12 ) recognizes specifically the human protein , which is transgenically expressed and does not recognize endogenous mouse MYC protein . Stained gels were imaged on Leica SP5 confocal microscope using 63x water lens and the LAS AF imaging software . Big Data Processor ( Tischer et al . , 2019a ) , a Fiji plugin for lazy loading of big image data , was used to visualize the images in 2D slicing mode , crop stacks in x , y , z , and t , bin images ( 3 × 3×1 in x , y , z ) , perform chromatic shift correction between channels and convert . h5 files from the InVi SPIM into an Imaris compatible multi-resolution file format ( . ims ) for further analysis ( Figure 3—figure supplement 3 ) . The oncogenic cells ( H2B-GFP channel ) displayed heterogeneous morphologies as well as varying intensity textures , making it difficult to segment them using conventional thresholding approaches . We thus used a trainable segmentation approach to convert the raw intensity values into pixel probability maps , using the Fiji plugin CATS ( Tischer and Pepperkok , 2019b ) ( Context Aware Trainable Segmentation ) . Using the H2B-GFP channel images as input , we trained three pixel classes: background , nucleus center and nucleus boundary . For training we drew about 20 ( background ) , 120 ( nucleus center ) , 100 ( nucleus boundary ) labels distributed across the different time-frames of the movie . After feature computation and training of a Random Forest classifier the whole dataset was processed on EMBL’s high performance computer cluster . The segmentation of one dataset -typically 100 timepoints- is distributed across few hundred jobs , each job using 32 GB RAM , 16 cores , and running for about 30 min . The nucleus center probability maps were then exported from CATS and added as an additional channel to the converted intensity data ( Figure 3—figure supplement 3 ) . The data were then loaded into Imaris ( Bitplane AG , 2020 , Software available at http://bitplane . com ) for 3D visualization and further processing . Using the Imaris’ Surfaces function , we segmented the nucleus center probability maps into objects . To do so , probability maps were manually thresholded , using a surface smoothening parameter of 0 . 3 µm; the minimum quality parameter for seed points was set to 0 . 1 , and object splitting was applied for objects larger than 5 . 5 µm . Objects with volumes less than 20 µm3 were excluded . Next , all objects were tracked over time using Imaris’ Lineage tracking algorithm with a maximum distance between objects in subsequent time points limited to 10 µm and a maximum gap size between identification of the object in a particular track limited to 10 time points . Analysis of segmentation results is shown in Figure 4—figure supplement 1 . Most of the errors in the object segmentation were false merges , where two cells were segmented as one . This kind of error is frequently not sustained in the previous or following time points and the maximum gap size parameter of the tracking algorithm thus frequently provides correct tracks nonetheless . The resulting lineage trees of proliferating tumor cells within the acinus were corrected manually within Imaris , for example , excluding apoptotic cells and auto-fluorescent debris . Center of mass coordinates of each cell were measured and exported from Imaris for subsequent feature analysis ( Figure 4b ) . Observations suggest that tumors in acini originate from clusters of oncogene-expressing cells produced by independent transduction events . To identify these clusters , we computed the pairwise Euclidean distances between all oncogene-expressing cells in an acinus at the start of the experiment and applied hierarchical clustering with complete linkage . Clusters were identified automatically by cutting the branches of the trees using the dynamic tree cut algorithm ( Langfelder et al . , 2008 ) . This defined a cluster as a group of oncogene-expressing cells that are closer to each other than to other oncogene-expressing cells of the same acinus . Note that a cluster can be composed of a single cell if this cell is comparatively isolated from other transduced cells . For each cluster we identified the following features as possibly linked to tumor formation: ( 1 ) number of cells in the acinus ( 2 ) cell density expressed as the ratio of number of cells to acinus surface area computed by assuming the acinus is a sphere with diameter equal to the distance between the two most distant cells ( 3 ) number of oncogene-expressing cells in the acinus ( 4 ) number of cells ( including both oncogene-expressing and normal cells ) in the cluster volume defined as the sphere centered at the center of mass of the cluster with diameter equal to the distance between the two farthest oncogene-expressing cells of the cluster ( 5 ) number of oncogene-expressing cells in the cluster ( 6 ) average pairwise distance between all cells in the cluster volume ( 7 ) average pairwise distance between oncogene-expressing cells in the cluster ( 8 ) fraction of oncogene-expressing cells in the cluster volume ( 9 ) number of contacts between oncogene-expressing cells in the cluster . Two cells are presumed in contact if they are less than the average cell diameter + 2 standard deviation apart . Oncogene-expressing cells were tracked over time and a cluster was associated with a tumor outcome if any of its cells lead to tumor formation . Here , ‘tumor formation’ is defined as the phenotypic observation of increased proliferation rate and disrupted polarity observed in transduced cell clusters upon induction with doxycycline . Regions displaying ‘tumor formation’ within the acinus often form multi-layer cell clusters with fast-dividing cells and increased apoptosis . To identify which features were linked to this outcome , we took an information-theoretic approach to model selection . We fitted a logistic regression model for all possible linear combinations of features and selected the best model based on the Akaike information criterion ( with correction for small sample sizes ) ( Calcagno and Mazancourt , 2010 ) . This model included the following features: number of oncogene-expressing cells in the cluster , number of oncogene-expressing cells in the acinus , number of cells in the cluster and number of contacts between oncogene-expressing cells in the cluster . Of these features , only the number of oncogene-expressing cells in the cluster contributed significantly to tumor formation with an odds ratio of 8 . 96 ( Figure 4b ) . Of note , all the models within 3 information criterion units of the best model also included the number of oncogene-expressing cells in the cluster as a significant contributor to tumorigenesis . Relative importance of different features can be estimated by the sum of the information criterion values of the models in which each feature appears ( Buckland et al . , 1997 ) . Computing relative variable importance across all models also indicated this feature as the most important . To control for acinus of origin , we also evaluated a mixed effect model including the three most important fixed effect features ( number of transduced cells in a cluster , number of cells in a cluster and number of contacts between transduced cells in a cluster ) , a random effect for acinus and interaction terms for the starting number of normal cells . In this case again , the number of oncogene-expressing cells in the cluster appeared as the main contributing factor to tumorigenesis with an odds ratio of 6 . 73 ( Figure 4—figure supplement 2b ) . | There are now drugs to treat many types of cancer , but questions still remain around how these diseases start in the first place . Researchers think that tumor growth begins when a single cell suffers damage to certain sites in its DNA that eventually cause it to divide uncontrollably . That damaged cell , and its descendants , go on to form a lump , or tumor . The trouble with proving this theory is that it is hard to watch it happening in real time . Doctors usually only meet people with cancer when their tumors start to cause health problems . By this point , the tumors contain millions of cells . A way to watch the very beginnings of a cancer could reveal risk factors within a tissue that foster the growth of a tumor . But first , researchers need to test their theory about how the disease begins in the first place . One way to do this is to surround a single cancer cell with healthy cells and watch what happens next . To do this , Alladin , Chaible et al . took healthy cells from the breast tissue of mice and grew them in the laboratory into mini-organs called organoids . These organoids share a lot of features with actual mouse breast tissue; they can even make milk if given the right hormones . Once the organoids were ready , Alladin , Chaible et al then started modifying a small number of single cells inside them by switching on genes called oncogenes , which are known to drive cancer formation in humans . Using fluorescent proteins and a sheet of laser light it was possible to watch what happened to the cells over time . This revealed that , even though all the oncogene-driven single cells received the same signals , not all of them started to divide uncontrollably . In fact , a single modified cell had a low chance of forming a tumor on its own . The more oncogene-driven cells there were near to each other , the more likely they were to form tumors . Alladin , Chaible et al . think that this is because the healthy tissue interacts with the modified , oncogene-driven cells to suppress tumor formation . It is only when a larger number of modified cells group together and start to communicate with each other that they can override the inhibitory messages of the healthy tissue . How healthy tissue stops single modified cells from forming tumors is not yet clear . But , with this new mini-organ system , researchers now have the tools to investigate . In the future , this could lead to new strategies to stop cancer before it has a chance to get started . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"tools",
"and",
"resources",
"cancer",
"biology"
] | 2020 | Tracking cells in epithelial acini by light sheet microscopy reveals proximity effects in breast cancer initiation |
The dynamics of predator-prey pursuit appears complex , making the development of a framework explaining predator and prey strategies problematic . We develop a model for terrestrial , cursorial predators to examine how animal mass modulates predator and prey trajectories and affects best strategies for both parties . We incorporated the maximum speed-mass relationship with an explanation of why larger animals should have greater turn radii; the forces needed to turn scale linearly with mass whereas the maximum forces an animal can exert scale to a 2/3 power law . This clarifies why in a meta-analysis , we found a preponderance of predator/prey mass ratios that minimized the turn radii of predators compared to their prey . It also explained why acceleration data from wild cheetahs pursuing different prey showed different cornering behaviour with prey type . The outcome of predator prey pursuits thus depends critically on mass effects and the ability of animals to time turns precisely .
Animals that feed on mobile prey have evolved a repertoire of anatomical features and behavioural strategies to maximize capture success ( Lima , 2002 ) , matched by prey in an evolutionary arms race in their attempts to avoid capture ( Randall et al . , 1995; Walker et al . , 2005; Cortez , 2011 ) . Numerous authors have examined strategies for both predators and prey in such interactions ( Weihs and Webb , 1984; Domenici et al . , 2011 ) , with a particular wealth of literature on animals operating in air ( Warrick , 1998; Hedenström and Rosén , 2001 ) and water ( Domenici , 2001 ) with considerations ranging from turn capacity ( Fish , 1999 ) , through random turning behaviour by prey ( Jones et al . , 2011; Combes et al . , 2012 ) to speed accuracy tradeoffs during decision making ( Chittka et al . , 2009 ) . Such considerations appear less well developed in terrestrial , cursorial interactions with , in particular , no discussion of how varying mass between parties might affect strategy , although this has been examined in the aquatic literature ( Domenici , 2001 ) . Elliot et al . , ( 1977 ) provided a conceptual model for prey acquisition by terrestrial carnivores describing four major elements; the search , the stalk , the attack , and the subdue . Of these , the attack is the most power-demanding ( Williams et al . , 2014 ) , typically involving complex high speed manoeuvres ( Van Damme and Van Dooren , 1999; Kane and Zamani , 2014 ) , underpinned by apparently complicated behavioural options for both predators and prey ( Estes and Goddard , 1967; Wilson et al . , 2013b ) . In fact , operating in a planar terrestrial environment , options for both parties are restricted , being based primarily on the choice of speed ( Elliot et al . , 1977 ) and/or trajectory ( Howland , 1974 ) , with actual performance in these being determined by physical constraints that determine maximum attainable speed and limits on turn radius as a function of velocity . Maximum speed in terrestrial running animals tends to increase to an asymptote with body mass ( Garland , 1983 ) and this process is driven by increases in leg length with body size , which facilitates higher speeds ( Garland , 1983; Bejan and Marden , 2006 ) , being ultimately limited by the scaling of body mass relative to leg strength ( Alexander , 2002b ) . Turn radius as a function of velocity in terrestrial animals is modulated by a number of critical elements , notably the extent to which leg strength can support the forces required to generate the centripetal acceleration for the turn ( Greene and McMahon , 1979; Greene , 1985; Alexander , 2002b; Tan and Wilson , 2011 ) . In addition , important elements influencing turning mechanics and maximum turn speed are; the interplay of limb force limits ( Chang and Kram , 2007 ) and friction limits ( Usherwood and Wilson , 2005 , 2006 ) , and morphological factors ( Walter , 2003; Jindrich et al . , 2007; Jindrich and Qiao , 2009 ) . Across these studies on terrestrial animals , however , the most consistent factor limiting speed during turns is force limits , with the total force demands relating to the combined effects of supporting the body mass and providing the necessary acceleration . This means that there must be a broad scaling trend in turning performance with body mass . To our knowledge though , although such scaling trends have been considered for animals operating in water and air ( Domenici , 2001; Hedenström and Rosén , 2001 ) , this has not been explicitly examined in the literature for terrestrial animals . This trend is due to the mismatch between the linear relationship of increased force demands with body size ( proportional to Mass1 . 0 ) , and the non-linear scaling of leg strength ( roughly proportional to Mass0 . 66 ( Schmidt-Nielsen , 1984 ) ( but see Biewener , 1989 ) , leading to a relative decrease in strength , and therefore turn capacity , with increasing size . The seminal work on turn performance ( Greene and McMahon , 1979; Greene , 1985; Usherwood and Wilson , 2005 , 2006; Tan and Wilson , 2011 ) has , to date , essentially concentrated on individual species , where variation with mass is of little consequence except in an absolute sense . However , where species interact , we would expect this mismatch to have profound implications , and perhaps nowhere more than in predator-prey interactions during pursuit . This work investigates the implications of mass in modulating options for terrestrial mammalian predators during the attack phase of attempts at prey capture . We limit ourselves to terrestrial predation because mass effects have already been examined for organisms in fluid media ( Domenici , 2001; Van den Hout et al . , 2010 ) and because dealing with animals that operate in a 2-dimensional surface obviates many of the complexities associated with the energetics of changing height in aerial animals ( Weihs and Webb , 1984 ) . In order to tease out expected trends , we use a simple model that isolates only the features of mass relevant to force demands , and mass-linked capacity for force production . Thus , we assume , among other things , that both predators and prey are geometrically similar ( which ignores compensating mechanisms such as upright posture and more robust limbs of some large animals as well as variation in traction ) and that the chase environment is flat , open and homogeneous ( Howland , 1974; Wilson et al . , 2011 ) . We propose that the attack phase of pursuit predators is essentially what has been treated within theory encompassing the ‘homicidal chauffeur’ game ( Marec and Van Nhan , 1977 ) , where a car driver attempts to hit a pedestrian in a defined open space . In this , we divide trajectories of both parties into straight line and turn sections and , by adopting a game theory-based approach ( Dugatkin and Reeve , 1998; Lima , 2002 ) , we distil out simple objectives for both parties; predators should attempt to minimize their distance to the prey , while prey attempt to maximize this distance ( Weihs and Webb , 1984 ) . To date , the most notable attempt to define the predator-prey chase scenario is the work by Howland ( 1974 ) , who defined many of the rules and consequences for a single turn gambit and we use this work as a starting point . However , a large body of theory also exists ( notably Wei et al . , 2009 and references therein ) , particularly that dealing with the best strategy for missiles to engage with their targets to maximize strike probability ( Shneydor , 1998; Siouris , 2004 ) and this is also considered within our terrestrial pursuit scenario . Specifically , our work has four elements . Firstly , for predator and prey mass equivalences , we examine how the initiation of a turn by a prey being pursued by a predator affects the change in predator-prey distance according to speed and timing , before extending the single turn scenario up to multiple turns . Secondly , we consider how differential masses between predators and prey affect the outcome of single turn manoeuvres . Third , we then use acceleration data acquired from tags deployed on free-living cheetahs pursuing prey of varying masses ( Hayward et al . , 2006 ) to consider whether our model predictions are broadly manifest in the wild . Finally , we compile data on the masses of mammalian predators and their prey to examine whether our model explains general patterns in prey size selection ( Carbone et al . , 2007 ) . Our approach reveals that the dynamics of movements by predators and prey of varying body masses can be treated within a single framework where the classifications and likely outcomes of pursuits , as well as the relative sizes of predators and prey , seem largely dependent on simple physical rules .
Our model for deriving the characteristics of a defined turn as a function of speed in equivalent-sized predators and prey predicted that single turns initiated by the prey lead to one of primarily two phenomena: Either the predator cuts the corner , reducing the predator-prey distance ( Figure 1A ) benefitting the predator , or it overshoots the corner , increasing the predator-prey distance ( Figure 1B ) benefitting the prey . A third scenario might be where the predator follows the prey trajectory precisely , in which case there is no change in benefit to either party although in this circumstance , the predator , with its higher speed , will eventually converge on the prey ( Figure 1C ) . Although a single turn may not lead to capture or escape of the prey , multiple turns with consistent undershooting by the predator can do so ( Figure 1 ) . For any given turn , the change in predator-prey distance over time was predicted to be critically dependent on; ( i ) the distance between the parties at the moment of the turn ( Figure 2A ) , with shorter distances ( for distances >0 ) at the moment of the turn leading to greater overshoot by the predator , ( ii ) the difference in speed between parties ( Figure 2B ) , with greater speed differences leading to greater overshoot by the predator , and ( iii ) the reaction time of the predator ( Figure 2C ) , with slower reaction times leading to greater overshoot . 10 . 7554/eLife . 06487 . 003Figure 1 . Two sequential prey turns during predator-prey pursuits showing the trajectories ( left-hand figures ) adopted by both prey ( blue lines ) and predator ( red lines ) during instances of turns that are ( A ) too early or ( B ) correctly timed by the prey , leading to corner-cutting or overshooting , respectively , by the predator . ( C ) shows the scenario where the predator and prey adopt identical trajectories . The right hand figures show how the distance between the predator and prey varies with time for the shown trajectories . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 00310 . 7554/eLife . 06487 . 004Figure 2 . Predator-prey distance during pursuit as a function of time during an execution of an unexpected 90° turn by the prey followed by straight-line travel , for; ( A ) different predator-prey distances at the moment of the execution of the turn by the prey ( indicated by the dashed vertical line ) [reaction time of the predator = 0 . 3 s , predator and prey speeds 20 and 15 m/s , respectively] , ( B ) different predator speeds ( black line shows a speed of 18 m/s and increasingly pale lines show speeds of 19 and 20 m/s respectively ) [reaction time of predator = 0 . 3 s ] ( C ) different reaction times by the predator ( black line shows a reaction time of 0 . 5 s and increasingly pale lines show reactions times of 1 . 0 , 1 . 5 and 2 . 0 s , respectively ) [predator and prey speeds = 20 and 15 m/s , respectively] . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 004 Where predators and prey have different masses , the model predicted mass-dependent distances travelled in a given turn , with larger animals having to run farther during cornering ( Figure 3 ) . Thus , during a single 90° turn , a 250 kg predator ( an example of which may be a lion or tiger ) is predicted to have to run farther than all considered prey ( potential prey ranging between a 3 kg and a 200 kg ) , whereas a 30 kg predator ( e . g . , a cheetah or a wolf ) is predicted to travel shorter distances during a turn than 3 of the 5 prey masses considered ( Figure 3 ) . Thus , cornering is predicted to be more advantageous as an escape manoeuvre as the prey size decreases relative to that of the predator and we would expect to see evidence of that in wild animal data . 10 . 7554/eLife . 06487 . 005Figure 3 . Predicted difference in distance travelled during a 90° turn by two different predators , one of mass 30 kg ( e . g . , a cheetah or wolf—continuous grey lines ) and one of mass 250 kg ( e . g . , a lion or tiger—black dashed lines ) compared to that travelled by prey of various masses ( indicated by different symbols ) as a function of running speed . Prey masses might correspond to for example , 3 kg—a hare , 10 kg—a steenbok , 40 kg—a springbok , 100 kg—a white-tailed deer , 200 kg—a hartebeest . Positive values show a greater distance run by the predator , negative values show greater distance run by the prey . Note that not all speeds reach 15 m/s due to the smallest prey not being predicted to reach this maximum ( see text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 005 This prediction could be examined in the data derived from the fieldwork on free-living cheetahs , where we observed 36 pursuits involving 7 prey species ( Table 1 ) , of which 33 had corresponding acceleration data for 5 species totalling 899 s . 10 . 7554/eLife . 06487 . 006Table 1 . Summary of the characteristics of cheetah-prey pursuits with prey nominally ranked in order of mass ( top smallest to bottom largest ) DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 006PreyChase parametersNo . huntsSuccessTotal duration ( s ) No . of turnsTurn rate ( Hz ) N%MeanSDMeanSDMeanSDHare210010 . 8–50 . 50 . 26Steenbok195328 . 711 . 95 . 472 . 970 . 320 . 19Duiker1100––––––Springbok75728 . 376 . 173 . 060 . 320 . 17Ostrich1100––––––Wildebeest250353 . 26 . 50 . 710 . 220 . 09Gemsbok47518 . 610 . 62 . 331 . 530 . 150 . 05Two species were pursued where no corresponding acceleration data were available . Identifying cornering behaviour via lateral g-forces , we documented a total of 547 turns within all chases from all animals . In these , there was a significant interaction between prey species and turn number ( turns were sequentially numbered within each chase ) on the rate of turn: specifically , different prey species had different turn rates as the turn numbers progressed ( Figure 4 ) ( χ2 = 11 . 13 , p = 0 . 03 , df = 4 ) ( Table 1 ) . We also noted that cheetah turns became shorter as the chase progressed ( Spearman's ρ = −0 . 347 , p < 0 . 0001 , although g-values reached during turns did not change significantly over time ( ρ = 0 . 031 , p = 0 . 72 ) . Critically , these dynamics were different in successful and unsuccessful pursuits . Turn duration decreased with increasing turn number within any particular chase ( χ2 = 26 . 52 , df = 1 , p < 0 . 0001 ) and turn durations were shorter in successful pursuits ( χ2 = 6 . 24 , df = 1 , p = 0 . 013 ) . 10 . 7554/eLife . 06487 . 007Figure 4 . The duration of turns ( expressed as the length of time between the onset of the turn and the point at which maximum g was reached within the turn ) made by cheetahs pursuing different prey with time into the chase . Joined points show chases referring to individual prey . Longer turns will tend to have larger turn radii so while , overall , pursuit of larger prey is characterized by larger turn radii , turn radii diminish as the chase progresses . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 007 Extending our modelling exercise to derive maximum cornering ability as a function of mass and speed for a suite of theoretical mammalian cursorial predators and their prey indicated that , within the animal mass range considered , mass affects maximum speed ( by a factor of less than 6 ) much less than it affects minimum turning radius at maximum speed ( which affects it by a factor of up to 260 ) . This is due to the combined effect of generally increasing maximum speeds with mass ( see above ) and the additional effect of mass on turn radius ( Figure 5 ) . Importantly though , the surface describing comparative turn abilities showed that predators turn tighter relative to prey in a specific area of the surface defined in terms of predator and prey mass ratios ( Figure 5 ) . Insertion of the mean mass of 54 species of canids and felids and their ( predominantly mammalian ) prey from an extensive database ( Carbone et al . , 1999 ) into this plot , shows that all predators were located in the area where predator turn capabilities were maximized compared to their prey ( Figure 5 ) . The implication from this is that there is strong selection pressure for turning ability in predators and that ( i ) predators evolve to take particular sized prey with a mass that results in the minimum turn radius ratio most in favour of the predator , and/or ( ii ) prey sizes that have minimum turn radii that most closely accord with those of the predators tend to be caught more often than other-sized prey . 10 . 7554/eLife . 06487 . 008Figure 5 . Relationship between mass and performance for predators and prey derived from the model . Performance is expressed as maximum speed ratio ( Vmax pred/Vmax prey; higher values indicate relatively faster predators ) and minimum turn radius ( Tmin prey/Tmin pred; higher values indicate relatively tighter turning predators ) . The grey spheres indicate the range of possible options computed systematically , the blue spheres represent single predator species computed against their mean prey masses , while the cyan spheres indicate the specific case of the cheetah , taking variously sized prey items across the size range . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 008
The outcome of a predator-prey pursuit depends on performance , both in terms of speed and cornering capacity ( Wilson et al . , 2013a , 2013b ) , and how these relate to power requirements and timing . Our work echoes that of Howland ( 1974 ) , which dealt with the same issue using a different approach , emphasising that high speed , aside from being necessary for the predator to gain on the prey , also increases the likelihood that the predator will overshoot the prey turn trajectory ( Figure 1B ) ( Howland , 1974; Alexander , 2003 ) . Conversely , as speed decreases , the difference in distance covered during any turn between predator and prey decreases ( Figures 1–3 ) , fulfilling the fundamental game rule for the predator . The optimum strategy for a pursuit predator should , therefore , be to attempt to elicit turns by the prey which , if the prey are working close to maximum power , will result in a reduction in their speed ( Shubkina et al . , 2012 ) , because energy is needed for the turn ( Wilson et al . , 2013c ) . Invoking multiple rapid turns might therefore be a strategy that predators seek to promote . Here , decreasing turn duration over time is expected during successful chases , as we observed ( Figure 4 ) , coupled with generally decreasing speed , as reported by Wilson et al . ( Wilson et al . , 2013b ) . However , turning is also critical for survival of prey because maintenance of a straight line trajectory leads to inevitable capture if the predator is faster . This seemingly contradictory situation of whether the predators or prey benefit most from prey turns is clarified by timing . When the predator is far from the prey , the prey should not turn since to do so allows the predator to cut the corner of the prey's trajectory and decrease the distance between itself and the prey more rapidly ( Eilam , 2005 ) ( Figure 1 ) . However , turns by the prey do benefit the prey if the timing is correct because this leads to an overshoot by the predator ( Figure 1 , Figure 2 ) . Where such overshoot turns are consistent , they should generally lead to rarefying turn rates with reduced chances of capture ( Wilson et al . , 2013b ) . The outcome of extended pursuits is also likely influenced by endurance . Both parties will have power limitations restricting their options on instantaneous performance , as well as ultimately limiting how long the chase can continue before exhaustion . A multiple turn terrestrial chase has energetic costs associated with straight line travel that are a ( linear ) function of speed ( Taylor and Heglund , 1982 ) so , all other things being equal , at such times the predator should be expending energy faster if it is to gain on the prey . However , there are substantial costs to turning above those of straight line travel ( Wilson et al . , 2013c ) so the angular extent of the turn and the time spent turning will both affect the rate of energy expenditure ( Wilson et al . , 2013c ) . Thus , where the predator cuts the corner compared to the trajectory taken by the prey ( Figure 1 ) , the reduced turn costs should act to reduce its overall rate of energy expenditure , making the predator energy expenditure closer to that of the prey . This will tend to lead to similarity in power use between parties resulting in similar giving-up times , assuming both parties can allocate similar amounts of energy to a pursuit ( Figure 6 ) . However , where a predator overshoots the cornering trajectory of the prey ( Figure 1 ) , it has to contend with the increased energetic demands of travelling farther , and with a greater turn angle , than the prey ( Figure 2 ) . Thus , where overshooting occurs consistently , it will tend to make cumulative energy expenditure between the two parties more disparate resulting in the predator reaching endurance limits earlier than the prey ( Figure 6 ) . In reality , multiple turn pursuits , such as we observed in our cheetah-prey interactions , will consist of a both corner-cutting and overshooting by the predators with the proportion of either perhaps biasing the likelihood of prey capture or the chase being abandoned . 10 . 7554/eLife . 06487 . 009Figure 6 . Schematic figure to show how power use is expected to vary during the course of a predator-prey chase consisting of four straight-line trajectories interspaced with three turns ( turns made by the prey are shown between arrows ) by both parties . The prey ( dashed line ) has lower energy expenditure than the predator during straight-line sections because it is travelling slower . However , because extra power is required for a turn ( Wilson et al . , 2013c ) , a predator that consistently cuts the corner ( dark grey line—‘predator with undershoot’—cf . Figure 1 ) spends less time cornering , expending less energy for the corner , and may maintain energy expenditure at levels similar to those of the prey despite travelling faster: Here both parties may reach limits to endurance performance at a similar time . However , a predator that consistently overshoots the corner ( light grey line—‘predator with overshoot’—cf . Figure 1 ) spends longer turning , expending markedly more energy than the prey at all times , reaching endurance limits earlier . ( In this depiction , predators and prey are assumed to have the same geometry , performances and masses ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 009 Our data and theoretical considerations based on the literature highlight the extent to which mass drives physical abilities in predator-prey pursuits . The fact that maximum speed generally increases with mass ( Garland , 1983 ) has been variously linked to factors such as absolute and relative leg length , stride length and stride frequency ( cf . Schmidt-Nielsen , 1984; Biewener , 1989 , 2003; Alexander , 2002a ) . This is complicated by the degree of geometric similarity between animals and models for elastic similarity ( cf . Garland , 1983 and refs therein ) , all factors which might also help explain the modest reverse trend in the maximum speed vs mass relationship for animals exceeding about 70 kg ( Garland , 1983 ) . That greater mass leads to greater turn radii , because the forces that animals can exert for a turn relate to a mass exponent of less than one , whereas the forces required for a turn scale to a mass exponent of one has not , to our knowledge , been previously discussed . Both these mass-dependent attributes lead to a tendency for larger animals to be faster , but less able to turn than smaller animals , which presumably has profound consequences for strategies adopted by predators and prey during pursuits . Mass-linked performance explains why , for example , smaller cheetah prey confer a greater size-derived speed advantage to the predator , which should lead to a more rapid closure of the cheetah-prey distance during straight line sections of the pursuit and result in a predicted , and observed , higher turning frequency in cheetah pursuing smaller prey ( Figure 4 , Table 1 ) . Similarly , our mass-dependent model of performance points to how predator size in relation to that of the prey results in differential distances travelled by both parties during turns . So the distance run during any given turn by a predator with similar mass to its prey should not be markedly greater than its prey , while a larger predator chasing the same prey must contend with covering a substantial increase in the distance travelled ( Figure 3 ) . However , the scaling of turning radius with speed is complex because either party may elect to travel more slowly to produce a tighter turning radius , which also reduces the chances of mistakes ( Chittka et al . , 2009 ) , although the predator must always presumably travel faster than the prey . Certainly , workers have suggested ( Wilson et al . , 2013a ) , and found ( Wilson et al . , 2013b ) , that cheetahs tend to reduce speed in the final stages of their chases ( cf . Shubkina et al . , 2012 ) . Thus , varying masses between predators and prey changes the nature of pursuits substantially . Pursuit predators up to body masses of about 70 kg would seem to need to be larger than their prey in order to be able to catch them because , up to this size , larger animals can run faster ( Garland , 1983; Bejan and Marden , 2006 ) . But larger-bodied predators must modulate speed carefully to compensate for their turn radius disadvantages because , although they can outrun , they are less likely to out-manoeuvre their prey ( Figures 3 , 4 ) . Therefore , chases are expected to be defined by more frequent turns with increasing mass differential between parties , which is what we observed in our cheetah data ( Table 1 ) . Where predators pursue prey larger than themselves , however , and assuming that the predator can travel faster than their prey , there is no advantage to be gained by the prey executing the sudden turns characteristic of small prey ( cf . Figure 4 ) . This explains the straight-line trajectories of for example , gemsbok Oryx gazelle being pursued by spotted hyaenas Crocuta crocuta ( Mills , 1990 ) . The primary deterrent to smaller predators hunting larger animals may come from the danger of injury ( Mills , 1990 ) , or the predator not having the strength to overcome its prey , something that can be mitigated to an extent by co-operative hunting or following the prey to exhaustion ( Estes and Goddard , 1967 ) . Our treatise assumes that both predators and prey interact on a homogeneous , flat surface but we expect any variation in the topography and vegetation to modulate the tactics adopted by both predators and prey . Shepard et al . , ( 2013 ) note how substrate type affects the costs of animals moving over it ( hence their term ‘energy landscape’ ) , which is predicted to affect route choice in a general sense . We expect high power pursuits of the type discussed here to be subject to the same rules , with changes in the energy landscape that differentially affect predators and their prey to be exploited by the relevant party . Of particular note is the work by Taylor et al . , ( 1972 ) , who noted that larger animals incur a proportionately greater increase in metabolic rate for movement up inclines , and this has been shown to affect area use in some species ( Wall et al . , 2006 ) . Correspondingly , we would expect smaller prey to favour selection of uphill gradients during pursuits . Similarly , smaller prey are expected to ‘run for cover’ ( Domenici et al . , 2011 ) , partly because such cover may represent an impossibly high-cost energy landscape for the predators . In addition , vegetation in patches , such as bushes or trees , may constrain turn radii , precluding small prey from perhaps turning as tightly as they might , to the advantage of the predator while , conversely , such features may allow prey to execute a turn without giving the predator the option of cutting the corner ( Figure 7 ) . 10 . 7554/eLife . 06487 . 010Figure 7 . Pursuit of a steenbok which has turned as it ran past a bush , constraining the cheetah to follow almost the same trajectory . The use of environmental features such as this makes the timing of the turn less critical since the cheetah cannot cut the corner substantially , even if the prey turns too early . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 010 Within the general context of predator-prey pursuits , some authors have noted that unpredictable ( protean ) movement by prey can enhance their chances of escape ( Jones et al . , 2011 ) . The theory is that unpredictable movement may catch the predator by surprise ( Humphries and Driver , 1970 ) , not least because , in our case of cursorial predator-prey interactions , the prey may execute a turn when the predator-prey distance is not yet critical . Clearly , if this occurs in planar environments , choices for changing trajectory can only amount to movement that is either left or right , and should not be executed too early where there are substantial corner-cutting benefits to be gained by the predator . However , since the timing of corners would appear so critical ( see above ) , we would expect selection pressure for random turns just before the critical phase . Future work would do well to consider this . Finally , predator-prey pursuit options by terrestrial animals will undoubtedly be altered when the condition changes from single to multiple predators . Here , observed manoeuvres are complex and include fanning out of predators ( Kelley , 1973; Mills , 1990 ) which should allow larger predators to mitigate for the effects of their reduced turning radii with respect to those of their prey . Indeed , understanding the precise advantages of group hunting as a function of group size , predator and prey masses , and the distribution of all parties in space in relation to environmental variability promises to be a major challenge .
There are a number of different strategies recognised within ‘pursuit’ scenarios ( e . g . , athletes chasing balls , children’s games of ‘tag’ , missiles , and insects involved in territorial disputes ) including , primarily ‘pure’ pursuit , and more ‘predictive’ strategies such as ‘constant bearing’ pursuit , and ‘constant absolute target direction’ ( CATD ) pursuit ( for definitions see Shneydor , 1998; Wei et al . , 2009; Nahin , 2012 ) . Constant bearing is only applicable if the speeds of both parties are constant ( Reddy , 2007 ) , which is generally inappropriate for predator-prey terrestrial pursuits ( Wilson et al . , 2013b ) although its use has been demonstrated in some predatory insects ( Olberg et al . , 2000 ) , fish ( Lanchester and Mark , 1975 ) and humans ( McBeath et al . , 1995 ) , while CATD has been shown in some birds , as has ‘pure pursuit’ ( Kane et al . , 2015 ) . We expect , however , pursuit strategies for flying animals ( and some latter-day missiles ) to differ from those of terrestrial animals . This is because initial engagement between two aerial parties invariably involves both moving prey and the pursuer having an approach that tends to occur at a tangent to the target ( Hedenström and Rosén , 2001; Ghose et al . , 2006 ) , rather than from directly behind . Part of the reason for this tangent relates to both predators and prey operating in 3-d , rather than 2-d , space so that the simple probability of a line of attack being directly along and behind the prey flight path at first encounter is correspondingly reduced . Tangential approach provides various options for pursuers to calculate trajectories and compute intersection points ( Olberg et al . , 2000 ) that minimize time and/or distance ( Ghose et al . , 2006 ) . It also helps mitigate manoeuvring problems likely to occur during pure pursuit as the inter-party distance approaches zero ( particularly where pursuer speed is much greater than that of the prey ) ( Siouris , 2004 ) . The development of the chase for a cursorial predator operating in the planer environment of the ground , however , invariably starts with a stalking phase on essentially non-moving prey followed by the rush ( Estes and Goddard , 1967; Elliot et al . , 1977; Mills , 1990; Williams et al . , 2014 ) . This situation is very different from some modelled interactions where both predators and prey have constant speed ( e . g . , Weihs and Webb , 1984 ) . Two options would seem appropriate for predator-prey encounters that start prey at , or close to , zero speed in planer environments faced with a predator moving rapidly towards them . One , which is more appropriate where the predator speed is markedly higher than prey and the distance between them small , involves prey escaping in a trajectory that is perpendicular to that of the predator ( Weihs and Webb , 1984; Stankowich and Coss , 2007; Domenici et al . , 2011 ) . In part , this ‘side step’ strategy capitalizes on the speed-linked smaller turn radius of the prey compared to the predator . Otherwise , if prey react to a distant , but approaching , predator by generally moving away , the speeds of both parties will tend to become better matched over time . Here , the best strategy is for prey to continue to move directly away because any other direction will reduce the inter-party distance correspondingly ( cf . Domenici et al . , 2011 ) . This explains observations reporting that prey typically do indeed initially run directly away from the predators ( cf . Finney et al . , 1997; Broom and Ruxton , 2005; Eilam , 2005 ) although this is not always the case ( see review in Domenici et al . , 2011 ) . This manoeuvre puts the pursuer directly behind the prey . During the ensuing pursuit phase , straight line pursuit by the predator towards the prey is the only viable predator option when the prey travels in a straight line ( ignoring topographic complexity —see ‘Discussion’ ) , which is also what has been observed ( Shubkina et al . , 2012 and refs therein ) . As the predator-prey distance tends towards zero , the prey must turn to avoid capture , and may even have particular strategies such as stotting to enhance this ( Stankowich and Coss , 2007 ) . We demonstrate later that there is considerable selection pressure on the prey to time this turn so that it occurs close to the contact point ( but not so close that the predator catches it ) . This scenario reduces differences , in both trajectories and time to contact , between the different pursuit strategies ( see above ) although any predictive pursuit options typically have a reduced trajectory length and time to contact than pure pursuit ( see below ) . Differences between the pursuit strategies are also likely reduced by prey reaction to the approaching predator based on the prey having properly informed assessment of predator trajectory . In this regard , typical terrestrial ‘prey’ , such as ungulates , have large panoramic visual fields with laterally facing orbital margins ( Heesy , 2004 ) and so may be able to assess the movements of their pursuers as well as their pursuers can assess them . Indeed , escape trajectories may be modulated to keep predators within the prey's visual ( or other sensory ) fields ( Domenici et al . , 2011 ) . In any event , the substantive differences between terrestrial and fluid media interactions may partly account for the reason that Shneydor ( 1998 ) considers that many terrestrial cursorial predators adopt pure pursuit ( hence the name ‘hound-hare pursuit’ ) , something that is backed up by observation ( Schaller , 2009; Shubkina et al . , 2010 , 2012 ) . Importantly though , quantitative differences in trajectories and pursuit times between pure- and predictive pursuit strategies do not change qualitative patterns . Thus , the essential message of the work , which considers mass effects , does not depend on our choice of pursuit strategy . In developing a model that highlights the effect of mass on turn performance , we make a number of critical assumptions . These are;A . That predator-prey interactions occur in a flat , homogenous area , otherwise lacking physical structure ( Marec and Van Nhan , 1977; Melikyan and Bernhard , 2005 ) and with no differential power requirements according to trajectory direction ( Shepard et al . , 2013 ) . B . That only the pursuit phase of a hunting predator is considered ( Elliot et al . , 1977 ) . C . That distance between predator and prey is chosen as the primary focus of the model , with prey maximizing instantaneous distance ( Weihs and Webb , 1984 ) . D . That both predator and prey are solitary . E . That both predator and prey are geometrically similar . F . That prey operate close to maximizing energy input to locomotion during the chase although the predators are not required to do so . We represent the pursuit of prey by plotting trajectories based on vectors representing a predator moving directly towards the prey at a speed greater than that of the prey . We then modify trajectories according to the model ( see below ) . We define the pursuit as consisting of two fundamental stages; stage 1 , which involves straight line travel by the predator towards the prey and , if the prey is not caught immediately , straight line travel by the prey in response to this , and stage 2 , where the prey initiates turns to reduce the chances of being caught and the predator responds by turning during pursuit . We examined the perspectives from our models with respect to prey pursuit behaviour observed directly , and quantified using collar-fitted tri-axial accelerometers ( G6a , Cefas , UK —recording rate 30 Hz ) , on six free-ranging cheetahs in the Kgalagadi Transfrontier Park ( 25°46′S 20°23′E ) , southern Africa . Cheetahs , which varied in mass between 30 and 45 kg , were equipped for a total of 66 animal days during which 36 pursuits of prey were observed . The orthogonal , tri-axial acceleration data were first sorted to identify the periods of active pursuit by matching times with periods when the animals were observed to hunt . Data from the sway and heave axes were somewhat interchangeable due to a partially rotating collar , but co-varied directly and changed with lateral acceleration and thus cornering behaviour ( the surge axis varied with longitudinal acceleration or deceleration ) . Thus , the mediolateral acceleration data ( heave and sway axes ) were initially smoothed over 2 s to derive a measure of the static ( nominally gravity-based ) acceleration ( Shepard et al . , 2008 ) and were found to show obvious waveforms due to the bounding gait of the cheetahs ( Figure 10 ) . These data were thus further smoothed over 0 . 5 s to minimize the influence of these waveforms while not overly compromising smoothed values ( Figure 10 ) . Turns were identified as increasing departures of the doubly smoothed data from 1×g . The specific points of their maxima were identified by values that departed maximally from 1 . 0 . Immediately adjacent ( and lower ) peaks which occurred as a result of the bounding behaviour ( which could be readily identified in the single-smoothed data—see Figure 10 ) were precluded . 10 . 7554/eLife . 06487 . 013Figure 10 . Data showing the vectorial sum of the static heave and sway acceleration axes ( expressed as a departure from 1 ) during the pursuit phase of a cheetah hunting a steenbok . The grey lines show the values using a running mean window of 2 s , which still shows appreciable signal noise due to the bounding movement of the animal . These values have been further smoothed over 0 . 5 s to give the black line which shows the main features of the lateral acceleration during turns with maximum g-forces developed during turns displayed as peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 06487 . 013 Analyses of these data were performed in R version 3 . 0 . 2 ( R Core Team , 2013 ) . The relationship between prey species , hunt success and turn number on turn rate was determined using a general linear model ( Bates et al . , 2013 ) . Turn number was entered as a covariate , prey species and hunt success as factors and cheetah ID as a random factor to account for repeated measurements within animals . The relationship between turn number and hunt success on turn duration was determined using a similar model with turn number entered as a covariate , hunt success as a factor and cheetah ID a random factor . For both models , function ‘lmer’ was used in the package lme4 . Wald χ2 statistics and p values were obtained using the function ‘Anova’ in the package ‘car’ . Data were tested for normality and homoscedasticity of variance using Shapiro–Wilk and Levene's tests . We used the approach outlined in our model to undertake a broad-based analysis to examine how the mass of terrestrial predators and prey ( using ranges between 0 . 05 and 5000 kg ) affected the interplay of maximum speed and minimum turn radius at that speed . Predictions for maximum speed were derived from an allometric relationship between mass ( M , kg ) and speed ( V , km/h ) for mammals of ( Garland , 1983 ) ;log10Vmax=1 . 478+0 . 2589 ( log10M ) −0 . 0623 ( log10M ) 2 . We then took data from an extensive data base on the mass of mammalian predators and their mammal prey ( Carbone et al . , 1999 ) so as to place data from wild animals in this context to determine whether any trends were apparent . | A pursuit between a predator and its prey involves complex strategies . Prey often make sudden sharp turns when running to evade a predator . Any predator that cannot turn quickly enough will have to run further to catch up with the prey again , thus potentially allowing the prey to pull away from the predator . The timing of these turns is crucial; if the prey turns when the predator is too far away , the predator can cut the corner off the turn and catch up with the prey more easily . The speed at which animals can turn depends on the forces involved in cornering , and larger animals need to produce greater forces for any given turn . However , larger animals can apply relatively less force than smaller animals for turns and so cannot turn as rapidly . The effect of the relationship between mass and turning ability on the strategies used during land-based pursuits had not been investigated . Wilson et al . have now created a mathematical model that considers how the mass of a predator and its prey influences the course and strategies used in a land-based pursuit . The model is based in part on a mathematical problem called the ‘homicidal chauffeur game’ , where a car driver attempts to run over a pedestrian . Wilson et al . 's model predicts that chases between large predators and smaller prey should feature frequent sharp turns , as the prey try to exploit their superior turning ability . However , when the predators and prey are of similar size , the prey gain little or no advantage from executing high-speed turns . Indeed , as turning slows the prey down , turning may often be disadvantageous , and so fewer turns should be seen during a pursuit . The predictions of the model were compared with the pursuit strategies of wild cheetahs , which were studied using collars equipped with tags to measure acceleration as the predators chased prey of different sizes—from hares to large antelopes called gemsboks . The tracking data confirmed the predictions of the model; thereby revealing that body mass and the ability of animals to choose when best to turn strongly determine the outcome of predator-prey pursuits . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology"
] | 2015 | Mass enhances speed but diminishes turn capacity in terrestrial pursuit predators |
Cytoplasmic dynein is the predominant minus-end-directed microtubule ( MT ) motor in most eukaryotic cells . In addition to transporting vesicular cargos , dynein helps to organize MTs within MT networks such as mitotic spindles . How dynein performs such non-canonical functions is unknown . Here we demonstrate that dynein crosslinks and slides anti-parallel MTs in vitro . Surprisingly , a minimal dimeric motor lacking a tail domain and associated subunits can cause MT sliding . Single molecule imaging reveals that motors pause and frequently reverse direction when encountering an anti-parallel MT overlap , suggesting that the two motor domains can bind both MTs simultaneously . In the mitotic spindle , inward microtubule sliding by dynein counteracts outward sliding generated by kinesin-5 , and we show that a tailless , dimeric motor is sufficient to drive this activity in mammalian cells . Our results identify an unexpected mechanism for dynein-driven microtubule sliding , which differs from filament sliding mechanisms described for other motor proteins .
Cytoplasmic dynein is a 1 . 2 MDa , multisubunit microtubule motor complex that belongs to the AAA family of molecular machines ( Ogura and Wilkinson , 2001; Allan , 2011 ) . The dynein complex is composed of a dimer of two heavy chains . Each heavy chain contains a C-terminal motor domain , and an N-terminal tail domain that binds to accessory chains and adaptor proteins , which are needed to link the motor to its cargo ( Kardon and Vale , 2009 ) . Dynein transports a plethora of cargoes towards MT minus-ends , including many types of organelles , mRNAs and proteins . In addition to its well-studied role in cargo transport , cytoplasmic dynein has been implicated in the organization of the MT cytoskeleton itself , particularly during cell division . When mammalian cells enter mitosis , dynein is needed to remodel the prophase MT network ( Rusan et al . , 2002 ) , and at later stages dynein assists in organizing MTs to form focused spindle poles ( Heald et al . , 1996; Merdes et al . , 1996 ) . Dynein also generates an inward force within the spindle that counteracts an outward force generated by kinesin-5 and kinesin-12 motors ( Mitchison et al . , 2005; Tanenbaum et al . , 2008 , 2009; Ferenz et al . , 2009; Vanneste et al . , 2009 ) . The balance of these forces is important for normal spindle assembly ( Tanenbaum and Medema , 2010 ) . In cell extracts , dynein also has been shown to organize MTs into aster-like structures ( Verde et al . , 1991; Gaglio et al . , 1996 ) , drive the fusion of two closely positioned spindles into a single bipolar spindle ( Gatlin et al . , 2009 ) , and transport stabilized MT seeds to the spindle pole ( Heald et al . , 1996 ) . Collectively , these results show that dynein plays an important role in organizing the MT network during cell division . In contrast to cargo transport , which involves a well-studied walking mechanism of the two dynein motor domains along a MT ( Gennerich and Vale , 2009; DeWitt et al . , 2012; Qiu et al . , 2012 ) , the mechanism by which dynein organizes MTs in a MT network has not been established . One possibility is that dynein is anchored at fixed subcellular sites through its tail domain and moves processively along MTs through the cytoplasm . However , since the ‘cargo’ is sufficiently large in this instance , the MT themselves would move , rather than the cargo ( Wühr et al . , 2009 ) . Such a mechanism may also function during spindle positioning , where cortically-anchored dynein pulls on spindle MTs ( Galli and van den Heuvel , 2008; Kiyomitsu and Cheeseman , 2012; Kotak et al . , 2012; Laan et al . , 2012 ) , as well as during centrosome separation , where dynein is anchored to the nuclear envelope ( Splinter et al . , 2010; Bolhy et al . , 2011; Raaijmakers et al . , 2012 ) . Alternatively , dynein could physically crosslink two MTs and slide them relative to each other . To generate sliding between two MTs , an individual dynein motor , a dimer of two heavy chains , could use its two motor domains to walk along one MT and employ a second MT binding domain to transport a ‘cargo MT’ . Axonemal dyneins function in this manner to produce MT sliding in cilia and flagella . Several classes of kinesin utilize a similar mechanism for crosslinking and sliding MTs ( Straube et al . , 2006; Braun et al . , 2009; Fink et al . , 2009; Seeger and Rice , 2010; Su et al . , 2011; Weaver et al . , 2011 ) . However , there is currently no evidence suggesting that cytoplasmic dynein contains a secondary MT binding site outside of the motor domain . A second model is that dynein could crossbridge and slide MTs by forming small oligomers or co-complexes with other proteins . Kinesin-5 , which forms a bipolar tetramer of motor domains , crossbridges and slides MTs by such a mechanism ( Kashina et al . , 1996; Kapitein et al . , 2005 ) . A third model is that dynein’s two motor domains bind to and move along two separate MTs . Here using an in vitro assay , we demonstrate that both native rat , and recombinant yeast cytoplasmic dynein can drive sliding of anti-parallel MTs in the absence of additional proteins . Single molecule data of dynein in a MT overlap zone is most consistent with a model in which the two dynein motor domains walk along different MTs to generate sliding . We also show in vivo that a minimal dynein dimer , lacking expected cargo binding interactions , can substitute for native dynein in generating an inward force within the spindle that antagonizes outward MT sliding forces generated by kinesin-5 and kinesin-12 . Together , these results show that dynein can slide and organize MTs , using a sliding mechanism that differs from that described for other motor proteins .
Previous work has reported that purified brain cytoplasmic dynein crosslinks and bundles MTs ( Amos , 1989; Toba and Toyoshima , 2004 ) , but the mechanism has remained unclear . We similarly found that purified rat brain dynein induces the formation of large bundles of purified MTs ( Figure 1A–D ) . Since cytoplasmic dynein can crossbridge MTs , we next investigated whether it can slide two MTs relative to each other . To test for MT sliding , we incubated brain dynein in solution with both green- and red-labeled fluorescent MTs; the green MTs in the MT bundles were also biotinylated , allowing their immobilization onto a streptavidin-coated coverslip ( Figure 1D , E ) . Addition of ATP into the assay chamber induced the red MTs to slide within the bundles relative to the stationary green biotin-MTs ( Figure 1E; Video 1 ) . 10 . 7554/eLife . 00943 . 003Figure 1 . Dynein crosslinks and slides MTs in bundles . ( A ) Schematic overview of the dynein constructs used in this study . The N-terminal tail is shown in gray , the linker in purple , the six numbered AAA+ domains are in light blue and the stalk and MT binding domain are depicted in orange . GFP and GST tags are shown in green and blue , respectively . The Halo tag ( DHA , Promega ) is shown in red . ( B ) Coomassie brilliant blue stained gels showing purified dynein constructs used in this study . The associated subunits of the brain cytoplasmic dynein complex are labeled; HC–heavy chain , IC–intermediate chain , LIC–light intermediate chain , LC–light chain . Recombinant yeast dynein constructs do not contain associated subunits . Molecular weight markers are indicated . ( C ) MTs incubated in the absence or presence of dynein are visualized by attachment to a streptavidin-coated coverslip via a biotin tag . Brain dynein and GST-Dyn1331kDa crosslink MTs into large bundles , while the dynein monomer , Dyn1331kDa does not . Scale bar , 10 µm . ( D ) Cartoon depicting two different mechanisms by which dynein could crosslink MTs , either using its two motor domains or through the tail domain . Alexa-568 and Alexa-488 labeled MTs are crosslinked by dynein . The green MTs are attached to the coverslip through a biotin-streptavidin linkage and perfusion of 1 mM ATP induces sliding between the MTs . ( E and F ) Example of rat ( E ) and GST-Dyn1331kDa ( F ) dynein-driven sliding of red-labeled MTs within the bundle after 1 mM ATP addition . Arrowhead tracks the sliding MT within the bundle . The time relative to the start is noted in min:s at the bottom of each image . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 00310 . 7554/eLife . 00943 . 004Video 1 . Rat brain cytoplasmic dynein bundles and slides MTs . Red-labeled MTs are slid and extruded from a bundle of red- and green-labled MT bundles by rat brain cytoplasmic dynein upon addition of ATP . Total time of the video is 551 s . Playback is 30 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 004 To crosslink and slide MTs , dynein could bind one MT with its motor domains and potentially another with a non-motor MT-binding domain ( Figure 1D ) . Alternatively , the two motor domains of the dynein dimer could each bind a different MT ( Figure 1D ) . To distinguish between these possibilities , we tested a well characterized , GST-dimerized , truncated yeast dynein construct , GST-Dyn1331kDa , which contains the motor domain but lacks the non-motor tail domain and other dynein subunits ( Reck-Peterson et al . , 2006 ) ( Figure 1A , B ) . This minimal , dimeric dynein was affinity purified , followed by gel filtration to remove any potential oligomers or aggregates . GST-Dyn1331kDa crosslinked MTs , while a monomeric version lacking the GST dimerizing domain did not ( Figure 1A–C ) . Similar to rat brain dynein , the GST-Dyn1331kDa motor induced sliding of MTs within the bundles , when ATP was added ( Figure 1F; Video 2 ) . These results provide the first direct in vitro demonstration that cytoplasmic dynein can slide MTs within bundles . They also reveal that sliding does not require dynein associated subunits or any potential second MT binding domain in the dynein heavy chain tail domain . 10 . 7554/eLife . 00943 . 005Video 2 . GST-Dyn1331kDa slides MTs within bundles . Red-labeled MTs are extruded from a bundle of red- and green-labeled MTs by GST-Dyn1331kDa upon addition of ATP , indicating the dynein motor domains alone are sufficient for this activity . Total time of the video is 84 s . Playback is 5 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 005 We next sought to develop an improved MT sliding assay in which the relative movement of two MTs could be more easily observed . To achieve this , we first bound GST-Dyn1331kDa in the absence of ATP to an immobilized ‘track’ MT on the coverslip surface , and then introduced ‘transport’ MTs , which became crosslinked to the ‘track’ MT by dynein . After addition of ATP , GST-Dyn1331kDa drove robust sliding of transport MTs along the immobilized track MTs ( Figure 2A , Video 3 ) . The transport MTs frequently moved until they reached the end of the track MT , where they swiveled around an end point on the track MT ( asterisk Figure 2A ) . The speed of MT-MT sliding ( 53 ± 24 nm/s; mean ± SD; Figure 2A ) was similar to the velocity observed in MT gliding assays ( 47 ± 11 nm/s; Figure 2A ) with this dynein construct . We found that purified rat brain dynein also produced MT-MT sliding using this same assay , although the movement was slower than surface gliding ( 10 ± 5 nm/s vs 615 ± 200 nm/s ) and characterized by more frequent pausing than MT gliding by surface-bound brain dynein ( Figure 2B , Video 4 ) . The slower motility may be a result of the much weaker processivity and directionality observed for individual mammalian dyneins in vitro compared to yeast dynein ( Ross et al . , 2006; Cho et al . , 2008; Miura et al . , 2010; Ori-McKenney et al . , 2010; Trokter et al . , 2012 ) . Other proteins ( such as LIS1 and Ndel1 ) that directly modify these properties of mammalian dynein may be needed for more robust processivity and sliding , consistent with the requirement of these proteins for dynein-dependent MT organization in vivo ( Raaijmakers et al . , 2013 ) . Nonetheless , the finding that rat dynein alone is capable of crosslinking and sliding MTs in our assays , shows that this type of motility is an intrinsic ability of the motor . 10 . 7554/eLife . 00943 . 006Figure 2 . Dynein crosslinks and slides single MT overlaps . ( A ) Example of single MT-MT sliding driven by GST-Dyn1331kDa . Successive frames , separated by 52 s , from the video and corresponding kymograph show the sliding . The transport MT is captured and aligned onto the track MT by GST-Dyn1331kDa . Arrowheads follow the transport MT as it slides along the track MT . Asterisk marks where the transport MT remains attached at the end of the track MT . Right , histograms of the MT-MT sliding and surface gliding velocities driven by GST-Dyn1331kDa with Gaussian fitting . ( B ) Example of rat brain dynein driven sliding in a single MT-MT overlap . Successive frames , separated by 46 s from the video are shown with corresponding kymograph below . The transport MT pauses before reaching the end of the track MT , which was frequently observed for rat dynein-driven movement . Velocity histograms for rat dynein driven MT-MT sliding and surface gliding are shown to the right with Gaussian fitting . ( C ) Polarity-marked MTs with long , brightly labeled plus-ends were used to determine the orientation of MT-MT sliding . The plus- and minus-ends of both MTs are indicated . Arrowhead shows the red transport MT slides with its minus-end away from the plus-end of the track MT , indicating that sliding is anti-parallel . Kymograph analysis of the anti-parallel sliding event is shown to the right . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 00610 . 7554/eLife . 00943 . 007Video 3 . Rat brain cytoplasmic dynein slides single MT overlaps . A red-labeled ‘transport’ MT is slid over a green-labeled ‘track’ MT by rat brain cytoplasmic dynein . The transport MT pauses before reaching the end of the track MT . Total time of the video is 271 s . Playback is 30 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 00710 . 7554/eLife . 00943 . 008Video 4 . GST-Dyn1331kDa slides single MT overlaps . A red-labeled ‘transport’ MT is slid over a green-labeled ‘track’ MT by GST-Dyn1331kDa , indicating that the dynein motor domains alone are sufficient for this activity . The transport MT slides to the end of the track MT and swivels around a nodal attachment point . Total time of the video is 306 s . Playback is 30 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 008 To determine the orientation of MT-MT sliding driven by dynein , we repeated the assay described above using polarity-marked MTs and GST-Dyn1331kDa . These experiments revealed that the large majority of MT sliding events involved MTs in an anti-parallel configuration ( 18 of 21 events , Figure 2C ) . To confirm this , we also analyzed MT polarity using the direction of movement of single GFP-tagged dynein motors on the track and transport MT , in cases where the transport MT moved past the end of the track MT . In 21 of 22 cases , the transport and track MTs were anti-parallel to one another . This result is consistent with a model in which dynein binds to an anti-parallel MT overlap with one motor domain on each MT in the overlap . In this configuration , both motor domains walk towards the minus-end of their respective MT , thereby driving sliding of the crosslinked pair of MTs ( Figure 1D ) . The above results demonstrate that GST-dimerized yeast dynein , and the native rat dynein complex , both can crosslink and slide MTs . However , in the case of the minimal , GST-dimerized dynein ( Cho and Vale , 2012 ) , it is possible that the artificial , anti-parallel dimerization may allow the two motor domains to crosslink MTs in a manner that would not occur in native dynein dimer . Additionally , the native dynein dimer may utilize a different mechanism , in which it crossbridges MTs through an additional MT binding site , either in the tail of the heavy chain or in one of the dynein accessory subunits . To rigorously examine these possibilities , we tested a truncated dynein molecule that includes its native dimerization domain , but lacks any of its associated subunits . GST-Dyn1331kDa , is monomeric in the absence of GST ( Reck-Peterson et al . , 2006 ) , so we reasoned that a larger portion of the tail domain was necessary for dynein heavy chain dimerization . Prior unpublished observations from our lab indicated that a dynein with a slightly longer N-terminus ( Dyn1387kD , Figure 3A ) was processive when fused to GFP ( S Reck-Peterson and A Carter , unpublished observations ) , and processivity is an attribute of a dynein dimer ( Reck-Peterson et al . , 2006 ) . We overexpressed Dyn1387kD from a Gal promoter , and the purified protein revealed no co-purifying accessory subunits by silver staining after SDS-PAGE ( Figure 3B ) . In sucrose gradients , purified Dyn1387kD sedimented at approximately 19S ( Figure 3C ) , which is similar in size to the full dynein complex ( Paschal et al . , 1987 ) , as well as a truncated dimeric dynein ( Trokter et al . , 2012 ) . Consistent with it being a dimer , single molecule assays revealed robust processive motion of Dyn1387kD along microtubules with an average speed of 75 ± 40 nm/s ( n = 95 ) ( Figure 3D ) . Analysis of moving Dyn1387kD molecules revealed a very similar single molecule intensity distribution to the entire Dyn1387kD population ( Figure 3E ) , indicating that the moving particles were not a minor subset of aggregated or oligomerized molecules . Further excluding the presence of multimers , the fluorescence intensities of single gel-filtered Dyn1387kD molecules were comparable or even lower in brightness to GST-Dyn1331kD ( Figure 3F ) . The reason for the lower brightness of GFP in Dyn1387kD is not clear , although it might be due to fluorescence quenching of nearby fluorphores . Regardless , this result argues that the molecules are not aggregated , and movement is due to dimers and not higher order molecular entities . 10 . 7554/eLife . 00943 . 009Figure 3 . Natively dimerized dynein can drive MT-MT sliding in the absence of associated factors . ( A ) Schematic overview of the Dyn1387kD construct compared to Dyn1 and GST-Dyn331kDa , with labels as in Figure 1A . ( B ) Gel-filtered Dyn1387kD was run on a denaturing gel and proteins were visualized by silver staining . Note the lack of detectable bands at the lower molecular weights expected for dynein accessory subunits . ( C ) Coomassie blue stained gel of sucrose gradient fractions showing sedimentation behavior of Dyn1387kD . Position of 19S and 4 . 3S standards and sucrose concentrations are indicated . ( D ) Kymograph of single Dyn1387kD molecule motility along a MT immobilized on the glass surface . ( E ) Cy5-labeled MTs were surface-immobilized in flow chambers . Dyn1387kD was added either in the presence of ATP or apyrase . Fluorescence intensities of single motors was determined either of all MT-bound molecules in the presence of apyrase ( light gray bars ) , or of the motile motors in the presence of ATP ( dark gray bars ) . ( F ) Flow chambers were prepared as in ( E ) , but MTs were incubated with either Dyn1387kD or GST-Dyn1331kD for 5 min in the presence of apyrase . Intensities of GFP spots were quantified after background subtraction . ( G ) Example of Dyn1387kD driven sliding in a single MT-MT overlap . Successive frames , separated by 34 s from the video are shown with corresponding kymograph below . Velocity histograms for MT-MT sliding and surface gliding are shown to the right with Gaussian fitting . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 00910 . 7554/eLife . 00943 . 010Figure 3—figure supplement 1 . Further characterization of GFP-Dyn1387kD . ( A ) Dyn1387kD induces MT bundling in the absence of ATP . Scale bar is 5 μm . ( B ) Successive frames from Video 5 showing sliding of red MTs in the presence of ATP within a red- and green-labeled MT bundle induced by Dyn1387kD . Scale bar , 5 μm . ( C ) Schematic showing the construct Dyn1387kD ΔMTBD in which the canonical MTBD has been removed . ( D ) MT co-pelleting assay demonstrates that GFP-Dyn1387kD ΔMTBD does not specifically bind to MTs . Supernatents ( S ) and MT pellets ( P ) are shown at various concentrations of MTs . A small fraction of Dyn1387kD ΔMTBD ( ≤10% ) is found in the pellet at each concentration of tubulin tested . The amount of dynein in the pellet remains constant with increasing tubulin concentration , and is thus assumed to be non-specific binding . ( E ) Dyn1387kD ΔMTBD does not induce MT bundling in the absence of ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 010 Dyn1387kD bundled MTs in the absence of ATP ( Figure 3—figure supplement 1A ) , demonstrating that it can crosslink MTs , similar to GST-Dyn1331kD . Addition of ATP to MT bundles resulted in MT sliding of MTs in the bundle ( Figure 3—figure supplement 1B; Video 5 ) . Similarly , Dyn1387kD drove efficient MT-MT sliding in single MT overlaps , at comparable speeds to its surface gliding motility ( Figure 3G; Video 6 ) . Thus , Dyn1387kD behaves very similarly to GST-Dyn1331kD . Further , the fluorescent dynein molecules appeared homogenous during the sliding event ( Video 7 ) , ruling out the presence of aggregates within the sliding overlaps . 10 . 7554/eLife . 00943 . 011Video 5 . Dyn1387kDa slides MTs within bundles . Red-labeled MTs are extruded from a bundle of red- and green-labeled MTs by Dyn1387kDa upon addition of ATP . Total time of the video is 250 s . Playback is 12 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 01110 . 7554/eLife . 00943 . 012Video 6 . Dyn1387kDa slides single MT overlaps . A red-labeled ‘transport’ MT is slid over a green-labeled ‘track’ MT by Dyn1387kDa . Total time of the video is 198 s . Playback is 12 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 01210 . 7554/eLife . 00943 . 013Video 7 . Observation of single Dyn1387kDa molecules in sliding MT-MT overlaps . Single GFP-labeled Dyn1387kDa molecules visualized during a MT-MT sliding event show homogenous fluorescence , further ruling out protein aggregation as a cause of the sliding behavior . Track MT is blue , transport MT is red . Total time of the video is 120 s . Playback is 6 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 013 Finally , we wished to determine if Dyn1387kD might cause MT-MT sliding by employing a second type of MT binding site outside of the canonical MT binding domain ( MTBD ) in its motor domain ( Figure 3A ) ( Carter et al . , 2008; Redwine et al . , 2012 ) . To test this possibility , we deleted the canonical MTBD and expressed the protein ( Dyn1387kD ΔMTBD ) . Dyn1387kD ΔMTBD showed no specific association with MTs in a cosedimentation assay ( Figure 3—figure supplement 1C , D ) and was unable to crosslink MTs ( Figure 3—figure supplement 1E ) . Thus , Dyn1387kD ΔMTBD does not possess a MT binding region outside of the motor domain . Together , these results show that a natively dimerized dynein can crosslink and slide MTs in vitro , and that associated chains or an additional MT binding site is not required for this activity . These results strongly support a model in which dynein slides MTs in vitro using only its two motor domains ( Figure 1D ) . The previous experiments suggest that a dimer of two motor domains can bind two different MTs simultaneously and generate anti-parallel sliding . This model would predict that individual dimeric dynein motors would be approximately stationary in a MT overlap zone , as each motor domain generates opposing pulling forces on the two MTs . To test this model , we imaged individual , fluorescently-labeled dynein molecules within an anti-parallel MT overlap . To generate stable anti-parallel overlaps , we crosslinked MTs with GFP-Ase1 ( Janson et al . , 2007 ) , a protein that crosslinks MTs into anti-parallel bundles with an inter-MT spacing of around 15–40 nm ( Gaillard et al . , 2008; Roque et al . , 2010; Subramanian et al . , 2010 ) . This inter-MT distance is within the size range expected for MT crosslinking by a dimeric dynein motor ( Vallee et al . , 1988; Burgess et al . , 2003 ) . We found that GFP-Ase1 specifically localized to regions of MT overlap ( data not shown ) , as previously described ( Loïodice et al . , 2005; Yamashita et al . , 2005; Janson et al . , 2007 ) . Quantitative imaging of MT fluorescence revealed that 2–3 MTs were typically present in each bundle . As expected from prior work ( Reck-Peterson et al . , 2006 ) , individual TMR-labeled GST-Dyn1331kDa moved unidirectionally and highly processively on single MTs . However , GST-Dyn1331kDa appeared largely immobile in MT overlaps ( Figure 4A ) . Interestingly , when individual molecules moving processively along a single MT reached the MT overlap zone , they would appear to stop moving ( Figure 4B ) . However , when individual molecules were tracked with ∼10 nm precision , it was apparent that individual dynein molecules were not completely immobile in MT overlap zones , but rather frequently moved bidirectionally with long pauses between each run ( Figure 4C ) . While yeast dynein occasionally takes backwards steps ( Reck-Peterson et al . , 2006 ) , such long distance directional switching and frequent prolonged pausing were never observed when dynein moved along single MTs ( Figure 4C ) . 10 . 7554/eLife . 00943 . 014Figure 4 . Dynein can bind two MTs simultaneously . ( A–E ) Biotinylated Cy5-MTs were incubated with 2 nM GFP-Ase1 or with buffer for 5 min . MTs were then bound to glass slides using surface-bound streptavidin . After washing unbound MTs , TMR-labeled GST-Dyn1331kDa ( A–D ) or GFP-tagged Dyn1 ( E ) was introduced into the flow cell . ( A ) Kymographs of TMR-GST-Dyn1331kDa on single MTs ( left panel ) and on anti-parallel MT overlaps ( right panel ) . ( B ) Images from a time series in which individual TMR-GST-Dyn1331kDa molecules can be observed walking into an overlap ( visualized by GFP-Ase1 ) and halting their unidirectional movement . Dotted line indicates the MT . Arrowheads indicate motors that walk from a single MT into an MT overlap . Scale bar is 2 μm . ( C ) High resolution tracking of individual motors on either single MTs ( left ) or in an anti-parallel MT overlap ( right ) . Arrows indicate long pauses in motility . Tracking precision was ∼10 nm . ( D ) Slides were prepared as in ( A ) , but 100 nM of GFP-Ase1 was added to single MTs together with TMR-labeled GST-Dyn1331kDa . GST-Dyn1331kDa shows unidirectional processive movement on Ase1-coated single MTs , but not in Ase1-generated MT overlaps . ( E ) Slides were prepared as in ( A ) , but full-length yeast Dyn1 was used . Kymographs show dynein moving on a single MT ( left ) or in a MT overlap created by Ase1 ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 01410 . 7554/eLife . 00943 . 015Figure 4—figure supplement 1 . Analysis of single Dyn1387kD molecules in MT overlaps . Biotinylated Cy5-MTs were incubated with 2 nM GFP-Ase1 for 5 min . MTs were then bound to glass and GFP-Dyn1387kDa was introduced into the flow cell . A kymograph of GFP-Dyn1387kDa within the MT overlap is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 015 To rule out that the unusual motility of dynein in MT overlaps was due to the presence of GFP-Ase1 on MTs , we loaded single , surface-immobilized MTs with 10- to 20-fold higher concentrations of GFP-Ase1 than was present in MT overlaps . Despite the high concentration of GFP-Ase1 , single dynein molecules moved unidirectionally and processively ( Figure 4D ) . Thus , we conclude that bidirectional switching and pausing are due to the presence of a MT overlap , not to the presence of GFP-Ase1 on MTs . To exclude the possibility that the artificial GST-dimerization of the dynein motor domains was affecting dynein motility within the MT overlaps , we repeated these experiments using GFP-tagged full-length yeast dynein ( Dyn1 ) ( Gennerich et al . , 2007 ) and Dyn1387kD . Dyn1 showed processive , unidirectional motility along single MTs ( Figure 4E ) , but transitioned to frequent bidirectional switching and pausing within MT overlaps ( Figure 4E ) . Similar results were found for Dyn1387kD ( Figure 4—figure supplement 1 ) . Taken together , these results strongly suggest that the dynein motor domains have sufficient flexibility to allow two distinct modes of stepping . The two motor domains of the dimer can bind to the same MT track , most likely in a compact , side-by-side configuration of the two AAA ring domains ( Cho and Vale , 2012 ) , and exhibit persistent unidirectional motion . Alternatively , they can bind to neighboring MTs within a MT overlap and produce a force that acts to slide the two MTs relative to each other ( Figure 1D ) . We next set out to test whether dynein can slide anti-parallel MTs in vivo using only its motor domains . In human cells , kinesin-5 , and kinesin-12 motors promote spindle bipolarity by generating an outward force within the spindle ( Blangy et al . , 1995; Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) . In the absence of both kinesins , a dynein-dependent inward force causes the rapid collapse of the metaphase spindle to a monopolar structure ( Tanenbaum et al . , 2008 , 2009; Ferenz et al . , 2009; Vanneste et al . , 2009 ) . While it is still unknown how dynein produces this inward force , it has been speculated to do so by sliding apart anti-parallel MTs ( Tanenbaum et al . , 2008; Ferenz et al . , 2009 ) . If this is true , then a minimal dynein dimer that elicits anti-parallel MT sliding in vitro might be able to antagonize the outward forces of kinesin-5 and kinesin-12 motors . To test this hypothesis , we generated a GST-dimerized , GFP-tagged human dynein construct lacking the tail domain ( GST-hDyn ) , similar to yeast GST-Dyn1331kDa construct . GST-hDyn lacks the consensus binding sites for other dynein subunits ( Vallee et al . , 2012 ) , which we confirmed by immunoprecipitation ( Figure 5—figure supplement 1A ) . When expressed in HEK293 cells , GST-hDyn partially localized to the spindle ( Figure 5—figure supplement 1B ) , indicating that it is a functional MT-binding protein . To test whether GST-hDyn is able to generate an inward force within the spindle , we designed two types of assays . In the first assay , we evaluated dynein’s ability to drive monopolar spindle formation when kinesin-5 is inhibited before the onset of mitosis . Kinesin-5 inhibition results in the formation of monopolar spindles ( Blangy et al . , 1995; Mayer et al . , 1999 ) , but this effect is prevented and bipolar spindle formation is restored when dynein is depleted by RNAi ( Tanenbaum et al . , 2008; Ferenz et al . , 2009 ) . This result has been interpreted as kinesin-5 and dynein producing counteracting sliding forces during spindle formation; when kinesin-5 is inhibited , dynein-induced sliding forces produce an unbalanced inward force that results in the formation of monopolar spindes . We tested whether GST-hDyn could produce this inward sliding force when expressed in cells depleted of dynein heavy chain by RNAi . We monitored dynein RNAi by blotting for the intermediate chain ( IC ) , which is co-depleted with the heavy chain upon knockdown and found that dynein was robustly depleted from cells upon RNAi ( Figure 5A ) . Strikingly , GST-hDyn expression substantially increased the number of monopolar spindles in dynein-depleted , kinesin-5 inhibited cells ( Figure 5B ) . These results reveal that GST-hDyn can recapitulate the inward force , normally produced by the endogenous dynein , providing further support that the motor domains alone are sufficient for this function . 10 . 7554/eLife . 00943 . 016Figure 5 . Dynein’s motor domains are sufficient to produce an inward force in the spindle . ( A ) HEK293 cells were either mock transfected or transfected with siRNA targeting the dynein heavy chain . After 24 hr cells were re-transfected with siRNA . 72 hr after initial transfection , cells were harvested and the level of dynein expression was determined by western blot . Note that the blot was probed for IC , which is co-depleted when dynein heavy chain is depleted . ( B ) Cells were either mock transfected or transfected with siRNA targeting the N-terminus of dynein . After 24 hr , cells were washed and transfected with GFP-tagged GST-hDyn . 48 hr after the first transfection , cells were re-transfected with dynein siRNA . 68 hr after the first transfection , cells were treated with 1 µM STLC for 6 hr and were then fixed and stained for α-tubulin . The fraction of mitotic cells with monopolar spindles was then scored . ( C and D ) Cells were transfected with GFP-tagged GST-hDyn . After 24 hr , cells were treated with MG132 for 1 hr and subsequently with 20 µM STLC for 1 hr where indicated . Cells were then fixed and stained for α-tubulin and the percentage of mitotic cells with monopolar spindles was scored . ( C ) shows representative images and ( D ) shows the quantification . ( E ) Cells were treated as in ( C ) , except GFP-tagged FKBP-hDyn was transfected instead of GST-hDyn . In addition , 200 nM AP20187 was added together with STLC where indicated . Scale bars , 5 µm . Error bars represent standard deviations . All graphs are averages of three independent experiments with 40–120 cells scored per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 01610 . 7554/eLife . 00943 . 017Figure 5—figure supplement 1 . Analysis of GST-hDyn activity in vivo . ( A ) HEK293 cells were transfected with GST-hDyn . 24 hr after transfection cells were lysed and pulldowns were performed with either beads alone ( 1 ) , an HA antibody ( 2 ) or a GFP antibody ( 3 ) to isolate the GST-hDyn protein . Input , bound and unbound fractions were probed for HA , IC or LIC1 . Note the smear in lane 2 is due to antibody heavy chain cross-reactivity . ( B ) HEK293 cells were transfected with GFP-tagged GST-hDyn . Cells were fixed and stained for α-tubulin . GST-hDyn is observed in the cytoplasm and on the spindle . ( C ) Cells were transfected with GFP-tagged GST-hDyn . After 24 hr , cells were treated with MG132 for 1 hr . Cells were then fixed and stained for α-tubulin and the percentage of mitotic cells with bipolar spindles was scored . ( D ) Cells were either mock transfected or transfected with siRNA targeting the N-terminus of dynein . After 24 hr , cells were washed and transfected with GFP-tagged GST-hDyn . 48 hr after the first transfection , cells were re-transfected with dynein siRNA . 68 hr after the first transfection , cells were fixed . Cells were stained for α-tubulin and the degree of spindle pole focusing was determined . Left panel shows representative images . Arrows indicate additional pole . Right panel shows quantification . Scale bars , 5 µm . Error bars represent standard deviations . All graphs are averages of three independent experiments with 30–120 cells scored per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 017 In a second assay , we tested dynein-driven MT sliding forces in metaphase-arrested spindle ( cells treated with the proteosome inhibitor MG132 ) . During metaphase , two kinesins ( kinesin-5 and kinesin-12 ) provide an outward force on the spindle , which is counterbalanced by an inward force produced by dynein ( Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) . Thus , unlike in early stages of mitosis , when a kinesin-5 inhibitor is applied at metaphase , it only results in a collapse of the spindle to a monopolar structure in a low percentage of the cells ( 38 ± 5% ) since kinesin-12 can still effectively oppose dynein ( Figure 5C , D ) . However , in metaphase-arrested cells expressing GST-hDyn , kinesin-5 inhibition resulted in a much greater percentage ( 88 ± 2% ) of cells with monopolar spindles ( Figure 5C , D ) . While treatment of cells with MG132 might alter the abundance of other MT binding proteins as well in this assay ( Song and Rape , 2010 ) , this result suggests that the expression of GST-hDyn increased the total inward force , tipping the balance of forces and resulting in spindle collapse . Furthermore , overexpression of GST-hDyn , did not result in spindle collapse in the presence of both kinesin-5 and kinesin-12 activity ( Figure 5—figure supplement 1C ) . Our two assays together suggest that GST-hDyn is sufficient to generate an inward force , which is most likely the product of anti-parallel MT sliding . To determine if dimerization of the motor domains is required for the inward force generation by dynein in vivo , we fused dynein to FKBP ( FKBP-hDyn ) , which homodimerizes upon addition of the small molecule dimerizer AP20187 ( inducing processivity of yeast dynein [Reck-Peterson et al . , 2006] ) , and then performed RNAi rescue experiments . Control cells treated with STLC to inhibit kinesin-5 mostly formed monopolar spindles ( 87 . 7 ± 4 . 5% ) . As described earlier , depletion of endogenous dynein by RNAi in the presence of STLC decreased the number of monopolar spindles to 46 . 5 ± 1 . 5% , consistent with a role for dynein in antagonizing kinesin-5 in bipolar spindle assembly ( Figure 5E ) . As a control , the dimerizer AP20187 did not significantly affect spindle bipolarity ( 50 . 7 ± 1 . 8% monopolar spindles , p=0 . 16 ) in STLC-treated , dynein-depleted cells in the absence of FKBP-hDyn ( Figure 5E ) . Next , cells expressing FKBP-hDyn were depleted of endogenous dynein by RNAi and treated with STLC . Expression of FKBP-hDyn in the absence of AP20187 did not alter the fraction of monopolar spindles ( 47 . 8 ± 2 . 7%; mean ± SD ) , indicating that the monomeric version of dynein was unable to reconstitute the function of native dynein . However , addition of AP20187 to cells expressing FKBP-hDyn substantially increased in the fraction of cells with monopolar spindles ( Figure 5E , 75 . 5 ± 2 . 5% ) . This result demonstrates that dimerization of two dynein motor domains robustly activates the ability of dynein to generate an inward force in the spindle . Together , our results suggest that a minimal , dimeric motor can slide anti-parallel MTs to generate an inward force during spindle assembly . In addition to its role in generating an inward force in the spindle , dynein also is important for focusing MT minus ends at the spindle pole ( Walczak and Heald , 2008 ) . To test whether the GST-hDyn was able to support spindle pole focusing , we compared spindle pole focusing in control cells , cells depleted of dynein , and cells depleted of dynein that express GST-hDyn . Depletion of dynein increased the percentage of spindles with unfocused poles and detached centrosomes; however , this pole focusing defect was not rescued by GST-hDyn ( Figure 5—figure supplement 1D ) . These results show that a minimal dynein is able to generate an inward force in the spindle , which likely involves anti-parallel MT sliding , but is unable to fulfill other mitotic functions of dynein .
In this study , we show that cytoplasmic dynein can slide anti-parallel MTs in vitro and provide evidence that this mechanism can occur in vivo . Several kinesin motors have the ability to crosslink and slide MTs in vitro . However , these motors either form a tetrameric complex ( Kashina et al . , 1996; Kapitein et al . , 2008 ) , or contain a second MT binding domain within their tail domain , which facilitates MT crosslinking ( Jolly et al . , 2010 ) ( Figure 6 ) . In contrast , our results show that MT crosslinking and sliding by dynein dimers does not require its tail domain , accessory subunits , regulatory proteins , or further oligomerization . Rather , our results indicate that the two motor domains of a dynein dimer can bind to separate MTs and each motor domain can walk independently on the MT to which it is bound . To accomplish this , the two motor domains likely splay apart , allowing the dimer to bind to the two MTs simultaneously ( Figure 6 ) . 10 . 7554/eLife . 00943 . 018Figure 6 . Mechanisms of MT-MT sliding . ( A ) Kinesin-5 forms tetrameric molecules , allowing them to crosslink and slide MTs using motor domains at opposite ends of the tetramer . ( B ) Kinesin-1 and kinesin-14 utilize a secondary , non-motor , MT binding site located in their tail domains to crosslink and slide MTs . ( C ) Our data suggest that cytoplasmic dynein utilizes a novel mechanism for crosslinking and sliding MTs . The motor domains of a single dynein dimer splay apart and bind to separate MTs . The motor domains then move on opposite MTs , causing sliding of anti-parallel MTs . DOI: http://dx . doi . org/10 . 7554/eLife . 00943 . 018 Our single molecule results also reveal that the two dynein motor domains can switch abruptly from walking along the same MT to walking on separate MTs when the motor encounters a MT-MT overlap . Inside the overlap , dynein often pauses and switches directions , suggesting the motor spends a fraction of time stepping with both motor domains on a single MT in the overlap , and some fraction with each motor domain bound simultaneously to opposite MTs . This behavior was observed for both GST-dimerized dynein , as well as two natively dimerized dyneins ( Dyn1 and Dyn1387kD ) , indicating that it was not an effect of artificial dimerization . These results suggest that the two motor domains have considerable flexibility , allowing them to explore space on a rapid time scale , making frequent transitions between one and two MT bound states . The ability of native or artificially dimerized dynein to walk on the same or different MTs suggests that the dynein motor domains work in a more autonomous fashion than kinesin-1 . Consistent with this notion , recent data suggests that dynein’s two motor domains step in a relatively stochastic and independent fashion ( DeWitt et al . , 2012; Qiu et al . , 2012 ) . Furthermore , a dynein dimer made of one active and one inactive motor domain , still moves processively on a single MT ( DeWitt et al . , 2012 ) , suggesting that a single active head can step forward , provided that it remains tethered to the MT track . The relatively uncoupled motion of dynein’s two motor domains likely allows for them to step independently on oppositely oriented MTs in an overlap and promote MT-MT sliding . Dynein has a key role in minus-end-directed cargo transport along MTs in mammalian cells . In addition , a large body of evidence suggests that dynein controls MT organization during mitosis by sliding MTs within the spindle ( Verde et al . , 1991; Rusan et al . , 2002; Tanenbaum et al . , 2008; Ferenz et al . , 2009; Gatlin et al . , 2009 ) . In these earlier studies , this MT sliding in vivo was speculated to occur between anti-parallel MTs , generating an inward force that opposes the outward forces of kinesin-5 and kinesin-12 acting upon these overlapping MTs . This model agrees well with our in vitro data showing that dynein virtually always slides MTs in an anti-parallel configuration . Our in vivo experiments also demonstrate that a minimal dimer is able to generate an inward force in the spindle that can reconstitute the function of the endogenous motor . This activity was observed using two different dimerization methods , indicating that the MT-MT sliding ability is independent of the method of dimerization . As we cannot directly observe MT-MT sliding in the spindle , we cannot exclude that the minimal dynein collapses the spindle through an alternative mechanism , for example through interaction with additional proteins . However , our in vivo analysis is consistent with the notion that dynein slides MTs in the spindle using only its motor domains , which is further supported by the in vitro experiments . These results suggest that sliding of MTs in the spindle reflects an inherent activity of the dynein motor domains alone , consistent with a model in which the two motor domains of dynein walk on two distinct MTs in the spindle , driving their relative movement . However , the situation with native dynein may be more complicated , since other dynein subunits , as well as additional dynein-associated proteins , such as Lis1 and Nde1 contribute to inward force generation in vivo ( Tanenbaum et al . , 2008; Raaijmakers et al . , 2013 ) . These factors could change the fundamental sliding mechanism of native dynein in vivo , so that the two motor domains do not crossbridge two overlapping microtubules , as described in our model . However , we feel that this is unlikely and rather these factors preserve the fundamental mechanism described ( Figure 6 ) , but may be necessary for promoting dynein stability , efficient targeting to the spindle , and/or modify dynein’s force production ( McKenney et al . , 2010; Trokter et al . , 2012 ) . In addition to the sliding mechanism described in this study , dynein may act through multiple parallel mechanisms to control proper MT organization in the spindle . For example , Wühr and Mitchison ( Wühr et al . , 2010 ) provide evidence that dynein anchored at cytoplasmic sites may also contribute to MT sliding in the spindle in large embryonic cells . Additionally , we show that spindle pole focusing cannot be restored by expression of the minimal dynein dimer . While generation of an inward force likely involves sliding of anti-parallel MTs in the middle of the spindle , pole focusing occurs in a region where most microtubules are expected to have a parallel orientation , perhaps explaining why the minimal dynein is unable to support pole-focusing activity . It is currently unclear how dynein controls pole focusing; however interactions of dynein with NuMA may be important for this activity ( Merdes et al . , 1996 , 2000 ) , Kinesin-14 motors play an important role in pole focusing through MT crossbridging and MT sliding as well ( Figure 6 ) ( Mountain et al . , 1999; Braun et al . , 2009; Fink et al . , 2009 ) . MT sliding has also been observed in other cellular systems . While kinesin-1 is best characterized for such activities ( Navone et al . , 1992; Straube et al . , 2006; Jolly et al . , 2010 ) , dynein may contribute as well . Anterograde transport of MTs was observed in nerve axons and shown to be dependent upon dynein activity ( He et al . , 2005 ) , although the mechanism by which dynein moved MTs in the axon was unclear . Similarly , loss of dynein in Drosophila neurons results in a loss of uniform MT polarity within axons , suggesting that dynein actively sorts MTs based on their polarity ( Zheng et al . , 2008 ) . The mechanism of MT-MT sliding described here provides a model for how dynein may achieve such MT sorting in neuronal cells . Thus , dynein driven MT-MT sliding could be a more generally used mechanism to organize and sort MTs in the cell .
Rat brain cytoplasmic dynein was purified as described ( Paschal et al . , 1991 ) . The peak dynein-containing fractions from the final sucrose gradient were collected , pooled and frozen in LiN2 . This preparation of dynein contains no detectable dynactin ( Figure 1A ) ( Ori-McKenney et al . , 2010 ) . Recombinant yeast dynein constructs were purified and labeled with the Halo-TMR ligand as described ( Reck-Peterson et al . , 2006 ) . Dyn1387kD was created by inserting the Gal promoter and ZZ-TEV-HA-GFP cassette ( Reck-Peterson et al . , 2006 ) into the yeast dynein heavy chain sequence immediately prior to the sequence 732SYTFYTN . The construct encodes amino acids 732–4092 of the dynein heavy chain . Yeast dynein was gel filtered using a Superose 6 column in gel filtration buffer ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM K-acetate , 2 mM Mg-acetate , 1 mM EGTA , 10% glycerol , 0 . 1 mM ATP ) , except for the full-length dynein which was too low of a concentration and was thus used directly after TEV release . Glass coverslips were acid washed as described ( http://labs . bio . unc . edu/Salmon/protocolscoverslippreps . html ) . A ∼10 µl flow chamber was assembled using double-sided sticky tape and a glass slide . The chamber was coated sequentially with the following solutions: 5 mg/ml BSA-biotin ( Sigma , St . Louis , MO ) , 60 µl BC buffer ( BRB80 , 1 mg/ml BSA , 1 mg/ml casein , 0 . 5% Pluronic F-168 , pH 6 . 8 ) , 20 µl 0 . 5 mg/ml streptavidin ( Vector Labs , Burlingame , CA ) , and 60 µl BC buffer to remove excess streptavidin . The chamber was finally washed into assay buffer ( 30 mM Hepes pH 7 . 4 , 50 mM K-acetate , 2 mM Mg-acetate , 1 mM EGTA , 10% glycerol ) containing an oxygen scavenging system ( 0 . 5 mg/ml glucose oxidase , 0 . 1 mg/ml catalase , 25 mM glucose , 70 mM β-mercaptoethanol ) , 0 . 2 mg/ml κ-casein , and 0 . 1% Pluronic F-168 . Pig brain tubulin was purified and labeled as described ( Castoldi and Popov , 2003 ) . MTs were assembled using GMP-CPP , centrifuged at 16 , 000×g and resuspended in BRB80 buffer containing 10 µM taxol . MTs were labeled with ∼10% fluorescent tubulin and ∼10% biotin tubulin where applicable . For bundling experiments , equal volumes of red- or green-labeled MTs were mixed together and then incubated with dynein constructs for 10 min at room temperature . The solution was then perfused into a streptavidin-coated chamber and incubated for 10 min to allow binding to the coverslip surface . The chamber was then washed with BC buffer to remove unbound MTs and imaged using TIRF microscopy . For sliding experiments , assay buffer with 1 mM ATP was added to the chamber and videos were acquired to capture the MT sliding . For MT sliding on single overlaps , chambers were first coated with BSA-biotin and streptavidin . Biotinylated track MTs were bound to the surface and the chamber was washed extensively to remove unbound MTs . Dynein was then introduced and allowed to bind to the track MTs for 5 min at room temperature . Rat brain dynein at ∼100 nM or GST-Dyn1331kDaa at ∼200 nM was used . Unbound dynein was removed by washing with BC buffer , followed by introduction of transport MTs diluted in BC buffer . After a 10 min incubation , unbound track MTs were washed out with BC buffer and the chamber was equilibrated in assay buffer containing 2 mM ATP and an oxygen scavenger system . Sliding was imaged in TIRF mode . Data was acquired and subsequently analyzed by making kymographs of the sliding events using µManager . Polarity-marked MTs were made essentially as described previously ( Goodwin and Vale , 2010 ) , but the GMP-CPP seeds were first capped at their minus ends by incubation with 0 . 3 µM NEM-treated tubulin . Sliding between MTs was observed in the same buffer with a final concentration of 100 mM K-acetate . At 200 mM K-acetate , no annealing between transport and track MTs was observed . To track single molecules of dynein with high resolution , the Localization Microscopy plugin of µManager ( developed by Nico Stuurman ) was used ( http://valelab . ucsf . edu/∼MM/MMwiki/index . php/ Localization_Microscopy ) . To express a minimal human dynein constructs in mammalian cells , the pTON vector ( a modified version of pcDNA4TO with N-terminal GFP ( Tanenbaum et al . , 2011 ) was modified to include either N-terminal GFP-Halo-HA-GST tags ( GST-hDyn ) or GFP-FKBP ( F36V ) tags ( FKBP-hDyn ) . The C-terminal ∼380 kDa motor domain of human dynein ( nucleotides 3850-13 , 938 of clone KIAA1997 ) was then inserted downstream of either the GST or the FKBP . This boundary was chosen to be similar to the previously published motor domain constructs of rat ( Höök et al . , 2005 ) , Dictyostelium , and yeast dyneins ( Reck-Peterson et al . , 2006 ) . HEK293 cells were transfected with either GST-hDyn or FKBP-hDyn and cells were analyzed ∼24 hr after transfection . Expression of FKBP-hDyn at very high expression levels disrupted chromosome alignment and resulted in defects in spindle morphology , independently of AP20187 addition . Very high expressing cells also had an increased tendency to form monopolar spindles . Therefore , for all experiments we focused our analysis on the ∼70% of cells within the population with low to moderate expression levels of FKBP-hDyn . Step gradients were made by carefully layering 250 μl each of 8 , 16 , 24 , or 32% sucrose in Pipes-Hepes buffer ( 50 mM Pipes , 50 mM Hepes , 2 mM MgSO4 , 1 mM EDTA , pH 7 . 0 ) in a TLS-55 tube . A 250 μl solution of either TEV released GFP-Dyn1387kD , or gradient standards , was layered on top . The gradients were centrifuged in a TLS-55 rotor at 55K rpm ( 200 , 000×g ) for 3 hr at 4°C . Standards used were thyroglobulin ( 19S; Sigma ) or BSA ( 4 . 3S; Sigma ) . Gradients were fractionated by carefully pipetting 100 μl from the top . Cy5-and biotin-labeled MTs were surface-immobilized using biotin-BSA and streptavidin . GFP-tagged motors were incubated with or without ATP or apyrase in the flow chamber for 5 min , after which images were taken using identical microscopy settings . The density of motors was kept at around 2–4 motors per MT to ensure spots represented a single motor . GFP fluorescence was measured in ImageJ and background signal was subtracted . HEK293 cells were cultured in DMEM supplemented with 10% FCS and antibiotics . siRNA transfections were performed with Hiperfect ( Qiagen , Valencia , CA ) according to manufacturer’s guidelines . DNA was transfected using polyethyleneimine ( PEI ) . MG132 ( Sigma ) was dissolved in DMSO and was used at 5 µM final concentration . STLC ( Sigma ) was dissolved in DMSO and was used at indicated concentrations . AP20187 ( Clontech , Mountain View , CA ) was dissolved in ethanol and used at a final concentration of 200 nM . Cells were grown in glass bottom 96-well plates . At the time of fixation , culture medium was removed and cells were fixed in PBS with 3 . 7% formaldehyde and 1% Triton X-100 for 5 min . Cells were then washed with PBS and post-fixed with cold methanol for 5 min . Fixed cells were incubated with anti-α-tubulin antibody ( 1:5000; Sigma ) overnight . Secondary antibody was goat-anti-mouse-AlexaFluor555 ( 1:1000; Invitrogen , Grand Island , NY ) , which was incubated for 1 hr . Cells were imaged on a Zeiss spinning disc confocal with a 100 × 1 . 45 NA objective and a Hamamatsu EM-CCD camera . The microscope was controlled by µManager software ( Edelstein et al . , 2010 ) . | When cells divide , they must also divide their contents . In particular , both ‘mother’ and ‘daughter’ cells require full sets of chromosomes , which must first be duplicated , and then evenly distributed between the cells . Protein filaments called microtubules form a network that helps to accurately segregate the chromosomes . Microtubules emanate from structures at each end of the dividing cell known as spindle poles; after the chromosomes have duplicated , the microtubules latch onto them and align the pairs in the middle of the cell . As the two cells separate , microtubules at opposite spindle poles reel in one chromosome from each pair . Microtubules are composed of alternating copies of two different types of a protein called tubulin , and have ends with distinct properties . The ‘minus’ ends are directed outwards , away from the chromosomes; the ‘plus’ ends—which can actively add tubulin—grow toward the middle of the cell , and can also bind to chromosomes . Microtubules can be manipulated by motor proteins that ‘walk’ along them carrying cargoes , which can include other microtubules . The combined actions of many motor proteins rearrange the microtubule network into a configuration that enables the chromosomes , and other cellular structures , to partition equally between the mother and daughter cells . Motor proteins such as dynein and kinesin transport cargoes along microtubules; each motor is composed of two identical copies of the protein bound to each other . Kinesin walks toward the plus end of a microtubule , propelling itself using ‘feet’ that are called motor domains; it binds cargoes ( including other microtubules ) through additional regions located at the opposite end of the protein . In contrast , dynein walks toward the minus end of a microtubule . Although dynein is known to carry certain cargoes through regions outside its motor domain , how it transports other microtubules is not well understood . Tanenbaum et al . now show that regions outside the motor domain of dynein are unnecessary to transport microtubule cargoes . When two dynein motor domains are isolated and linked to each other in vitro , each can bind to a separate microtubule . By walking toward the minus ends of their respective microtubules , the motor domains drive the microtubules in opposite directions , sliding them apart . These studies thus provide insight into the mechanism through which dynein works with additional motor proteins ( such as kinesin ) to rearrange microtubules during cell division—and also to ensure that chromosomes segregate evenly between mother and daughter cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2013 | Cytoplasmic dynein crosslinks and slides anti-parallel microtubules using its two motor domains |
Morphogenesis emerges from complex multiscale interactions between genetic and mechanical processes . To understand these processes , the evolution of cell shape , proliferation and gene expression must be quantified . This quantification is usually performed either in full 3D , which is computationally expensive and technically challenging , or on 2D planar projections , which introduces geometrical artifacts on highly curved organs . Here we present MorphoGraphX ( www . MorphoGraphX . org ) , a software that bridges this gap by working directly with curved surface images extracted from 3D data . In addition to traditional 3D image analysis , we have developed algorithms to operate on curved surfaces , such as cell segmentation , lineage tracking and fluorescence signal quantification . The software's modular design makes it easy to include existing libraries , or to implement new algorithms . Cell geometries extracted with MorphoGraphX can be exported and used as templates for simulation models , providing a powerful platform to investigate the interactions between shape , genes and growth .
Morphogenesis of multicellular organisms occurs through multiscale interactions of genetic networks , cell-to-cell signaling , growth and cell division . Because of the complexity of temporal and spatial interactions involved , computer simulations are becoming widely used ( Dumais and Steele , 2000; Jönsson et al . , 2006; Sick et al . , 2006; Lecuit and Lenne , 2007; Merks et al . , 2007; Stoma et al . , 2008; Kondo and Miura , 2010; Varner et al . , 2010; Kennaway et al . , 2011; Santuari et al . , 2011; Aegerter-Wilmsen et al . , 2012; Kierzkowski et al . , 2012; Bassel et al . , 2014; Milde et al . , 2014; Sampathkumar et al . , 2014; Yoshida et al . , 2014 ) in what is now being called Computational Morphodynamics ( Chickarmane et al . , 2010 ) . Key to this methodology is the combination of time-lapse microscopy to quantify changes in cell geometry and gene expression with dynamic spatial modeling ( Jönsson et al . , 2012 ) . Confocal microscopy is frequently the tool of choice for data collection , as the proliferation of fluorescence markers and variations in the method make it possible to visualize proteins , organelles , cell boundaries , and even protein–protein interaction and protein movement in vivo . Other technologies such as serial block-face scanning electron microscopy ( SEM ) ( Denk and Horstmann , 2004 ) make it possible to study sub-cellular structures at a much higher resolution on fixed samples . However , despite the rapid advancement of 3D imaging technologies , there is a lack of methods and software to process and quantify these data and to integrate them within simulation environments . Most simulation models of morphogenesis operate on 2D templates ( Dumais and Steele , 2000; Jönsson et al . , 2006; Sick et al . , 2006; Merks et al . , 2007; Stoma et al . , 2008; Kondo and Miura , 2010; Varner et al . , 2010; Kennaway et al . , 2011; Santuari et al . , 2011; Aegerter-Wilmsen et al . , 2012; Kierzkowski et al . , 2012; Sampathkumar et al . , 2014 ) . This is not surprising since many key biological processes occur on surfaces , for example in epithelial layers ( Lecuit and Lenne , 2007; Savaldi-Goldstein et al . , 2007; Heller et al . , 2014 ) . Morphogenesis involves complex 3D deformation , such as folding during gastrulation in animal systems or bulging out of new lateral organs in plants , causing significant curvature in the tissues controlling these events . It is therefore essential to be able to quantify cell shapes and fluorescence-based reporters on curved surface layers of cells . The simplest method to achieve this is to take several image slices and project them onto a single plane ( Butler et al . , 2009; Chickarmane et al . , 2010; Kuchen et al . , 2012 ) . However , when trying to quantify cell shape change , division orientations , or growth , distortions due to the projection quickly become too large as the angle between the surface and the projection plane increases . Even small amounts of tissue curvature can hinder the accurate imaging of a single cell layer over an entire sample . To alleviate some of these issues , methods have been developed to determine the 3D position of cell junctions on the surface , while the segmentation into cells is still performed on flat 2D images ( Dumais and Kwiatkowska , 2002; de Reuille et al . , 2005; Routier-Kierzkowska and Kwiatkowska , 2008 ) . However these approaches are labor intensive , limited to tissues that can be visualized as a flat 2D image , and are not accurate when the angle of the tissue with the projection plane becomes too large . Furthermore , methods based on tissue casts combined with stereo reconstruction of SEM images ( Dumais and Kwiatkowska , 2002; Routier-Kierzkowska and Kwiatkowska , 2008 ) need to be combined with methods using fluorescent markers ( Uyttewaal et al . , 2012 ) if gene expression is to be monitored . Here we present a method and the open-source software MorphoGraphX ( www . MorphoGraphX . org , Box 1 ) to quantify the temporal evolution of cellular geometry and fluorescence signal on curved 2D surface layers of cells over multiple time points in both plants and animals . In addition to 2D curved surfaces , MorphoGraphX also possesses a rich set of tools for full 3D image processing and cell segmentation , and can be used to easily transfer realistic cell geometries and fluorescent marker data into computational modeling environments . MorphoGraphX is built from a collection of loadable modules ( shared libraries ) , centered around an interactive visualization core that exploits the latest features of modern Graphics Processing Units ( GPUs ) . This design allows the software to be easily adapted to changing research needs , and facilitates the integration of algorithms from other open-source imaging processing libraries into a custom work flow . The software is the first of its kind specialized to process curved surface layers of cells , and here we demonstrate its capabilities both in plant and animal systems .
Modern imaging technologies today provide us with an abundance of data from a variety of sources: Confocal Laser Scanning Microscopy , Magnetic Resonance Imaging and block-face SEM all provide full 3D volumetric data that can be rendered in MorphoGraphX ( Figure 1 , Video 1 ) . Our software can also process surfaces , which can be imported from 3D scanners , reconstructions from Stereo-SEM images ( Routier-Kierzkowska and Kwiatkowska , 2008 ) , focus stacking microscopes and scanning probe methods such as Cellular Force Microscopy ( Routier-Kierzkowska et al . , 2012 ) ( CFM ) , or extracted within MorphoGraphX from full 3D data sets ( Figure 1 ) . MorphoGraphX contains a highly optimized rendering engine that is capable of accurate rendering of both semi-transparent surfaces and volumetric data simultaneously . Surfaces are represented by an oriented triangular mesh , which is typically extracted from the surface of an object , and thus represents the outermost tissue layer ( Figure 1A , C , D ) , or the boundaries of individual 3D objects ( e . g . , cells ) in the case of full 3D segmentation ( Figure 1B ) . Once processed , surfaces and associated data can be exported in a variety of file formats suitable for loading into modeling or analysis softwares , allowing the direct use of sample geometry in computer simulations ( Santuari et al . , 2011; Kierzkowski et al . , 2012; Bassel et al . , 2014; Sampathkumar et al . , 2014; Yoshida et al . , 2014 ) . 10 . 7554/eLife . 05864 . 004Figure 1 . MorphoGraphX renderings of 3D image data and surfaces . ( A ) Extraction of a brain surface ( gray , semi-transparent surface colored by signal intensity ) from a Magnetic Resonance Angiography scan of an adult patient ( IXI dataset , http://www . brain-development . org/ ) . Surrounding skull and skin ( green ) have been digitally removed prior to segmentation . Voxels from the brain blood vessels are colored in purple . ( B ) Serial block-face scanning electron microscopy ( SEM ) images of mouse neocortex ( Whole Brain Catalog , http://ccdb . ucsd . edu/index . shtm , microscopy product ID: 8244 ) . Cutaway view ( gray ) shows segmented blood vessels ( dark purple ) and five pyramidal neurons colored according to cell label number . ( C ) Topographic scan of onion epidermal cells using Cellular Force Microscopy ( Routier-Kierzkowska et al . , 2012 ) , colored by height . ( D ) 3D reconstruction of Arabidopsis thaliana leaf from stereoscopic SEM images ( Routier-Kierzkowska and Kwiatkowska , 2008 ) , colored by cell size . Scale bars: ( A ) 2 cm , ( B and C ) 20 μm , ( D ) 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 00410 . 7554/eLife . 05864 . 005Video 1 . User interface and rendering in MorphoGraphX . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 005 A key strength of MorphoGraphX is the ability to summarize 3D data as a curved surface image . After extracting the shape of an organ , 3D data just below the surface can be projected onto it , creating a curved image of the outer layer of cells ( Figure 2 ) . This enables the extraction of precise cell outlines without the distortions associated with a flat 2D projection ( Figure 2—figure supplement 1 ) . We have found that many algorithms designed for 2 and 3D image processing can be adapted to our curved surface images . Feature extraction in MorphoGraphX typically follows a pattern: ( i ) volumetric data ( often a cell outline marker ) is pre-processed to remove noise or obstructions; ( ii ) the object of interest is turned into a mask ( binary image ) ; ( iii ) the object is extracted as a surface mesh; ( iv ) volumetric data is projected onto the surface; ( v ) the projection is used for segmentation of the surface into cells ( Figure 2 , Video 2 ) . The segmentation can be fully automatic ( Video 3 ) or directed by manually placed seeds . Steps ( i–iii ) can be repeated as surfaces of interest will often be used to help pre-processing the volumetric data . For example , surfaces can be used to trim the 3D image ( Figure 2—figure supplement 2 ) , or to select regions of interest for data analysis . 10 . 7554/eLife . 05864 . 006Figure 2 . Feature extraction and 3D editing of confocal image stacks . A sample workflow from raw data to segmented cells is presented for an A . thaliana flower ( A–F ) . ( A and B ) After removing noise with 3D filters , the stack ( green ) is converted into a mask using edge detection ( yellow ) . ( C ) A coarse representation of the surface is extracted with marching cubes , then smoothed and subdivided . ( D ) After subdivision , a thin band of signal representing the epidermal layer ( purple ) is projected onto the mesh , giving a clear outline of the cells . Note that the projection is perpendicular to the curved surface and its depth is user-defined ( in this case , from 2 to 5 μm ) . ( E ) The surface is then segmented with the watershed algorithm , which we adapted to work on unstructured triangular meshes . ( F ) Closeup of adaptive subdivision , with finer resolution near cell boundaries . A similar process flow was used to segment shoot apical meristem in tomato ( Kierzkowski et al . , 2012; Nakayama et al . , 2012 ) and A . thaliana ( Kierzkowski et al . , 2013 ) , as well as Cardamine hirsuta leaves ( Vlad et al . , 2014 ) . ( G ) 3D editing tools can be used to expose internal cell layers prior to surface extraction . Cell shapes extracted from the curved pouch of a Drosophila melanogaster wing disc , after removing signal from the overlying peripodial membrane ( Aegerter-Wilmsen et al . , 2012 ) . Alternatively , the stack can be cleaned by removing voxel data above an extracted mesh or conserving only the signal at a defined distance from the mesh , as shown in purple in ( D ) and Figure 2—figure supplement 2 . ( H ) MorphoGraphX also provides tools to project signal on arbitrary curved surfaces defined interactively by moving control points ( red ) . A Bezier surface is highly bent to cut through the cortical cells of a mature A . thaliana embryo . Scale bars: 2 μm in ( F ) , 20 μm in all other panels . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 00610 . 7554/eLife . 05864 . 007Figure 2—figure supplement 1 . Maximal projection vs projection of signal on curved surface . ( A ) Flat maximal projection of the membrane protein PIN1::GFP in an Arabidopsis inflorescence meristem using ImageJ . ( B ) The same confocal stack analysed in MorphoGraphX was used to extract a surface mesh first , then project the signal normal to the surface at a depth from 1 to 5 μm . In the case of curved organs like the shoot apical meristem of Arabidopsis , maximal projection ( close up inset A ) results in distortions that make it difficult to interpret the image ( i . e . , determine PIN1 polarization ) or track changes in cell shape . The projection method used in MorphoGraphX ( inset B ) , on the other hand , is less prone to artefacts . Scale bars 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 00710 . 7554/eLife . 05864 . 008Figure 2—figure supplement 2 . Mesh-volume interaction . The original volumetric data ( A ) can be trimmed using a surface , either to clean up the data for further segmentation or to display the part of the data which was projected . Here we kept only the volumetric data located between 2 to 5 μm from the curved surface ( B ) , showing what part of the signal was projected onto the surface to obtain the cell outline in main Figure 2D . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 00810 . 7554/eLife . 05864 . 009Video 2 . Manual segmentation of a tomato shoot apex . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 00910 . 7554/eLife . 05864 . 010Video 3 . Automatic segmentation of a tomato shoot apex . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 010 MorphoGraphX allows user-defined surfaces to interact with volumetric data both for visualization and feature extraction . The researcher can interactively define Bezier surfaces to visualize curved slices through an object . By manipulating the Bezier control points it is possible to fit almost any shape to a surface of interest within the sample . An extreme example of this is shown in Figure 2H where the surface has been shaped to display the cortical cells of a mature Arabidopsis embryo . The Bezier surface can be converted to a triangular mesh , and segmented into cells with the same procedure used for Figure 2A–E . The extracted tissue geometry can be then used , for example , as template for simulations ( Santuari et al . , 2011 ) . Once a surface is segmented into cells , data collected simultaneously on a different channel , such as a GFP fusion to a protein of interest , can then be projected onto the segmented surface ( Figure 3 ) . This allows the quantification of genetic expression and protein localization at the cellular , or sub-cellular scale . As with the cell outlines , the projection creates a curved image of the data that can be processed in a similar way as a planar 2D image . Many tools commonly used for the analysis of flat images ( for example in softwares such as Adobe Photoshop , Gimp and ImageJ ) have been adapted for use on curved surfaces in MorphoGraphX . This includes Gaussian blur , erosion , dilation , morphological closing , local minima detection , normalization , etc . The flexibility of this approach is demonstrated by our implementation of more complex algorithms , such as the watershed transform for cell segmentation and our adaptation of an algorithm based on signal gradients to compute the orientation of microtubules ( Figure 3A , Figure 3—figure supplement 3 ) that was previously implemented in 2D ( Boudaoud et al . , 2014 ) . 10 . 7554/eLife . 05864 . 011Figure 3 . Quantification of signal projected on the mesh surface . ( A ) Microtubule orientation ( red line ) determined in epidermal cells of C . hirsuta fruits . Signal for TUA6-GFP ( green ) at a maximal depth of 1 . 5 μm was projected on the curved surface and processed with a modified version of a 2D image analysis algorithm ( Boudaoud et al . , 2014 ) to compute fiber orientation . Line length indicates strength of orientation . ( B ) Quantification of vestigial ( left ) and wingless ( right ) transcription in the wing disc of D . melanogaster at 0–20 μm depth . Data from ( Aegerter-Wilmsen et al . , 2012 ) . ( C and D ) Quantification of PIN1::GFP signal in Arabidopsis shoot apical meristem at different depths . A projection between 0 and 6 μm away from the surface corresponds to the epidermal ( L1 ) layer ( C ) , while a depth of 6–12 μm reflects the sub-epidermal ( L2 ) layer . ( E ) Sub-cellular localization of PINFORMED1 ( PIN1 ) in the L1 is assessed by quantification of the projected signal for each cell wall , as in ( Nakayama et al . , 2012 ) . The projected PIN1 signal can be compared with other markers of organ initiation , such as the curvature . While projected PIN1 signal from the L1 ( C and E ) shows a clear accumulation of signal at the incipient primordium ( star ) , there is no sign of up-regulation in the deeper layer ( D ) nor visible bulge yet ( see Figure 3—figure supplement 1 ) . ( C–E ) Data from ( Kierzkowski et al . , 2013 ) . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01110 . 7554/eLife . 05864 . 012Figure 3—figure supplement 1 . PIN1 expression levels in L1 and L2 vs curvature in Arabidopsis inflorescence meristem . ( A and B ) show PIN1::GFP signal quantification respectively in the epidermal ( L1 ) and sub-epidermal ( L2 ) layers , as in main Figure 3 . ( C ) Tissue curvature for a neighborhood of 10 μm , with positively curved areas in red and negatively curved on blue . Phyllotactic order ( P6 , P5 , . . . , I1 , I2 ) is indicated , based on PIN1 expression and curvature . I2 marks the youngest incipient primordium , with no apparent bulging nor peak of PIN1 expression in the L2 yet , but PIN1 up-regulation already visible in L1 layer . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01210 . 7554/eLife . 05864 . 013Figure 3—figure supplement 2 . Quantification of PIN1-GFP signal localized to close to the membrane vs internal signal . ( A ) Projection of PIN1-GFP signal on the curved mesh surface extracted from a young flower bud . ( B ) Quantification of the average internal signal in each cell , in arbitrary units . ( C ) Quantification of membrane signal , in arbitrary units . ( D ) Ratio of internal vs total signal per cell . All signals ( total , internal and membrane ) are here averaged over the total area of the triangles used for the computation . Membrane and internal signal are distinguished based on the distance to the cell border . The signal from vertices closer than 0 . 7 µm to the cell outline is considered as membrane signal , while vertices farther than this threshold are considered internal . Note that the resolution of classical confocal images ( A ) does not allow the separation of plasma membrane signal from two adjacent cells in this case where the cell walls are very thin . It is only possible to distinguish PIN1 localization , not polarity . The use of other imaging techniques , such as hyper-resolution microscopy , could potentially allow a precise quantification of PIN1 polarization at the individual cell level in MorphoGraphX . In organs with larger cells , a more sophisticated analysis of the signal near the walls , such as that used in the CellSeT software ( Pound et al . , 2012 ) , could be implemented for curved surfaces . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01310 . 7554/eLife . 05864 . 014Figure 3—figure supplement 3 . Quantification of microtubule orientation . We adapted the FibrilTool ImageJ plugin ( Boudaoud et al . , 2014 ) to extract fibril orientations in MorphoGraphX . While the original algorithm was written for 2D unsegmented images , our implementation works on our curved surface images . This presents an advantage for highly curved organs or cells , as is the case with the Cardamine fruit epidermis ( A ) . In addition , the use with segmented images allows the border exclusion zone to be assigned automatically , substantially reducing the clicking required and greatly increases throughput . Fiber orientation is determined by finding the principal component of color gradient within each cell . The main orientation of fibers ( B , white segments ) is perpendicular to the direction of maximal gradient . The degree of fiber orientation ( B , colormap ) , or anisotropy , is given by the formula: ( gradientMax/gradientMin -1 ) . The adapted algorithm in MorphoGraphX allows us to directly combine various types of quantifications , for example , growth direction , PIN localization and MT orientation , on the same dataset . Scale bars: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 014 Signal coming from different tissue layers can be visualized and quantified by adjusting the depth of projection ( Figure 3B–E ) . This is particularly useful to distinguish protein expression levels at different depths within an organ . As an illustration , in the shoot apical meristem of Arabidopsis thaliana we can observe that the auxin efflux carrier PINFORMED1 ( PIN1 ) is first upregulated in the epidermis at the site of incipient primordium initiation before being activated in deeper layers ( Bayer et al . , 2009; Kierzkowski et al . , 2013 ) ( Figure 3C , D and Figure 3—figure supplement 1 ) . Quantification can also be performed at the sub-cellular scale ( Pound et al . , 2012 ) . The amount of fluorescence signal projected onto the triangle mesh can be divided into a membrane localized portion and a cell interior portion ( Figure 3E and Figure 3—figure supplement 2 ) . This is accomplished by summing all the signal within a fixed distance from a cell border and considering it as being associated with the membrane , while all the signal further away from the cell outline is called internal . The process can be used to quantify what portion of a tagged protein , for example , the auxin efflux carrier PIN1 , is localized to the plasma membrane or internalized ( Nakayama et al . , 2012 ) . Projection of the signal on the surface allows to summarize essential information from several channels of a large confocal data set into a very compact form . For example , the global shape of the sample can be extracted from an autofluorescence signal , while the cell wall or membrane marker collected in another channel is used to segment cells and obtain their geometry . The expression level of a protein from a third channel may then be quantified at the cellular level based on the segmentation . Finally , several samples in a time lapse experiment can be compared to obtain information about the temporal evolution of shape and gene expression . In addition to data from single image stacks , MorphoGraphX is able to process and compare multiple time points . This enables the analysis of stacks before and after experimental treatments , or time-lapse data . This capability relies on an efficient method to co-segment samples from two time points . One approach is to segment both stacks separately and then to use an automated algorithm to match the points ( Fernandez et al . , 2010 ) . However , automatic segmentation and matching can be prone to errors that have to be checked and corrected by hand , which can be very time-consuming depending on the error rate . For this we have developed a user-friendly interface in MorphoGraphX to manually identify cell lineages on curved surfaces representing the tissue at different time points ( Video 4 ) . Errors in lineage are detected automatically by comparing the neighborhoods of daughter cells and their parents . Once the co-segmentation is complete , changes in cell area or gene expression over the interval between two time points can be computed and visualized as a heatmap ( Figure 4 ) . Cell proliferation can also be visualized as a colormap ( Vlad et al . , 2014 ) , along with marking of the new walls ( Figure 4 ) . Pairwise correspondence between time points can be combined in longer time series ( Figure 4—figure supplement 3 ) , for example to perform clonal analysis over several days ( Vlad et al . , 2014 ) . The data can be output in various formats for further processing , such as the comparison of growth rates with protein expression levels or microtubule orientations . 10 . 7554/eLife . 05864 . 015Video 4 . Lineage tracking and growth analysis of time lapse data on tomato shoot apex . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01510 . 7554/eLife . 05864 . 016Figure 4 . Growth in the tomato shoot apex over 22 hr . ( A ) Expression of the auxin activity reporter pDR5::VENUS visualized underneath the semi-transparent mesh . ( B ) Average curvature ( μm−1 ) for a neighborhood of 20 μm , with positive values in red , and negative values in blue . ( C ) Shoot apex surface colored by cell proliferation rate as in ( Vlad et al . , 2014 ) . New cell walls are indicated in dark red . ( D ) Top and side views of the heat map of areal expansion over the first 11 hr interval . Principal directions of growth ( PDGs ) are indicated for cells displaying an anisotropy above 15% , with expansion in white and shrinkage in red . Note the rapid anisotropic expansion of the developing primordium P1 and of the peripheral zone close to the incipient primordium I1 , while cells in the boundary between P1 and the meristem contract in one direction ( red lines ) . Arrows indicate the correspondence between top and side views . Raw confocal data from ( Kierzkowski et al . , 2012 ) . Scale bars 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01610 . 7554/eLife . 05864 . 017Figure 4—figure supplement 1 . Simplification of mesh . ( A ) After several rounds of signal projection , segmentation with watershed and mesh sub-division , the final mesh has a finer resolution at the cell borders than in the rest of the cell . ( B ) A simplified version of the mesh can be created , to conserve only some of the vertices from the cell outline ( in red ) and the cell centers ( in orange ) . This simplified mesh can be used to compute cell shape and tissue curvature , or to study the neighborhood information within a tissue . ( C ) The mesh can be even more simplified to keep only the cell junctions ( green dots ) , cell boundaries represented by edges ( red ) and centers ( not shown ) . This representation is used in the computation of PDGs . Simplified meshes are also useful to export for use as starting geometry for simulation models . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01710 . 7554/eLife . 05864 . 018Figure 4—figure supplement 2 . Computation of PDGs in case of anisotropic deformation . In ( A and B ) the cell outline is simplified to keep only the junctions ( see Supplementary Figure 4C ) . Mother cells ( A ) on the first time point and their daughters ( B ) on the second time point are identified using the lineage tracking tool . The same junctions ( green dots ) are identified on both time points using the cell neighborhood information . Notice that new junctions generated by cell division on the second time point have no match . Pairs of matched vertices between the two time points are then used to computed the PDGs for each cell . The result can be visualized either on the first ( C ) or second ( D ) time points of the simplified meshes . PDGs can also be saved and re-loaded for display on the original , fine meshes ( E and F ) . Different colors can be used to visualize expansion ( in white ) and contraction ( in red ) on the axis ( G and H ) . The results of the PDG computation can also be visualized as heat maps , by coloring each cell according to the magnitude of deformation in the maximal ( G ) or minimal direction ( H ) , or by other quantities ( i . e . , anisotropy , etc ) . Blue cells in ( H ) shrink along one axis and belong to the boundary region of the tomato shoot apex shown in main Figure 4D , while the cells colored in red in ( G ) and ( H ) belong to the fast growing peripheral zone adajcent to the boundary . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 01810 . 7554/eLife . 05864 . 019Figure 4—figure supplement 3 . Analysis of time lapse series of tomato shoot apical growth over 48 hr ( 5 time points , 12 hr intervals ) . First column on left: Quantification of pDR5::3xVENUS-N7 signal over 20 μm depth . Arbitrary units . Second column: Average tissue curvature for a neighborhood of 15 μm , given in μm−1 . Third column: Cell proliferation , given in number of daughter cells . New cell walls are marked in dark red . Right column: Areal expansion ( in % ) for each 12 hr interval , displayed on the second time point . The axis of PDGs are displayed only for cells with high anisotropy . White axis represent expansion , red axis shrinkage . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 019 In addition to extracting areal growth rates from time-lapse data , MorphoGraphX can also be used to quantify growth directions and anisotropy . The cell junctions ( Figure 4—figure supplement 1 ) are used as landmarks to compute the two dimensional Principal Directions of Growth ( PDGs ) of the surface associated with each cell ( Dumais and Kwiatkowska , 2002; Kwiatkowska and Dumais , 2003; Routier-Kierzkowska and Kwiatkowska , 2008 ) . The cell lineage information is used to account for cell division and identify only the junctions that are conserved in between time points ( Figure 4—figure supplement 2 ) . Principal growth directions and their associated magnitudes can be displayed on the surface of the simplified mesh used for the computation , or stored to be later displayed on the original mesh . The growth anisotropy is computed from the magnitudes of the growth directions ( Figure 4—figure supplement 2 ) . For visual clarity , growth directions can optionally be displayed only in cells in which the anisotropy is above a user-defined threshold ( Figure 4D ) . Other directional quantities can also be computed , stored and displayed in MorphoGraphX on a cellular basis . For example , the local tissue curvature ( Goldfeather and Interrante , 2004 ) can be calculated based on the position of the neighbors closer than a given radius and displayed in a manner similar to the growth directions ( Figure 4B and Figure 4—figure supplement 3 ) , making it a convenient tool for precise staging of fast growing organs such at the shoot apical meristem ( Kwiatkowska and Dumais , 2003; Kwiatkowska , 2004 ) . We demonstrate the capabilities of MorphoGraphX by quantifying growth of the stem cell niche and surrounding tissue in the shoot apex of tomato with time lapse imaging over several days ( Kierzkowski et al . , 2012 ) ( Figure 4 and Figure 4—figure supplement 3 ) . The shoot apex is the source of all the aerial structure of the plant . At the summit , a slow growing central zone harbors the stem cell niche , surrounded by a fast growing peripheral zone where organ initiation occurs ( Steeves and Sussex , 1989; Dumais and Kwiatkowska , 2002 ) . The analysis in MorphoGraphX starts with surface extraction followed by manual or automatic segmentation ( Videos 2 , 3 ) , and lineage matching ( Video 4 ) of all of the time points in the series . We observed similar patterns of growth , cell proliferation and organ geometry in the tomato shoot apex as those reported in other species ( Kwiatkowska and Dumais , 2003; Grandjean et al . , 2004; Kwiatkowska , 2004; Reddy et al . , 2004; Kwiatkowska and Routier-Kierzkowska , 2009 ) . The first geometrical indicator of primordium initiation we noted was a slightly elevated curvature at the corner of the meristem ( Kwiatkowska and Dumais , 2003; Kwiatkowska , 2004 ) . This early change in shape coincided with increased growth in the peripheral zone . The peripheral zone itself displayed differential growth dependent on the dynamics of primordium initiation . Regions adjacent to older primordia exhibited fast , highly anisotropic expansion ( Figure 4 and Figure 4—figure supplement 3 ) . In contrast , the part of the meristem closest to the newly separated primordium ( P1 in Figure 4 ) was not distinguishable based on growth rates . As previously observed in Anagallis arvensis ( Kwiatkowska and Dumais , 2003; Kwiatkowska and Routier-Kierzkowska , 2009 ) , this accelerating growth of the peripheral zone progressively pushed away newly formed organs as they differentiated , making more space available on the meristem for further initiation and suggesting a possible feedback between lateral organ growth and meristem expansion . In addition to changes in geometry and growth , we used an activity reporter of the growth hormone auxin , pDR5::3xVENUS-N7 ( Heisler et al . , 2005 ) , to follow primordium development . Interestingly , while auxin activity is already visible at the first sign of primordium initiation , DR5 expression does not strictly correlate with growth . In particular , no DR5 signal is detected in the fast expanding regions close to older primordia . We also found that DR5 expression is present in the crease separating young primordia from the meristem , an area where the cells exhibited a slight decrease in surface area ( Figure 4D ) . As shown in previous studies ( Kwiatkowska and Dumais , 2003; Kwiatkowska , 2004; Kwiatkowska and Routier-Kierzkowska , 2009 ) , the quantification of growth anisotropy shows that cells in the boundary displayed a small increase in length only in the direction parallel to the border between meristem and primordium , suggesting compression by the growing organ ( Hamant et al . , 2008 ) ( Figure 4D and Figure 4—figure supplements 2 , 3 ) . The extraction of cellular 3D shape is of paramount importance for different purposes , such as to study volumetric deformation , quantify fluorescence expression in 3D , or generate cellular templates for 3D simulation models ( Bassel et al . , 2014; Yoshida et al . , 2014 ) ( Figure 5D ) . However , volumetric segmentation requires very high quality of signals , since the cell outlines must be visible from all angles . For plant tissues , which often display autofluorescence , 3D segmentation of cells from confocal images is therefore mainly used in the case of cleared , fixed samples ( Bassel et al . , 2014; Yoshida et al . , 2014 ) ( Figure 5B–D ) or limited to the outermost layers of cells ( Figures 5A , 6D ) . The penetration of confocal images for 3D segmentation of live samples could be improved by using multi-photon confocal microscopy . Another possibility is to combine confocal stacks acquired from different angles ( Fernandez et al . , 2010 ) . Currently it is possible to assemble data from multiple angle acquisition within MorphoGraphX . 10 . 7554/eLife . 05864 . 020Figure 5 . 3D segmentation for growth tracking and modeling templates . ( A ) Volume segmentation of trichomes from time-lapse confocal imaging in Capsella rubella leaf colored by cell label number . ( B ) Full 3D segmentation of developing Arabidopsis embryos , colored by cell label number . Data from ( Yoshida et al . , 2014 ) . ( C ) False colored projection of the average growth rate along the main axis of an Arabidopsis embryo . Data from ( Bassel et al . , 2014 ) . ( D ) Mechanical model of embryo based on a 3D mesh showing cell wall expansion due to turgor pressure , as published in ( Bassel et al . , 2014 ) . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 02010 . 7554/eLife . 05864 . 021Figure 6 . Validation of the method . ( A and B ) Control for viewing angle . ( A ) A shoot apex imaged from different directions . A first image stack ( in red ) was acquired before tilting the Z axis ( dashed lines ) by approximately 30° and acquiring a second stack ( in green ) . Cells were then segmented on both stacks and their areas compared ( B ) . Note that the pairwise cell size differences are random , with no obvious trend related to the viewing angle . Average error per cell is less than 2% . Colorbar: relative surface area increase in percent . Panels ( A ) and ( B ) adapted from Figure 5 of Kierzkowski et al . ( 2012 ) . ( C and D ) Comparison between projected areas and actual 3D volumes . ( C ) The epidermal cells of the apex were projected on the surface and segmented . Heatmap shows percent increase in area over 11 hr of growth . ( D ) The same data was segmented in 3D . Heatmap shows the percent increase in volume of cells , same color scale as in ( C ) . Note the close correspondence in cell expansion extracted from surface and volumetric segmentations . ( E ) Difference in size between automatically and manually segmented cells on a tomato shoot apex . Cells fused by auto-segmentation are in bright red , split cells are in dark blue . ( F ) Cell sizes ( in μm2 ) from manual ( top ) and automatic ( bottom ) segmentation on a fragment of Drosophila wing disc . Scale bars: 40 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05864 . 021 MorphoGraphX uses the auto-seeded , morphological watershed algorithm available in the Insight Segmentation and Registration Toolkit ( Yoo et al . , 2002 ) ( ITK ) for 3D segmentation . We have developed a collection of user-friendly 3D voxel editing tools allowing the researcher to correct segmentation errors . Alternatively , cells and other objects which are not in contact with each other can be segmented by extracting the surface of the individual objects ( Figure 1B , Video1 ) . As with 2D surfaces of cells , geometrical properties ( surface area , wall length , volume ) and fluorescent signal ( e . g . , total signal per cell , membrane localization ) of the 3D cells can be quantified ( Figure 5 ) and exported to spreadsheet files for further analysis ( Bassel et al . , 2014 ) . Cells segmented in 3D can also be exported for use in simulation models , where highly realistic geometries are required ( Figure 5D ) . When projecting data onto surface meshes several sources of error should be considered . Since the Z dimension in confocal images is typically considerably lower in resolution than in XY , it is possible that the view angle affects the results . To estimate the error introduced by this effect , we imaged the same sample twice from different angles ( Kierzkowski et al . , 2012 ) . Co-segmentation with approximately 30° difference in view angle lead to small segmentation differences , averaging to less than 2% ( Figure 6A , B ) . Note that there is no obvious bias from the view angle . Another potential source of error comes from representing 3D cells as a 2D surface . To estimate the error introduced by this abstraction , we co-segmented two time points of growth in the same tomato shoot apex as that shown in Figure 4 . The cells were segmented on the curved 2D surface , and the process was repeated from the same sample by segmenting the surface layer of cells in full 3D . The heat maps of volume increase in 3D show the same areas of slow and fast growth as the surface segmentation . In cases when the tissue thickness is preserved over growth , as in the epidermal layer of the shoot apex , tracking cell expansion on the surface is therefore a reasonable approximation for volumetric cell expansion ( Figure 6C , D ) . MorphoGraphX offers the possibility to segment cells automatically ( Video 3 ) or to place the seeds for watershed segmentation manually ( Video 2 ) . While automatic segmentation is faster in the case of high quality data , manual seeding is recommended in regions where part of the signal is too faint or blurry , partially masking the cell outline . To estimate the error in our auto-segmentation method , we compared the number of cells segmented automatically vs manually on the same region of two high quality samples . For a tomato shoot apex sample , the auto-segmentation error rate was about 2% , with only 12 cells under-segmented ( fused ) and 1 cell over-segmented ( split ) over a total of 576 cells ( Figure 6E ) . Once detected , segmentation errors can be easily fixed by the researcher ( Video 3 ) . Automatic seeding considerably shortens the time needed to segment large samples , such as a Drosophila wing disc ( Figure 6F ) . The total number of cells varied by about 3% ( 6304 autosegmented vs 6510 manually seeded cells ) .
A key strength of our MorphoGraphX software is the ability to accurately extract curved surface meshes from 3D volumetric data and perform image processing on the resulting curved ( 2 . 5D ) surface images . This has wide application , since many biological processes happen on surfaces , and the method has been proven in both animal ( Aegerter-Wilmsen et al . , 2012 ) and plant ( Santuari et al . , 2011; Chitwood et al . , 2012; Kierzkowski et al . , 2012; Nakayama et al . , 2012; De Rybel et al . , 2013; Kierzkowski et al . , 2013; Wabnik et al . , 2013; Sampathkumar et al . , 2014; Vlad et al . , 2014; Yoshida et al . , 2014 ) systems , in embryonic as well as mature tissues . The method is especially powerful for time-lapse confocal imaging , where laser exposure has to be kept to a minimum , limiting penetration to the outermost layers of the sample . In addition to curved surface image processing , MorphoGraphX provides an intuitive and user-friendly interface for the visualization and editing of 3D volumetric data , making it possible to digitally remove obstructing objects from the surface of interest , such as the peripodial membrane overlying the Drosophila wing disc ( Aegerter-Wilmsen et al . , 2012 ) . We have also included a range of standard 3D image processing tools , similar to those available in many other softwares ( Fernandez et al . , 2010; Peng et al . , 2010; Sommer et al . , 2011; Federici et al . , 2012; Mosaliganti et al . , 2012; Schmidt et al . , 2014 ) . These can be used for 3D segmentation ( De Rybel et al . , 2013; Bassel et al . , 2014; Yoshida et al . , 2014 ) , or to pre-process data before surface extraction . The modular design of MorphoGraphX allows the integration of existing libraries and the creation of custom processing ‘pipelines’ , going from the raw microscopy image to feature extraction and fluorescence quantification . MorphoGraphX is implemented as a collection of shared libraries , and new libraries can be added or removed without recompiling MorphoGraphX . This combines the functionality of plugins with the computational efficiency of C++ . The most common operations for 3D visualization , filtering and editing have been written to exploit the massively parallel architecture of modern graphics cards , which can have thousands of processing cores . As a result , 3D operations that would normally be very slow to run on a common PC take seconds to perform , making use of the computational power of inexpensive consumer graphics cards . Many of the more complex operations use the multi-core capabilities of the CPU . This makes most operations interactive and user-friendly , allowing the researcher to easily experiment with new work flows , algorithms and parameters . The flexibility of MorphoGraphX also simplifies the development of modules to import 3D voxel data and cellular or surfaces meshes from other custom imaging platforms . Such bridges have been created to import data from recently published growth tracking softwares including the MARS-ALT multi angle reconstruction pipeline ( Fernandez et al . , 2010 ) , and the stereo SEM reconstruction software ( Routier-Kierzkowska and Kwiatkowska , 2008 ) . MorphoGraphX has been used to quantify cell size ( Aegerter-Wilmsen et al . , 2012 ) , growth and proliferation ( Kierzkowski et al . , 2012; Vlad et al . , 2014 ) , mechanical properties ( Kierzkowski et al . , 2012 ) and protein localization ( Nakayama et al . , 2012; Kierzkowski et al . , 2013 ) , as well as 3D cell geometry ( De Rybel et al . , 2013; Bassel et al . , 2014; Yoshida et al . , 2014 ) . In addition to quantification , a current challenge in understanding development is to integrate these new data with computational models . Cellular geometry extracted from biological samples can be easily exchanged with modeling tools , such as Organism ( Sampathkumar et al . , 2014 ) and VVe ( Bassel et al . , 2014 ) . Meshes extracted in MorphoGraphX can be used directly for realistic simulation templates ( Bassel et al . , 2014 ) , or simplified depending on modeling requirements ( Santuari et al . , 2011; Wabnik et al . , 2013; Sampathkumar et al . , 2014 ) . Examples include simulation models of hormone transport ( Santuari et al . , 2011; De Rybel et al . , 2013; Wabnik et al . , 2013 ) , cell division plane analysis ( Yoshida et al . , 2014 ) and 3D cellular models of tissue mechanics ( Bassel et al . , 2014; Sampathkumar et al . , 2014 ) . MorphoGraphX was developed by researchers and designed to be easily adaptable to new research requirements . Its user interface was built in close collaboration with experimentalists , with features and techniques added to address research problems and bottlenecks in work flows as they have arisen . Fully automatic tools are complemented with intuitive methods for interactive correction ( Peng et al . , 2011 ) and validation , greatly increasing the utility of new and existing algorithms . Streamlined data exchange with modeling tools allows cell geometry and gene expression data to be used as model inputs , and facilitates the validation of simulation results . These features combine to make MorphoGraphX a significant step towards an interdisciplinary computational morphodynamics platform to study the interactions between growth , mechanics and gene expression .
Live confocal time-lapse series of developing flower of A . thaliana Col-0 ( Figure 2A–F and Figure 2—figure supplement 2 ) , shoot apical meristem of tomato ( Solanum lycopersicum ) DR5 reporter line ( Shani et al . , 2010 ) ( Figure 4—figure supplement 3 ) and leaf trichomes of Capsella rubella ( Figure 5A ) were acquired using SP8 or SP5 Leica confocal microscopes , as described previously ( Kierzkowski et al . , 2012; Vlad et al . , 2014 ) . After dissection samples were stained with 0 . 1% propidium iodide ( PI ) and grown in vitro on medium ( Bayer et al . , 2009 ) . Confocal imaging was performed with a 63× long distance water immersion objective and an argon laser emitting at the wavelength of 488 nm . PI signal was collected at 600–665 nm . In the case of tomato shoot apex , pDR5::3xVENUS-N7 signal was also collected , at 505–545 nm . Distance between stacks was 0 . 5 μm . Time intervals were 11 hr for tomato and 24 hr for A . thaliana and C . rubella time lapse series . Mature A . thaliana embryos ( Figure 2H ) were fixed and stained as previously described ( Bassel et al . , 2014 ) and imaged using a Zeiss LSM710 confocal microscope with a 25× oil immersion lens . Confocal stacks of microtubule marker line TUA6-GFP ( Ueda et al . , 1999 ) in live Cardamine hirsuta fruits ( Figure 3A ) were acquired using a SP2 Leica microscope , with a 40× long working distance water immersion objective and an argon laser emitting at 488 nm . GFP signal was collected at 495–545 nm . The z step between stack slices was 0 . 2 μm . The sequential replica method ( Williams and Green , 1988 ) was used to acquire a stereopair of SEM images from an Arabidospsis leaf surface ( Figure 1D ) as described in ( Elsner et al . , 2012 ) . Stereoscopic reconstruction ( Routier-Kierzkowska and Kwiatkowska , 2008 ) was then performed for the stereo pair and converted into a triangular mesh using a custom MorphoGraphX module . All other data presented in this manuscript were acquired for previously published work or available through on-line catalogs . MorphoGraphX is written in C++ and has been developed on GNU/Linux . For GPU processing , MorphoGraphX uses CUDA ( https://developer . nvidia . com/cuda-zone ) via the Thrust template library ( http://thrust . github . io ) . Multi-threaded host processing is done using OpenMP ( http://openmp . org/wp/ ) . CUDA requires a compatible nVidia ( http://www . nvidia . com ) graphics card . The user interface is designed in Qt4 ( http://qt-project . org/ ) , and OpenGL is used for 3D rendering ( http://www . opengl . org ) . MorphoGraphX can be extended using either C++ modules or Python scripts . C++ modules can be loaded at the start of MorphoGraphX through a plug-in system , inspired by the shared library loading architecture of Lpfg in VLab ( Federl and Prusinkiewicz , 1999 ) . C++ processes can access all the internal data structures used in MorphoGraphX and modify them as needed . They can also call other processes or query for their existence , and get parameter values in a uniform way from the graphical user interface . The last parameter values used for each process are stored in the project ( . mgxv ) file for future reference . All process calls and their parameters are logged to a re-playable python script log file created in the current directory . Each process is represented as a light C++ object defining the name , parameters and code of the process and is bundled in shared libraries for easy distribution . The shared library is placed into a system or user process folder , and the processes it contains are loaded upon startup . Python scripts can also be written and executed within MorphoGraphX using the Python Script process . This option offers a more limited interaction with MorphoGraphX as a script is only able to launch other processes and not directly interact with the data structure . However , it allows use of the wealth of modules existing for Python 2 . 7 for file interactions and data analysis . Most data analysis processes import/export their data as CSV files to facilitate the writing of Python modules for complex or ad-hoc data analysis . Surfaces are represented by vertex–vertex systems ( Smith et al . , 2004 ) , which implement graph rotation systems . Properties can be stored in the mesh , such as the label attributed to an individual vertex , the normal associated to it or a value for the projected signal . The rendering uses a modified front-to-back depth peeling technique ( Everitt , 2001 ) interweaving the volumetric rendering between peels of translucent surfaces . The volumetric rendering itself is done using volume ray casting ( Levoy , 1990 ) , using the depth of the successive pair of peels to limit the ray casting to the region currently being rendered . This method allows for correct polygon–polygon and polygon-volume intersections . Combined with occlusion detection , we implemented early ray termination when the total opacity of the current fragment becomes too high for subsequent peels to be visible . MorphoGraphX can be easily extended to import and export voxel and triangle mesh data in various formats . For voxel data , MorphoGraphX can read and write the tiff format compatible with ImageJ or Fiji ( Schindelin et al . , 2012 ) . 3D data can also be loaded from series of 2D images using any of the various image formats supported by the C++ Template Image Processing Toolkit ( CImg ) ( Tschumperlé , 2012 ) . The Visualization Toolkit ( VTK ) ( Wills , 2012 ) is used to import and export VTK triangle meshes . Various other formats , such as the Stanford Polygon File format ( . ply ) , Wavefront's Object format ( . obj ) or 3D Systems StereoLithography format ( . stl ) , are also supported directly . For many of the mesh imports , polygons with be converted to triangles upon loading by generating a center point and making a triangle fan . The first step in processing the data stacks is to remove noise and then identify the which voxels belong inside of the organ ( Figure 2A , B ) . 3D image processing filters for noise reduction are followed by edge detection combined with feature filling . Once the inside of the organ is identified it is represented as a binary image ( Figure 2B ) . Next the surface is extracted using a variant of the marching cubes algorithm ( Bloomenthal , 1988 ) . Fairly large cubes are used , creating a relatively coarse mesh and avoiding the extraction of very small features due to surface noise ( Figure 2C ) . Once a coarse surface mesh is extracted , it is uniformly subdivided . The resolution of this initial mesh has to be sufficient for a first segmentation , which can be subsequently refined . After the surface is extracted and subdivided , a column of signal normal to the surface is projected onto the mesh at every vertex , creating a 2D curved image of the cell outlines on the surface layer ( see Figure 2D , Video 2 ) . The image is segmented into cells using a seeded watershed segmentation algorithm . After blurring the image , auto-seeding is performed by finding local minima of signal within a given radius . Seeds are then propagated with watershed . Depending on the radius used for detecting the local minima , several seeds can be placed within a single cell , resulting in over-segmentation . The cells are later merged , based on the relative strength of signal on the walls separating them ( Video 3 ) . Normalization of the signal with a radius greater than that of the largest cell typically improves merging results . For convenience , the processes are chained together in a single auto-segmentation process . The final segmentation is then manually corrected . The amount of manual correction required can vary depending on signal quality , and in some cases it can be more efficient to perform some or all of the seeding manually . We have placed emphasis on designing the user interface for MorphoGraphX to streamline the process of manual seeding and segmentation correction ( Videos 2 , 3 ) . After the initial segmentation , the edges of the cells will often look rough , as there are not enough points to describe them correctly . To extract the geometry more precisely , the mesh can be subdivided specifically at the interfaces between cells ( Figure 2F ) or in areas of high signal intensity . After subdivision the signal is re-projected , and the surface segmented again . The seeds are retained during this process so that re-seeding is not required . Several steps of subdivision and re-segmentation can be applied until the desired precision is achieved ( Video 2 ) . The resulting mesh will be dense around the areas of interest ( e . g . , the interface between cells ) , while keeping the areas of low interest ( the inside of cells ) coarse , thus limiting the total size of the mesh file . Once the cells have been segmented from two different time points , the cells and their progeny can be identified manually . Each mesh is loaded in a separate channel and roughly aligned manually so that the cells outlines match . For each cell in the second time point , the user identify a mother cell with a mouse click ( Video 4 ) . The lineage information is then used to compare cell size ( areal growth ) or the projected signal intensity in the original cells and their daughters . A segmented mesh contains information about the cells neighborhood , that is , which are the cell walls shared by two cells and where do the cell walls intersect . The mesh can be simplified to contain only vertices necessary to describe each cell contour and the connections between neighbor cells ( Figure 4—figure supplement 1 ) . Plant cells do not slide with respect to each other , therefore the junction between cell walls can be used as landmarks to track tissue deformation over time series ( Green et al . , 1991 ) . Combined with the cell lineage information , the simplified cellular mesh ( Figure 4—figure supplement 1 ) is used to find the correspondence between cell junctions in meshes extracted from different time points ( Figure 4—figure supplement 2 ) . After identifying pairs of junctions conserved in both meshes using the lineage information , we project for each cell the junctions on the average cell plane and compute a best fit of the 2D transformation ( translation , rotation , anisotropic scaling ) that will minimize the squared distance between pairs of junctions ( Goodall and Green , 1986; Routier-Kierzkowska and Kwiatkowska , 2008 ) . Decomposing the transformation into singular vectors and values gives the PDGs and associated scaling values ( PDGmax , PDGmin ) , that describe anisotropic growth . Anisotropy values used in ( Figure 4 and Figure 4—figure supplements 2 , 3 ) were computed according to the following definition: anisotropy = PDGmax/PDGmin . The cellular mesh can also be used to compute other quantities , such as the tissue curvature ( Figure 3—figure supplement 1 and Figure 4—figure supplement 3 ) . In that case the vertices belonging to the cell outline are used to compute the principal curvatures for each cell center , within a given periphery . Color maps resulting from the computation of growth , curvature , signal quantification , etc . can be written to a spreadsheet giving easy access for further processing . Similarly , cell axis vectors can also be exported to be either re-rendered in MorphoGraphX or loaded for further analysis using other software , such as Matlab or Python . The ITK ( Yoo et al . , 2002 ) auto-seeded watershed segmentation algorithm implemented in MorphoGraphX was used to segment the cells in 3D in Figures 5 , 6D . After segmentation the cell surface is extracted using marching cubes and labeled . In some cases individual cells can also be segmented using a custom edge detect function from multiple angles ( Figure 1B , Video 1 ) . MorphoGraphX also provides the possibility to stitch stacks or combine multi angle stacks in 3D . However , this is not a pre-requisite for 3D segmentation in MorphoGraphX . | Animals , plants and other multicellular organisms develop their distinctive three-dimensional shapes as they grow . This process—called morphogenesis—is influenced by many genes and involves communication between cells to control the ability of individual cells to divide and grow . The precise timing and location of events in particular cells is very important in determining the final shape of the organism . Common techniques for studying morphogenesis use microscopes to take 2-dimensional ( 2D ) and 3-dimensional ( 3D ) time-lapse videos of living cells . Fluorescent tags allow scientists to observe specific proteins , cell boundaries , and interactions between individual cells . These imaging techniques can produce large sets of data that need to be analyzed using a computer and incorporated into computer simulations that predict how a tissue or organ within an organism grows to form its final shape . Currently , most computational models of morphogenesis work on 2D templates and focus on how tissues and organs form . However , many patterning events occur on surfaces that are curved or folded , so 2D models may lose important details . Developing 3D models would provide a more accurate picture , but these models are expensive and technically challenging to make . To address this problem , Barbier de Reuille , Routier-Kierzkowska et al . present an open-source , customizable software platform called MorphoGraphX . This software extracts images from 3D data to recreate curved 2D surfaces . Barbier de Reuille , Routier-Kierkowska et al . have also developed algorithms to help analyze growth and gene activity in these curved images , and the data can be exported and used in computer simulations . Several scientists have already used this software in their studies , but Barbier de Reuille , Routier-Kierzkowska et al . have now made the software more widely available and have provided a full explanation of how it works . How scientists can extend and customize MorphoGraphX to answer their own unique research questions is also described . It is anticipated that MorphoGraphX will become a popular platform for the open sharing of computational tools to study morphogenesis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"developmental",
"biology",
"tools",
"and",
"resources"
] | 2015 | MorphoGraphX: A platform for quantifying morphogenesis in 4D |
The 14th–18th century pandemic of Yersinia pestis caused devastating disease outbreaks in Europe for almost 400 years . The reasons for plague’s persistence and abrupt disappearance in Europe are poorly understood , but could have been due to either the presence of now-extinct plague foci in Europe itself , or successive disease introductions from other locations . Here we present five Y . pestis genomes from one of the last European outbreaks of plague , from 1722 in Marseille , France . The lineage identified has not been found in any extant Y . pestis foci sampled to date , and has its ancestry in strains obtained from victims of the 14th century Black Death . These data suggest the existence of a previously uncharacterized historical plague focus that persisted for at least three centuries . We propose that this disease source may have been responsible for the many resurgences of plague in Europe following the Black Death .
The bacterium Yersinia pestis is among the most virulent pathogens known to cause disease in humans . As the agent of plague it is an existing threat to public health as the cause of both emerging and re-emerging rodent-derived epidemics in many regions of the world ( Duplantier et al . , 2005; Vogler et al . , 2011; Gage and Kosoy , 2005 ) . This , and its confirmed involvement in three major historical pandemics , have made it the subject of intense study . The first pandemic , also known as the Justinian Plague , occurred from the 6th through the 8th centuries; the second pandemic spanned the 14th to the 18th centuries; and the third pandemic started in the 19th century and persists to the present day . Attempts to date the evolutionary history of Y . pestis using molecular clocks have been compromised by extensive variation in nucleotide substitution rates among lineages ( Cui et al . , 2013; Wagner et al . , 2014 ) , such that there is considerable uncertainty over how long this pathogen has caused epidemic disease in human populations . In addition , there has been lively debate as to whether or not it was the principal cause of the three historical pandemics ( Cohn , 2008; Scott and Duncan , 2005 ) . It is well established that extant lineages of Y . pestis circulated during the third pandemic ( Achtman et al . , 2004; Morelli et al . , 2010 ) , and various ancient DNA studies have now unequivocally demonstrated its involvement in the early phase of the first pandemic in the 6th century ( Harbeck et al . , 2013; Wagner et al . , 2014 ) and the Black Death ( 1347–1351 ) , which marks the beginning of the second plague pandemic ( Bos et al . , 2011; Haensch et al . , 2010; Schuenemann et al . , 2011 ) . All three pandemics likely arose from natural rodent foci in Asia and spread along trade routes to Europe and other parts of the world ( Morelli et al . , 2010; Wagner et al . , 2014 ) . The strains associated with the first and second pandemics represent independent emergence events from these rodent reservoirs in Asia . The on-going third pandemic also originated in Asia , although genetic evidence suggests that it may be derived from strains that descended from those associated with the second wave that spread back to Asia and became re-established in rodent populations there ( Wagner et al . , 2014 ) . The impact of the second pandemic was extraordinary . During this time period there were hundreds and perhaps thousands of local plague outbreaks in human populations throughout Europe ( Schmid et al . , 2015 ) . It is very likely that some of these outbreaks were caused by the spread of plague via the maritime transport of humans and cargo , as was undoubtedly the case during the global spread of plague during the third pandemic ( Morelli et al . , 2010 ) . However , the processes responsible for the potential long-term persistence of plague in Europe during the second pandemic are still subject to debate . It is possible that once introduced to Europe around the time of the Black Death , Y . pestis persisted there for centuries , cycling in and between rodent and human populations and being introduced or reintroduced to various regions throughout Europe ( Carmichael , 2014 ) . Another possibility is that plague did not persist long-term in European rodent populations , but rather was continually reintroduced from rodent plague foci in Asia ( Schmid et al . , 2015 ) . To address this key issue in Y . pestis evolution and epidemiology , we investigated plague-associated skeletal material from one of the last well-documented European epidemics , the Great Plague of Marseille ( 1720–1722 ) , that occurred in the Provence region of France at the end of the second pandemic . In particular , we compared the evolutionary relationship of these historical Y . pestis strains to those sampled from other time periods , both modern and ancient . Our results demonstrate that the strains responsible for the Black Death left descendants that persisted for several centuries in an as yet unidentified host reservoir population , accumulated genetic variation , and eventually contributed to the Great Plague of Marseille in mid-eighteenth century France . Although these lineages no longer seem to be represented within the genetic diversity of extant sampled Y . pestis , they may have been involved in additional , earlier second-pandemics in the Mediterranean region and beyond .
Using the pan-array design , we were able to identify a ~15 kb Genomic Island ( GI ) in the OBS strains as previously observed in the Justinian and Black Death strains ( Wagner et al . , 2014 ) and referred to as DFR 4 ( difference region 4 ) . Notably , this region has been lost in some Y . pestis strains such as CO92 . This island is a striking example of the decay common to the highly plastic Y . pestis genome .
The Observance lineage of Y . pestis identified by our analysis of material from the Great Plague of Marseille ( 1720–1722 ) is clearly phylogenetically distinct and has not been identified in any extant plague foci for which full genome data are available . That this lineage is seemingly confined to these ancient samples suggests that it is now extinct , or has yet to be sampled from extant reservoirs . Based on available data , the Y . pestis strains most closely related to the Observance lineage are those obtained from other ancient DNA studies of victims of the Black Death in London and other areas ( Haensch et al . , 2010; Bos et al . , 2011 ) that occurred some 350 years earlier . The most important conclusion from the current study is that Y . pestis persisted and diversified in at least two lineages throughout the second pandemic , one represented by our Observance strain , and the other by London St . Mary Graces individual 6330 . The absence of currently sequenced modern Y . pestis genomes that share an immediate phylogenetic relationship with the Observance lineage suggests that it may no longer exist . The length of the branch leading to the Observance lineage is consistent with a long history of circulation . Of note , this new lineage does not share the single derived mutation that is common to the London St . Mary Graces genome and the Bergen op Zoom plague victim from the Netherlands ( Haensch et al . , 2010 ) ; hence , multiple variants of Y . pestis evidently circulated in London and elsewhere in the 14th century , and at least one of these variants followed a path that differs from the lineage that later gave rise to the 18th century plague epidemic in Provence . Since the novel lineage of Y . pestis identified here was obtained from an active Mediterranean port city that served as a main commercial hub and entry point into Western Europe from various origins , the precise location of the disease’s source population cannot be easily determined . Several scenarios can explain these data . First , all Y . pestis diversity documented in the Black Death and later plague outbreaks could have stemmed from foci in Asia , where European epidemics were the result of successive introductions from this distant , albeit prolific , source population ( Schmid et al . , 2015 ) . This model would not require the back migration of plague into Asia for the third pandemic proposed elsewhere ( Wagner et al , 2014 ) , since all preexisting diversity would already be present . A multiple wave theory has previously been proposed to explain the presence of a reportedly different 14th century lineage in the Low Countries ( defined by their single 's12' SNP ) compared to those circulating in England and France ( Haensch et al . , 2010 ) . However , the presence of this branch 1 position in both the ancestral and subsequently the derived state in two temporally close London outbreaks ( Bos et al . , 2011; Figure 3B ) suggests that it became fixed during the Black Death or its immediate aftermath , rather than having been introduced via a separate pulse from an Asian focus . To date , all of the Y . pestis material examined from the second pandemic has been assigned to Branch 1 in the Y . pestis phylogeny , indicating a close evolutionary relationship . As the large “'big bang' radiation of Y . pestis at the beginning of the second pandemic ( Cui et al . , 2013 ) established at least four new lineages , it seems unlikely that independent waves of plague from Asia into Europe during the Black Death would have involved strains that group on the same branch and differ by so few positions . As China harbours many branch 1 descendent lineages that also share this 's12'” SNP , our observation adds weight to the notion that the branch 1 lineage of Y . pestis spread east sometime during the second pandemic after it entered Europe , and subsequently became established in one or more East Asian reservoir species ( Wagner et al . , 2014 ) . Here it remained before radiating to other locations in the late 19th century , giving rise to the current worldwide 'third wave' Y . pestis pandemic . In our view , a far more plausible option to account for the distinct position of the Observance lineage is that a Y . pestis strain responsible for the Black Death became established in a natural host reservoir population within Europe or western Asia . Once established , this Y . pestis lineage evolved locally for hundreds of years and contributed to repeated human epidemics in Europe . A similar scenario is possible for the lineage ( s ) identified in the London St . Mary Graces and Bergen op Zoom material: like Observance , both descend from the Black Death and together they represent at least one source of European plague distinct from that which later caused the Great Plague of Marseille . Plague’s temporary persistence somewhere on the European mainland is a compelling possibility ( Carmichael , 2014 ) . Alternatively , reservoir populations located in geographically adjacent regions , such as in western Asia or the Caucasus , may have acted as regular sources of disease through successive westerly pulses during the three century-long second pandemic ( Schmid et al . , 2015 ) . Since our study used material exclusively from a highly active port with trade connections to many areas in the Mediterranean and beyond , it is impossible to identify a source population for the Great Plague of Marseille at our current resolution . Our analysis reveals a previously uncharacterized plague source that could have fuelled the European epidemics from the time of the Black Death until the mid-eighteenth century . This parallels aspects of the third pandemic , that rapidly spread globally and established novel , long-lived endemic rodent foci that periodically emerge to cause human disease ( Morelli et al . , 2010 ) . The current resolution of Y . pestis phylogeography suggests that the historical reservoir identified here may no longer exist . More extensive sampling of both modern rodent populations and ancient human , as well as rodent , skeletal remains from various regions of Asia , the Caucasus , and Europe may reveal additional clues regarding past ecological niches for plague . The reasons for the apparent disappearance of this historic natural plague focus are , however , unknown , and any single model proposed for plague’s disappearance is likely to be overly simplistic . Pre-industrial Europe shared many of the same conditions that are correlated with plague dissemination today including pronounced social inequality , and poor sanitation coupled with high population density in urban centers ( Vogler et al . , 2011; Barnes , 2014 ) . This constellation of anthropogenic factors , along with the significant social and environmental changes that occurred during the Industrial Revolution must be considered alongside models of climate and vector-driven dynamics as contributing to the rapid decline in historical European plague outbreaks , where genomic data provide but one piece of critical information in this consilient approach .
A phylogenetic tree of Y . pestis strains was inferred using the maximum likelihood ( ML ) method available in PhyML ( Guindon et al . , 2010 ) , assuming the GTR model of nucleotide substitution ( parameter values available from the authors on request ) and a combination of NNI and SPR branch-swapping . The robustness of individual nodes was assessed using bootstrap resampling ( 1000 pseudo-replicates ) using the same substitution model and branch-swapping procedure as described above . The phylogeny comprised data from the five Observance ( OBS ) strains , one sequence from Black Death victims who died in London between 1348 and 1350 ( a combined pool of identical strains 8291 , 11972 , and 8124 ) , one sequence from a subsequent outbreak in London , 1350 – 1400 ( individual 6330 ) , one sequence from the Plague of Justinian , AD540 ( strain A120 ) , and 130 modern Y . pestis strains . A single sequence of Y . pseudotuberculosis ( strain IP32953 ) was used as an outgroup to root the tree , although with all derived SNPs removed to assist branch-length scaling . | A bacterium called Yersina pestis is responsible for numerous human outbreaks of plague throughout history . It is carried by rats and other rodents and can spread to humans causing what we conventionally refer to as plague . The most notorious of these plague outbreaks – the Black Death – claimed millions of lives in Europe in the mid-14th century . Several other plague outbreaks emerged in Europe over the next 400 years . Then , there was a large gap before the plague re-emerged as threat in the 19th century and it continues to infect humans today , though on a smaller scale . Scientists have extensively studied Y . pestis to understand its origin and how it evolved to become such a deadly threat . These studies led to the assumption that the plague outbreaks of the 14–18th centuries likely originated in rodents in Asia and spread along trade routes to other parts of the world . However , it is not clear why the plague persisted in Europe for 400 years after the Black Death . Could the bacteria have gained a foothold in local rodents instead of being reintroduced from Asia each time ? If it did , why did it then disappear for such a long period from the end of the 18th century ? To help answer these questions , Bos , Herbig et al . sequenced the DNA of Y . pestis samples collected from the teeth of five individuals who died of plague during the last major European outbreak of plague in 1722 in Marseille , France . The DNA sequences of these bacterial samples were then compared with the DNA sequences of modern day Y . pestis and other historical samples of the bacteria . The results showed the bacteria in the Marseille outbreak likely evolved from the strain that caused the Black Death back in the 14th century . The comparisons showed that the strain isolated from the teeth is not found today , and may be extinct . This suggests that a historical reservoir for plague existed somewhere , perhaps in Asia , or perhaps in Europe itself , and was able to cause outbreaks up until the 18th century . Bos , Herbig et al . ’s findings may help researchers trying to control the current outbreaks of the plague in Madagascar and other places . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Eighteenth century Yersinia pestis genomes reveal the long-term persistence of an historical plague focus |
We perform a comprehensive integrative analysis of multiple structural MR-based brain features and find for the first-time strong evidence relating inter-individual brain structural variations to a wide range of demographic and behavioral variates across a large cohort of young healthy human volunteers . Our analyses reveal that a robust ‘positive-negative’ spectrum of behavioral and demographic variates , recently associated to covariation in brain function , can already be identified using only structural features , highlighting the importance of careful integration of structural features in any analysis of inter-individual differences in functional connectivity and downstream associations with behavioral/demographic variates .
Understanding individual human behavior has attracted the attention of scientists and philosophers since antiquity . The first quantitative approach intended to deepen such understanding dates to the first half of the 19-th century when skull measures were related to human behavior or cognitive abilities ( Simpson , 2005; Fodor , 1983 ) . Technical , intellectual and clinical advances in the last two centuries allow us to now accurately quantify brain structure and function ( Lerch et al . , 2017; Huettel et al . , 2004; Friston et al . , 2002; Woolrich et al . , 2004; Rorden et al . , 2007 ) , and to summarize certain ‘aspects’ of human behavior by means of standardized tests . Such advances facilitate exploratory statistical learning analyses to uncover previously hidden relationships between brain features and human behavior , demographics or pathologies ( Poldrack and Farah , 2015 ) . These developments are expected to be pushed even further with the emergence of the big data magnetic resonance imaging ( MRI ) epidemiology phenomenon ( Van Essen et al . , 2013; Collins , 2012 ) , and some examples of such expectations have already reported associations with blood-oxygen-level dependent ( BOLD ) brain function ( Finn et al . , 2015; Smith et al . , 2015 ) ; for example , functional connectivity patterns can be used to identify individuals ( Finn et al . , 2015 ) , predict fluid intelligence ( Finn et al . , 2015 ) , or describe a mode of functional connectivity variation that relates to lifestyle , happiness and well-being ( Smith et al . , 2015 ) . Although the brain’s structural-functional relationships are not yet fully understood , linking structure to behavior is essential for either type of imaging modality to be fully interpretable as an imaging phenotype . Furthermore , given the long-term character of some demographic variables ( e . g . overall happiness ) , we hypothesize that different brain structural features , such as regional variation in the density of gray matter or subject-dependent degree of cortical expansion , should also reflect these relationships . To test these hypotheses , in this work we make use of the large quantity of high quality behavioral and neuroimaging data collected by one of the big data initiatives , the Human Connectome Project ( Van Essen et al . , 2013 ) ( HCP ) . The HCP sample includes detailed structural imaging , diffusion MRI , resting-state and several different functional MRI tasks for each subject . Furthermore , the availability of more than 300 behavioral and demographic measures ( Van Essen et al . , 2012 ) allows the post-hoc exploration of a wide range of associations ( Groves et al . , 2011 ) . We further hypothesize that behavioral variations can be explained by more general brain structure variations than isolated single feature variations ( e . g . cortical thickness variations ) ; we consequently extract multiple structural features from the different MR modalities and perform a simultaneous analysis by linked independent component analysis ( Linked ICA; Groves et al . , 2011; Groves et al . , 2012 ) . Linked ICA is a Bayesian extension of Independent Component Analyses developed for multi-modal data integration , where multiple ICA factorizations are simultaneously performed and all of them share the same unique mixing matrix . Such analyses increase statistical power by evidence integration across different features ( Wolfers et al . , 2017; Doan et al . , 2017 ) and have been shown to be powerful in identifying correlated patterns of structural and diffusion spatial variation that can then be studied in relation to individual behavioral and demographic measures ( Doan et al . , 2017; Douaud et al . , 2014; Francx et al . , 2016 ) . Although similar analyses have been previously performed ( Douaud et al . , 2014 ) , in this work we benefit from the unique characteristics of the data sample; we consider brain and behavioral data from close to 500 ‘healthy young adults’ which reduces common pathology- and age-related variance and increases the power to detect associations due to normal cross-sectional variability . Our results support the hypothesis that structural brain features are strongly associated with demographic and behavioral variates . Interestingly , the most relevant mode of inter-individual variations across brain structural measures identified through the multi-modal data fusion approach maps on to recent findings obtained using functional MRI data from the same HCP cohort . In particular , our findings closely resemble the ‘positive-negative’ set of behavioral measures identified in Smith et al . ( 2015 ) on the basis of functional ( co- ) variations . Using post-hoc analysis of the functional and structural modes we show that inter-individual differences attributed to brain function need to be reconsidered taking into account variations in brain structure across the cohort .
The multi-modal structural brain data analyses ( Figure 1 , operations A and B ) resulted in a total of 100 collections of component maps , each of which can be represented by a collection of 7 spatial maps covering the gray-matter space ( voxel-based morphometry feature ( VBM ) ) , diffusion skeleton space ( Fractional Anisotropy ( FA ) , Mean Diffusivity ( MD ) and Anisotropy Mode ( MO ) features ) , cortical vertex space ( cortical thickness ( CT ) and pial area ( PA ) features ) and a voxel-wise map of the Jacobian deformation ( JD ) . In addition , each collection of maps is associated with a single vector of contributions that describe the degree to which a given collection is ‘driven’ by the different modalities ( feature loadings ) . Finally , each collection is associated with a single vector that describes how each individual subject contributes to the component ( subject loadings ) . Post-hoc linear correlation analyses of these subject contributions with behavioral measures identified , after FDR correction ( Figure 1 , operations C , D and E , FDR corrected q < 2 . 2 × 10-4 ) , a total of 155 significant brain-behavior correlations , summarized by 30 components reflecting at least one significant relationship to behavior . We provide the full results in Supplementary file 2 and a brief summary in the bottom left panel of Figure 1 where we color code the significant Pearson correlation values for the components showing at least one Bonferroni corrected ( Bonferroni corrected q < 1 . 4 × 10-6 ) significant correlation to a behavioral or demographic measure . Only a single component ( number 6 ) shows strong associations across a broad set of behavioral domains ( 48 measures ) and across all structural modalities ( i . e . is not dominated by one of the structural data types in that no single modality contributes >50% to the total variance of the component ) . The relative contributions from the different modalities are 22% for VBM , 6% JD , 15% FA , 23% MD , 20% MO , 7% CT and 4% PA ( Appendix 1—figure 1 ) , reflecting a dominance of gray matter densities and diffusion measures and a lower involvement of the purely morphometric and cortical measures . Relating the behavioral associations , in Figure 1 bottom right panel we provide a summary of the behavioral measures significantly correlating with component six as well as the corresponding Pearson correlation values . Note that in the cases where several measures are grouped together we report their mean correlation value - full results are given in Supplementary file 2 . We observe that component number six relates to various behavioral scores including working memory , language function and general wellbeing ( life satisfaction , social support ) . In Figure 2 we present the associated spatial maps: VBM measures are most heavily weighted in bilateral orbitofrontal cortex , temporal pole , lingual gyrus and the putamen ( first row ) . Morphometric differences ( JD features ) load into temporal lobes , caudate and brainstem ( second row ) , and white matter tracts do most heavily weigh onto the internal capsule , anterior thalamic radiation and the anterior corona radiata ( 3rd , 4th , and 5th rows ) . Cortical effects ( 6th and 7th rows ) are largely associations with multi-modal association cortex that show effects whereas primary sensory cortices are not implicated . Note that the involvement of areas such as the putamen and lingual gyrus are relevant to explain the behavioral relationships found with working memory and word processing . Furthermore , the involvement of structural connections between subcortical and prefrontal areas as well as the orbitofrontal cortex and temporal poles could explain the link to more complex functions such as emotional support or life satisfaction . Note that each component considers structural multi-modal characterizations of the brain where each modality contains unique information and together builds into a multi-modal multivariate component . Consequently , although these results are hard to interpret as being nested into the same space as ( functional ) canonical brain networks , the structural weighting in gray matter modalities in orbitofrontal and temporal cortex , in conjunction with the white matter tracts that connect those regions , is a clear indication of an underlying network structure relating to component six and consequently , to its behavioral associations . Several other components also reflect behavioral patterns worth recognizing . In Figure 3 we report spatial maps associated with the components showing at least one Bonferroni corrected significant relationship with any behavioral measure ( p<1 . 4×10−6 ) . For components 1 , 2 and 89 we show spatial maps for all modalities contributing ( Appendix 1—figure 1 ) . In order to provide a clearer interpretation , for the other components we decided to show a selection of the relevant modalities and full NIfTI maps are separately available as supplementary material . Component one relates mainly to gender , physical strength and language and it is defined by significant changes in gray matter density ( VBM measures ) and cortical areal expansion ( PA measure ) . Its associated spatial patterns appear to in fact reflect brain size and cortical area differences in both temporal lobes . Similarly , component two is driven by VBM maps and correlates with variations in gender , age , height , weight and strength . Its spatial extent includes the paracingulate gyrus and bilateral insular and opercular cortex . Components 7 , 24 , 25 , 29 and 55 are driven by at least three feature modalities and they map into gender , weight , body mass and height . Component seven maps into gender and shows cingulate gyrus and insular cortex . Component 24 maps into weight and body mass and is mapped into putamen , intracalcarine cortex and thalamus . Components 25 and 29 relate height with the inferior temporal gyrus and the cerebellum together with strong DWI weightings in the brainstem . Component 55 relates to weight and maps into the precentral gyrus and asymmetric differences in DWI measures . Number 89 maps VBM and JD into hematocrit and involves the lingual and occipital fusiform gyrus . Finally , component 20 maps JD and VBM in the posterior midline into age and relationship status . Although another set of components show associations to behavior , these are limited to a single modality and/or a small set of behavioral variates ( Supplementary file 2 ) . Many of these components show simple relationships to overall size measures such as weight , body mass ( BMI ) or height , and the associations are weaker than those reported above; we consequently decided to not further discuss their spatial extent in this work and we provide full NIfTI images as supplementary material . To validate the robustness of the presented results to the model order choice we performed analyses at different dimensionalities and observe that especially lower indexed components are highly reproducible . In particular , component number six is recovered at dimensionalities 90 and 110 with a subject mode correlation value of around r ~ 0 . 9 ( details can be found in Appendix 1 , section ‘Robustness: model order’ ) . Regarding the influence of purely morphometric differences in the analyses , a comparative analysis excluding the JD revealed essentially unaltered brain-behavior associations . Analysis of the JD feature in isolation showed that no fully corrected significant association to the reported positive-negative structural mode is found when considering uniquely morphometric differences , even if considering several components together . However , uncorrected statistics suggests that information of the positive-negative mode could already be present at the morphometric level . These results are presented in Appendix 1 , section ‘Robustness: analyses without Jacobians’ , and ‘Robustness: Analyzing morphometric differences’ . Given the similar associations to behavior found between the presented structural mode ( component 6 ) and the ‘positive-negative’ functional mode reported in Smith et al . ( 2015 ) , and since both results are obtained using the HCP sample , we quantified the linear relation between them . With the analysis restricted to the 421 subjects common to both studies , we found that the structural and the functional subject modes are significantly correlated ( r = 0 . 4643 , r2 = 0 . 21 , permutation p<10−5 ) . Post-hoc correlation analyses to behavior replicated the original functional positive-negative mode by identifying 60 functional-behavioral relationships ( Smith et al . , 2015 ) ; we found that 22 of these behavioral measures are also associated to the structural mode and that there is no significant difference in the correlation values provided by the functional or the structural analyses at these intersecting behavioral measures ( details can be found in Appendix 1 , section: ‘On the power of structural and functional associations to behavior’ ) . To identify the linear dependence between the behavioral/demographic modes obtained from functional and structural data we used a generalized linear model ( GLM ) . We regressed the structural mode from the functional one and performed post-hoc linear correlation analysis of the residualised functional mode relative to behavioral variates as in Smith et al . ( 2015 ) . Note that structural features – due to the necessary co-alignment within the functional pipelines – acts as a mediator and therefore could induce significant imaging-to-behavior associations ( also see Bijsterbosch et al . , 2018 ) . Conversely , however , the structural features enter into the cross-subject analysis without any possible cross-talk from functional data , so that there is no possible interference from functional to structural features . The post-hoc correlation analysis of the residualised functional mode to behavior revealed a significant decrease in correlation ( mean r decrease = 0 . 078 , p<0 . 01 ) that result in the structural mode removing 73% of the 60 associations originally found using functional data . The remaining 16 significant relationships involve measures as handedness , education , tobacco use , list sorting , delay discount , and intelligence . As such , the two modes are significantly overlapping . We also performed a structural-functional Linked-ICA analyses where we added partial correlation matrices obtained from resting state fMRI to the set of originally considered structural features . We selected functional fMRI features to match Smith et al . ( 2015 ) ; details on data availability and processing are provided in Appendix 1 , section ‘Individual features pre-processing’ . This structural-functional analysis recovered the positive-negative mode reported in the originally reported multi-modal structural analyses . More concretely , we found a component , number 16 , significantly correlating ( r = 0 . 89 , p<10^−5 ) with the mainly reported structural mode ( component 6 ) . The contribution of each modality to this mode equals 20% for VBM , 15 . 6% for FA , 24 . 4% for MD , 23 . 9% for MO , 7% for CT , 3% for PA , 5% for JD and 0 . 0012% for the functional partial correlation feature . While all structural features reflect approximately the same contribution as in the original structural analyses , it is interesting that the functional data does marginally contribute to the found mode , suggesting that structure in its own can explain the positive-negative behavioral mode . As a final step in our analysis we performed a causal analysis between the structural mode ( component 6 ) and the functional mode reported in Smith et al . ( 2015 ) , on the basis of calculating pairwise likelihood ratios ( Hyvarinen and Smith , 2013 ) between the function-to-structure and structure-to-function model . This analysis estimated a likelihood ratio of ~0 . 04 , that is a significant structure to function causation effect ( p<0 . 0025 , using permutation testing ) ( Hyvarinen and Smith , 2013 ) ( for details see Appendix 1 , section ‘Testing structure-function causal effects’ ) . Replication of these causal results was achieved by performing an analogous multi-modal structural Linked ICA analyses considering this time the HCP1200 sample , and including only the subjects not used in the originally considered HCP500 sample . The analyses revealed a component that , restricted to the 331 intersecting subjects , showed once more a significant correlation with the positive-negative mode reported in Smith et al . ( r = 0 . 1773 , p<0 . 002 ) . Furthermore , consecutive causal analyses from the structural mode to the functional mode estimated a likelihood ratio of ~0 . 09 , that is a significant structure-to-function causation effect ( p<0 . 005 ) .
We present a simultaneous analysis of brain structural measures that reveals how several types of behavior and demographics link to variations in such measures of brain structure . Several components detect simple associations between brain size ( encoded in gray matter density and cortical area ) being related to gender , strength , endurance or language function . More interestingly , we encounter a single pattern of gray and white matter covariation that is strongly associated with several measures relating to cognitive function including working memory and language function , while also being strongly related to several measures of wellbeing including life satisfaction or emotional support . Accordingly , the spatial organization of the component that relates to these measures predominantly includes regions and connections that are relevant to working memory and word processing such as the putamen and lingual gyrus ( Mechelli et al . , 2000; Arsalidou et al . , 2013 ) . Additionally , the inclusion of regions such as the orbitofrontal cortex and temporal poles , as well as structural connections from subcortical to prefrontal regions , could explain the link to more complex functions such as emotional support and life satisfaction . Furthermore , the mode of structural variation we report here relates to several recently reported results obtained using functional MRI . In particular , our results relate to the ones presented in Finn et al . ( 2015 ) since it identifies fluid intelligence measures and it also shares many behavioral measures also identified by the ‘positive-negative’ mode reported in Smith et al . ( 2015 ) . Clearly , the functional analyses presented in Smith et al . ( 2015 ) and the one we present here , while using entirely different MRI measurements , are both able to get at the core of the same behavioral spectrum; in fact , the structural mode and the functional mode are strongly correlated subject measures ( r = 0 . 46 ) . Our analyses reliably augment the spectrum of behavioral variables reported by the functional analyses by extending it with many working memory , language , relational task , ASR and DSM measures ( Figure 1 bottom right and Supplementary file 2 ) . It is to note here that while the statistics reported in Smith et al . ( 2015 ) were obtained from a Canonical Correlation Analyses ( CCA ) between partial correlation matrices and all behavioral measures at once , the statistics we present here involve simple linear correlations . While the former type of analysis can benefit from the multi-variate type of analysis through the application of CCA , ensuing results can be hard to interpret . The straight-forward individual linear correlation analysis against the behavioral/demographic measures separately instead affords simple interpretation . These findings directly look into the relationship between brain structure and function . In fact , the functional mode of variation is strongly associated with connectivity in brain areas approximately resembling the Default Mode Network ( Smith et al . , 2015 ) and , given the spatial extent and the strong weight of the DWI data in the structural mode we report , it seems reasonable to assume that these white matter structure variations could contribute to the functional connectivity changes reported in Smith et al . ( 2015 ) . Further , we found no clear spatial overlap between the reported structural mode and the cortical functional extent of the ‘positive-negative’ mode , suggesting that integrated functional-structural analyses should increase the sensitivity of both functional and structural analyses . Further , these results might question whether group functional connectivity measures using fMRI provide direct measures of brain connectivity or are biased due to individual structural differences that may become ‘visible’ in the analysis of functional cross-subject . An analogous multimodal analysis excluding the JD feature provided equivalent results to those presented here ( Supplementary file 3 ) and unimodal analysis of only the JD features ( using simple ICA-based decomposition ( Beckmann and Smith , 2004 ) of the single JD modality ) did not provide significant correlation to the behavioral mode at the level of fully corrected statistics . These extra analyses confirm that the structural features relating to the behavioral mode are not uniquely driven by morphometric differences . The post-hoc correlation analysis of the residualised functional mode to behavior revealed a significant decrease in correlation ( mean r decrease = 0 . 078 , p<0 . 01 ) that result in the structural mode removing 73% of the 60 associations originally found using functional data . The remaining 16 significant relationships involve measures as handedness , education , tobacco use , list sorting , delay discount , or intelligence . As such , our results confirm that many associations previously attributed to functional connectivity are already present at the structural level . This could be interpreted in terms of a specialization of functional imaging towards a specific subset of behavioral measures for which it provides strong effects even after linear accounting for the structural findings , implying that not all previously identified associations can be explained through inter-individual differences in brain structure . As such , these two modes are significantly overlapping measures that are not fully reflected in the Jacobian deformation field . The presence of residual functional associations to behavior suggests that these associations - although possibly influenced by structural variation - cannot uniquely be attributed to simple morphometric differences . These results align with recent findings by Bijsterbosch et al . ( 2018 ) who show that individual spatial configurations extracted from functional MRI rather than the connectivity profiles between areas seem to stronger relate to the positive-negative mode . While the presence of residual associations could be interpreted as evidence for functional-structural integration , care needs to be taken with regards to the interpretation of these associations and changes thereof . First , note that all of these methods interrogate the linear relationships between variates . It is entire possible that the association between imaging phenotypes and behavioral/demographic measures involve non-linear relationships that remain at best incompletely accounted for within these analytical frameworks . Second , the implicit symmetry of linear correlations implies that a corresponding residualised analysis ( where we regress functional variations from the structural mode ) similarly removes significant associations . Indeed , in such a case only 7 out of 48 associations remain significant ( relating to weight , antisocial behavior ( DSM ) , family structure problems , relational task or adult self-report ( ASR ) questions , see Appendix 1 , section ‘On the power of structural and functional associations to behavior’ ) . These results suggest a segregation of different structural and functional specializations towards different behavioral measures with for example , intelligence being , not only , but more related to brain function , and antisocial or relational task measures relating more strongly to brain structure . Finally , a causal analysis revealed a significant structural to functional mode causation ( Hyvarinen and Smith , 2013 ) where the likelihood of the structural mode causally influencing the functional mode ( from Smith et al . , 2015 ) is >20 times higher than the likelihood of the reverse causation . Although the causal model introduced in Hyvarinen and Smith ( 2013 ) considers the residuals after linear modeling of a pair of signals , care is advised when considering causal inference on two vectors of observations , as we cannot exclude the possibility that unobserved underlying processes simultaneously influence brain structure and function ( ‘hidden causation’ ) . Nevertheless , these causal findings align with the fact that cross-subject analysis of functional data typically necessitates processing of structural data ( e . g . through co-registration into a common space ) . As such , structural variations will enter as mediating factor in any functional analysis pipeline and need to be accounted for suitably . However , there is no reverse influence of functional variations in the analysis of structural measures . Such dependencies remain poorly modeled in current analysis procedures and future work will have to focus on robustifying functional MRI analysis with regard to cross-subject variations in brain structure , for example by more advanced alignment procedures and/or through derivation of functional measures that are invariant under variations in structure . This will have important implications for the interpretation of future finding across neuroimaging ‘big data’ studies and will help improve our understanding of the functional-structural integration and its relation to behavioral associations .
To uncover relationships between the behavioral/demographic measures and the components obtained from the Linked-ICA decomposition we perform a correlation analysis between each independent component subjects’ contribution and each available behavioral measure . This operation is schematically summarized in Figure 1 operation C . To take into account the family structure present in the HCP sample while assessing significance we use the Permutation Analysis of Linear Models ( PALM ) ( Winkler et al . , 2014; Winkler et al . , 2015 ) and use 106 permutations per tested correlation ( Figure 1 operation D ) . We define significance at p<0 . 05 and address the multiple comparison by applying FDR correction ( Benjamini and Hochberg , 1995 ) as well as full Bonferroni correction ( Figure 1 operation E ) . | For years , scientists have tried to explain human behavior by measuring brain characteristics . During the first half of the 19th century , craniometry , the science of taking measurements of the skull , was a popular field of research and cognitive abilities as well as many behaviors were associated with different skull sizes and shapes . Although craniometry has been broadly discredited as a science , the study of brain structure and function , and their correlation to human behavior , continues to this day . Currently , one of the most powerful tools used in the study of the brain is magnetic resonance imaging ( MRI ) , which relies on strong magnetic fields and radio waves to produce detailed imaging . These images can provide functional information , by measuring changes in blood flow to different parts of the brain , as well as structural information such as the amount of gray or white matter or the size of different brain regions . Many studies have shown correlations between functional MRI ( fMRI ) data and behavioral and demographic traits , such as years of education , lifestyle habits or stress . Another advance in the study of the relationship between behaviors and the brain has been the emergence of better statistical analysis tools thanks to increasing computing power . These tools have made it possible to integrate data from different sources and analyze many variables at the same time , allowing patterns to emerge that would have been previously missed . Llera et al . have analyzed a large dataset from young healthy volunteers to show that changes in behavioral traits can be predicted by brain structure , and not just by brain function as previously shown . Different types of brain structural data , including what the surface of the brain looks like and relative volumes of gray and white matter , were integrated and analyzed , and correlations between changes in these variables and changes in the demographic and behavioral traits of the subjects were found . Previously , a robust relationship had been established between specific patterns of connections and activity in the brain and a group of characteristics such as life satisfaction , working memory , weight and strength , loneliness , family history of drugs and alcohol use , etc . Llera et al . show that this relationship also holds between the traits and structural brain data . As an example , there is a positive correlation between changes in the number of years of education and the income of the subjects and changes in a pattern of integrated structural data that include the amount of gray matter , white matter integrity and size of specific brain structures . Given these findings it becomes important to reconsider whether differences between individuals previously attributed to brain function could simply explained by the shape or size of the brain and its parts . These findings show that physical brain characteristics , including its size or the shape of its surface , could predict information such as individuals’ lifestyle decisions or their income; also implying that these characteristics are not simply a product of brain function . The results also demonstrate the power of combining different types of brain data to predict patterns in behavior . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Inter-individual differences in human brain structure and morphology link to variation in demographics and behavior |
Mechanical loading , such as caused by exercise , stimulates bone formation by osteoblasts and increases bone strength , but the mechanisms are poorly understood . Osteocytes reside in bone matrix , sense changes in mechanical load , and produce signals that alter bone formation by osteoblasts . We report that the ion channel Piezo1 is required for changes in gene expression induced by fluid shear stress in cultured osteocytes and stimulation of Piezo1 by a small molecule agonist is sufficient to replicate the effects of fluid flow on osteocytes . Conditional deletion of Piezo1 in osteoblasts and osteocytes notably reduced bone mass and strength in mice . Conversely , administration of a Piezo1 agonist to adult mice increased bone mass , mimicking the effects of mechanical loading . These results demonstrate that Piezo1 is a mechanosensitive ion channel by which osteoblast lineage cells sense and respond to changes in mechanical load and identify a novel target for anabolic bone therapy .
Mechanical signals play critical roles in bone growth and homeostasis ( Turner et al . , 2009; Ozcivici et al . , 2010 ) . Mechanical stimuli increase bone mass by stimulating the activity and production of bone forming osteoblasts ( Meakin et al . , 2014; Klein-Nulend et al . , 2012 ) . In contrast , loss of mechanical signals decreases bone mass by reducing bone formation and stimulating production of bone resorbing osteoclasts ( Kondo et al . , 2005; Nakamura et al . , 2013; Xiong et al . , 2011 ) . Osteocytes , which are cells buried in the bone matrix and derived from osteoblasts , are able to sense changes in mechanical load and orchestrate bone resorption and formation ( Bonewald , 2011; Klein-Nulend et al . , 2013 ) . However , the molecular mechanisms by which osteocytes sense changes in mechanical loads remain unclear . A variety of cell surface proteins and structures , including integrins , focal adhesions , and primary cilia , have been proposed to sense mechanical signals in bone cells ( Litzenberger et al . , 2010; Nguyen and Jacobs , 2013; Rubin et al . , 2006 ) . In addition , several lines of evidence suggest that ion channels are involved in the sensing of mechanical signals by osteocytes ( Hung et al . , 1995; Lu et al . , 2012; Lewis et al . , 2017; Li et al . , 2002 ) . For example , calcium influx is an early event following mechanical stimulus in osteocytes ( Hung et al . , 1995; Lu et al . , 2012 ) . Several calcium channels , including transient receptor potential channels ( TRPV ) and multimeric L-type and T-type voltage-sensitive calcium channels ( VSCC ) are expressed in osteoblasts and osteocytes ( Li et al . , 2002; Abed et al . , 2009; Shao et al . , 2005 ) . TRPV4 is perhaps the most studied calcium channel in bone ( Lee et al . , 2015; Masuyama et al . , 2008; Mizoguchi et al . , 2008; Suzuki et al . , 2013 ) . Although conditional deletion of Trpv4 in the osteoblast lineage has not yet been reported , Trpv4 germline knockout mice exhibit high bone mass , which is opposite of what would be expected with loss of mechanical responsiveness ( Masuyama et al . , 2008; van der Eerden et al . , 2013 ) . Histological analysis of these mice revealed decreased osteoclast number and a normal bone formation rate ( Masuyama et al . , 2008; van der Eerden et al . , 2013 ) , arguing against a role for TRPV4 as a mechanosensor in bone . Although mice with germline deletion of the L-type VSCC Cacna1d have reduced cross-sectional area in long bones , these mice respond normally to mechanical loading ( Li et al . , 2010 ) . Thus , heretofore , a definitive role for a specific calcium channel in the response of the skeleton to mechanical loading has not been demonstrated . Herein we sought to identify calcium channels involved in mechanosensation in osteocytes . We found that Piezo1 , a mechanosensitive ion channel , is highly expressed in osteocytes and its expression and activity were increased by fluid sheer stress . In addition , conditional deletion of Piezo1 in osteoblasts and osteocytes decreased cortical thickness and cancellous bone volume . Moreover , the skeletal response to anabolic loading was significantly blunted in mice lacking Piezo1 in osteoblasts and osteocytes . Importantly , administration of Yoda1 , a Piezo1 agonist , increased bone mass in vivo . Overall , our results suggest that osteoblasts , osteocytes , or both , sense and respond to changes in mechanical signals in part via activation of the Piezo1 calcium channel and identify activation of Piezo1 signaling as a potential therapeutic approach for osteoporosis .
To identify calcium channels that respond to mechanical signals in osteocytes , we compared gene expression profiles of the osteocytic cell line MLO-Y4 under static and fluid flow conditions by RNA-seq . Principal components analysis and volcano plot of transcripts indicated that a significant number of genes were differentially expressed in MLO-Y4 cells under static versus fluid shear stress ( Figure 1—figure supplement 1A , B ) . GO-enrichment analysis revealed enrichment in genes known to respond to mechanical signals , thereby validating the fluid flow experiment ( Figure 1—figure supplement 2A ) . We then identified differentially expressed genes related to calcium channels . Piezo1 was the most highly expressed among 78 calcium channels detected in MLO-Y4 cells under static condition ( Figure 1—figure supplement 2B ) . In addition , Piezo1 was also highly up-regulated by fluid flow in MLO-Y4 cells as determined by RNA-seq ( Figure 1A ) and RT-qPCR ( Figure 1B ) . The Piezo ion channel family consists of two members , Piezo1 and Piezo2 . While Piezo2 is expressed predominately in neurons , Piezo1 is mainly expressed in non-neuronal cells ( Murthy et al . , 2017 ) . Consistent with this previous evidence , the expression of Piezo1 was approximately 200-fold higher than that of Piezo2 in MLO-Y4 cells ( Figure 1B ) . Piezo1 expression was also much higher than Piezo2 in osteocyte-enriched cortical bone isolated from 12-week-old mice ( Figure 1C ) . Therefore , we focused our remaining analysis on Piezo1 . Knock-down of Piezo1 mRNA in MLO-Y4 cells significantly blunted the increase in intracellular calcium induced by fluid-flow ( Figure 1D ) . Knock-down of Piezo1 also blunted fluid-flow stimulation of Ptgs2 and Tnfrsf11b ( Figure 1E ) , two well-known targets of fluid shear stress in osteocytes ( Wadhwa et al . , 2002; Zhao et al . , 2016 ) . Conversely , overexpression of Piezo1 in MLO-Y4 cells increased the expression of Ptgs2 and Tnfrsf11b and enhanced their induction by fluid shear stress ( Figure 1F ) . These results demonstrate that Piezo1 contributes to the response of MLO-Y4 cells to fluid shear stress . To determine the role of Piezo1 in osteocytes in vivo , we deleted Piezo1 by crossing Piezo1f/f mice ( Cahalan et al . , 2015 ) with Dmp1-Cre transgenic mice , which express the Cre recombinase in osteoblasts and osteocytes ( Bivi et al . , 2012; Xiong et al . , 2015 ) . Deletion of the Piezo1 gene was confirmed by qPCR of genomic DNA isolated from osteocyte-enriched cortical bone ( Figure 2A ) . Mice lacking the Piezo1 gene in osteoblasts and osteocytes , hereafter referred to as Dmp1-Cre;Piezo1f/f mice , exhibited normal body weight compared to their control Piezo1f/f littermates ( Figure 2—figure supplement 1A ) . Both female and male Dmp1-Cre;Piezo1f/f mice exhibited low bone mineral density ( BMD ) at 5 , 8 , and 12 weeks of age as measured by dual energy x-ray absorptiometry ( DXA ) and the difference increased as the mice matured ( Figure 2B and Figure 2—figure supplement 1B ) . Since the three control groups , including wild-type ( WT ) , Dmp1-Cre , and Piezo1f/f littermates , displayed similar BMD , we used Piezo1f/f littermates as controls in the remaining studies . Spontaneous fractures were observed in the tibia of conditional knockout mice at a frequency of 0 . 16 ( Figure 2C ) . Detailed analysis of the skeletal phenotype of these mice at 12 weeks of age by micro-CT revealed that femoral cortical thickness was lower in Dmp1-Cre;Piezo1f/f mice compared with controls in both sexes ( Figure 2D , E and Figure 2—figure supplement 1C ) . Periosteal and endocortical circumferences were also decreased in the femur of Dmp1-Cre;Piezo1f/f mice ( Figure 2E ) . In line with these changes , the total cross sectional area , cortical bone area , and medullary area were reduced in the conditional knockout mice ( Figure 2—figure supplement 1D ) . In contrast to the changes in bone width , the length of the femurs was not different between genotypes indicating that longitudinal bone growth was normal in conditional knockout mice ( Figure 2—figure supplement 1E ) . A decrease in cortical bone thickness was also detected in vertebrae of Dmp1-Cre;Piezo1f/f female and male mice ( Figure 2F and Figure 2—figure supplement 1F ) . Analysis of cancellous bone in the femur and vertebra revealed that bone volume over tissue volume , trabecular number , and trabecular thickness were decreased , while trabecular separation was increased in female Dmp1-Cre;Piezo1f/f mice compared to their control littermates ( Figure 2G , H and Figure 2—figure supplement 1G , H ) . Similar results were obtained in male mice ( Figure 2—figure supplement 1I , J ) . Biomechanical testing by 3-point bending showed that the femurs from Dmp1-Cre;Piezo1f/f mice had reduced stiffness and ultimate force ( Figure 2I ) . However , the Young’s modulus and ultimate stress did not change , suggesting that the lower strength was due to differences in size and mass rather than changes in bone material properties ( Figure 2I ) . Consistent with this , the tissue mineral density of femoral cortical bone was unaffected by deletion of Piezo1 ( Figure 2J ) . To evaluate the cellular changes underlying the skeletal phenotype of the conditional knockout mice , we performed bone histomorphometry of femoral cortical bone and found that periosteal and endocortical mineralizing surfaces were significantly reduced in Dmp1-Cre;Piezo1f/f mice at 5 weeks , an age of rapid bone growth ( Figure 2K and Figure 2—figure supplement 2A ) . Bone formation at the outer ( periosteal ) surfaces of bone is a critical process for the enlargement of the skeleton . While double labels were easily seen in control mice , double labels were not observed in the conditional knockout mice , indicating that the bone formation rate at the periosteum of Dmp1-Cre;Piezo1f/f mice was extremely low . Histomorphometric analysis of vertebral trabecular bone also revealed a decrease in mineralizing surface , mineral apposition rate , and bone formation rate in the conditional knockout mice ( Figure 2L ) . In line with these changes , osteoblast number was lower in Dmp1-Cre;Piezo1f/f mice ( Figure 2M ) . In addition , we observed an increase in osteoclast number in the conditional knockout mice ( Figure 2M ) . To evaluate whether cell death could account for the changes seen with Piezo1 deletion , we measured the percentage of empty osteocyte lacunae and osteocyte number . We did not observe changes in the percentage of empty osteocyte lacunae or the number of osteocytes normalized to bone area in Dmp1-Cre;Piezo1f/f mice compared to their littermate controls ( Figure 2—figure supplement 2B , C ) . Consistent with these results , we did not observe any apparent morphological changes in osteocytes in the conditional knockout mice ( Figure 2—figure supplement 2D ) . In addition , knock-down of Piezo1 in MLO-Y4 cells decreased , rather than increased , Capase3 activity ( Figure 2—figure supplement 2E ) . These results indicate that Piezo1 deletion does not increase osteocyte death in vitro or in vivo . We also analyzed osteoblastogenesis in vitro and found normal osteoblast differentiation of bone marrow stromal cells from Dmp1-Cre;Piezo1f/f mice , as indicated by Alizarin Red staining ( Figure 2—figure supplement 2F ) . Since the Dmp1-Cre transgene also leads to recombination in a sub-population of muscle cells ( Lim et al . , 2017 ) , we measured Piezo1 deletion in gastrocnemius muscle , lean body weight , and gastrocnemius muscle mass to determine whether altered muscle mass could have contributed to the skeletal phenotype . We detected about 20% deletion of the Piezo1 gene in the conditional knockout mice ( Figure 2—figure supplement 3A ) . In addition , Piezo1 expression in gastrocnemius muscle was about 10 times lower than that in bone ( Figure 2—figure supplement 3B ) . More importantly , we did not observe any difference in lean body weight or gastrocnemius muscle mass between the conditional knockout mice and their control littermates ( Figure 2—figure supplement 3C , D ) . These results demonstrate that Piezo1 in osteoblasts , osteocytes , or both , is essential for normal bone size and mass . To determine whether Piezo1 in osteoblasts or osteocytes is required for the skeletal response to increased mechanical loading , we loaded the left tibia of 16-week-old female Dmp1-Cre;Piezo1f/f mice and their control littermates with +1200µε peak strain at the midshaft , as illustrated in Figure 3A . Two weeks of anabolic loading increased tibial cortical thickness in control mice but not in conditional knockout mice ( Figure 3B ) . Consistent with the changes in bone mass , loading increased periosteal bone formation rate in control mice , due to increases in both mineralizing surface and mineral apposition rate ( Figure 3C , D ) . The load-stimulated bone formation was significantly blunted in conditional knockout mice ( Figure 3C , D ) . These results suggest that Piezo1 in osteoblasts , osteocytes , or both , plays an essential role in the response of the skeleton to mechanical loads . To understand the molecular mechanisms by which Piezo1 increases bone mass , we compared expression of genes known to influence bone formation and resorption between Dmp1-Cre;Piezo1f/f mice and control littermates . Production of Wnt1 or the Wnt signaling inhibitor Sclerostin ( Sost ) by osteocytes represent critical stimulatory or inhibitory signals to bone formation , respectively ( Luther et al . , 2018; Li et al . , 2008 ) . Wnt1 mRNA was lower in cortical bone shafts of conditional knockout mice at both 5 and 12 weeks of age while the expression of Sost was unaffected ( Figure 4A , B ) . Consistent with increased osteoclast number , expression of the essential pro-osteoclastogenic cytokine RANKL ( Tnfsf11 ) was higher in the conditional knockout mice ( Figure 4B ) . In contrast , expression of OPG ( Tnfrsf11b ) , a secreted decoy receptor for RANKL , was not different between the genotypes ( Figure 4B ) , despite our observation of reduced Tnfrsf11b expression in MLO-Y4 cells lacking Piezo1 ( Figure 1E ) . The expression of Wnt1 can be stimulated by mechanical loading in mice ( Holguin et al . , 2016 ) . Therefore , we determined whether mechanical signals increase Wnt1 expression via Piezo1 . Fluid shear stress increased Wnt1 expression in MLO-Y4 cells but this was blunted after knock-down of Piezo1 ( Figure 4C ) . Basal expression of Wnt1 was also reduced by Piezo1 knock-down ( Figure 4C ) . YAP1 and TAZ are two related transcriptional cofactors that can be activated by mechanical signals , including fluid flow and matrix rigidity , and recently Piezo1 has been shown to control their activity ( Wang et al . , 2016; Dupont et al . , 2011; Pathak et al . , 2014 ) . We have shown previously that deletion of Yap1 and Taz using Dmp1-Cre decreases bone mass , due to both reduced bone formation and increased osteoclast number ( Xiong et al . , 2018 ) . Here , we analyzed the diaphysis of femurs of these mice and found that cortical thickness , periosteal circumference , and endocortical circumference were significantly decreased in Dmp1-Cre;Yap1f/f , Tazf/f mice compared to their Yap1f/f , Tazf/f littermates ( Figure 4—figure supplement 1A ) . Because these changes were similar to the ones seen in cortical bone of Dmp1-Cre;Piezo1f/f mice , we examined whether Piezo1 controls Wnt1 expression via YAP1 and TAZ . Silencing the Piezo1 gene in MLO-Y4 cells decreased the expression of Cyr61 , a YAP1 and TAZ target gene , and blunted fluid shear stress induction of Cyr61 expression ( Figure 4C ) . We then silenced the Yap1 and Taz genes in MLO-Y4 cells to examine whether these factors are required for the stimulation of Wnt1 by fluid shear stress . We found that lack of Yap1 and Taz blunted the response to fluid flow including the increase in Ptgs2 , Wnt1 , and Cyr61 expression ( Figure 4D ) . Knock-down of Piezo1 and Yap1/Taz was confirmed by mRNA abundance ( Figure 4—figure supplement 1B , C ) . Importantly , silencing Piezo1 blunted YAP1 activation caused by fluid shear stress , indicated by blunted nuclear translocation of YAP1 ( Figure 4E , F ) . Similarly , we deleted Piezo1 in UAMS-32 cells , a murine osteoblastic cell line , using CRISPR/Cas9 and found that expression of Ptgs2 , Wnt1 , and Cyr61 induced by fluid flow were blunted in Piezo1 knock out cells ( Figure 4—figure supplement 2 ) . To determine whether Piezo1 is required for Wnt1 expression induced by mechanical loading in vivo , we applied one bout of compressive loading on the tibia of Dmp1-Cre;Piezo1f/f mice and their Piezo1f/f littermates with +1200µε peak strain at the midshaft . Mechanical loading increased Wnt1 and Cyr61 expression in control mice ( Figure 4G ) . However , these increases were blunted in Dmp1-Cre;Piezo1f/f mice ( Figure 4G ) . Taken together , these results indicate that stimulation of Piezo1 by mechanical signals increases Wnt1 expression at least in part via activation of YAP1 and TAZ . Finally , we determined whether activation of Piezo1 is sufficient to mimic the effects of mechanical stimulation in osteocytes and bone . Treatment of MLO-Y4 cells with Yoda1 , a small molecule agonist of Piezo1 ( Syeda et al . , 2015 ) , increased intracellular calcium concentration ( Figure 5A ) , and stimulated expression of Ptgs2 , Wnt1 , and Tnfrsf11b ( Figure 5B ) , similar to the effect of fluid flow on these cells . Importantly , silencing of Piezo1 completely prevented the increase of intracellular calcium ( Figure 5A ) , as well as the changes in gene expression induced by Yoda1 ( Figure 5B ) . Likewise , silencing Yap1 and Taz in MLO-Y4 cells significantly blunted the increase of Ptgs2 , Wnt1 , and Tnfrsf11b by Yoda1 , indicating that the response to Yoda1 also requires YAP1 and TAZ ( Figure 5C ) . Yoda1 also promoted expression of Ptgs2 , Wnt1 , Tnfrsf11b , Cyr61 , and decreased Sost in cortical bone organ cultures from C57BL/6J mice ( Figure 5D ) . Importantly , Yoda1 increased Wnt1 expression in osteocyte-enriched cortical bone in vivo ( Figure 5E ) . These results demonstrated that Yoda1 mimics the response to fluid flow in authentic osteocytes . To determine whether Yoda1 is able to increase bone mass in vivo , we administered Yoda1 to 4-month-old female WT C57BL/6J mice for 2 weeks ( Figure 5F ) . Yoda1 did not alter body weight ( Figure 5—figure supplement 1A ) but increased cortical thickness and cancellous bone mass in the distal femur ( Figure 5G ) . Yoda1 also increased cortical thickness in the vertebra ( Figure 5H ) . However , we did not detect changes in cancellous bone volume in vertebrae ( Figure 5H ) . Consistent with the effect on bone mass , the serum levels of osteocalcin , a bone formation marker , were increased in Yoda1-treated mice ( Figure 5I ) . In contrast , we did not observe changes in the serum levels of CTX , a bone resorption marker ( Figure 5—figure supplement 1B ) . Our results demonstrate that activation of Piezo1 by Yoda1 mimics the effects of fluid shear stress on osteocytes and increases bone mass in mice .
Loss of function studies in epithelial cells have shown that Piezo1 responds to various forms of mechanical forces , including membrane stretch , static pressure , and fluid shear stress ( Li et al . , 2014; Gudipaty et al . , 2017; Miyamoto et al . , 2014 ) . Moreover , Piezo1 can be activated by mechanical perturbations of the lipid bilayer alone , demonstrating its role in mechanosensation ( Syeda et al . , 2016 ) . Here , the rapid response of MLO-Y4 cells to fluid shear stress is blunted by knocking-down Piezo1 indicating its important role in mechanosensation in bone cells . In addition , the basal skeletal phenotype of mice lacking Piezo1 in osteoblasts and osteocytes suggests that they have a reduced ability to respond to mechanical stimulation . Direct testing of this idea by performing an anabolic loading regime confirmed that the bones of the conditional knockout mice were less responsive to mechanical signals than controls . This decrease cannot be attributed to intrinsic cell defect since cell survive is not affected by Piezo1 deletion . Thus , our studies demonstrate that Piezo1 plays a critical role in sensing mechanical signals and maintaining bone homeostasis . In humans , truncation mutations in Piezo1 cause a recessive form of generalized lymphatic dysplasia but a musculoskeletal phenotype has not been reported ( Fotiou et al . , 2015 ) . Nonetheless , SNPs in the human Piezo1 locus are associated with low bone mineral density and increased fracture risk ( Morris et al . , 2019 ) . While preparing the revision of this manuscript , Sun et , al published a similar study in which Piezo1 was deleted from osteoblast lineage cells using BGLAP-Cre transgenic mice ( Sun et al . , 2019 ) . Similar to our studies , deletion of Piezo1 in osteoblast lineage cells resulted in a low bone mass phenotype . Importantly , loss of Piezo1 in osteoblast lineage cells blunted the bone loss caused by hind-limb suspension , supporting the idea that Piezo1 contributes to the skeletal response to mechanical stimulation . Deletion of Piezo1 from osteoblasts and osteocytes did not completely abolish the response of skeleton to mechanical stimulus . Thus Piezo1 is not the sole mechanosensor in osteoblasts and osteocytes . Other cell surface proteins and structures including integrins , focal adhesions , and primary cilia , also likely contribute to sensing mechanical signals in bone . Possible crosstalk between Piezo1 and these other sensors will need to be addressed in future studies . It is also possible that cells other than osteoblasts and osteocytes , such as osteoblast progenitors , sense changes in load and contribute to the increase in bone formation . It is important to note that , in addition to osteoblasts and osteocytes , the Dmp1-Cre transgene used in our study also causes recombination in skeletal muscle cells ( Xiong et al . , 2011; Lim et al . , 2017; Xiong et al . , 2015 ) . Therefore , it is possible that loss of Piezo1 in muscle cells also contributed to the skeletal phenotype we observed in the conditional knockout mice . However , lean body weight and muscle mass in the conditional knockout mice were unchanged , arguing against a role for muscle cells in the skeletal phenotype . In addition , the potent effects of Piezo1 gain- and loss-of-function in MLO-Y4 cells suggest that its effects are at least partly due to actions in osteocytes . Nonetheless , to distinguish between the possible contributions of Piezo1 in osteoblasts versus osteocytes , further studies using a Cre driver strain that is active in osteocytes but not in osteoblasts will be required . We identified Wnt1 as a potential downstream effector of Piezo1 . Previous studies have shown that mechanical loading increases Wnt1 expression in murine bone ( Holguin et al . , 2016; Kelly et al . , 2016 ) . Importantly , deletion of Wnt1 in osteoblasts and osteocytes using a Dmp1-Cre transgene produced a skeletal phenotype that resembles the one we observed by deletion of Piezo1 using the same Cre driver strain ( Joeng et al . , 2017 ) . Taken together , these results suggest that mechanical signals stimulate Wnt1 expression via activation of Piezo1 . The molecular pathways by which Piezo1 controls gene expression are only partially understood . Nonetheless , cell culture studies demonstrate that Piezo1 is required for YAP1 nuclear localization in neural stem cells ( Pathak et al . , 2014 ) . Consistent with this , we found that Piezo1 controls nuclear translocation of YAP1 induced by fluid flow in MLO-Y4 cells . YAP1 and TAZ have been implicated as mediators of the response to mechanical signals in a variety of cell types ( Dupont et al . , 2011; Hansen et al . , 2015 ) . Our finding that YAP1 and TAZ are required for stimulation of Wnt1 by fluid flow or Yoda1 suggests that mechanical activation of Piezo1 stimulates Wnt1 expression in osteocytes , at least in part , by activating YAP1 and TAZ . Consistent with this idea , deletion of Yap1 and Taz in mature osteoblasts and osteocytes caused a skeletal phenotype that was similar to deletion of Piezo1 , albeit less pronounced ( Xiong et al . , 2018 ) . The milder bone phenotype of Yap1/Taz conditional knockout mice suggests that YAP1 and TAZ are not the only downstream effectors of Piezo1 in osteoblast lineage cells . Similar to unloading , deletion of Piezo1 in osteoblasts and osteocytes led to not only decreases in bone formation , but also increases in RANKL expression and bone resorption . Indeed , increased RANKL expression as well as osteoclast number have been observed in hind-limb unloaded mice ( Xiong et al . , 2011 ) . In our previous studies , we detected an increase in osteoclast number in mice that lack Yap1 and Taz in osteoblasts and osteocytes ( Xiong et al . , 2018 ) , suggesting that YAP1 and TAZ are downstream effectors of Piezo1 in controlling osteoclast formation . Thus , loss of Piezo1 in osteoblasts and osteocytes mimics the overall effect of unloading on the skeleton , further supporting the idea that Piezo1 is a mechanosensor in bone . Activation of Piezo1 using the small molecule Yoda1 mimics the effects of fluid flow in various cell types including endothelial cells , erythrocytes , platelets , and smooth muscle cells ( Cahalan et al . , 2015; Li et al . , 2014; Ilkan et al . , 2017; Rode et al . , 2017 ) . In addition , Yoda1 administration promotes lymphatic valve formation during development ( Choi et al . , 2019 ) . Here , we showed that Piezo1 activation by Yoda1 mimics the impact of mechanical stimulation in cultured osteocytic cells as well as ex vivo bone organ cultures . More importantly , administration of Yoda1 to mice increased bone mass and elevated a bone formation marker in the circulation , demonstrating that activation of Piezo1 is a potential target for anabolic bone therapy . One possible limitation of such an approach would be the functions of Piezo1 in other tissues , such as the vasculature . However , it is important to note that bone anabolism requires only transient mechanical stimulation of the skeleton in rodents or humans ( Vlachopoulos et al . , 2018; Hinton et al . , 2015; Kato et al . , 2006 ) . Therefore , it is possible that selectivity for bone anabolism may be achieved by administration regimes that result in only transient activation of Piezo1 by ligands such as Yoda1 . In summary , our studies demonstrate a critical role for Piezo1 in the maintenance of bone homeostasis and suggest that this occurs via mediation of mechanosensation in osteoblasts , osteocytes , or both . Our finding that activation of Piezo1 mimics the effects of mechanical stimulation on bone cells and increases bone mass in mice sets the stage for exploration of this pathway as a therapeutic target for osteoporosis .
The generation of mice harboring Piezo1 conditional allele , termed Piezo1f/f mice , was described previously ( Cahalan et al . , 2015 ) . Mice harboring both Yap1 and Taz conditional alleles , termed Yap1f/f;Tazf/f mice were kindly provided by Eric N . Olson ( UT Southwestern Medical Center , Texas ) and were described previously ( Xin et al . , 2013 ) . The 8 kb Dmp1-Cre transgenic mice were described previously ( Bivi et al . , 2012 ) . To generate Dmp1-Cre; Piezo1f/f mice and littermates , we mated Piezo1f/f mice ( crossed into C57BL/6J for more than 10 generations ) and Dmp1-Cre mice ( crossed into C57BL/6J for more than 10 generations ) . Dmp1-Cre; Yap1f/f , Tazf/f mice and littermates were obtained by mating Yap1f/f , Tazf/f mice ( mixture of 129/Sv and C57BL/6J ) and Dmp1-Cre mice ( crossed into C57BL/6J for more than 10 generations ) . We housed all mice in the animal facility of the University of Arkansas for Medical Sciences . Animal studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . Animal use protocols ( 3782 , 3805 , and 3897 ) were approved by the Institutional Animal Care and Use Committee ( IACUC ) of the University of Arkansas for Medical Sciences . All of the animals were handled according to approved protocols . To quantify cancellous bone formation , we injected mice with calcein ( 20 mg/kg body weight ) intraperitoneally 7 and 3 days before harvest . To quantify periosteal and endocortical bone formation , we injected mice with calcein ( 20 mg/kg body weight ) and Alizarin Red ( 20 mg/kg body weight ) 10 and 3 days before harvest . For gene expression , we injected Yoda1 into 4-month-old female C57BL/6J mice one time and harvested tibiae 4 hr later for RNA extraction . For bone mass evaluation , we injected Yoda1 into 4-month-old female C57BL/6J mice five consecutive days per week for 2 weeks ( day 1–5 and day 8–12 ) and harvested the mice at day 15 for analysis . Yoda1 ( Sigma , St . Louis , MO ) was dissolved in DMSO at 40 mM as a stock , diluted in 5% ethanol , and injected intraperitoneally at 5 µmol/kg body weight . Mice were rank-ordered by body weight and then assigned to Veh or Yoda1 groups to give identical group means . All investigators involved in data collection were blinded as to the genotype and group of the mice . HEK 293 T cells were authenticated by ATCC . MLO-Y4 cells were created and authenticated in Dr . Lynda Bonewald’s lab ( Kato et al . , 1997 ) . We tested the MLO-Y4 cells by morphology and osteocytic gene expression such as RANKL and OPG . UAMS-32 cells were created and authenticated by Dr . Charles O’Brien ( O'Brien et al . , 1999; Fu et al . , 2002 ) . Cells were treated with plasmocin to prevent potential mycoplasma contamination . MLO-Y4 cells were cultured in α-MEM supplemented with 5% FBS , 5% BCS , and 1% penicillin/streptomycin/glutamine . Fifteen dynes/cm2 oscillatory fluid shear stress was applied on MLO-Y4 cells at 1 Hz for 2 hr using an IBDI pump system ( IBIDI , Germany ) . For Yoda1 treatment , cells were cultured in the presence of 10 µM Yoda1 ( Sigma , St . Louis , MO ) or DMSO for 2 hr . Immediately after the treatments , we isolated RNA from cells using RNeasy mini kit ( Qiagen , Germany ) for qPCR or RNA-seq analysis . To silence Piezo1 , we generated Piezo1 shRNA expression plasmid using the following oligonucleotides in the pLKO . 1-TRC cloning vector ( Addgene Plasmid #10878 , a gift from David Root ) : forward oligo: 5’-CCGGTC-GGCGCTTGCTAGAACTTCACTCGAGTGAAGTTCTAGCAAGCGCCGATTTTTG-3’; reverse oligo: 5’- AATTCAAAAATCGGCGCTTGCTAGAACTTCACTCGAGTGAAGTTCTAGCAAGCGC-CGA-3’ ( Zhang et al . , 2017 ) . Yap1 shRNA ( TRCN0000238432 ) and Taz shRNA ( TRCN0000095951 ) were purchased from Sigma ( St . Louis , MO ) . A shRNA against firefly luciferase was used as a control ( Sigma , St . Louis , MO ) . For virus production , HEK293T cells were cultured in a 6-well culture plate and co-transfected with a total 3 μg of lentiviral shRNA vector , pMD2G ( Addgene plasmid #12259 , a gift from Didier Trono ) , and psPAX2 ( Addgene plasmid # 12260 , a gift from Didier Trono ) at the ratio of 2:0 . 9:0 . 4 using TransIT-LT1 transfection reagent ( Mirus , Madison , WI ) . Culture media was changed 12 hr after transfection and viral supernatants were collected 48 hr after media change . Viral supernatants were filtered through a 0 . 45 µm filter and used immediately to transduce cells cultured in a 10 cm dish . Cells were then subjected to selection with G418 ( 100 µg/ml ) or puromycin ( 25 µg/ml ) for 5 days before treatment . To overexpress Piezo1 in MLO-Y4 cells , we transfected mPiezo1-IRES-eGFP ( Addgene plasmid # 80925 , a gift from Ardem Patapoutian ) into MLO-Y4 cells using TransIT-LT1 transfection reagent ( Mirus , Madison , WI ) and then treated these cells with 15 dynes/cm2 oscillatory fluid shear stress at 1 Hz for 2 hr . Plasmids for expression of Cas9 and sgRNAs for knocking out Piezo1 in UAMS-32 cells were prepared by inserting oligonucleotides encoding the desired sgRNA sequence into the pX458 vector using the protocol recommended by the Zhang laboratory ( Cong et al . , 2013 ) . Plasmids expressing Cas9 and Piezo1 sgRNAs were transfected into UAMS-32 cells using TransIT-LT1 transfection reagent ( Mirus , Madison , WI ) . Cells were sorted into 96-well plates for single cell cloning 48 hr after transfection . Single cell colonies were then screened for homozygous deletion using the following primers: Forward: 5’-GCTGTCAGGGTAAGCAGTATC-3’ , Reverse: 5’-GGAATATGAGGACAGCAGTCC-3’ . All homozygous mutant cell colonies were then pooled together for further analysis . Cas9 transfected cells were used as a control . All in vitro cell culture experiments were performed three times with three technical replicates . Female mice at 5 weeks of age were euthanized in a CO2 chamber . Femurs were dissected and both ends were removed in a culture hood . Bone marrow was then flushed out using PBS and the periosteal surface was scraped to remove periosteal cells . Femoral shafts were then cultured in a 12-well-plate with 1 ml of α-MEM supplemented with 10% FBS and 1% penicillin/streptomycin/glutamine for 24 hr . We then treated femur shafts with 10 µM Yoda1 ( Sigma , St . Louis , MO ) or DMSO for 4 hr . Femur shafts were then collected for RNA isolation and qPCR analysis . Ex vivo femoral organ culture was repeated twice with three biological replicates . Bone marrow stromal cells were flushed out from long bones , collected into a 50 ml cubical tube , and filtered through a 40 µm cell strainer to obtain a single cell suspension . Bone marrow stromal cells were then seeded into a 12-well-plate at 5 × 106 cells/well and cultured in α-MEM containing 10% fetal bovine serum , 1% penicillin/streptomycin/glutamine , 1% ascorbic acid , and 10 mM β-glycerolphosphate . Culture medium was changed every 3 days . After 21 days , the cultures were fixed with 10% buffered formalin and stained with an aqueous solution of 40 mM Alizarin Red to evaluate osteoblastogenesis . Purified RNA was used as input for sequencing library preparation and indexing using the TruSeq stranded mRNA kit ( Illumina , CA ) , following the manufacturer’s protocol . The libraries were then pooled and sequenced using a NextSeq sequencer with 75 cycles of sequencing reaction . Data handling and processing were performed on the basis of a previous bioinformatics pipeline ( Nookaew et al . , 2012 ) . The high-quality reads ( phred quality score , >25; length after trimming , >20 bases ) were obtained with the dynamic trimming algorithm in the SolexaQA++ toolkits ( Cox et al . , 2010 ) , and aligned with the mouse genome version GRCm38 using BWA software ( Li and Durbin , 2009 ) . Then the alignment files ( . bam ) were used to generate read counts for statistical analysis . The differential gene expression analysis was performed using negative binomial based statistic ( Love et al . , 2014 ) . The adjusted p-values were used for gene enrichment analysis based on Gene Ontology using the piano package ( Väremo et al . , 2013 ) . Raw RNA-seq results have been deposited in GEO database under BioProject PRJNA551282 with accession numbers: SRR9598498 , SRR9598497 , SRR9598496 , SRR9598495 , SRR9598494 , and SRR9598493 . Detailed RNAseq analysis was shown in Supplementary file 1 . For intracellular calcium concentration measurement under fluid flow condition , 1 × 105 MLO-Y4 cells were seeded in a µ-Slide I Luer ( 0 . 4 mm ) fluid chamber slide ( IBIDI , Germany ) overnight . One hour before initiating fluid flow , the culture medium was removed and 100 µl Hank's Buffer with Hepes ( HHBS ) containing 4 µM Fluo-8 ( Abcam , Cambridge , MA ) was added to the culture , as described by the manufacturer . The cells were then cultured at 37°C for 30 min . After additional incubation at room temperature for 30 min , the chamber slide was placed under a confocal microscope in order to record the intensity of fluorescence of MLO-Y4 cells . Fluorescence was recorded for 3 min before starting fluid flow using HHBS and then recorded for another 10 min . The increase of intracellular concentration was calculated by subtracting the initial mean fluorescence . For measuring intracellular calcium concentration in cells with Yoda1 treatment , we cultured 4 , 000 MLO-Y4 cells per well in a 96-well-plate . We preloaded the cells with Fluo-8 as described by the manufacturer and read the intensity of the fluorescence using a Victor X3 multi-label plate reader ( Perkin Elmer , Waltham , MA ) immediately after the treatment . We measured the fluorescence for 5 min at an interval of 20 s . The percentage of increase in intracellular calcium concentration was calculated as ( Fx-F0 ) /F0 . Tibial X-rays were obtained using an UltraFocus X-ray machine ( Faxitron Bioptics , Tucson , Arizona ) and BMD of the lumbar spine and femur were measured by dual-energy X-ray absorptiometry using a PIXImus Densitometer ( GE-Lunar Corp . ) Three dimensional bone volume and architecture of L4 vertebra , femur , and tibia were measured by µCT ( model μCT40 , Scanco Medical , Wayne , PA ) . The femur , vertebrae ( L4 ) , or tibia , were cleaned of soft tissues and fixed in 10% Millonig’s formalin for 24 hr . Bone were then gradually dehydrated into 100% ethanol . Bone samples were loaded into a 12 . 3 mm diameter scanning tube and images acquired in the μCT40 . The scans were integrated into 3D voxel images ( 1024 × 1024 pixel matrices for each individual planar stack ) and a Gaussian filter ( sigma = 0 . 8 , support = 1 ) was used to reduce signal noise . Scanco Eval Program v . 6 . 0 was used for measuring bone volume . A threshold of 220 mg/cm3 was applied to all scans at medium resolution ( E = 55 kVp , I = 145 µA , integration time = 200 ms ) for trabecular bone measurements . The cortical bone and the primary spongiosa were manually excluded from the analysis . Trabecular bone measurements at the distal femur were made on 151 slices beginning 8–10 slices away from the growth plate and proceeding proximally . Trabecular bone measurements in the vertebra was determined using 100 slices ( 1 . 2 mm ) of the anterior ( ventral ) vertebral body immediately inferior ( caudal ) to the superior ( cranial ) growth plate . All trabecular measurements were made by drawing contours every 10 to 20 slices and voxel counting was used for bone volume per tissue volume and sphere filling distance transformation indices , without pre-assumptions about the bone shape as a rod or plate for trabecular microarchitecture . Femoral cortical thickness , periosteal circumference , and endocortical circumference were measured at the mid-diaphysis . For tibial cortical thickness , we analyzed 18 slices 5 mm proximal from the distal tibiofibular junction . Vertebral cortical bone thickness was determined on the ventral cortical wall using contours of cross-sectional images , drawn to exclude trabecular bone . Cortical analysis were measured at a threshold of 260 mg/cm3 . Calibration and quality control were performed weekly using five density standards and spatial resolution was verified monthly using a tungsten wire rod . We based beam-hardening correction on the calibration records . Corrections for 200 mg hydroxyapatite were made for all energies . Lumbar vertebrae were fixed for 24 hr in 10% Millonig’s formalin , dehydrated into 100% ethanol , embedded in methyl methacrylate , and then 5 μm longitudinal sections were obtained . After removal of plastic and rehydration , we stained sections for TRAP activity and counter-stained with T-blue . Quantitative histomorphometry was performed to determine osteoblast and osteoclast number using Osteomeasure system ( OsteoMetrics , Decatur , GA ) interfaced to an Axio image M2 ( Carl Zeiss , NY ) . Bone formation rate was measured using unstained sections in Osteomeasure system . We used terminology recommended by the Histomorphometry Nomenclature Committee of the American Society for Bone and Mineral Research ( Dempster et al . , 2013 ) . For quantification of periosteal and endocortical bone formation , femurs or tibiae were fixed in 10% Millonig’s formalin for 24 hr , dehydrated into 100% ethanol , embedded in methyl methacrylate , and then 80 μm cross sections were obtained at the femoral mid-diaphysis for femoral sections and 5 mm proximal from the distal tibiofibular junction for tibial sections . We then measured mineralizing surface and mineral apposition rate using the Osteomeasure system . A cyclic axial load was applied to left tibia of mice to achieve +1200 µε peak strain at the tibial midshaft using an Electroforce TA 5500 ( TA Instruments , New Castle , DE ) . To determine the required load to achieve +1200 µε peak strain for each genotype of experimental mice , axial loading was applied to harvested tibiae ex vivo . A single-element strain gauge ( C2A-06-015LW-120 , VPG Micro-Measurements , Wendell , NC ) was attached to the antero-medial surface of the tibia located 5 mm proximal from the distal tibiofibular junction using M-Bond 200 adhesive kit ( VPG Micro-Measurements ) . We recorded the force-strain regressions using Electroforce TA 5500 software . We then applied the same amount of load to mice in vivo according to their genotype ( 8 . 5 Newton for Piezo1f/f mice and 7 . 5 Newton for Dmp1-Cre; Piezo1f/f mice ) . The left tibia of each mouse was loaded for five consecutive days per week for 2 weeks ( day 1–5 and day 8–12 ) , and the load was applied in 1200 cycles with 4 Hz triangle waveform and 0 . 1 s rest time between each cycle , a protocol shown to be anabolic ( Sun et al . , 2018 ) . We injected calcein ( Sigma , St . Louis , MO ) and Alizarin Red ( Sigma ) intraperitoneally into mice 10 days and 3 days before euthanasia to label new bone formation . We euthanized the mice and collected tissues at day 15 for skeletal analysis . For gene expression analysis , we loaded left tibia of 4-month-old female mice with one bout of loading and harvested tibiae 5 hr after loading for RNA extraction . We performed three-point bending test on femurs at room temperature using a miniature bending apparatus with the posterior femoral surface lying on lower supports ( 7 mm apart ) and the left support immediately proximal to the distal condyles . Load was applied to the anterior femoral surface by an actuator midway between the two supports moving at a constant rate of 3 mm/min to produce a physiological in vivo strain rate of 1% for the average murine femur . Maximum load ( N ) and displacement ( mm ) were recorded . The external measurements ( length , width and thickness ) of the femora were recorded with a digital caliper . We measured the moment of inertia in the midshaft of femur using µCT ( model μCT40 , Scanco Medical ) . The mechanical properties were normalized for bone size and ultimate strength and stress ( N/mm2; in megapascals and MPa ) was calculated . Organs and whole bones were harvested from animals , removed of soft tissues , and stored immediately in liquid nitrogen . We prepared osteocyte-enriched bone by removing the ends of femurs and tibias and then flushing the bone marrow with PBS . We then scraped the bone surface with a scalpel and froze them in liquid nitrogen for later RNA isolation , or decalcified them for genomic DNA isolation . We isolated total RNA using TRIzol ( Life Technologies , NY ) , according to the manufacturer’s instructions and prepared cDNA using High Capacity first strand cDNA synthesis kit ( Life Technologies ) . We performed quantitative RT-PCR using the following Taqman assays from Applied Biosystems: Piezo1 ( Mm01241549_m1 ) ; Piezo2 ( Mm01265861_m1 ) ; Ptgs2 ( Mm00478374_m1 ) ; Cyr61 ( Mm00487498_m1 ) ; Wnt1 ( Mm01300-555_g1 ) ; Yap1 ( Mm011432-63_m1 ) ; Taz ( Mm01289583_m1 ) ; Tnfsf11 ( Mm00441906_m1 ) ; Tnfrsf11b ( Mm00435452_m1 ) ; Sost ( Mm00470479_m1 ) ; and ribosomal protein S2 ( Mrps2 ) ( Mm00475529_m1 ) . We calculated relative mRNA amounts using the ∆Ct method ( Livak and Schmittgen , 2001 ) . We isolated genomic DNA from decalcified cortical bone after digestion with proteinase K and phenol/chloroform extraction . We obtained two custom Taqman assays from Applied Biosystems for quantifying Piezo1 gene deletion efficiency: one specific for sequences between the loxP sites and the other specific for sequences downstream from the 3′ loxP site . Cultured cells were fixed in 4% freshly prepared paraformaldehyde for 15 min . Slides were washed in PBST for 5 min , pretreated with PBS containing 0 . 1% Triton X-100 for 20 min , and blocked in 2 . 5% normal goat serum for one hour . Anti-YAP1 antibody ( 14074S , Cell Signaling , Danvers , MA ) was diluted 1:200 in PBST containing 2 . 5% normal goat serum and incubated with the chamber slides overnight at 4°C followed by rinsing and additional incubation for 1 hr with goat anti-rabbit IgG H and L ( Alexa Fluor 488 ) ( 1:200 ) ( ab150077 , Abcam , Cambridge , MA ) . Non-immune goat IgG was used as a negative control . Slides were mounted with aqueous mounting medium ( H-1000 , VECTOR LABORATORIES , INC . , Burlingame , CA ) . Stained slides were imaged using Axio imager M2 fluorescence microscope ( Carl Zeiss , NY ) . Mean fluorescence intensity was quantified using ImageJ ( NIH , Bethesda , Maryland ) . Circulating osteocalcin and CTX in serum was measured using a mouse Osteocalcin enzyme immunoassay kit ( Thermo Fisher ) and RatLaps ( CTX-I ) EIA kit ( Immunodiagnostic Systems , Boldon , UK ) respectively according to the manual provided by manufacturers . Blood was collected by retro-orbital bleeding into 1 . 7 mL microcentrifuge tubes . Blood was then kept at room temperature for one hour and centrifuged at 1500 x g for 10 min to separate serum from cells . GraphPad Prism seven software ( GraphPad , San Diego ) was used for statistical analysis . Two-way analysis of variance ( ANOVA ) or Student’s t-test were used to detect statistically significant treatment effects , after determining that the data were normally distributed and exhibited equivalent variances . All t-tests were two-sided . P-values less than 0 . 05 were considered as significant . Error bars in all figures represent s . d . . | Bone size and strength depend on physical activity . Increased forces on the skeleton , such as those that occur during exercise , trigger more bone formation and make bones stronger . Conversely , reduced forces , caused for example by the lack physical activity , cause bone loss and increase the risk of fractures . Bones contain cells called osteocytes . These cells sense the forces exerted on bone and orchestrate bone formation in response . Calcium channels are one type of molecule that has been proposed to help osteocytes to sense forces . Calcium channels reside in the cell membrane and can change their structure to allow calcium ions to flow into the cell . Some of them allow calcium ions into the cell in direct response to physical forces , communicating to the cell that a force has been applied . These are called mechanosensitive ion channels . Until now , however , no specific calcium channels involved in force sensing had been identified in osteocytes . Li et al . searched for calcium channels in osteocytes , and found high levels of a mechanosensitive ion channel called Piezo1 . Then , Li et al . made genetically modified mice that did not have any Piezo1 in these cells . The skeleton of these mice was small and weak . Moreover , the bones of these modified mice did not respond to forces like the bones of normal mice . To demonstrate this , Li et al . applied a short bout of increased force to the leg bones of unmodified mice and to those of the Piezo1 deficient mice . After two weeks , the bones of the unmodified mice had increased in thickness , whereas the bones lacking Piezo1 had not . A separate study by Sun , Chi et al . showed similar results when Piezo1 was removed from bone cells grown in the laboratory . Finally , Li et al . tested the impact of a chemical called Yoda1 on bones . Yoda1 makes the Piezo1 channel open , thus mimicking a physical force . These experiments showed that mice treated with Yoda1 had thicker bones than untreated mice . The ability of human bone to become stronger in response to exercise decreases with age , which contributes to the development of osteoporosis . Conditions that require severely restricted exercise , such as disability or extended bedrest , also lead to bone loss . These experiments show that Piezo1 allows bone to respond to physical force , and suggest Piezo1 as a promising therapeutic target to help curtail bone loss in these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2019 | Stimulation of Piezo1 by mechanical signals promotes bone anabolism |
How pancreatic β-cells acquire function in vivo is a long-standing mystery due to the lack of technology to visualize β-cell function in living animals . Here , we applied a high-resolution two-photon light-sheet microscope for the first in vivo imaging of Ca2+activity of every β-cell in Tg ( ins:Rcamp1 . 07 ) zebrafish . We reveal that the heterogeneity of β-cell functional development in vivo occurred as two waves propagating from the islet mantle to the core , coordinated by islet vascularization . Increasing amounts of glucose induced functional acquisition and enhancement of β-cells via activating calcineurin/nuclear factor of activated T-cells ( NFAT ) signaling . Conserved in mammalians , calcineurin/NFAT prompted high-glucose-stimulated insulin secretion of neonatal mouse islets cultured in vitro . However , the reduction in low-glucose-stimulated insulin secretion was dependent on optimal glucose but independent of calcineurin/NFAT . Thus , combination of optimal glucose and calcineurin activation represents a previously unexplored strategy for promoting functional maturation of stem cell-derived β-like cells in vitro .
Pancreatic β-cells secrete insulin to regulate glucose metabolism . Insufficient functional β-cell mass leads to glucose intolerance and diabetes . Researchers have intensively studied β-cell development for the last two decades to generate new therapeutic approaches for diabetes . Although many of the mechanisms regulating the early development of pancreatic progenitor cells have been discovered ( Pan and Wright , 2011 ) , the mechanisms regulating β-cell functional acquisition are still poorly defined ( Kushner et al . , 2014 ) . The β-cells reside in islets containing endocrine , vascular , neuronal and mesenchymal cells . Various signals arising from this neurovascular milieu , such as gap junctions , neuronal transmitters , endothelial factors and hormones , have been reported to be involved in β-cell development ( Borden et al . , 2013; Carvalho et al . , 2010; Cleaver and Dor , 2012; Omar et al . , 2016 ) . These factors may also directly control β-cell functional acquisition , and the related mechanisms need to be studied in vivo . However , the lack of a method to evaluate β-cell function in vivo has hindered the exploration of these fundamental questions . Specifically , β-cell functional development was suggested to be heterogeneous , and this heterogeneous process cannot be studied by simply measuring glucose-stimulated insulin secretion ( GSIS ) in vivo ( Aguayo-Mazzucato et al . , 2011; Bader et al . , 2016; Blum et al . , 2012; Qiu et al . , 2018; van der Meulen et al . , 2017 ) . Development of a method to visualize individual β-cell function in vivo will overcome this problem ( Benninger and Hodson , 2018 ) . The primary function of a mature β-cell is to quickly secrete stored insulin in response to increases in blood glucose concentrations . Glucose-induced Ca2+ influx , which triggers insulin secretion from β-cells , is often used as a functional marker of β-cells ( Pagliuca et al . , 2014; Rezania et al . , 2014; Singh et al . , 2017 ) . The extent of Ca2 + influx has been used to assess the level of β-cell maturation: the greater the amount of Ca2+ influx in , the more mature the β-cells will be ( Rezania et al . , 2014 ) . In isolated islets whose β-cells are labeled with genetically encoded fluorescent Ca2+ indicators , Ca2+ influx in primary β-cells is imaged as an increase in the fluorescent intensity of the indicators due to their conformational change upon Ca2+ binding ( Singh et al . , 2017; van der Meulen et al . , 2017 ) . However , noninvasively imaging Ca2+ transients in β-cells in vivo has not been achieved yet because of the nontransparent pancreas in mammals and the technical challenges in developing high-resolution imaging tools . Here , we used zebrafish as a model animal because its β-cell development is conserved with mammals in general ( Supplementary file 1 ) ( Avolio et al . , 2013; Qiu et al . , 2018; Reinert et al . , 2014; Shah et al . , 2011; Shih et al . , 2013; Tiso et al . , 2009; Yamaoka and Itakura , 1999 ) and because β-cell functionality can be observed in its transparent , externally developing embryos ( Huang et al . , 2001 ) . We generated transgenic ( Tg ) ( ins:Rcamp1 . 07 ) zebrafish , in which every β-cell was labeled with Rcamp1 . 07 ( a red fluorescent Ca2+ indicator ) and used a homemade high-resolution , two-photon , three-axis , digital scanning light-sheet microscope ( 2P3A-DSLM ) to visualize for the first time the glucose-stimulated Ca2+ responses of individual β-cells in vivo . We revealed that a gradually increased glucose concentration , delivered through local diffusion or islet microcirculation , finely triggered the embryonic β-cells to acquire and enhance their function by activating the calcineurin/nuclear factor of activated T-cells ( NFAT ) signaling . We further demonstrated that this mechanism was conserved in neonatal mouse β-cell maturation ex vivo , which may promote the functional acquisition and optimal maturity of stem cell-derived β-like cells cultured in vitro . Finally , the variable functions of neonatal mouse islets cultured under different concentrations of glucose speaks for the importance of being able to image β-cell function in vivo , and this technology can be used to study other mechanisms in islet biology , including transdifferentiation , dedifferentiation and regeneration .
To visualize β-cell function in vivo , we created a transgenic zebrafish line , Tg ( ins:Rcamp1 . 07 ) , in which the red fluorescent calcium indicator Rcamp1 . 07 ( Ohkura et al . , 2012 ) is expressed under the control of the insulin promoter . We demonstrated that Rcamp1 . 07 was exclusively expressed in β-cells , as confirmed by the cellular co-localization of Rcamp1 . 07 with EGFP in Tg ( ins:Rcamp1 . 07 ) ;Tg ( ins:EGFP ) double transgenic fish ( Huang et al . , 2001 ) ( Figure 1—figure supplement 1A–F ) and with immunofluorescently labeled insulin in Tg ( ins:Rcamp1 . 07 ) fish ( Figure 1—figure supplement 1G–J ) . As Rcamp1 . 07 exhibited a very bright basal fluorescence at basal calcium concentrations , all β-cells within the islet were lighted up and readily detectable even in the absence of glucose stimulation ( Figure 1—figure supplement 1C–F ) . We then attempted to record the glucose-induced fluorescent intensity change in Rcamp1 . 07 in a live Tg ( ins:Rcamp1 . 07 ) embryo at 72 hpf , at which stage zebrafish islets have been suggested to regulate glucose ( Jurczyk et al . , 2011 ) . To stimulate β-cells in vivo , glucose was added to the E3 medium to a final concentration of 20 mM to incubate the Tg ( ins:Rcamp1 . 07 ) fish embryos . Within 3 min of stimulation , we observed a robust transient increase in the fluorescence intensity of Rcamp1 . 07 , which indicates glucose-stimulated Ca2+ influx in vivo , under a spinning-disc confocal microscope ( Figure 1—figure supplement 2 ) . However , individual β-cells were difficult to discern under either the confocal microscope or a single-photon selective-plane illuminative microscope ( 1P-SPIM ) ( Figure 1—figure supplement 3 and Video 1 ) . Using a two-photon microscope ( TPM ) , we resolved individual β-cells in the XY plane , but the cell boundaries along the Z-axis were blurred because of a low axial resolution and a high scattering of the illumination light in deep tissues ( Figure 1—figure supplement 3 and Video 1 ) . Previously , based on 2P-DSLM and the tunable acoustic grin device ( Dean and Fiolka , 2014; Duocastella et al . , 2012; Duocastella et al . , 2014; Olivier et al . , 2009; Truong et al . , 2011 ) , we have developed a 2P3A-DSLM ( Zong et al . , 2015 ) . Using this homemade system , we were able to reconstruct a clear three-dimensional ( 3D ) structure of the islet in live fish embryos ( Figure 1A–B and Videos 1–3 ) . Under 2P3A-DSLM , glucose-responsive β-cells first appeared at 48 hpf in vivo , 24 hr earlier than reported previously ( Jurczyk et al . , 2011; Singh et al . , 2017 ) . Next , we quantified insulin-positive cells and glucose-responsive β-cells in Tg ( ins:Rcamp1 . 07 ) ;Tg ( ins:EGFP ) fish embryos at different stages . Interestingly , we identified a progressive increase in glucose-responsive β-cells ( from 1 . 56 ± 0 . 27 to 11 . 38 ± 0 . 49 ) during the hatching period ( 48–72 hpf ) following an increase in the number of total β-cells ( from 14 . 5 ± 1 . 45 to 21 . 88 ± 0 . 95 ) during 36–48 hpf ( Figure 1C and Figure 1—figure supplement 4A ) . We then focused on the hatching period to study the mechanisms by which the embryonic β-cells acquire their function . We used the maximal amplitude ( Max ΔF/F0 ) of glucose-stimulated Ca2+ transients that is a key parameter of β-cell function ( Bruin et al . , 2015; Rezania et al . , 2014 ) , as the criteria to evaluate the functional status of individual β-cells in vivo ( Figure 1—figure supplement 4B ) . Compared with the glucose-responsive β-cells appeared at 48 hpf , β-cells at 56 hpf exhibited larger average Ca2+ responses ( Max ΔF/F0: 93 . 1% ± 9 . 7% versus 53 . 4% ± 8 . 9% ) upon glucose stimulation . These average responses were further enhanced at 72 hpf , at which point β-cells responded with an approximately 150% maximal amplitude of the evoked Ca2+ transients ( Figure 1D–E and Figure 1—figure supplement 4C ) . The histogram of maximal amplitudes of calcium transients from glucose-responsive β-cells shifts right gradually from 48 to 72 hpf , indicating that the increase in calcium signaling of β-cells from 56 to 72 hpf was from the same cells with increased activity ( Figure 1—figure supplement 4C ) . Moreover , some neighboring β-cells showed synchronized Ca2+ responses in response to glucose ( Figure 1—figure supplement 4D–E and Video 2 ) . Therefore , we witnessed a gradual development of β-cell functionality in vivo from 48 to 72 hpf in zebrafish . Interestingly , by analysing 3D reconstructed images of the Tg ( ins:Rcamp1 . 07 ) zebrafish islets at different stages , we found that β-cells acquired glucose responsiveness sequentially from the mantle to the core of the islet ( Figure 1D , Figure 2A and Figure 2—figure supplement 1 ) . From 48–60 hpf , β-cells in the mantle initiated and enhanced their functionality earlier than those in the core and exhibited a higher level of glucose responsiveness ( Figure 2B ) and higher maximum Ca2+ transients ( Figure 2C ) . From 60 to 72 hpf , β-cells in the core started to initiate and accelerate their functional development . At 72 hpf , β-cells in the mantle and core were indistinguishable in all related parameters ( Figure 2B–C ) . These results indicate β-cell functional heterogeneity from islet mantle to the core during the hatching period . To identify the potential mechanisms underlying the temporal and spatial heterogeneity of β-cell functional development , we visualized β-cells and their adjacent vessels in double transgenic zebrafish Tg ( ins:EGFP ) ;Tg ( flk1:mCherry ) or Tg ( ins:Rcamp1 . 07 ) ;Tg ( flk1:GFP ) , in both of which β-cells and vascular endothelial cells were labeled ( Wang et al . , 2013 ) . We found that blood vessels initiated contacts with the islet mantle from 48 to 60 hpf and further penetrated into the inner layers of the islet from 60 to 72 hpf ( Figure 2—figure supplement 2A–B and Video 4 ) . Most of the glucose-responsive β-cells were located proximal to blood vessels ( Figure 2—figure supplement 2C–E ) , and their longitudinal increase was paralleled by an increase in the number of β-cells located adjacent to the vessels ( Figure 2—figure supplement 2F ) . These results indicate that β-cell functional development temporally and spatially correlates with islet vascularization . To explore whether the heterogeneous development of β-cell function is caused by islet vascularization , we examined β-cell function in the Tg ( ins:Rcamp1 . 07 ) ; cloche-/- mutant embryos that have a normal number of β-cells but no vascular endothelial cells or blood cells ( Figure 2D ) ( Field et al . , 2003 ) . At 56 hpf , glucose-responsive β-cells in cloche-/- embryos were indistinguishable from those in age-matched controls ( Figure 2F ) . In contrast , at 72 hpf , cloche-/- mutants contained fewer glucose-responsive β-cells in the islet core ( 1 . 28 ± 0 . 47 versus 5 . 51 ± 0 . 43 ) and exhibited smaller maximum Ca2+ transients in glucose-responsive β-cells ( Max ΔF/F0: 59 . 4% ± 7 . 8% versus 145 . 6% ± 8 . 3% ) than the controls ( Figure 2F ) . To exclude the possibility that the phenotypes observed above was because β-cells in the islet core did not have access to the glucose stimulation , we incubated embryos with supra-physiological dose of 2- ( N- ( 7-nitrobenz-2-oxa-1 , 3-diazol-4-yl ) Amino ) −2-deoxyglucose ( 2-NBDG , 20 mM ) to visualize this fluorescent deoxyglucose analog penetration in the fish embryos . The 2-NBDG ( 20 mM ) efficiently penetrated into the whole islets within 5 min in 56 hpf and 72 hpf cloche-/- mutants even in the absence of blood circulation ( Figure 2—figure supplement 3A ) , indicating that acutely applied high glucose is able to reach all β-cells within the islet independent of islet circulation . Thus , the defective function of β-cells in the islet core of 72 hpf cloche-/- mutants is due to an arrested maturity of these cells rather than a limited access to high glucose . Next , we transiently stopped circulation using 2 , 3-butanedione monoxime ( 2 , 3-BDM ) ( Bartman et al . , 2004 ) in wild-type fish for a 9 hr treatment either from 44 to 53 hpf or from 60 to 69 hpf and evaluated β-cell function under 20 mM glucose stimulation at 56 hpf and 72 hpf respectively ( Figure 2E–F ) . Although blood circulation was completely recovered during functional evaluation , the blockade of circulation from 60 to 69 hpf significantly impaired β-cell maturity in the islet core ( glucose-responsive β-cell number: 1 . 75 ± 0 . 29 versus 5 . 51 ± 0 . 43 ( control ) ) ; Max ΔF/F0: 73 . 1% ± 9 . 9% versus 145 . 6% ± 8 . 3% ( control ) ) to an extent similar to that observed in cloche-/- mutants at the same age ( Figure 2F ) . Therefore , blood circulation , but not the vascular endothelial cells per se , provides a key inductive signal for the initiation and enhancement of β-cell function in the islet core . On the other hand , given that the blockade of circulation from 44 to 53 hpf did not affect β-cells in the islet mantle to acquire glucose responsiveness ( Figure 2F ) , blood circulation is not required for the initiation of β-cell functional acquisition in the islet mantle . Nevertheless , we could not exclude the possibility that β-cell functional maturation may cause these cells to secrete factors that promote angiogenesis , or completely eliminate the possible involvement of vascular endothelial cells in β-cell functional development . Glucose has been reported to regulate embryonic pancreatic endocrine cell differentiation ( Guillemain et al . , 2007 ) . Thus , we investigated whether this major nutrient in the circulatory system also plays a role in the functional development of β-cells . We used 3-mercaptopicolinic acid ( 3 MPA ) , an inhibitor of gluconeogenic phosphoenolpyruvate carboxykinase 1 ( pck1 ) , to suppress endogenous glucose synthesis ( Jurczyk et al . , 2011 ) for 9 hr from 44 to 53 hpf or from 60 to 69 hpf . This treatment severely inhibited β-cell function in the mantle and the core of the islet under both conditions ( Figure 3 and Video 5 ) . Therefore , endogenous glucose is essential for the development of β-cell function in the mantle and the core of the islet . Because yolk syncytial layers , located very close to the islet , already express pck1 before islet vascularization ( Jurczyk et al . , 2011 ) , locally synthesized glucose may diffuse to the islet mantle to initiate the function of peripheral β-cells in the islet . However , β-cells in the islet core started to acquire function only after the establishment of intra-islet vascularization , indicating that the delivery of inductive concentrations of glucose to β-cells in the islet core may require blood circulation . Indeed , a physiological dose of 2-NBDG ( 8 mM ) did not efficiently reach the islet core at 56 hpf when islet circulation has not been established completely , but fully penetrated into the whole islet at 72 hpf with islet vascularization ( Figure 2—figure supplement 3B ) . These data are in direct contrast to the penetration of supra-physiological dose of 2-NBDG ( 20 mM ) ( Figure 2—figure supplement 3A ) and clearly indicate the role of islet vascularization in delivering a physiological dose of glucose to β-cells within the whole islet . To explore the optimal glucose concentration for the induction , we incubated zebrafish embryos with a combination of 3 MPA and different concentrations of glucose ( 3 mM , 5 mM , 8 mM or 20 mM ) for 9 hr ( Figure 3 ) . Treatment with 3 mM or 5 mM glucose from 44 to 53 hpf completely rescued the β-cell functional defects caused by 3 MPA , whereas 8 mM glucose partially restored the number of glucose-responsive β-cells in the islet mantle ( Figure 3 ) . In contrast , treatment with 8 mM glucose from 60 to 69 hpf completely rescued the phenotype , while 3 mM or 5 mM glucose was insufficient to maintain a normal functional β-cell mass in the islet core ( Figure 3 ) . In both cases , 20 mM glucose exhibited the worst performance in the rescue experiments , indicating the glucotoxicity of this high concentration under long-term exposure , which has been reported for mammalian β-cells ( Poitout et al . , 2006 ) . However , it is interesting that glucotoxicity in fish embryo caused by chronic incubation of 20 mM is less severe than expected , suggesting that living organism may have some buffering mechanisms to antagonize the glucotoxicity of 20 mM glucose to β-cells . Collectively , glucose triggers the initiation and enhancement of β-cell functionality in both the core and the mantle of the islet . Moreover , the inductive concentration of glucose is gradually increased with time and mediates the heterogeneous development of β-cell functionality in vivo . The activation of calcineurin/NFAT by a glucokinase activator was proposed to mediate β-cell development and function in mice ( Goodyer et al . , 2012; Heit et al . , 2006 ) . Calcineurin is a calcium- and calmodulin-dependent serine/threonine protein phosphatase that activates cytoplasmic NFAT by dephosphorylating it . The activated NFAT is then translocated into the nucleus to regulate downstream gene transcription . To investigate whether calcineurin/NFAT is the signaling pathway downstream of glucose that controls the functional development of β-cells in vivo , we examined β-cell function after preincubating embryos with calcineurin/NFAT inhibitors or activators . Inhibition of calcineurin with FK506 ( Goodyer et al . , 2012; Heit , 2007 ) for 9 hr from 36 to 45 hpf prevented the appearance of glucose-responsive β-cells at 48 hpf without affecting the total β-cell number ( Figure 4A and Figure 4—figure supplement 1A ) . Inhibition of calcineurin with FK506 or NFAT with VIVIT ( Demozay et al . , 2011 ) from 44 to 53 hpf or from 60 to 69 hpf significantly reduced the number of glucose-responsive β-cells and their maximal Ca2+ transient amplitude at 56 hpf and 72 hpf ( Figure 4B–C and Figure 4—figure supplement 1B–C ) . On the other hand , activation of calcineurin with CGA or NFAT with ProINDY ( Ogawa et al . , 2010 ) for 9 hr rescued the defective function of β-cells caused by 3 MPA , 2 , 3-BDM or the cloche-/- mutants ( Figure 4B–C , Figure 4—figure supplement 1B–C and Video 6 ) . These results suggest that calcineurin/NFAT mediates the role of glucose on triggering β-cell functional development . To exclude the possible off-target effects of the pharmacological reagents , we did a control experiment by testing the effects of the pharmacological reagents ( 10 mM BDM , 3 mM 3 MPA , 10 μM FK506 and 141 . 2 μM CGA ) on calcium activities of neurons in the central nervous system ( CNS ) using Tg ( elavl3:Gcamp6s ) zebrafish in which neurons are labeled by the calcium indicator Gcamp6s ( Dunn et al . , 2016 ) . The result showed that the calcium levels of the CNS neurons were not affected by any of the pharmacological reagents , indicating that the pharmacological treatment in our study specifically affected β-cells ( Figure 4—figure supplement 2 ) . In addition , we designed a dominant-negative zebrafish calcineurin A ( dn-zCnA ) using a similar strategy described previously ( Faure et al . , 2007 ) and generated transient Tg ( ins:EGFP-GSG-T2A-dn-zCnA ) zebrafish embryos on the Tg ( ins:Rcamp1 . 07 ) genetic background . In these embryos , dn-zCnA is co-expressed with EGFP through a GSG-T2A linker under the control of the insulin promoter . In total , 75 . 8% ± 2 . 5% of Rcamp1 . 07-positive cells co-expressed EGFP , indicating that 70%–80% of β-cells expressed dn-zCnA in Tg ( ins:Rcamp1 . 07 ) ;Tg ( ins:EGFP-GSG-T2A-dnCnA ) double transgenic zebrafish ( Figure 4D ) . However , in these embryos glucose-responsive β-cells appearing at 48 hpf were all EGFP negative , and their number was similar with that in the age-matched controls ( Figure 4E ) . In contrast , the number of glucose-responsive β-cells ( 70%–80% expressing dn-zCnA ) was reduced ( 4 . 13 ± 1 . 28 versus 8 . 87 ± 0 . 88 ( control ) ) and the maximum Ca2+ transients were lower in embryos expressing dn-zCnA at 72 hpf than in age-matched controls ( 78 . 7% ± 6 . 9% versus 112 . 3% ± 10 . 3% ) ( Figure 4E ) , suggesting that genetic perturbation of calcineurin/NFAT signaling prevents the glucose-responsiveness of β-cells . To rule out the possibility of the non-specific effects slowing the appearance of markers in the cells , we examined the numbers of glucose-responsive β-cells and their maximal calcium responses to glucose in late ( seven dpf ) embryos . From 72 to 7 dpf , the glucose-responsiveness of β-cells in dn-zCnA zebrafish did not catch up but were still much less than that in 72 hpf control ( Figure 4E ) . Therefore , all of these results demonstrate that calcineurin/NFAT signaling , downstream of glucose , is critical for β-cell functional development . To explore the effect of calcineurin/NFAT signaling in β-cell functional development in mammals , we isolated neonatal islets from postnatal day 0 ( P0 ) mice and cultured them for 3 days in media supplemented with different concentrations of glucose in the presence or absence of CGA . Then , we measured insulin secretion sequentially in response to either low ( 3 mM ) or high ( 20 mM ) glucose stimulation . When islets had been cultured in medium with 5 . 6 mM or 7 mM glucose , in the presence or absence of CGA , neonatal islets secreted significantly more insulin than adult islets under 3 mM glucose stimulation ( Figure 5A ) , which agreed with previous results that the neonatal β-cells are more sensitive to low glucose ( Blum et al . , 2012 ) . For neonatal β-cells cultured in medium with 11 mM glucose , their insulin secretion at low glucose dropped to a level similar to that of adult β-cells ( Figure 5A ) . This finding indicates that optimal glucose may trigger β-cell maturation by reducing their sensitivity to low glucose through a calcineurin/NFAT-independent pathway . On the other hand , no matter what glucose concentration was used in the culture medium , neonatal islets cultured in vitro secreted substantially less insulin than adult islets under 20 mM glucose stimulation . However , supplementing CGA in the culture medium of 11 mM glucose increased the high-glucose-stimulated insulin secretion of the neonatal β-cells ( 7 . 29 ± 0 . 86 ng/ml; Glucose Stimulation Index ( GSI ) = 22 . 74 ± 2 . 68 ) to a level similar to that of adult β-cells ( 7 . 44 ± 0 . 53 ng/ml; GSI = 27 . 29 ± 3 . 29 ) ( Figure 5B–C ) . Taken together , our results demonstrate that optimal glucose could trigger neonatal β-cells functional maturation by reducing low-glucose-induced insulin secretion and increasing high-glucose-stimulated insulin secretion , and the latter mechanism is through the activation of calcineurin/NFAT ( Figure 5A–C ) . Finally , we used mouse islets cultured in vitro to evaluate the correlation between glucose-stimulated calcium transients and GSIS . Compared with adult islets , neonatal islets cultured in medium with 5 . 6 mM glucose also exhibited a significantly higher ensemble of calcium transients under 2 . 8 mM glucose ( 130 . 3% ± 2 . 3% versus 102 . 2% ± 7 . 1% ) ; these calcium transients were abolished if the neonatal islets were cultured in 11 mM glucose ( 95 . 7% ± 10 . 3% versus 101 . 2% ± 4 . 3% ) , and the neonatal islets became indistinguishable from those of adult islets ( Figure 5D–E and Figure 5—figure supplement 1 ) . Moreover , although including CGA in the culture medium did not reduce low-glucose-triggered calcium transients in neonatal islets cultured in 5 . 6 mM glucose ( 126 . 4% ± 1 . 9% versus 130 . 3% ± 2 . 3% ) , it did enhance the maximal amplitude of calcium transients triggered by 16 . 7 mM glucose in neonatal islets cultured both in 5 . 6 mM glucose ( 176 . 4% ± 8 . 4% versus 147 . 5% ± 10 . 2% ) and 11 mM glucose ( 232 . 2% ± 12 . 5% versus 181 . 7% ± 9 . 5% ) ( Figure 5D–F and Figure 5—figure supplement 1 ) . Therefore , glucose-induced calcium response is a reliable and sensitive marker of β-cell functional status .
Visualizing the function of every β-cell in vivo has always been the holy grail for β-cell researchers ( Gotthardt et al . , 2014 ) . Unlike in vivo imaging of β-cells in mammalian species with conventional positron emission tomography and single-photon emission computed tomography that only provide poor spatial-temporal resolution , transparent zebrafish constitutes a unique model that allows high-resolution fluorescent imaging in vivo . In addition to this competitive advantage , we applied the high-resolution 2P3A-DSLM , which has been shown to outperform other microscopes , including the point-scanning TPM , in differentiating individual β-cells in live fish embryos ( Video 1 ) . With these tools , we achieved the first imaging of individual β-cell function in vivo . We further revealed a heterogeneous functional development of β-cells in vivo and identified its underlying mechanism , that is , islet vascularization regulates glucose delivery to control the heterogeneous functional development of β-cells . Interestingly , there was an increase in optimal glucose concentrations for inducing and enhancing β-cell function from the early to the late hatching period , while a suboptimal or supra-physiological concentration of glucose impaired β-cell functional development ( Figure 3 ) . Therefore , the precise delivery of the optimal inductive glucose concentrations via blood circulation is critical for the normal development of β-cell function in vivo . An increase in glucose concentration to induce the functional development of β-cells at different stages is likely to be a conserved mechanism in mammals . For example , the plasma glucose level in fetal rodents is relatively low but starts to increase immediately before birth and progressively reaches a plateau ( above 4 mM ) after P2 ( Rozzo et al . , 2009 ) . From P6 to P20 , the plasma glucose level further gradually increases from 6 to 12 mM ( Aguayo-Mazzucato et al . , 2006 ) . This period is accompanied by islet vascularization , the appearance of glucose-responsive β-cells and further maturation ( Hole et al . , 1988; Reinert et al . , 2014; Rozzo et al . , 2009 ) . On the other hand , inappropriately elevated blood glucose levels impaired the β-cell maturation process , as we observed in zebrafish ( Figure 3 ) . Maternal diabetes in rats leads to immature β-cells in new-borns ( Eriksson and Swenne , 1982 ) . In human , offspring of type two diabetic women have a higher prevalence of type two diabetes than the offspring of nondiabetic women ( Pettitt et al . , 1988 ) . Therefore , glucose , in addition to being the substrate for adult β-cells , also acts as the critical signaling molecule to induce immature β-cells to become more capable of metabolizing glucose and secreting insulin . This mechanism may represent a general principle that applies to the functional development of other cells . A previous study from mice indicates that β-cell functional maturation is marked by an increase in the glucose threshold for insulin secretion . However , the mechanisms underlying the increase in the glucose threshold for insulin secretion are not known . Here , through in vivo study in zebrafish and ex vivo study using neonatal mouse islets , we demonstrated that calcineurin/NFAT mediates the inductive roles of glucose , enhancing high-glucose-stimulated insulin secretion during maturation; however , the reduction in low-glucose-stimulated insulin secretion is mediated by a glucose-activated calcineurin/NFAT-independent pathway during neonatal mouse islet maturation . Multiple glucose-activated signaling pathways are potential ‘switches’ that enhance the glucose metabolism machinery during the development of β-cell function ( Eriksson and Swenne , 1982; Lawrence et al . , 2002; Vanderford et al . , 2007; Vaulont et al . , 2000; Zhang et al . , 2015 ) . These pathways may also involve enhancing high-glucose-stimulated insulin secretion . However , which pathway mediates the optimal glucose-induced reduction in low-glucose-stimulated insulin secretion awaits further investigation . Nevertheless , the combined application of a calcineurin activator and optimal glucose overcame the microcirculation deficit in isolated mouse islets and accelerated the ex vivo functional development of neonatal mouse β-cells ( Figure 5 ) . Similar to isolated mouse islets , stem cell-derived islet-like clusters also lack blood circulation . Different research groups have used a variety of glucose concentrations , ranging from 2 . 8 to 20 mM , in the final stage of differentiation ( Pagliuca et al . , 2014; Rezania et al . , 2014; Russ et al . , 2015 ) ; these glucose concentrations may be suboptimal . However , the lack of glucose delivery to all β-like cells within the clusters due to the absence of blood circulation may still be the critical missing factor for promoting high-glucose-stimulated insulin secretion of these cells , even at an optimal glucose concentration , such as 11 mM . Based on our results , the use of a calcineurin activator may overcome this problem and improve the functionality of stem cell-derived β-like cells in clusters . Although the calcineurin/NFAT signaling pathway has been implicated in β-cell development and maturation in mice ( Goodyer et al . , 2012; Heit et al . , 2006 ) , this pathway has never been manipulated directly in the trials of in vitro differentiation of stem cells into functionally mature β-cells . Our study offers a new strategy of using a calcineurin activator combined with an optimal glucose concentration to promote the functional maturity of stem cell-derived β-like cells within these islet-like clusters in vitro . Our study also provides a new direction for using in vivo imaging of β-cell mass and function to study mechanisms of islet biology such as regeneration , cell identity or functional changes during disease progression . An understanding of these principals may shed light on methods for developing cell replacement therapies to treat diabetes .
The Tg ( ins:Rcamp1 . 07 ) reporter zebrafish line was generated using meganuclease-mediated transgenesis as previously described ( Soroldoni et al . , 2009 ) . Briefly , zebrafish BAC_CH211_69I14 ( BACPAC Resources Center ) , which contains the zebrafish insulin gene , was modified using Red/ET recombineering technology to replace the coding sequence of insulin with Rcamp1 . 07 ( Fu et al . , 2010 ) . Rcamp1 . 07-tagged BAC underwent a second round of recombineering for the subcloning of the modified chromosomal locus into a plasmid backbone containing two I-SceI meganuclease sites . The following primers were used for BAC modification: For subcloning: The resulting constructs were co-injected with I-SceI meganuclease ( Roche , Indianapolis , IN ) at a DNA concentration of 100 ng/μl into one-cell stage zebrafish embryos . These F0 embryos were screened for the transient expression of Rcamp1 . 07 in the pancreatic islet at 48 hpf using a fluorescence stereomicroscope . The positive F0 founders were raised to adulthood and screened by visual inspection of their F1 progenies from outcrossing with the wild-type AB strain . Based on the intensity of the fluorescence signal , one founder was selected , and subsequent generations were propagated and expanded . The Tg ( ins:EGFP-GSG-T2A-dn-zCnA ) zebrafish line was generated using Tol2 transposase RNA-mediated transgenesis as previously described ( Kawakami , 2007 ) . Briefly , dn-zCnA was constructed by deleting the autoinhibitory and the calmodulin-binding domains through introducing a stop codon at the N396 amino acid and by mutating the histidine at the position 152 , a phosphatase-active site , to glutamine ( Zou et al . , 2001 ) . The GSG-T2A peptide was used to separate EGFP and dn-zCnA elements . The EGFP-GSG-T2A-dn-zCnA fragment was cloned downstream of 2 . 1 kb of the proximal insulin promoter and into a Tol2 plasmid using a ClonExpress system . The final construct was injected along with Tol2 transposase RNA into Tg ( ins:Rcamp1 . 07 ) eggs to generate mosaic Tg ( ins:EGFP-GSG-T2A-dn-zCnA ) ;Tg ( ins:Rcamp1 . 07 ) F0 fish for imaging analysis . The wild-type AB strain and transgenic fish were maintained and handled according to the institutional guidelines of animal usage and maintenance of Peking University . The Tg ( ins:EGFP ) fish were from Dr . Lin Shuo at UCLA; the Tg ( flk1:mCherry ) fish were from Dr . Zhang Bo at PKU; the Tg ( flk1:GFP ) fish were from Dr . Chen Jau-Nian at UCLA and the Tg ( elavl3:Gcamp6s ) fish were from Dr . Florian Engert at Harvard University . Tg ( ins:Rcamp1 . 07 ) was crossed with heterozygous cloche-/+ to obtain Tg ( ins:Rcamp1 . 07 ) ;cloche-/+ zebrafish . Tg ( ins:Rcamp1 . 07 ) ;cloche-/- embryos from the crossing of Tg ( ins:Rcamp1 . 07 ) ;cloche-/+ with cloche-/+ fish were used for the imaging experiments . Phenylthiourea ( 0 . 002% ) ( PTU , Sigma , St . Louis , MO ) was added at 12 hpf to prevent pigment synthesis . Prior to live imaging experiments , embryos were anesthetized with 0 . 01% tricaine ( Sigma ) . Heterozygous Tg ( ins:Rcamp1 . 07 ) embryos were used for pharmacological treatments and imaging analyses . All chemicals were prepared as high-concentration stocks and diluted in E3 medium to the final concentrations used for treatment , which were carefully selected to be nontoxic and effective . Embryos were treated with each chemical for 9 hr , either from 36 to 45 hpf , 44–53 hpf or from 60 to 69 hpf , followed by a 3 hr recovery after the chemicals were completely washed out with E3 medium to test the pharmacological effects of the chemicals on the functional development of pancreatic β-cells . Imaging analyses for functional evaluation were performed at 48 hpf , 56 hpf or 72 hpf . For the pharmacological treatment of zebrafish embryos , 10 mM 2 , 3-BDM ( Sigma ) was used to block blood circulation . 2 , 3-BDM is a well-characterized , low-affinity , noncompetitive inhibitor of skeletal muscle myosin-II that blocks myofibrillar ATPase in a dose-dependent manner . This compound decreases the myocardial force , thereby abolishing blood flow at the 10 mM concentration in our experiments ( Bartman et al . , 2004 ) . Zebrafish embryos were treated with 3 mM 3 MPA ( Santa Cruz Biotechnology , Dallas , Texas ) to inhibit endogenous glucose production , 141 . 2 μM chlorogenic acid ( CGA , Sigma ) or 10 μM FK506 ( Invivogen , San Diego , CA ) to activate or inhibit calcineurin , respectively , and 2 . 5 μM ProINDY ( Tocris , Minneapolis , MN ) or 20 μM VIVIT peptide ( Tocris ) to activate or inhibit NFAT , respectively . 2-NBDG ( Invitrogen , Waltham , MA ) is a fluorescent deoxyglucose analog that can be taken up by the cells through glucose transporters . Fluorescence generated by 2-NBDG is therefore used to measure glucose uptake with a fluorescence microscope , as its intensity is proportional to glucose uptake by the cells ( Yamada et al . , 2007 ) . 2-NBDG stock was diluted to working concentrations and then ready to be used for incubating live zebrafish embryos . After thoroughly washing , images were captured under a TPM ( Zeiss 710 ) . The 2-NBDG signal was excited at 920 nm and collected between 510 and 540 nm . For the confocal time-lapse imaging , images were captured using an Olympus spinning-disc confocal microscope with a 10×/0 . 4 objective and 1 . 6 × preamplifier . Tg ( ins:Rcamp1 . 07 ) embryos were embedded in a 1% ultra-pure agarose ( Invitrogen ) cylinder and immersed in an imaging chamber filled with E3 medium containing 0 . 01% tricaine . A D-glucose stock solution was added to the E3 medium to a final concentration of 20 mM during stimulation . Time-lapse images were captured once per second with a 100 ms exposure time . Images were collected using MetaMorph software and analyzed with Fiji software . High-resolution images of pancreatic β-cells and blood vessels in live zebrafish embryos were captured with a 2P3A-DSLM equipped with two 40×/0 . 8 water lenses , as previously described ( Zong et al . , 2015 ) . Briefly , anesthetized embryos were embedded in a 1% ultra-pure agarose cylinder and immersed in E3 medium containing 0 . 01% tricaine . When monitoring β-cell function within zebrafish embryos , glucose was added to the E3 medium to a final concentration of 20 mM as an acute stimulation . For the 2D time-lapse imaging experiments used in the statistical analysis , the islet was optically sectioned into 5–6 layers to ensure that the calcium transients were recorded in all β-cells within the islet . For the fast volumetric imaging and reconstruction of calcium transients within the whole islet , the islet was optically sectioned into 25 layers . Each layer was captured five times with an 8 ms exposure time and was averaged as one single image . Images were collected using the HCImage software ( Hamamatsu ) and processed with R-L deconvolution by the Fiji software . The volumetric calcium transients were reconstructed using Amira software ( FEI ) . For every β-cell , the relative change in the fluorescence intensity of Rcamp1 . 07 was indicated by the ΔF/F0 . The ΔF value was calculated as the difference between the absolute intensity values ( F ) at any time point and the initial intensity value ( F0 ) at the first time point . For 1P-SPIM imaging , we used a homemade SPIM setup equipped with a 40×/0 . 8 water lens . For TPM imaging , we used a fast resonant-scanned TPM equipped with a 40×/0 . 8 water lens . The same 72 hpf zebrafish samples were sequentially imaged with 1P-SPIM , TPM and 2P3A-DSLM . Under each configuration , the whole islet was optically sectioned by 100 planes ( z-step: 500 nm ) with an exposure time of 150 ms per frame . Images were captured using the HCImage software and processed with R-L deconvolution using Fiji software . The embryos were fixed in 4% paraformaldehyde ( PFA , AppliChem , St . Louis , MO ) at 4°C overnight . After washing , the embryos were dehydrated in 30% sucrose ( Sigma ) and then transferred to embedding chambers filled with OCT compound ( Tissue Tek , Sakura Finetek , Torrance , CA ) . After embedding , the samples were frozen in liquid nitrogen as soon as possible . Sectioning was performed using a Leica CM1900 Cryostat set to a 10 μm thickness and a −25°C chamber temperature . The sections were collected and kept at −20°C in a sealed slide box . Each section was gently transferred to room temperature ( RT ) by allowing the section to melt onto the slide for at least 1 hr prior to immunostaining . Sections were washed once and then permeabilized with acetone for 5–10 min at RT . After extensively washing and blocking with PBST ( PBS + 0 . 1% Tween-20 ) containing 0 . 2% bovine serum albumin ( BSA , AppliChem ) and 5% fetal bovine serum ( Gibco , Gaithersburg , MD ) for 1 hr at RT , the sections were incubated with primary antibodies overnight at 4°C . Then , secondary antibodies were applied at 4°C overnight after thorough washing . For nuclear staining , the sections were incubated with 2 μg/ml DAPI ( Solarbio , Beijing , CN ) for 10 min . After extensive washing , the sections were mounted in 80% glycerol ( Sigma ) and imaged under an Olympus spinning-disc confocal microscope . The primary antibodies included monoclonal rat anti-mCherry antibody ( 1:200 , Thermo , Waltham , MA ) used to detect Rcamp1 . 07 and polyclonal guinea pig anti-insulin antibody ( 1:200 , Dako , Carpentaria , CA ) . The secondary antibodies included Alexa Fluor 568 goat anti-rat IgG ( 1:500 , Thermo ) and DyLight 488 goat anti-guinea pig IgG ( 1:500 , Thermo ) . Adult islets from 8-week-old mice were isolated as previously described ( Wang et al . , 2016 ) . For islet isolation from P0 mice , pancreata were directly dissected without perfusion and digested with 0 . 5 mg/ml Collagenase P ( Roche ) . The isolated islets were cultured for 3 days in RPMI1640 media containing different concentrations of glucose ( 5 . 6 mM , 7 mM , 11 mM , 15 mM and 20 mM ) with or without CGA ( 56 . 48 μM ) . The culture medium was changed daily . To measure GSIS , 10 islets of similar sizes were selected and preincubated in KRB buffer for 2 hr at 37°C in a 5% CO2 incubator . The islets were then transferred to low-glucose ( 3 mM ) KRB buffer and incubated for 1 hr at 37°C , 5% CO2 . The supernatant was collected , and basal insulin secretion was measured . The same islets were transferred to high-glucose ( 20 mM ) KRB buffer for another 1 hr incubation at 37°C , 5% CO2 . The supernatant was collected and stored at −20°C , and later , the insulin content was measured using the rat/mouse insulin ELISA kit ( Millipore , St . Louis , MO ) . For ex vivo calcium imaging , the islets were washed with KRB buffer , incubated with 10 μM Cal-520 AM ( AATB ) for 60 min , washed twice with KRB buffer , and then incubated further at 37°C for another 15 min without the dye . Before imaging , islets were attached to chamber coated with Cell-Tak ( Corning , NY , USA ) and then immediately staged on spinning-disc confocal microscope for the acquisition of high-resolution images . Time-lapse images were captured once per second at a single-cell resolution of 40 × magnification . The progression of glucose challenges and the time of stimulation during imaging were as follows: 5 min in 0 . 5 mM glucose KRB buffer; 10 min in 2 . 8 mM glucose KRB buffer; and 10 min in 16 . 7 mM glucose KRB buffer . Images were analyzed using Fiji software . Based on both 2D images and 3D reconstructed images of Tg ( ins:Rcamp1 . 07 ) zebrafish islets , all β-cells were manually outlined according to their localization ( inside or outside , Figure 2A ) or glucose responsiveness ( responsive or nonresponsive , Figure 1D ) . The coordinate positions of the β-cells were added to the region of interest ( ROI ) manager in the Fiji software . Then , we loaded the raw data and the ROI information into our self-developed MATLAB program and transformed the grayscale map to hue-saturation value ( HSV ) color space , setting saturation to one and the proportionate value to the gray level distribution . The hue consists of 0 , 120 and 240 representing three colors . As shown in Figure 2A and Figure 1D , the inside or glucose-responsive cells are red colored ( hue = 0 ) , and the outside or glucose nonresponsive cells are blue colored ( hue = 240 ) or green colored ( hue = 120 ) . All data were analyzed using GraphPad Prism six software . The results are displayed as the mean value ±standard error of the mean ( SEM ) . Unpaired Student’s two-tailed t-tests were used to compare data between two indicated groups . One-way ANOVA followed by Dunnett’s test was used for multiple comparisons with the control group . The asterisks * , ** and *** indicate significance with p values less than 0 . 05 , 0 . 01 and 0 . 001 , respectively . Animal care , generation of transgenic zebrafish lines , in vivo imaging of living zebrafish embryos and all other experiments involving zebrafish and mouse islets were approved by the IACUC of Peking University in China ( reference number: IMM-ChenLY-2 ) . | When the amount of sugar in our body rises , specialised cells known as β-cells respond by releasing insulin , a hormone that acts on various organs to keep blood sugar levels within a healthy range . These cells cluster in small ‘islets’ inside our pancreas . If the number of working β-cells declines , diseases such as diabetes may appear and it becomes difficult to regulate the amount of sugar in our bodies . Understanding how β-cells normally develop and mature in the embryo could help us learn how to make new ones in the laboratory . In particular , researchers are interested in studying how different body signals , such as blood sugar levels , turn immature β-cells into fully productive cells . However , in mammals , the pancreas and its islets are buried deep inside the embryo and they cannot be observed easily . Here , Zhao et al . circumvented this problem by doing experiments on zebrafish embryos , which are transparent , grow outside their mother’s body , and have pancreatic islets that are similar to the ones found in mammals . A three-dimensional microscopy technique was used to watch individual β-cells activity over long periods , which revealed that the cells start being able to produce insulin at different times . The β-cells around the edge of each islet were the first to have access to blood sugar signals: they gained their hormone-producing role earlier than the cells in the core of an islet , which only sensed the information later on . Zhao et al . then exposed the zebrafish embryos to different amounts of sugar . This showed that there is an optimal concentration of sugar which helps β-cells develop by kick-starting a cascade of events inside the cell . Further experiments confirmed that the same pathway and optimal sugar concentration exist for mammalian islets grown in the laboratory . These findings may help researchers find better ways of making new β-cells to treat diabetic patients . In the future , using the three-dimensional imaging technique in zebrafish embryos may lead to more discoveries on how the pancreas matures . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2019 | In vivo imaging of β-cell function reveals glucose-mediated heterogeneity of β-cell functional development |
Transport of biologically active molecules across tight epithelial barriers is a major challenge preventing therapeutic peptides from oral drug delivery . Here , we identify a set of synthetic glycosphingolipids that harness the endogenous process of intracellular lipid-sorting to enable mucosal absorption of the incretin hormone GLP-1 . Peptide cargoes covalently fused to glycosphingolipids with ceramide domains containing C6:0 or smaller fatty acids were transported with 20-100-fold greater efficiency across epithelial barriers in vitro and in vivo . This was explained by structure-function of the ceramide domain in intracellular sorting and by the affinity of the glycosphingolipid species for insertion into and retention in cell membranes . In mice , GLP-1 fused to short-chain glycosphingolipids was rapidly and systemically absorbed after gastric gavage to affect glucose tolerance with serum bioavailability comparable to intraperitoneal injection of GLP-1 alone . This is unprecedented for mucosal absorption of therapeutic peptides , and defines a technology with many other clinical applications .
One of the major challenges for applying protein and peptide biologics to clinical medicine is the lack of rational and efficient methods to circumvent epithelial and endothelial cell barriers separating large molecules from target tissues . In the case of epithelial cells lining mucosal surfaces , the pathway for absorption of large solutes is by transcytosis – a process of transcellular endosome trafficking that connects one surface of the cell with the other ( Tuma and Hubbard , 2003; Mostov et al . , 2000; Garcia-Castillo et al . , 2017 ) . The same is true for transport of protein and peptide cargoes across tight endothelial barriers that separate blood from tissue - typified by the blood-brain barrier ( Abbott , 2013; Pardridge , 2015; Preston et al . , 2014; Lajoie and Shusta , 2015 ) . Here , we address these problems by testing structure-function of the glycosphingolipids for their intracellular trafficking in transcytosis , and for their use as vehicles to enable transcellular transport of therapeutic peptides . These studies were informed by our findings that the structure of the ceramide ( lipid ) domain plays a decisive role in the intracellular trafficking of the glycosphingolipid GM1 , the lipid receptor responsible for cholera toxin entry into the endoplasmic reticulum ( ER ) of host cells and required for disease ( Chinnapen et al . , 2012 ) . GM1 species containing ceramides with ‘kinked’ cis-unsaturated C18:1 or C16:1 fatty acids , or non-native ‘short chain’ C12:0 fatty acids , enter the sorting/recycling endosome of epithelial cells allowing for transport to various intracellular destinations: including the recycling pathway and retrograde pathway to the Golgi and ER . These lipids do not efficiently traffic into the late endosome-lysosome pathway . In contrast , GM1 sphingolipids with long saturated fatty acid chains ( C16:0 or longer ) sort almost exclusively into late endosomes and lysosomes ( Chinnapen et al . , 2012 ) . The sorting step separating the intracellular distributions of these closely related lipids emerges from the early sorting endosome , and we find it robust across all cell lines so far tested . Our observations are consistent with the two major models for lipid sorting: one by molecular shape ( Hao et al . , 2004; Mayor et al . , 1993; Mukherjee et al . , 1999 ) and the other by membrane microdomains ( lipid rafts ) ( Brown , 2006; Simons and Ehehalt , 2002; Simons and Vaz , 2004 ) . In polarized epithelial cells , another pathway emerges from the sorting endosome and leads to membrane transport across the cell by transcytosis . The same GM1 species with cis-unsaturated or short-chain fatty acids that efficiently enter the recycling endosome also sort into this pathway ( Saslowsky et al . , 2013; te Welscher et al . , 2014 ) . By analogy with the bacterial toxins and viruses that bind glycosphingolipids for trafficking into host cells ( Chinnapen et al . , 2007; Spooner and Lord , 2012; Ewers and Helenius , 2011; Cho et al . , 2012 ) , this result suggested a means for enabling the uptake and transepithelial transport of protein or peptide therapeutic cargoes for mucosal delivery . Our first attempt to test this idea showed that these glycosphingolipid species were capable of sorting a therapeutic cargo into the transcytotic pathway . But release into solution to effect transport across epithelial barriers in vitro , or absorption into the systemic circulation in vivo was not detectable ( te Welscher et al . , 2014 ) . To solve this problem , we conducted additional structure-function studies for the glycosphingolipids in intracellular sorting and discovered modifications of the ceramide and oligosaccharide domains that enable the lipids to act as molecular carriers for mucosal absorption of therapeutic peptides , achieving levels of bioavailability comparable to that of intraperitoneal injection .
To test if GM1 glycosphingolipids can be harnessed for biologic drug delivery , we first developed a non-degradable all D-isomer reporter peptide for structure-function studies on the ceramide domain . The reporter peptide was designed to contain two functional groups , a biotin for high-affinity streptavidin-enrichment , and an alkyne reactive group for chemical ligation to fluorophore molecules and quantitative detection . A C-terminal reactive aminooxy was used for coupling the reporter peptide to the oligosaccharide domain of the different GM1 species ( Figure 1A and Figure 1—figure supplement 1A ) ( te Welscher et al . , 2014 ) . The functional groups on the reporter peptide , that is biotin , alkyne , fluorophore and combinations of , were tested to verify the absence of confounding effects on GM1 trafficking ( Figure 1—figure supplement 1B ) . This was assessed by confocal microscopy for endosome sorting and transcytosis , using fluorescent cholera toxin B-subunit to label the GM1-peptide fusion molecules ( Figure 1—figure supplement 1C ) . In all cases , the peptide-coupled GM1 species containing cis-unsaturated or short fatty acid ceramide domains sorted into small cytoplasmic vesicles and basolateral membranes consistent with the recycling and transcytotic pathways , whereas the peptide-coupled GM1 species containing saturated long fatty acid ceramide domains did not; they were sorted into larger cytoplasmic puncta consistent with the late endosome/lysosome instead ( Figure 1—figure supplement 1C ) . Both events were blocked at 4°C consistent with uptake by endocytosis . These results are consistent with our previous studies ( Chinnapen et al . , 2012; te Welscher et al . , 2014 ) and validate the reporter construct . All glycosphingolipid-peptide fusion molecules subsequently prepared were coupled to Alexa Fluor-488 ( AF488 ) , purified by HPLC , and structures confirmed by mass spectrometry ( Figure 1—figure supplement 1A and Material and methods ) . When tested by pulse-chase in MDCK cells , the peptide-GM1 fusion molecules were internalized and sorted as predicted ( Chinnapen et al . , 2012 ) . The GM1 species containing long saturated fatty acids ( C16:0-GM1 ) were localized to intracellular puncta consistent with sorting to the lysosome ( Figure 1B , bottom panels ) , and the GM1 species containing short fatty acids were sorted into the recycling and transcytotic pathways as evidenced by localization to apical and basolateral plasma membranes and small intracellular vesicles ( Figure 1B middle panels ) . This interpretation was confirmed using lysotracker to mark the lysosome and the transferrin receptor to mark the recycling endosome ( Figure 1—figure supplement 1D ) . The peptide alone did not bind or enter cell monolayers ( Figure 1B , top panels ) . Because we use GM1 originally purified from bovine brain to synthesize the different GM1 species , the end products comprise two isoforms of the long chain base: one containing a sphingosine chain of C18:1 and the other of C20:1 . For the GM1 species containing C12:0 fatty acids , the two sphingosine-isoforms were purified and found to track identically in transcytosis ( Figure 1B , middle two panels ) . Thus , it is the structure of the fatty acid that dominates in the sorting reactions ( Chinnapen et al . , 2012 ) . To test structure-function of the ceramide fatty acid chain , we developed a quantitative assay for transcytosis ( Figure 1—figure supplement 2A–B and Material and methods ) . The assay is sensitive to picomolar concentrations and linear over a large 6-log dynamic range ( Figure 1—figure supplement 2B ) . Different GM1-peptide fusions ( 0 . 1 μM ) were applied to apical reservoirs of polarized epithelial cell monolayers and transport to basolateral reservoirs analyzed after 3 hr by streptavidin-capture and quantitative fluorometric read ( Figure 1C , Figure 1—figure supplement 2C ) . Defatted bovine serum albumin ( 1% w/v ) was added to the basolateral reservoir to amplify release of the lipid-peptide fusion molecules from membrane to solution after transcytosis . In all studies , conditions for equal loading of the different GM1-peptide fusion molecules were determined by quantitative fluorescence measurement of washed cells treated with trypsin to remove adherent glycosphingolipids not incorporated into the membrane bilayer ( Figure 1—figure supplement 2D ) . Transcytosis for the different GM1-peptide fusion molecules was quantified as an apparent permeability coefficient ( PAPP; cm/s ) and compared against both the unfused reporter peptide ( labeled peptide ) or untreated monolayers as negative controls ( Figure 1C , Figure 1—figure supplement 2C ) . When tested on human intestinal T84 cell monolayers , we found approximately a 10-fold increase in transepithelial transport ( PAPP ) for the GM1 ceramide species containing C6:0 , C4:0 , C2:0 , fatty acids or lyso-GM1 as compared to controls . Introduction of an unsaturated cis-double bond to the short chain ceramide fatty acids ( C12:1 and C6:1 ) had no effect on transcytosis in comparison to the saturated species ( C12:0 and C6:0 ) ( Figure 1C ) . This result is in contrast to the dramatic effect the cis-double bond induces in trafficking of the long fatty-acid chain GM1 glycosphingolipids ( Chinnapen et al . , 2012; Saslowsky et al . , 2013 ) . Transepithelial transport was dose-dependent for the C6:0-GM1-peptide fusion ( grey bars ) and greatly exceeded transport of the unconjugated reporter peptide ( white bars ) over a wide range of concentrations ( Figure 1D ) . Mixing experiments using unconjugated GM1 and reporter peptide as individual molecules confirmed that transcellular transport of the peptide cargo was dependent on fusion to the GM1 glycosphingolipid ( Figure 1E ) . Neither the unfused reporter peptide nor the GM1-peptide fusion had any detectable confounding effects on cell viability as determined by measurement of metabolic activity ( MTT assay ) , or monolayer integrity and tight junction function assessed as trans-epithelial resistance ( TEER ) or dextran flux ( Figure 1—figure supplement 2E–G ) . Several approaches were used to confirm that the mechanism of cargo transport across epithelial cell monolayers was by transcytosis and not by paracellular leak . First , we tested for transport across epithelial monolayers at 4°C . Such low temperature effectively stops all forms of membrane dynamics including transcytosis , but has minimal effects on paracellular solute diffusion . We found no detectable transport of GM1-peptide fusions across T84 cell monolayers at 4°C , consistent with transport via transcytosis ( Figure 2A ) . The same results were obtained when transcytosis was measured by live cell confocal microscopy . In these experiments , the apical membranes of epithelial cell monolayers were incubated with the C6:0-GM1-peptide fusion at 10°C for 45 min to allow for incorporation of the GM1 ceramide into the apical membrane with minimal uptake into the cell by endocytosis ( Figure 2B , x–z and y–z images ) . Monolayers were washed and then chased for 15 min at 37°C or kept at a restrictive temperature of 10°C . We found the C6:0-GM1-peptide fusion localized to basolateral membranes in cells chased at 37°C , but not at 10°C ( Figure 2B; left and middle panels respectively ) . Only after breaking open tight junctions by removal of extracellular Ca2+ ( EDTA treatment ) did the GM1-peptide fusion molecule gain access to the basolateral membrane at 10°C ( Figure 2B , right panel ) . In a third approach , we blocked endocytosis at physiologic temperature using the dynamin inhibitor Dyngo-4A . For the C6:0 and C12:0 GM1-peptide fusion molecules , transport into the basolateral reservoir was strongly inhibited by Dyngo-4A treatment , consistent with active transcellular transport by transcytosis ( Figure 2C ) . In contrast , Dyngo-4A had no detectable effect on transport of the reporter peptide alone , as expected for diffusion of solutes by paracellular leak . Similar results were obtained using a genetic approach . The exocyst complex is necessary for efficient receptor-mediated transcytosis of immunoglobulins ( Oztan et al . , 2007; Nelms et al . , 2017 ) , and esiRNA knock-down of EXOC2 subunit caused the predicted 50% decrease in trans-epithelial transport of the C6:0-GM1 peptide fusion molecule ( Figure 2D ) . In contrast , transport of the unfused reporter peptide was not affected by exocyst KD . Thus , fusion of a peptide cargo to certain GM1 species enables active transport of the peptide across the epithelial barrier by transcytosis . To explain how the very short chain fatty acids amplified transport across epithelial cell monolayers , we first measured the rate of transcytosis by pulse chase . Apical membranes of MDCK cell monolayers were loaded at 10°C with equal amounts of C12:0 and C6:0-GM1-peptide fusions , washed , and then incubated at 37°C and imaged by live-cell confocal microscopy over time . Transcytosis was measured as fluorescence at basolateral membranes . By this method , we found no detectable difference among the two GM1 species in the rate of transcytosis ( Figure 3—figure supplement 1A ) . In both cases , basolateral membranes were fluorescent after a 10 min chase . At longer chase times , however , we observed a dramatic difference between the C6:0- and C12:0-peptide fusion molecules ( Figure 3A ) . Monolayers loaded with the C12:0-GM1 peptide fusion stained brightly at both the apical ( bottom left panel ) and basolateral membranes ( bottom right panel ) . In stark contrast , monolayers loaded with the C6:0-GM1 peptide fusion showed no fluorescence ( middle panels ) . We interpreted this result as indicating a higher rate of release from cell membranes to the basolateral solution and emptying the cell of the peptide-GM1 fusion over time . To test this idea , we quantified the rate of release from cell membranes for the fluorescent GM1-peptide fusion molecules ( Material and Methods ) . We studied the rate of GM1 release into DMEM media alone ( Figure 3B ) , as well as into DMEM containing defatted bovine albumin ( BSA ) ( Pagano , 1989 ) ( Figure 3C ) . Results show a faster and more complete diffusion from membrane to solution for the C6:0-GM1 fusion molecule ( Figure 3B and C; red curve ) compared to the longer chain C12:0-GM1-peptide ( blue curve ) . Faster and more complete release into solution was also observed for the C2:0-GM1-peptide fusion molecule ( Figure 3B and C; purple curve ) . Thus , the greater efficiency for transepithelial transport by the short chain GM1 species is largely explained by their greater efficiency of diffusion from membrane to solution after transcytosis . Glycosphingolipids contain another major functional domain in addition to the ceramide , the extracellular oligosaccharide head group . These are structurally diverse and operate in a variety of biologic activities ( Cantù et al . , 2011 ) . In all cases , however , the oligosaccharide head group acts to trap sphingolipids in the outer leaflet of cell membranes , thus rendering the lipids dependent on membrane trafficking for their distribution across the cell . To test if the effects of ceramide structure on transcytosis and membrane-release were specific to GM1 glycosphingolipids , or could be generalized to other glycosphingolipid species , we fused our reporter peptide to a GM3 ganglioside synthesized to contain ceramide domains with either C12:0 or C6:0 fatty acids . The oligosaccharide domain of GM3 differs from GM1 by the absence of two sugars and thus lacks the terminal galactose ( and GalNAc ) that functions strongly as a lectin-binding site in GM1 . When tested for transcytosis , we unexpectedly found that the GM3-C12:0-peptide fusion molecule crosses epithelial monolayers far more efficiently that the closely related GM1-C12:0-peptide; and as efficiently as the GM1-C6:0 and C2:0 species ( Figure 3D ) . Similarly , transepithelial transport for the GM3-C6:0-peptide was approximately 2-fold greater than that observed for the GM1-C6:0-peptide when compared directly ( Figure 3—figure supplement 1B ) . Transport was strongly inhibited by pretreatment with the dynamin inhibitor Dyngo-4A , implicating active transcellular trafficking by transcytosis . In membrane-release studies , we found a higher rate of release to solution for the C12:0-GM3-peptide fusion ( green curve ) when compared to the GM1 fusion molecule ( Figure 3B and C ) . Thus , the GM1 glycosphingolipid species appear to be retained in the membrane more tightly than the GM3 species containing the same ceramide domains . Because GM3 lacks a free terminal galactose , we hypothesized the GM1 lipids , which contain the terminal galactose , might be further tethered to the membrane by a form of lectin-binding at the cell surface . To test this idea , we studied the rate of membrane release for the C12:0-GM1 species in the presence or absence of 100 mM lactose ( Glc-Gal disaccharide ) as a competitive ligand ( Figure 3E ) . These studies show enhanced release from the membrane in the presence of excess free lactose , but not excess mannitol , implicating interaction with a galactose-specific lectin membrane tether ( Figure 3E ) . To confirm this idea , we studied the rate of membrane release for the C12:0-GM3 species that lacks the terminal disaccharide galactosyl N-acetyl galactosamine ( Gal-GalNAc ) contained in GM1 linked ( 1→4 ) to the central galactose . Here we find that lactose at high 100 mM concentrations competed both the GM1 and GM3 species off the membrane ( not shown ) but at lower doses ( 5 mM ) lactose released only the GM1-fusion ( Figure 3—figure supplement 1C–D ) . Likewise , Gal-GalNAc ( 5 mM ) was effective at enhancing release of only the GM1-fusion molecules ( Figure 3—figure supplement 1E–F ) . Thus , the oligosaccharide domain of the glycosphingolipids can also affect the efficiency of transport across epithelial barriers , we propose by interacting with lectin-like molecules at the cell surface . The results also strengthen the implications and general principles for how the oligosaccharide head group of native glycosphingolipids may affect sorting and retention in specific regions of the cell . To test for glycosphingolipid-mediated transport across the intestine in vivo , the unfused reporter peptide or the C4:0-GM1-peptide fusion molecule were intragastrically gavaged to mice at equal doses ( 0 . 5 nmol/kg ) and absorption into the blood analyzed after 15 and 30 min using the streptavidin-capture assay . At both time points , we find evidence of absorption into the systemic circulation for the GM1-peptide fusion molecules ( nearly 3% of the applied dose ) , but not for the unfused peptide ( Figure 4A ) . The same results were obtained for the C12:0-GM3-peptide fusion molecule ( Figure 4B ) . We also measured uptake into the liver , where at 1 hr after gastric gavage we find the glycosphingolipid-peptide fusion molecule , but not the unfused peptide ( Figure 4C ) . Thus , fusion to the glycosphingolipids facilitated absorption of the peptide cargo across the intestine and into the two tissues we sampled , blood and liver . The reporter peptide on its own was not detectably absorbed . To test if these results can be generalized , we applied the C6:0- and C12:0-GM1-peptide fusions to the nasal epithelium , another tight epithelial mucosal surface . In this case , the C6:0-GM1-peptide ( green labeling; Figure 4D ) could be visualized by two-photon microscopy within the epithelial barrier in all regions of the nasal epithelium ( Figure 4D ) , including in areas of pseudostratified ( top right panels ) and simple columnar epithelial tissues ( bottom right panels ) . Uptake of the unfused peptide , applied at the same dose , was very rarely detected ( left panels ) . Absorption to the systemic circulation for the GM1-peptide fusion molecules was confirmed biochemically by measuring content in the blood 15 min after nasal administration ( Figure 4E ) . Here , we find approximately a 10-fold increase in blood levels for the GM1-peptide fusion molecules , compared to peptide alone ( which is close to background ) . Unexpectedly , in the nasal epithelium , we find evidence for efficient absorption of the C12:0-GM1-peptide fusion molecules , similar to our results with the C12:0-GM3-peptide species in the intestine . The result suggests that different tissues may interact in different ways with the oligosaccharide domains of glycosphingolipids . In this case , the nasal epithelium may not bind the GM1 oligosaccharide , thus allowing for more efficient release from cell membranes into solution after transcytosis and systemic absorption . Glucagon-like peptide-1 ( GLP-1 ) and related peptides have become important drugs in the management of type two diabetes mellitus , by both promoting weight reduction and sensitizing glucose-stimulated insulin release ( van Bloemendaal et al . , 2014; Heppner and Perez-Tilve , 2015; Tran et al . , 2017 ) . A major factor limiting the clinical utility in many individuals is the fact that all currently available preparations must be delivered by subcutaneous injection . To test if the properties of glycosphingolipid trafficking could be applied to enable oral absorption of GLP-1 , we coupled a long-half-life version of GLP-1 ( Figure 5A ) with C-terminal peptide linker ( termed here GLP-1 for simplicity ) to the C6:0-GM1 ceramide species as described ( te Welscher et al . , 2014 ) . The bioactivity of the glycolipid-GLP-1 fusion molecule was quantitatively assessed using HEK cells expressing the hGLP-1 receptor and CRE ( cAMP ) luciferase reporter ( te Welscher et al . , 2014 ) . As controls , the commercially available long-acting GLP-1 ( Exendin-4 ) , and the unfused GLP-1-peptide were assessed in parallel ( Figure 5—figure supplement 1A ) . The fusion of C6:0-GM1 to GLP-1 caused some loss of function , but the molecule remained highly potent as an incretin hormone , closely comparable to that of the controls . The GLP-1 peptide cargo is 40 residues , approximately 4-fold greater in size compared to the reporter peptide . We first studied transport across intestinal T84 cell monolayers in vitro to test if GM1 glycosphingolipids could transport such a larger cargo . In these studies , GLP-1 transport was quantified by luciferase bioassay as previously described ( te Welscher et al . , 2014 ) ( Figure 5B ) . Here , we find an even greater effect of the glycosphingolipids on transepithelial GLP-1 transport ( 20–100-fold above controls ) . This is explained by a much lower rate of paracellular leak for the larger sized 40-residue GLP-1 peptide . Such size-exclusion from tight junctions is a well-known determinant of paracellular solute diffusion across intact epithelial barriers . To test for absorption and biologic incretin activity in vivo , we gastrically gavaged equal doses ( 10 nmol/kg ) of the C6:0-GM1-GLP-1 fusion , the unfused GLP-1 peptide ( GLP-1 oral ) , or vehicle into wild-type mice and measured effects on glucose metabolism by glucose tolerance test ( Figure 5C ) . Here , we find a lower peak and more rapid return of blood glucose to normal levels in the animals gavaged the C6:0-GM1-GLP-1 fusion molecules compared to animals gavaged GLP-1 peptide ( Figure 5C and D ) . The effect on glucose tolerance by gastrically administered C6:0-GM1-GLP-1 was similar to the effect achieved by the intraperitoneal injection of GLP-1 peptide alone , implicating an equally high level of bioavailability for the gastrically-delivered GM1-fusion molecule ( Figure 5D ) . We confirmed intestinal absorption of the C6:0-GM1-GLP-1 into the systemic circulation in two ways . First , we measured GLP-1 activity in blood samples by streptavidin capture and quantitative luciferase bioassay ( Figure 5E ) . The results show absorption of the GM1-GLP-1 fusion molecule into the blood , but not for unfused GLP-1 . In a second approach , we synthesized an all D-amino acid ( non-degradable ) isomer of GLP-1 coupled to AF488 to allow for direct quantitative measurement of the 40-residue isomer in the blood using the same streptavidin capture assay as described for our reporter peptide ( Figure 5A ) . Again , we find evidence for absorption of the GLP-1 cargo when fused to the C6:0-GM1 transport vehicle , but no absorption for the unfused GLP-1 peptide ( Figure 5F ) . These experiments were performed in two different laboratories , using two different animal facilities with the same results . In all assays ( Figure 5C–5F ) , we find the efficiency of intestinal absorption enabled by fusion to C6:0-GM1 was again almost as efficient as for IP injection of the peptide alone , implicating a high level of bioavailability for the GM1-fusion molecules applied by gastric gavage . Notably , however , the C2:0-GM1-GLP-1 fusion molecule had no effect on glucose tolerance ( Figure 5D ) and was not detectably absorbed after gastric gavage ( Figure 5—figure supplement 1B ) , even though this molecule was readily transported across epithelial monolayers in vitro ( Figure 1C ) . This may be explained by lower affinity of the C2:0- ( and lyso- ) ceramide domains for incorporation into cell membranes , as inferred from membrane loading and release assays ( Figure 3B and Figure 1—figure supplement 2D ) . The difference in biology ( transcytosis in vitro versus absorption in vivo ) becomes apparent only in vivo where the conditions for epithelial uptake and transport are not optimized as they are in vitro . Thus , although it seemed at first glance that further shortening of the fatty acid beyond C4:0 should amplify transepithelial transport and thus clinical utility , this was not the case and the result informs further development of the technology . In summary , we find that fusion of therapeutic peptides to GM1 and GM3 glycosphingolipids with short fatty acids enables their active transport across tight epithelial barriers by transcytosis . In the case of the incretin hormone GLP-1 , fusion to the lipid carriers allows for gastric ( oral ) absorption with high bioavailability and the expected effects on blood glucose , highlighting the potential use of this technology in clinical applications .
Our findings delineate a novel synthetic method for enabling absorption of therapeutic peptides across mucosal surfaces in vivo . The approach is based on the natural biology of lipid sorting for the glycosphingolipids , which depends primarily on the structure of the ceramide domain to allow for trafficking in the transcytotic pathway , and thus active transport across mucosal surfaces without barrier disruption . For applications requiring systemic drug delivery , non-native glycosphingolipid carriers with ceramide domains containing short-chain fatty acids are required to allow for efficient release from cell membranes into the circulation after transcytosis . The apparent high level of intestinal bioavailability enabled by the glycosphingolipid carriers is unprecedented . The mechanism ( s ) for transcellular trafficking co-opted by the cis-unsaturated or short chain fatty acid glycosphingolipids are not fully understood . The most robust sorting event for GM1 glycosphingolipids appears to occur in the early endosome where long saturated chain ceramides are trafficked to the late endosome/lysosome , and the cis-unsaturated and short-chain glycoceramides are not ( [Chinnapen et al . , 2012] and Schmieder and Lencer unpublished results ) . It is possible the unsaturated and short-chain ceramide domains engage sorting mechanisms that dictate their trafficking to the recycling endosome and elsewhere , but we suggest it also possible that their trafficking might be stochastic after escape from the lysosomal pathway , essentially tracking along with bulk membrane flow . In other words , the robust sorting event may occur only for the long chain saturated glycosphingolipids , directing them to the lysosome . Another key structural feature enabling this technology must be the oligosaccharide head group . This domain traps the ceramide lipid in the outer membrane leaflet , preventing flip-flop between leaflets and thus rendering the molecule dependent on membrane dynamics for movement throughout the cell – an essential feature for a trafficking vehicle . As shown by our studies using GM3 , the extracellular oligosaccharide can in some cell types also affect the efficiency of transepithelial transport . One way , we suggest , may be by binding to adjacent membrane lectins , thus enhancing the tethering of the lipid to the membrane surface . Differences have also been reported between GM1 and GM3 with respect to plasma membrane localization and bilayer/curvature dynamics in vitro ( Cantù et al . , 2011; Janich and Corbeil , 2007 ) . In the case of transport across mucosal barriers , we envision several applications for the glycosphingolipids of relevance to clinical medicine . One would be as vehicles for systemic delivery of peptide hormones as demonstrated here , or for topical delivery of agonist or antagonist peptides to specific mucosal surfaces . Another would be for delivery of antigens or adjuvants to enable mucosal vaccination or oral tolerance . It is possible that these glycosphingolipids will transport therapeutic proteins in the same way . Finally , while the biology of endosomes in endothelial cells is much less well understood , we believe at least some of the basic principles for lipid sorting in epithelial cells will apply to this cell type; and the glycosphingolipid carriers defined here may also be used to enable transport of biologics across tight endothelial barriers .
T84 cells and MDCK cells were both obtained from ATCC ( American Type Culture Collection ) . T84 cells were cultured in a 1:1 mixture of Ham's F12 medium and Dulbecco's modified Eagle's medium with 2 . 5 mM L-glutamine , 95%; fetal bovine serum , 5% . MDCK cells were maintained in DMEM supplemented with 10% heat-inactivated FBS and penicillin and streptomycin , all obtained from Thermo Fisher Scientific . T84 and MDCK cells were routinely confirmed to be negative for mycoplasma using a PCR kit from MD Biosciences . T84 or MDCK-II cells plated on 24-well Transwell inserts ( polyester membranes , Costar ) were washed and equilibrated in DMEM without serum containing defatted-BSA ( df-BSA ) . Unconjugated reporter peptide or GM1-peptide fusions at 0 . 1 μM complexed to df-BSA in a 1:1 ratio were then added apically for 3 hr . An excess of BSA ( 1% wt volume ) was added basolaterally to aid in extraction of lipid from cell membranes . After a 3 hr continuous incubation , 1 mL basolateral media was collected and incubated with 10 µL magnetic streptavidin sepharose beads overnight at 4°C , washed with TBS-Tween , and eluted in 95% formamide/10 mM EDTA/0 . 4 mg/mL biotin . For detection of the reporter peptide or GM1-peptide fusion , fluorescence was read using an Infinite M1000 plate reader ( Tecan ) . For each biological replicate concentrations were calculated from standard curves for each compound . The apparent permeability coefficient ( PAPP , cm/s ) is calculated across cell monolayers grown in transwells based on the appearance rate of lipid-peptide fusions in the basolateral compartment over time: Papp ( cm/s ) = ( VD / ( A*MD ) ) * ( DMR/Dt ) Papp ( cm/s ) = ( cm3 / ( cm2 * mol ) ) * ( mol/s ) VD = apical ( donor ) volume ( cm3 ) MD = apical ( donor ) amount ( mol ) A = membrane surface area ( cm2 ) of apical ( donor ) chamber ( i . e . transwell surface area ) DMR/Dt = the amount of compound ( mol ) transferred to the basolateral ( receiver ) compartment over time ( s ) . The reader is referred to Bio-Protocol ( Garcia-Castillo et al . , 2018 ) . WT C57/BL/6 mice ( male 7–9 weeks old ) were purchased from Jackson Laboratory ( Maine USA ) and acclimatized for one week . For intestinal absorption experiments , mice that were fasted overnight were lightly anesthetized with isoflurane and gastrically gavaged with a 0 . 5 nmol/kg dose in a 200 µL volume . Compounds were diluted in PBS containing df-BSA in a 1:1 ratio prior to administration to mice . For analysis of systemic absorption , blood samples were obtained using standard cardiac puncture procedures at 15 or 30 min after compound administration . 100 µL blood was diluted with 400 µL RIPA buffer and incubated with 10 µL streptavidin Sepharose overnight at 4°C , washed , and eluted in 95% formamide/10 mM EDTA/0 . 4 mg/mL biotin as in our in vitro assay . Liver tissue was flash frozen in liquid nitrogen and ground with a chilled mortar and pestle on dry ice . After obtaining dry weight , samples were homogenized in 1 mL RIPA buffer , centrifuged , and supernatant incubated with 10 µL streptavidin Sepharose and bound molecules eluted with 95% formamide/10 mM EDTA/0 . 4 mg/mL biotin . Amount of compound accumulated in the liver was normalized per mg dry weight . For intraperitoneal glucose tolerance tests , a 10 nmol/kg dose was used to gavage overnight-fasted WT C57/BL/6 mice ( male 7–9 weeks old ) with GM1-GLP-1 fusion molecules or unfused GLP-1 . Glucose measurements following i . p administration of 2 mg/g glucose solution were obtained from tail vein blood applied directly to glucose strips as in ( te Welscher et al . , 2014 ) . MDCK-II cells were plated on 96-well plates the day prior to the experiment . Cells were washed with 10°C serum-free DMEM ( no phenol red ) and equilibrated with DMEM containing 0 . 1 μM df-BSA for 15 min . Cells were loaded for 45 min at 10°C with 0 . 1 μM GM1-peptide molecule with a molar ratio of 1:1 ( lipid:df-BSA ) . After loading , cells were washed , warmed to 37°C degrees in DMEM ( no phenol red ) to allow for proper lipid incorporation , and incubated with 0 . 25% trypsin in HBSS to release adherent glycosphingolipids not incorporated into the membrane bilayer . Cells were then incubated in DMEM alone or DMEM containing 1% df-BSA for 2 min , 15 min , or 1 hr . After the indicated time , media was collected and GM1-peptide molecules released into solution quantified using standards for each compound . Cells were subsequently lysed in RIPA buffer and the amount of cell-associated GM1-peptide remaining at that time point quantified using known standards . Amount of GM1-peptide released into the solution was calculated as a ratio of total lipid incorporated ( i . e GM1-peptide in media + cell associated GM1-peptide ) . Gangliosides of different fatty acid species were supplied by Prof . Sandro Sonnino ( U . Milan , Italy ) . Peptides containing modified functional residues were custom synthesized by Novo Nordisk ( DK ) . Synthesis of peptide-lipid conjugates was accomplished by a modified method previously published ( te Welscher et al . , 2014 ) . In a typical 2 mL reaction , 2 mg ( approximately 1300 nmoles depending on fatty acid ) of ganglioside was oxidized with sodium periodate ( 13 μmoles ) in oxidation buffer ( 100 mM sodium acetate pH 5 . 5 , 150 mM NaCl ) for 30 min on ice and protected from light . The reaction was quenched by addition of glycerol ( 5% final ) . The reaction was desalted by Bond Elut SepPak C18 cartridge ( Agilent , MA ) and methanol used to elute from the column was removed by Speed Vac concentration ( Savant ) . The oxidized product was then reconstituted in 2 mL PBS pH 6 . 9 in the presence of 10% DMF and reacted with 2700 nmoles of aminooxy-containing peptide in the presence of 10 mM aniline ( Dirksen and Dawson , 2008 ) . The reaction was incubated for 20 hr at room temperature with mixing on a nutator , where the GM1-peptide fusion product formed normally resulted in a white precipitate . The precipitate was separated from the solution by centrifugation , then resuspended in 400 μL 50% isopropanol/water after brief sonication . PBS pH 6 . 9 was added ( 200 μL ) along with 4 . 8 μmoles of sodium cyanoborohydride and incubated for 3 hr to reduce the oxime bond . Lipid-peptide conjugates were purified by semi-preparative HPLC , and confirmed by either MALDI-TOF ( AB Voyager ) , or ESI LC-MS ( Agilent , MA ) . With exception of the fluorescent peptide described in Extended Data Figure 1 that was done by maleimide linkage , the labeling of peptides with Alexa fluorophore was typically done via copper-mediated Click chemistry . 320 μM peptide-lipid fusions containing an N-terminal alkyne residue ( propargylglycine ) were reacted with equimolar concentrations of Alexa Fluor 488 -azide under the following conditions . 50 mM Tris-Cl , 5 mM copper ( II ) sulfate , 100 mM sodium ascorbate , 37 mM ( Tris[ ( 1-benzyl-1H-1 , 2 , 3-triazol-4-yl ) methyl]amine , TBTA in DMSO/t-butanol 1:4 ) 1 mM ( Tris ( 2-carboxyethyl ) phosphine hydrochloride , TCEP – Sigma ) and reacted for 16 hr at room temperature with mixing via nutator . Products were purified by HPLC and confirmed by mass spectrometry . Products were lyophilized and stored at −20°C . Compounds were resuspended in 33% DMF/water to make stock solutions for assays . m/z mass spectrometry values ± 3 Da were as follows: For GM1-C12:0 reporter conjugates with different functional groups on the peptide and d18:1 long chain base , alkyne = 2475 . 5 Da ( 1+ ) ; biotin = 2734 . 4 Da ( 1+ ) ; alkyne-biotin = 2829 . 4 ( Da ) ( 1+ ) ; Alexa Fluor 488 maleimide = 1552 . 2 Da and d20:1 = 1566 . 2 Da . For GM1-C16:0 species in this series , ( 2+ ) mass was observed at 1580 . 2 Da . Most of the structure-function studies with GM1 fatty acid species lyso to C12:1 , were detected with a 3 + charge . m/z values ± 5 Da for d18:1 and d20:1 sphingosine , respectively , were as follows: lyso = 1101 . 1 Da ( 3+ ) and 1110 . 5 Da ( 3+ ) ; C2:0 = 1115 . 1 Da ( 3+ ) and 1124 . 5 Da ( 3+ ) ; C4:0 = 1124 . 5 Da ( 3+ ) and 1133 . 8 Da ( 3+ ) ; C6:0 = 1134 . 1 Da ( 3+ ) and 1142 . 5 Da ( 3+ ) ; C6:1 = 1133 . 1 Da ( 3+ ) and 1143 . 5 Da ( 3+ ) ; C12:0 = 1161 . 8 Da ( 3+ ) and 1171 . 2 Da ( 3+ ) ; C12:1 = 11161 . 1 Da ( 3+ ) and 1170 . 5 Da ( 3+ ) . Free peptide was observed as a single ion peak at 2102 . 8 Da . For GM3 molecular species conjugates , m/z was observed at: C6:0 = 1212 . 1 Da ( 3+ ) and 1026 . 1 Da ( 3+ ) ; C12:0 = 1040 . 1 Da ( 3+ ) and 1054 . 1 Da ( 3+ ) . To generate bioactive GLP1 fusion lipids , two peptides were joined together via a triazole linkage . Long half-life GLP1 sequences were synthesized containing isobutyrate residues substituted at key dipeptidyl peptidase-4 ( DPP-4 ) cleavage sites , and a C-terminal azido-lysine ( Figure 5 and Extended Data Figure 1 ) . A Tobacco Etch Virus protease site ( ENLYFQS ) was originally designed into the sequence but was not used for the purposes of this paper . The peptide was joined to reporter peptide-lipid conjugates via N-terminal alkyne using Click chemistry as described above . To synthesize the all-D GLP1-lipid fusions , peptides were made as a complete chain on solid phase , and contained aminooxy and biotin groups ( Figure 5 ) . Linkage to oxidized ganglioside was performed as stated above . m/z values ± 8 Da for the biologic GLP1-fusion were observed at: C2:0 = 1808 . 4 Da ( 4+ ) ; C6:0 = 1822 . 2 with free peptide seen as a 3 + charge at 2251 . 0 Da . For the all-D isomer version of GLP1 , GLPD fused to GM1-C6:0 , m/z was seen as a 3 + charge at 2251 . 0 Da and the free peptide as a 2 + charge at 2726 . 7 Da . | To work properly , drugs need to be absorbed efficiently into the body . Medications that are injected directly into the bloodstream are often quickly transported to the organs or tissues they target . But injections are not always convenient , and many patients would instead prefer to swallow a pill or tablet . If a drug is swallowed , however , it must first be absorbed through the gut before it can enter the bloodstream . The lining of the gut consists of tightly linked layers of cells that readily take up small molecules , such as water and simple nutrients , but exclude almost all larger ones . Since several important types of drugs are large or poorly absorbed molecules , such as proteins , finding methods to help them cross the gut barrier is a major part of drug development . Originally from bacteria , cholera toxin is an example of a large , naturally occurring protein that does cross the gut lining . To do this , the toxin specifically attaches onto GM1 , a type of lipid molecule that is found on the outer surface of gut cells , and hijacks the system that moves this lipid within cells . Previous studies identified several key features of GM1’s structure that enable this movement; and , in 2014 , researchers tested GM1 as a ‘carrier’ to help the gut to absorb large therapeutic molecules . This approach was successful in cells grown in the laboratory , but not when the drugs were fed to animals . To overcome this issue , Garcia-Castillo , Chinnapen et al . – who include some of the researchers involved in the earlier studies – set out to further boost GM1’s ability to transport drugs across the gut lining . First several hybrid molecules were made , consisting of different structures of GM1 ( the ‘carrier’ ) fused to a reporter peptide ( the ‘cargo’ ) . Laboratory experiments with human intestinal cells and dog kidney cells , both of which form tightly-linked layers much like the actual lining of the gut , revealed specific structural variations of the GM1-derived carrier that transported the cargo across the cell barrier more efficiently . Garcia-Castillo , Chinnapen et al . went on to test the efficiency of these carriers further by switching the reporter cargo to a therapeutic hormone called GLP-1 . This hormone is used to treat people with type II diabetes but is currently given via an injection . The same structural variants of GM1 that enhanced delivery of the reporter cargo also worked for the larger GLP-1 hormone . Garcia-Castillo , Chinnapen et al . then fed the GM1-GLP-1 fusions to mice , and measured the amount of GLP-1 hormone absorbed into the blood . Crucially , the mice fed GM1-GLP-1 molecules absorbed the drug just as well as mice injected with the GLP-1 that is normally given to diabetes patients . Together these findings represent a major contribution to the pharmaceutical toolbox . They may also ultimately lead to more drugs that can be given as a patient-friendly pill or tablet , readily cross the gut barrier and achieve widespread drug delivery around the body . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2018 | Mucosal absorption of therapeutic peptides by harnessing the endogenous sorting of glycosphingolipids |
In many species , within-group conflict leads to immediate avoidance of potential aggressors or increases in affiliation , but no studies have investigated delayed post-conflict management behaviour . Here , we experimentally test that possibility using a wild but habituated population of dwarf mongooses ( Helogale parvula ) . First , we used natural and playback-simulated foraging displacements to demonstrate that bystanders take notice of the vocalisations produced during such within-group conflict events but that they do not engage in any immediate post-conflict affiliative behaviour with the protagonists or other bystanders . We then used another playback experiment to assess delayed effects of within-group conflict on grooming interactions: we examined affiliative behaviour at the evening sleeping burrow , 30–60 min after the most recent simulated foraging displacement . Overall , fewer individuals groomed on evenings following an afternoon of simulated conflict , but those that did groomed more than on control evenings . Subordinate bystanders groomed with the simulated aggressor significantly less , and groomed more with one another , on conflict compared to control evenings . Our study provides experimental evidence that dwarf mongooses acoustically obtain information about within-group contests ( including protagonist identity ) , retain that information , and use it to inform conflict-management decisions with a temporal delay .
Conflicts of interest are common in social species , with disagreements between group members arising over access to mates or food , synchronisation of group activities , and the direction of travel ( Aureli et al . , 2002; Conradt and Roper , 2009; Hardy and Briffa , 2013 ) . Within-group conflict , especially if it escalates to aggression , can be costly in terms of injury and mortality , time and energy expenditure , increased stress , and disrupted social relationships ( Aureli , 1997; Aureli et al . , 2002; de Waal , 2000 ) . Conflict-management strategies that minimise these costs , either by reducing the likelihood of aggressive escalation in the first place or by mitigating the consequences of such physical contests when they do arise , have therefore evolved in many species ( Aureli et al . , 2002; Aureli and de Waal , 2000 ) . Much of the early work on post-conflict behaviour focussed on interactions between the protagonists ( the aggressor and the target ) : many studies have documented increases in affiliation between former opponents in the aftermath of a contest ( reconciliation; Aureli et al . , 2002; de Waal , 2000; de Waal and van Roosmalen , 1979 ) , although there are also examples of victims avoiding aggressors ( wariness; Benkada et al . , 2020; Kutsukake and Clutton-Brock , 2008; Sommer et al . , 2002 ) . More recently , attention has shifted to the involvement of bystanders ( contest nonparticipants ) in post-conflict behaviour . Considering bystanders highlights the potentially groupwide effects of dyadic within-group conflicts and a wider range of post-conflict management strategies than would be apparent from a focus on just the protagonists ( De Marco et al . , 2010; Schino and Sciarretta , 2015 ) , thus providing additional insights into the dynamics of social relationships between groupmates ( Aureli and de Waal , 2000 ) . Multiple studies have now documented bystander-initiated affiliation with the victim as a means of avoiding redirected aggression ( self-protection ) or of providing substitute reconciliation or consolation ( Fraser et al . , 2009; Fraser et al . , 2008; Schino and Marini , 2012; Wittig and Boesch , 2010 ) . There is also some evidence of bystander-initiated affiliation with the aggressor , which could function as appeasement to reduce the likelihood of redirected aggression ( Cordoni and Palagi , 2015; Palagi et al . , 2008; Pallante et al . , 2018 ) , and group-wide post-conflict affiliation among bystanders , perhaps to reduce conflict-induced stress ( De Marco et al . , 2010; Judge and Mullen , 2005 ) . However , to the best of our knowledge , this research has focussed solely on interactions that occur in the immediate aftermath ( usually within 10 min ) of an aggressive within-group contest; the possibility of delayed post-conflict management behaviour has not been explored . There is increasing experimental evidence that nonhuman animals can remember past events and use information from them when making social decisions later ( Carter and Wilkinson , 2013; Kern and Radford , 2018; Seyfarth and Cheney , 1984; Wittig et al . , 2014 ) . This includes conflict-management decisions about whether to get involved in an aggressive interaction . For example , baboons ( Papio hamadryas ursinus ) were more likely to offer support in aggressive interactions to individuals they had groomed with earlier ( mean: 22 min before; range 10–55 min ) , evidenced by a move towards playbacks of grunt calls given during conflicts ( Cheney et al . , 2010 ) . Similarly , vervet monkeys ( Chlorocebus pygerythrus ) were more likely to offer coalitionary support to a groupmate in a conflict if they had groomed together within the last hour ( Borgeaud and Bshary , 2015 ) . Other studies have shown that individuals can use knowledge of previous agonistic interactions to inform how best to respond in subsequent aggressive encounters . For instance , chimpanzees ( Pan troglodytes ) that had been involved in an unreconciled conflict earlier in the day ( ca 2 hr before ) reacted aversively to the playback of an aggressive bark from their former opponent’s bond partner ( a third-party individual likely to offer aggressive support to the former opponent; Wittig et al . , 2014 ) . Moreover , it was recently shown that bystander wasps ( Polistes fuscatus ) were more aggressive towards individuals that they had observed to be less aggressive in a previous ( 10–30 min earlier ) fight with a third party ( Tibbetts et al . , 2020 ) . It is thus plausible that post-conflict decisions about the avoidance of protagonists and affiliation with groupmates could also occur some time after the relevant within-group contests . Investigating the capacity for delayed management behaviour is important because it is thought to be cognitively challenging to use social information gathered in the past , especially where there is reliance on memories of the actions of particular individuals ( Frith and Frith , 2012 ) , as other sources of personal and third-party information would likely arise in the interim ( Wittig et al . , 2014 ) . To make behavioural decisions , animals obtain information about social interactions using a variety of sensory modalities . Most research considering social monitoring of within-group conflicts has focussed on situations where individuals have seen the interaction; hence , bystanders are commonly defined as individuals who have observed the encounter ( Schino and Sciarretta , 2015 ) . But for those species living in visually occluded environments , those where group members can be scattered over large distances or those that forage in a way that prevents simultaneous vigilance , acoustic cues can be a valuable source of social information ( Bradbury and Vehrencamp , 2011 ) . Numerous species vocalise during or at the end of within-group contests ( Bertram et al . , 2010; Slocombe et al . , 2010 ) . For example , chimpanzees and rhesus monkeys ( Macaca mulatta ) produce screams when experiencing aggression ( Gouzoules et al . , 1984; Slocombe et al . , 2010 ) , whilst little blue penguins ( Eudyptula minor ) give specific calls after a contest is finished ( Waas , 1990 ) . These vocalisations likely provide bystanders with valuable information about the occurrence of within-group conflicts as well as about the group members that have potentially been involved and the outcome ( Gouzoules et al . , 1984; Slocombe and Zuberbühler , 2007; Szipl et al . , 2017; Whitehouse and Meunier , 2020 ) . Moreover , they can be used in playbacks to test post-conflict behaviour experimentally . Here , we investigate post-conflict management behaviour , including the possibility that it occurs with a delay ( ca 30–60 min later ) , in wild dwarf mongooses ( Helogale parvula ) ; the study population has been habituated to close human presence , facilitating detailed observations and field-based experiments ( Kern and Radford , 2018; Morris-Drake et al . , 2019 ) . Dwarf mongooses live in cooperatively breeding groups of up to 30 individuals , comprising a dominant breeding pair ( hereafter ‘dominant’ individuals ) and non-breeding subordinate helpers ( hereafter ‘subordinate’ individuals ) of both sexes ( Rasa , 1977 ) . The most prevalent affiliative behaviour in dwarf mongoose groups is allogrooming ( hereafter ‘grooming’ ) , which underpins the strength of relationships between group members ( Kern and Radford , 2021; Kern and Radford , 2016 ) , increases following stressful situations such as intergroup interactions ( Morris-Drake et al . , 2019 ) , and is traded as a reward for cooperative behaviour ( Kern and Radford , 2018 ) . Within-group aggressive interactions take two main forms: relatively rare targeted aggression , which usually acts to reinforce rank and is mainly due to reproductive conflict ( Rasa , 1977 ) , and relatively common foraging displacements , when a higher-ranking individual displaces a lower-ranking group member from a foraging patch and steals their prey ( Sharpe et al . , 2016; Sharpe et al . , 2013 ) . Foraging displacements generally involve the following behavioural sequence: the higher-ranking individual produces deep growls as it approaches the lower-ranking group member; the former then hip-slams the latter away from the food resource; and the displaced individual typically produces high-pitched squeals whilst it retreats ( Sharpe et al . , 2016; Sharpe et al . , 2013 ) . We determined whether vocal cues of within-group conflict elicit immediate or delayed behavioural responses ( avoidance or changes in affiliation ) by non-participant group members . We focussed on data collection of bystanders because it is not ecologically valid to consider how protagonists respond to their own calls .
We initially used both observational data and a playback experiment to investigate whether bystanders take notice of conflict between groupmates ( evidenced by an increase in vigilance ) and if they engage in affiliative interactions ( grooming ) as post-conflict management behaviour in the immediate aftermath ( full details in ‘Materials and methods’ ) . To collect data relating to natural foraging displacements ( which occur at a mean ± SE observer-detected rate of 2 . 6 ± 0 . 2 events per 3-hr observation session , range = 0–10 , N = 127 observation sessions across eight groups ) , we conducted focal watches on foraging subordinates in two situations: immediately after the human observer heard a foraging displacement ( conflict situation ) and on a matched occasion when there had been no foraging displacement for at least 10 min ( control situation ) . Paired data were collected from 16 subordinates in six groups , with conflict and control focal watches counterbalanced in order between individuals . To test experimentally the immediate responses of bystanders , and to isolate the importance of foraging-displacement vocalisations as a cue to conflict occurrence , we presented 17 foraging subordinates in eight groups with two playback treatments in a matched , counterbalanced design ( Experiment 1 ) . The conflict treatment entailed an initial playback of close calls from a dominant individual and a subordinate individual from the same group as the focal individual , followed by a playback of the dominant growling and the subordinate squealing ( simulating a foraging displacement ) ; the control treatment entailed the playback of close calls from the same two individuals for the same duration as a full conflict-treatment playback track ( Figure 1 ) . Foraging dwarf mongooses produce continuous low-amplitude close calls , which likely enable groupmates to stay in contact; there is no evidence that they have an aggressive function ( Kern and Radford , 2013; Sharpe et al . , 2013 ) . We chose for the playback the combination of a dominant individual as the aggressor and a subordinate individual as a target because this is the most common dyadic pairing observed in natural foraging displacements ( 74 . 3% of 740 events in 12 groups ) . In 2–3 min following both natural foraging displacements ( Wilcoxon signed-rank test: Z = 3 . 154 , N = 16 , Monte Carlo p<0 . 001; Figure 2a ) and those simulated by playbacks ( Z = 3 . 527 , N = 17 , p<0 . 001; Figure 2b ) , focal foragers spent a significantly greater proportion of time vigilant than in matched-control , non-conflict situations . The increased vigilance following foraging displacements indicates that bystanders take notice of conflict between groupmates; the experimental results demonstrate that the vocal cues are sufficient to trigger this reaction . However , the focal individual did not engage in any post-conflict grooming in the 5 min following either natural or simulated foraging displacements; grooming is generally rare ( ca 10% of bouts ) during foraging periods in dwarf mongooses ( Kern and Radford , 2018 ) . Thus , dwarf mongoose bystanders do not appear to engage in post-conflict affiliative behaviour in the immediate aftermath of hearing a within-group contest . To test if there were delayed effects of within-group conflict on affiliative behaviour ( grooming ) , we conducted a second repeated-measures playback experiment on eight groups ( Experiment 2 , Figure 3; full details in ‘Materials and methods’ ) . The general experimental design followed Kern and Radford , 2018 . In each trial session , we either simulated an increase in the conflict between a dominant ( aggressor ) and a subordinate ( target ) group member through playback of their foraging-displacement vocalisations ( conflict treatment ) or played back just the close calls of those individuals for an equivalent period ( control treatment ) . Trials were on separate days with treatment order counterbalanced between groups . In each trial , six to nine playbacks ( mean ± SE: 8 . 5 ± 0 . 2 , N = 16 trials ) were carried out over the course of 3 hr in the afternoon whilst the group were foraging and before they moved towards their evening sleeping refuge ( mean ± SE period between the final playback and first grooming bout at the sleeping refuge: 37 ± 5 min , N = 16 trials ) ; individual playbacks were as in Experiment 1 with different tracks played each time . At the refuge , we collected data ad libitum on all adult grooming interactions , including the identity of those involved and bout duration; each bout was always between just two individuals and generally mutual ( both parties approaching each other and grooming , without an obvious initiator ) . If within-group conflict does have delayed effects on affiliative behaviour , we expected an increase in the occurrence of foraging displacements to result in changes in evening grooming levels; 90% of grooming bouts occur at the sleeping refuge ( N = 6376 bouts , 174 individuals; Kern and Radford , 2018 ) . Overall , we found that group members were significantly less likely to be involved in grooming interactions in the evenings following conflict afternoons compared to control afternoons ( generalised linear mixed model [GLMM]: χ2 = 5 . 401 , df = 1 , p = 0 . 020; Table 1a; Figure 4a ) . However , when considering only those individuals that engaged in grooming , they spent a significantly greater proportion of time doing so on evenings when there had been an earlier simulated increase in conflict compared to control evenings ( χ2 = 15 . 873 , df = 1 , p<0 . 001; Table 1b; Figure 4b ) . This was because these individuals were grooming more frequently ( χ2 = 8 . 010 , df = 1 , p = 0 . 005; Table 1c ) and for longer per bout ( linear mixed model [LMM]: χ2 = 3 . 958 , df = 1 , p = 0 . 047; Table 1d ) after a simulated increase in conflict compared to control conditions . These results indicate that there is an overall response to simulated conflict within the group , but we also made some specific predictions . Assuming that aggressors and targets can be identified from their vocalisations—which has been demonstrated for dwarf mongoose close calls ( Sharpe et al . , 2013 ) , recruitment calls ( Kern and Radford , 2016 ) , and surveillance calls ( Kern and Radford , 2018 ) —we predicted that subordinates might engage in either less grooming ( due to wariness ) or more grooming ( as possible appeasement ) with aggressors , and that they might engage in more grooming with targets ( as possible consolation ) . We found evidence that simulating aggressive behaviour by a dominant individual during the afternoon resulted in subordinates engaging in less grooming with it at the sleeping refuge that evening . Following conflict trials , subordinates groomed with the dominant pair for a smaller proportion of time than after control trials ( Wilcoxon signed-rank test: Z = 2 . 240 , N = 8 , p = 0 . 021 ) . The reduced affiliative engagement by subordinates was driven by a change in behaviour towards the simulated aggressor specifically: subordinates engaged in significantly less grooming with the simulated aggressor on conflict evenings compared to control evenings ( proportion of time: Z = 2 . 521 , N = 8 , p = 0 . 008; Figure 5a; proportion of subordinates: Z = 2 . 201 , N = 8 , p = 0 . 033; Figure 5c ) , but there was no such treatment difference in the grooming of subordinates with the dominant whose calls were not played back ( proportion of time: Z = 0 . 105 , N = 8 , p = 1; Figure 5b; proportion of subordinates: Z = 0 . 813 , N = 8 , p = 0 . 499; Figure 5d ) . Moreover , on those occasions where individuals did groom , bout durations were somewhat shorter on conflict evenings compared to control evenings for grooming involving simulated aggressors ( mean ± SE duration , post-control: 34 ± 11 s; post-conflict: 23 ± 5 s; N = 4 pairs of trials ) , while the reverse was true for grooming involving the matched dominant ( post-control: 28 ± 8 s; post-conflict: 34 ± 8 s; N = 4 pairs of trials ) ; small sample sizes precluded statistical analysis . We also found some evidence that increasing within-group conflict during the afternoon resulted in more evening grooming between subordinates . When considering all bouts between subordinate group members , there was no significant treatment difference in the proportion of time spent grooming ( Wilcoxon signed-rank test: Z = 1 . 540 , N = 8 , p = 0 . 146 ) , but subordinate–subordinate grooming bouts were , on average , significantly longer on conflict evenings compared to control evenings ( Z = 2 . 366 , N = 7 , p = 0 . 015; Figure 6a ) . Considering bouts involving particular individuals , there were indications that targets might receive a conflict-driven increase in grooming from other subordinates not seen for preselected control subordinates ( those whose squeals had not been played back ) , but no statistically significant differences . The proportion of time grooming that involved the simulated target was doubled on conflict evenings ( mean ± SE: 0 . 31 ± 0 . 09 ) compared to control evenings ( 0 . 15 ± 0 . 06; Z = 1 . 572 , N = 8 , p = 0 . 156; Figure 6b ) , whereas there was , if anything , a decrease for the preselected control subordinate ( control: 0 . 37 ± 0 . 12; conflict: 0 . 28 ± 0 . 09; Z = 0 . 280 , N = 8 , p = 0 . 843; Figure 6c ) . The treatment difference in mean bout duration was also greater for grooming involving simulated targets ( 36 ± 14 s , N = 3 pairs of trials ) than that involving preselected control subordinates ( 22 ± 24 s , N = 3 pairs of trials ) , but too few matched evenings involved the relevant individuals to allow statistical testing .
Dwarf mongoose bystanders did not engage in any obvious post-conflict affiliation in the immediate aftermath of natural or simulated foraging displacements involving a dominant and subordinate group member , but did adjust their later grooming behaviour at the evening sleeping refuge following a simulated increase in within-group conflict during the afternoon . The increase in the average duration of later subordinate–subordinate grooming is in line with the increase in bystander–bystander grooming seen in some species in the immediate aftermath of a contest ( Judge and Mullen , 2005 ) . Such affiliation could reduce the group-wide social anxiety induced by aggression ( De Marco et al . , 2010; Judge and Bachmann , 2013; Schino and Sciarretta , 2015 ) . The later reduction in grooming of aggressors by bystanders is , to our knowledge , the first evidence for a change in this direction; some previous studies have documented increased grooming of aggressors by bystanders in the immediate aftermath of a single contest ( Cordoni and Palagi , 2015; Palagi et al . , 2008; Pallante et al . , 2018 ) , whilst a few others have found no evidence for such an increase ( Judge , 1991; Romero et al . , 2008; Verbeek and de Waal , 1997 ) . Subordinate bystanders could be avoiding the aggressor to reduce the likelihood of redirected aggression , which parallels the main strategy employed in the immediate aftermath of contests by meerkat ( Suricata suricatta ) and rook ( Corvus frugilegus ) targets attempting to avoid renewed aggression ( Benkada et al . , 2020; Kutsukake and Clutton-Brock , 2008 ) . Our results support previous research showing that within-group conflict can affect interactions beyond those between the protagonists and highlight that bystanders can employ different conflict-management strategies depending on the identity of the group members involved . Disagreements between two individuals can thus have wide-reaching and varied implications for affiliative behaviour , which underpins dyadic relationships and social structure in dwarf mongooses ( Kern and Radford , 2021; Kern and Radford , 2018 ) and many other group-living species ( Cameron et al . , 2009; Radford and Du Plessis , 2006; Silk et al . , 2009 ) . The changes in evening grooming patterns once the group had moved from their afternoon foraging areas demonstrate that individuals can retain information relating to earlier within-group conflict and use it when making later decisions about management-related behaviours . It is possible that hearing foraging displacements generates stress in bystanders , which could still be elevated on arrival at the sleeping burrow . But the specific reduction in grooming of perceived aggressors and not of non-playback dominants suggests decision-making about which groupmates to groom rather than a general byproduct of an altered physiological state . Numerous studies on a range of species have found changes in affiliative behaviour between various combinations of protagonists and bystanders in the minutes after within-group contests ( Aureli et al . , 2002; de Waal , 2000 ) . It remains unknown whether , in those species , there might be delayed effects ( as we have found ) of those contests that do not result in immediate changes in affiliative behaviour . By using call playbacks , we did not alter the state of the individuals who were simulated to be the aggressor and target , and thus can rule out the possibility that differences in grooming result from experimentally induced satiation effects ( which might have been the case if we had caused foraging displacements with the presentation of food items; Sharpe et al . , 2013 ) . Moreover , the use of playback simulations , rather than generation of actual contests , means that the delayed grooming effects are most likely driven by subordinate bystanders behaving differently towards perceived protagonists , rather than solicitation or rejection of grooming by the latter . Previous work has shown that dwarf mongooses can use vocal information to detect earlier cooperative contributions by groupmates and then reward them later ( Kern and Radford , 2018 ) ; we now provide experimental evidence for delayed post-conflict management behaviour . It is thus increasingly apparent that mongooses , as well as many other social species , are constantly monitoring the behaviour and interactions of groupmates , and are using memories of what they learn to inform later decisions ( Tibbetts et al . , 2020; Wittig et al . , 2014 ) . The cognitive demands of tracking individuals and their behaviours , remembering that information , and using it when making decisions explain why social interactions within ( Dunbar and Shultz , 2007 ) and between ( Ashton et al . , 2020 ) groups are believed to be strong drivers of animal intelligence . Our experiments show that dwarf mongooses can extract information about within-group conflicts , and the identity of at least some protagonists , from vocal cues alone . This adds to a growing body of work demonstrating the ability of social species to garner information acoustically about aggressive interactions ( Gouzoules et al . , 1984; Slocombe et al . , 2010; Slocombe and Zuberbühler , 2007 ) ; for example , male little blue penguins had an increased heartrate after hearing vocalisations produced by winners of a contest compared to those produced by losers ( Mouterde et al . , 2012 ) . Our findings also complement the small number of studies showing that social animals use vocalisations to assess the behaviour , such as the reliability ( Blumstein et al . , 2004 ) and cooperative contributions ( Kern and Radford , 2018 ) , of individually identifiable groupmates . Acoustic monitoring is beneficial as it allows information acquisition in environments where it would be difficult to do so visually ( eg , in low-light and dense vegetation ) or when group members are widely scattered and communication is needed over long distances ( Bradbury and Vehrencamp , 2011 ) . Moreover , acoustic information can be gathered at a relatively low cost: it can be done whilst still actively foraging ( Hollén et al . , 2008 ) and , in the case of aggressive encounters , at a safe distance that minimises the risk of the information-gatherer receiving any redirected aggression . Monitoring behaviours acoustically is likely not possible for all within-group interactions ( eg , grooming ) or in all social systems , but the calls commonly produced during and at the end of aggressive contests ( Bertram et al . , 2010; Slocombe et al . , 2010 ) provide a valuable means for bystanders to inform subsequent decision-making . We found clear evidence that subordinate bystanders engage in less grooming with simulated aggressors , but whether they increased their grooming with the simulated target is less clear-cut . There are several possible explanations for this difference in the strength of response exhibited to the two protagonists . First , all subordinates might be wary of the aggressor and so potentially reduce their grooming with that individual , whereas perhaps only those who are strongly bonded to the target might engage in extra grooming with it ( Fraser et al . , 2009; Fraser et al . , 2008 ) ; any such target-related effect might be diluted by considering all subordinates in analyses . Strong within-group relationships are apparent in dwarf mongoose groups ( Kern and Radford , 2021; Kern and Radford , 2016 ) , but we do not have the power in this study to consider how relationship quality influences delayed post-conflict grooming . Alternatively , there could be selective attention towards high-ranking individuals ( Chance , 1967 ) : there may be higher selective pressure to discriminate vocalisations from dominant individuals cf those from other subordinates if the former are more important in terms of social relationships and status . In many primate species , for example , individuals focus attention on higher-ranking groupmates or those with whom they have an antagonistic relationship ( Keverne et al . , 1978; McNelis and Boatright-Horowitz , 1998 ) , possibly to avoid aggression ( Schino and Sciarretta , 2016 ) . Since our simulated aggressors were dominants and our simulated targets were subordinates , the stronger effect of increased conflict on grooming with the former could reflect such an attention bias . Another possible reason for the difference in grooming responses to aggressors and targets could be differences in the natural acoustic properties of aggressive growls and submissive squeals ( Gustison and Townsend , 2015 ) . In principle , squeals might encode less identity information than growls ( Owren and Rendall , 2003; Rendall et al . , 1996 ) , although a number of studies have found that calls similar in structure and function to dwarf mongoose squeals are individually identifiable ( Cheney and Seyfarth , 1980; Fischer , 2004; Gouzoules et al . , 1984; Slocombe and Zuberbühler , 2005 ) . In addition , our playback contained three growls and one squeal ( to reflect natural foraging displacements ) , which could have made growls more salient or memorable and/or aided easier discrimination of the aggressor compared to the target . It might also be more cognitively demanding for receivers to discriminate the squeals from multiple subordinate individuals in a group , compared to growls , which are highly likely to come from one of the two dominant individuals . Finally , since contest-related vocalisations may vary depending on the severity of an attack ( Gouzoules et al . , 1984; Slocombe and Zuberbühler , 2007 ) , it is possible that we used less salient squeals than growls in our playbacks ( although both were recorded during natural foraging displacements ) . Future work is required to tease these possibilities apart . In summary , our results demonstrate that dwarf mongooses can obtain information about within-group contests ( including protagonist identity ) acoustically , retain that information , and use it to inform decisions about conflict management with a temporal delay . Such a delay might be most apparent in situations where there is little opportunity for immediate post-contest affiliation ( as is the case with foraging dwarf mongooses ) ; it may also be most apparent when there is a cumulative build-up of unresolved conflict . Our results come from a single population of dwarf mongooses , but we do not believe that there were any obvious biases introduced into the data collection that would limit the generalisability of the findings ( see the end of ‘Materials and methods’ for an assessment of the STRANGEness of our test sample ) . There is increasing experimental evidence that social animals can remember past events and take these into account when deciding whether to get involved in a contest ( Borgeaud and Bshary , 2015; Cheney et al . , 2010; Tibbetts et al . , 2020; Wittig et al . , 2014 ) ; we demonstrate that this ability extends to post-conflict affiliative behaviour . More generally , our results showcase the importance of considering group-wide consequences of dyadic within-group interactions and of looking for effects beyond the immediate aftermath .
We conducted our study on Sorabi Rock Lodge ( 24° 11′S , 30° 46′E ) , a private game reserve in the Limpopo Province , South Africa; full details are available in Kern and Radford , 2013 . This is the site of the Dwarf Mongoose Research Project ( DMRP ) , which has been studying a wild population of dwarf mongooses since 2011 . At the time of the study ( June to October 2019; non-breeding season ) , eight dwarf mongoose groups ( mean ± SE group size: 12 . 3 ± 1 . 7 , range: 5–16 ) were fully habituated to the close presence ( <5 m ) of human observers on foot . All the individuals in the population were identifiable , either through dye marks on their fur ( blond hair dye applied using an elongated paintbrush ) or natural features , such as scars . Individuals older than 1 year were classified as adults ( Kern et al . , 2016 ) data collection was focussed on adults as younger individuals are seldom involved in foraging displacements . Adults were sexed by observing ano-genital grooming ( Kern et al . , 2016 ) and classified as being either dominant ( the male and female breeding pair ) or subordinate; dominance status was established through observation of targeted aggression , scent marking , and reproductive behaviour ( Kern and Radford , 2013; Rasa , 1977 ) . To determine the natural frequency of foraging displacements in our experimental period , we recorded all observer-detected occurrences of such behaviour during observation sessions; this included displacements that were seen and heard . The calculated rate is likely a conservative estimate as an observer could have missed a foraging displacement ( particularly when the group was relatively widely scattered ) . We used data collected ad libitum as part of the long-term DMRP to assess the likelihood of particular dyads of individuals ( aggressor–target: dominant–dominant , dominant–subordinate , subordinate–subordinate , subordinate–dominant ) being involved in a foraging displacement . To collect data on responses to natural foraging displacements , we conducted paired focal watches ( conflict and control ) of 2–3 min duration on 16 subordinate group members in six groups whilst they were foraging; conflict and control focal watches did not differ significantly in their duration ( Wilcoxon signed-rank test: Z = 0 . 952 , N = 16 , p = 0 . 380 ) . A conflict watch was carried out immediately after a foraging displacement was heard by the observer , whilst a control watch was carried out when there had been no foraging displacement ( or any other agonistic interaction ) for at least 10 min . We only carried out focal watches when the relevant mongoose was in a medium-cover habitat ( 20–60% ground cover ) , weather conditions were calm ( still or light breeze ) , there had been no alarm call ( conspecific or heterospecific ) in the previous 10 min , there had been no predator encounter or inter-group interaction for at least 30 min , and the focal individual was not on the periphery of the group . We abandoned focal watches , and repeated them later , if the focal individual stopped general foraging activities or if there was an alarm call within the first 2 min . Otherwise , we aimed to collect 3 min of uninterrupted data , but if a behavioural change or alarm call occurred between the second and third minute , then the focal watch was retained . Pairs of watches on the same focal individual were completed within 1 month ( mean ± SE: 8 . 1 ± 2 . 7 days apart , range: 0–30 days ) ; group composition always remained the same between a pair of watches , and a minimum of 1 hr was left between watches that were conducted on the same day . We watched nine individuals first in control conditions and seven first following a foraging displacement . During each focal watch , we recorded behavioural data to a Dictaphone ( ICD-PX312 , Sony; Sony Europe Limited , Surrey , UK ) . Dwarf mongooses have two types of vigilance behaviour: vigilance scans , where individuals temporarily stop foraging in a head-down position to scan their surroundings ( Kern et al . , 2016 ) , and sentinel behaviour , where individuals cease foraging to scan from a raised position ( minimum 10 cm above the ground level; Kern and Radford , 2013 ) . Throughout each focal watch , we dictated the start and end point of each vigilance scan and sentinel bout , along with the occurrence of any grooming interaction with a groupmate . These data were used to calculate the proportion of time spent vigilant; no grooming occurred during these focal watches . No individual acted as a sentinel during the observational focal watches , and therefore the vigilance response measure was based on scan data only . We used a Wilcoxon signed-rank test to analyse the proportion of time vigilant in SPSS 24 ( IBM Corp , 2016 ) . Due to small sample sizes , we used the Monte Carlo repeated sampling method ( based on 10 , 000 samples ) to calculate an unbiased estimate of the exact p-value ( Mehta and Patel , 2011 ) . We conducted two field-based repeated-measures experiments using playbacks to simulate the occurrence of conflict between group members . Each experiment involved the playback of ‘conflict’ and ‘control’ tracks . We recorded all calls for track creation when weather conditions were calm using a Marantz PMD660 professional solid-state recorder ( Marantz America , Mahwah , NJ ) connected to a handheld Sennheiser ME66 directional microphone ( Sennheiser UK , High Wycombe , Buckinghamshire , UK; frequency response: 40–20 , 000 Hz ) with a Rycote softie windshield ( Rycote Microphone Windshields , Stroud , Gloucestershire , UK ) . The Marantz was set to record at 48 kHz with a 16-bit resolution , and files were saved in wav format . For conflict tracks , we recorded aggressive growls and submissive squeals opportunistically from natural foraging displacements or from conflicts induced by the presentation of a small amount of hard-boiled egg . Growls were recorded from either the dominant male or the dominant female in each group and squeals were recorded from a subordinate male or female in each group; all recorded calls came from foraging displacements where the dominant was the aggressor and the subordinate was the target . We recorded close calls , for use in both control and conflict tracks , from the same dominant and subordinate individuals whilst they were foraging . Recordings of all vocalisations were made 0 . 5–5 m from the relevant individual . We formed 40 s playback tracks in Audacity ( version 2 . 1 . 3 ) by extracting calls of good signal-to-noise ratios from original recordings and inserting them into ambient-sound recordings; ambient sound was recorded from the centre of the territory of the focal group on calm days and in the absence of dwarf mongooses . The first 36 s of each track ( conflict and control ) consisted of non-overlapping close calls from the relevant dominant and subordinate individual , with a rate of one close call every 6 s per individual . This rate of close calling falls within the natural range ( Kern and Radford , 2013 ) . For conflict tracks , the last 4 s consisted of a sequence of three growls from the dominant followed by one squeal from the subordinate; multiple growls and a single squeal reflect natural foraging displacements ( personal observation ) . In control tracks , the last 4 s consisted of three close calls from the dominant followed by one close call from the subordinate , to match the number of vocalisations in conflict tracks . Individual tracks always contained vocalisations from same-sex individuals . We created nine unique conflict and control tracks for each group . Given that the first 36 s of each track comprised close calls from the dominant and subordinate individual , we created three close-call sequences for each individual ( each sequence contained six close calls ) , resulting in nine unique close-call combinations . For the conflict tracks , in which the last 4 s contained growls and a squeal , we created three different growl sequences for the dominant ( each sequence consisted of three growls ) , which were each combined with three separate squeals from the subordinate . Lastly , for the final 4 s of the control tracks , we made three close-call sequences for the dominant ( each sequence contained three close calls to match the number of growls in conflict tracks ) and combined these with three different close calls from the subordinate . We applied a low-pass filter ( set to 200 Hz ) to all tracks to remove low-frequency disturbances . We played back tracks from an iPhone ( Apple , Cupertino , CA ) , connected to a Rokono B10 ( London , UK ) portable loudspeaker ( frequency response: 90–20 , 000 Hz ) concealed in vegetation . We set the amplitude to a sound-pressure level of 55 dB ( A ) at 1 m for close calls and growls , and 65 dB ( A ) at 1 m for squeals . This was the relevant amplitude of these vocalisations as determined by measurement of natural calls with a HandyMAN TEK 1345 sound-level meter ( Metrel UK Ltd , Normanton , UK ) . Experiment 1 was a complement to the observational focal watches ( see ‘Observational data collection’ ) , aiming to test whether bystanders might garner information about within-group conflict solely from vocalisations and then adjust their immediate affiliative behaviour . We randomly selected 17 subordinate individuals ( excluding those whose calls were used in the playback tracks ) to receive the two treatments ( conflict and control ) on separate days and in a counterbalanced order . Each treatment was repeated two to three times per individual during the same observation session , using a different playback track each time , with a minimum of 10 min between repeats; for one individual , it was possible to run one of the treatments only once . We completed the two treatments for the same individual within 2 weeks of each other ( mean ± SE: 2 . 8 ± 0 . 7 days apart , range: 1–11 days ) and at the same time of day ( either between 07:00 and 12:00 or between 12:30 and 17:30 ) . The 17 focal individuals were from eight groups; for groups where there was more than one focal individual ( N = 4 groups ) , we completed both treatments on one individual before moving on to the next . We conducted playbacks when the focal individual was foraging in a medium habitat with little or no breeze and when the callers in the playback were not the focal individual’s nearest neighbour ( other pre-requisites are detailed in ‘Observational data collection’ ) . Where possible , we placed the loudspeaker in the general direction of the playback individuals . As soon as the playback finished , we conducted a 2–3 min focal watch; the mean duration of focal watches was not significantly different between treatments ( Wilcoxon signed-rank test: Z = 1 . 397 , N = 17 , p = 0 . 168 ) . Collection of vigilance and grooming data was identical to that for observational focal watches . As for the natural foraging displacements ( see 'Observational data collection' ) , we analysed the proportion of time spent vigilant; no grooming occurred in any focal watches . Since each treatment was repeated two to three times on an individual , we analysed the mean proportion of time spent vigilant with a Wilcoxon signed-ranks test . In 5 out of 94 trials , an individual acted as a sentinel . We therefore ran the vigilance response measures including and excluding this sentinel behaviour . The data reported in the ‘Results’ section are those excluding sentinel bouts , but qualitatively similar results were found for those including this behaviour . Experiment 2 aimed to test whether there was a delayed effect of within-group conflict on affiliation between group members . We gave eight groups two treatments each on separate days , with treatment order counterbalanced between the groups . On conflict days , the perceived level of within-group conflict was increased during the afternoon by a playback of up to nine conflict tracks . On control days , perceived levels of within-group conflict were unmanipulated; up to nine control tracks were played back during the afternoon instead . There was no treatment difference in the number of natural foraging displacements that occurred throughout the afternoon ( Wilcoxon signed-rank test: Z = 1 . 725 , N = 8 , p = 0 . 158 ) . We completed the two treatments for the same group within 2 weeks of each other ( mean ± SE: 3 . 3 ± 1 . 0 days apart , range: 1–9 days ) . Trials were only attempted when the weather conditions were suitable ( not too windy or cold ) and were abandoned if any major disturbances occurred during the afternoon ( eg , predation attempts , inter-group interactions , multiple latrine events ) . On a trial afternoon , we played back tracks from the centre of the foraging group approximately every 20 min during the 3 hr period before the group started moving to an evening sleeping refuge . There were five trials ( two conflict , three control ) where circumstances ( eg , groups on the move , individuals foraging too far apart ) prevented us from completing all nine planned playbacks in an afternoon ( mean ± SE number of playbacks per trial: 8 . 5 ± 0 . 2 , range: 6–9 ) before the group headed to their sleeping refuge . Once at the refuge ( always termite mounds ) , we recorded all instances of adult grooming behaviour ad libitum until the mongooses went below the ground for the night; it is possible to collect data on all group members simultaneously because they are within a small area around the refuge compared to being scatted more widely when foraging ( ie , in Experiment 1 ) . Data collection involved dictating the identity of grooming partners and the start and end point of each bout . Periods of grooming data collection at the refuge ( mean ± SE: 15 . 5 ± 2 . 3 min , range: 2–37 min ) were not significantly different in duration between treatments ( Wilcoxon signed-rank test: Z = 1 . 332 , N = 8 , p = 0 . 209 ) . To analyse the overall grooming data at the refuge ( including grooming bouts >5 s; Kern and Radford , 2018 ) , we constructed mixed models in RStudio 3 . 6 . 2 ( R Core Team 2019 ) using the packages lme4 ( Bates et al . , 2015 ) and glmmTMB ( Brooks et al . , 2017 ) . For all models , we included treatment as a fixed effect and nested Individual ID within Group ID as random effects to account for data from the same individuals and groups . Error distributions were chosen such that there were no deviations from normality or homoscedasticity , as checked by graphical examination of residual plots; certain response variables were transformed to meet the assumptions of parametric testing . To assess the significance of treatment ( our one fixed effect ) , we compared a model containing treatment to a model without it ( null model ) using a likelihood ratio test ( analysis of variance ( ANOVA ) model comparison , χ2 test ) . All tests were two-tailed and considered significant below an alpha level of 0 . 05 . We first ran a GLMM to assess whether there was a difference in the likelihood that adult individuals participated in grooming behaviour; our response measure was a binary term—whether the individual engage in any grooming ( Yes or No ) For those individuals that did participate in grooming , we ran additional models to understand this behaviour further . We first analysed in a GLMM the proportion of time that individuals spent grooming ( summed grooming durations for each individual divided by the time available for grooming at the refuge , with the latter defined as the duration between the first and last grooming bout ) . We then considered whether the increase in proportion of time grooming was driven by a greater frequency ( GLMM analysing the number of grooming interactions each individual was involved in , with log ( duration ) as an offset term to account for differences in the time available for grooming ) or an increase in mean bout duration ( LMM ) . We subsequently ran Wilcoxon signed-rank tests in SPSS 24 ( as in ‘Observational data collection’ and ‘Experiment 1 protocol’ ) to consider the grooming behaviour between specific categories of group members ( see ‘Results’ ) . We have evaluated the STRANGEness of our test sample ( Webster and Rutz , 2020 ) and believe that for the research topic in question there was minimal introduced bias . We worked with free-living animals from a wild population of dwarf mongooses , so no trapping or housing was involved in the study; all members of the study groups were habituated to close observer presence , and so no bias in random selection occurred due to variation in the ability to approach potential subjects . Focal individuals for observational data collection and Experiment 1 were randomly selected subordinate adults of both sexes from the study groups . Subordinates were chosen since the majority of foraging displacements occur between a dominant individual and a subordinate , and in Experiment 2 , we were interested in comparing how bystanders groomed a perceived aggressor ( one of the dominant pair ) and the other dominant individual . For Experiment 2 , we recorded all instances of adult grooming behaviour in the study groups . The population had not been exposed to these experiments previously . | Social animals that live in groups often have disagreements over access to mates and food . Even fleeting in-group disputes can be costly , disrupting relationships , wasting time and energy , or causing injury if aggression escalates . So , much like humans , many social animals , including primates , birds and dogs , have evolved conflict management strategies to prevent and resolve in-group disagreements . In the immediate aftermath of a conflict , this usually involves changes in the interactions between those involved in the disagreement , or between bystander groupmates and either the victim or aggressor . Less is known about whether social animals can recall past disputes and if they can use conflict management strategies some time after a quarrel has occurred . That is , do aggressive interactions between groupmates influence later social decisions of bystanders in the group ? To investigate , Morris-Drake et al . studied groups of wild dwarf mongooses ( Helogale parvula ) that have become accustomed to living alongside humans in Limpopo Province , South Africa . Dwarf mongooses live in groups of up to 30 individuals , with one dominant breeding pair and lower-ranked helpers . When disagreements arise over food , an aggressor growls deeply and hip-slams the victim away from their foraging patch , stealing the victim’s prey in the process . Victims often produce high-pitched squeals in retreat . Using recordings of these calls , Morris-Drake et al . devised a field experiment to investigate how mongooses responded to nearby conflicts between other group members . Recordings simulating a conflict over food were played to groups of foraging mongooses over the course of an afternoon , so that group members effectively heard what sounded like repeated squabbles between two out-of-sight individuals . Similar to natural conflicts , the mongooses did not engage in any obvious conflict management behaviour immediately after hearing these disputes . But when the group returned to their sleeping burrow that evening , subordinate group members shunned the perceived aggressors from grooming , a key social activity . This work provides evidence that dwarf mongooses keep tabs on conflicts that occur between groupmates . It shows these animals can extract information about conflicts from vocal cues alone and that bystanders use this information when making later social decisions impacting group dynamics . It also adds to growing evidence from baboons , monkeys and chimpanzees that social animals can remember past events and take these into account when interacting with groupmates . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"evolutionary",
"biology"
] | 2021 | Experimental evidence for delayed post-conflict management behaviour in wild dwarf mongooses |
The final identity of a differentiated neuron is determined by multiple signaling events , including activity dependent calcium transients . Non-canonical Frizzled2 ( Fz2 ) signaling generates calcium transients that determine neuronal polarity , neuronal migration , and synapse assembly in the developing vertebrate brain . Here , we demonstrate a requirement for Fz2/Ca2+ signaling in determining the final differentiated state of a set of central brain dopaminergic neurons in Drosophila , referred to as the protocerebral anterior medial ( PAM ) cluster . Knockdown or inhibition of Fz2/Ca2+ signaling during maturation of the flight circuit in pupae reduces Tyrosine Hydroxylase ( TH ) expression in the PAM neurons and affects maintenance of flight . Thus , we demonstrate that Fz2/Ca2+ transients during development serve as a pre-requisite for normal adult behavior . Our results support a neural mechanism where PAM neuron send projections to the α' and β' lobes of a higher brain centre , the mushroom body , and function in dopaminergic re-inforcement of flight .
Genetically encoded developmental programs and neuronal activity together shape the neurotransmitter identity of developing neural circuits . In vertebrates , calcium transients generated by neuronal activity can influence neurotransmitter specification during development and in adults ( Spitzer , 2012; Borodinsky et al . , 2014 ) . One mechanism of generating Ca2+ transients is non-canonical Wnt/Ca2+ signaling initiated by membrane bound Frizzled receptors and a trimeric G-protein ( Slusarski et al . , 1997 ) . Such calcium signals are known to affect neuronal polarity , migration as well as synapse assembly in the developing and mature vertebrate brain ( Varela-Nallar et al . , 2010; Ciani et al . , 2011 ) . Wnt signaling was first identified in Drosophila where multiple genes encode Wnt and Fz proteins ( van Amerongen and Nusse , 2009 ) . However , the role of non-canonical Wnt/Ca2+ signaling during neural development and circuit maturation is poorly understood in invertebrates and its ability to stimulate Ca2+ transients during circuit maturation is unknown . In a screen for G-protein coupled receptors required for flight circuit maturation in Drosophila we identified dFrizzled2 ( dFz2 ) and found that flight deficits upon dFz2 knockdown can be suppressed by over-expression of the intracellular endoplasmic reticular Ca2+ sensor dSTIM ( Agrawal et al . , 2013 ) . Adult neural circuits in Drosophila , including the flight circuit , form in the pupal stages ( Consoulas et al . , 2002 ) , and it is known that maturation of the Drosophila flight circuit requires intracellular Ca2+ signaling ( Banerjee et al . , 2004; Agrawal et al . , 2013 ) . To understand the molecular and cellular basis for such flight deficits , we set out to map neurons that require dFz2 receptor signaling in the context of flight circuit maturation . Insect flight requires computation of multiple sensory inputs and their integration with the flight motor system . This computation and integration presumably occurs in central neurons and allows for control of initiation , maintenance and cessation of voluntary flight bouts ( Gotz , 1987; Strauss , 2002; Strausfeld and Hirth , 2013 ) . Recent work has shown that central dopaminergic neurons in the ventral ganglion modulate a pair of direct flight muscle motor neurons required for wing coordination during flight initiation and cessation ( Sadaf et al . , 2015 ) . In addition , central neurons that compute sensory information in real time and control the timing of a flight bout must exist but remain unknown . Most complex insect behaviors , including flight , are modulated by various monoamines and neuropeptides and in Drosophila , flight can be modulated by octopamine , serotonin and dopamine as well as several neuropeptides ( Taghert and Nitabach , 2012; Sadaf and Hasan , 2014; Van Breugel et al . , 2014 ) . Here , we show for the first time that dFz2 signaling drives the expression of Tyrosine Hydroxylase ( TH ) , the rate-limiting enzyme in dopamine synthesis ( Friggi-Grelin et al . , 2003 ) , during circuit maturation , in a specific set of central brain dopaminergic neurons , called the protocerebral anterior medial ( PAM ) neurons . The PAM cluster consists of approximately 90 dopaminergic neurons , which project to different regions of a higher brain structure called the mushroom body ( MB ) . PAM-MB connectivity has been studied for its role in olfactory associative learning and memory ( Aso et al . , 2012; Liu et al . , 2012 ) where it is thought to signal reward reinforcement . More recently , a PAM-MB circuit was shown to control negative geotaxis behavior in flies ( Riemensperger et al . , 2013 ) . Our studies demonstrate the presence of a novel PAM-MB flight circuit and support a role for PAM-MB synapses in dopaminergic re-inforcement of flight bouts .
To identify neurons , which require dFz2 function for flight , an RNAi strain ( dFz2-IR ) was expressed in independent neurotransmitter domains with the help of the UASGAL4 system for cell and tissue specific expression ( Brand and Perrimon , 1993 ) . Amongst the neuronal domains tested , significant flight deficits were observed upon knockdown of dFz2 in aminergic neurons ( 60% flight time; DdcGAL4 ) and in dopaminergic neurons ( 45% flight time; THGAL4 ) ( Figure 1A , Video 1 ) . Because , DdcGAL4 drives expression in serotonergic and dopaminergic neurons , we tested flies with knockdown of dFz2 in serotonergic neurons ( TRHGAL4 ) . These flies exhibit normal flight bouts in the tethered flight assay ( Figure 1A ) . Similarly , normal flight bouts were observed in flies with knockdown of dFz2 by OK371GAL4 ( mainly glutamatergic neurons ) , GADGAL4 ( many GABA-ergic neurons ) , and P386GAL4 ( a peptidergic neuron subset that expresses in cells with the neuropeptide processing enzyme amontillado; Figure 1—figure supplement 1 ) . Thus , dFz2 function is required primarily in flight circuit neurons , which express GAL4 under control of the pale ( ple ) gene promoter that codes for TH . TH catalyzes a rate-limiting enzymatic step in the synthesis of dopamine ( Friggi-Grelin et al . , 2003 ) . Therefore , expression of this enzyme is considered a reliable marker of dopaminergic neurons . The THGAL4 strain marks a high proportion of TH-positive neurons in the Drosophila central nervous system ( Friggi-Grelin et al . , 2003 ) . 10 . 7554/eLife . 07046 . 003Figure 1 . dFz2 function is required in dopaminergic neurons during development for normal adult flight . ( A ) Percentage flight times of individuals after knockdown of dFz2 in aminergic neurons ( DdcGAL4 ) , serotonergic neurons ( TRHGAL4 ) , dopaminergic neurons ( THGAL4 ) . Knockdown of dFz2 in aminergic neurons ( DdcGAL4 , first bar in red ) and in dopaminergic neurons ( THGAL4 , third bar in red ) showed reduced flight . Knockdowns were compared to their respective GAL4 controls ( gray bars; *p < 0 . 001 , Mann–Whitney U-test ) . ( B ) Percentage flight times of dFz2-IR heterozygotes ( gray bars ) and flies with knockdown of dFz2 in dopaminergic neurons ( red bars ) at specific developmental stages by temperature controlled THGAL4; GAL80ts expression are shown . Flies with knockdown during pupal development exhibit reduced flight similar to knockdown post-egg laying ( PEL ) as compared to controls ( *p < 0 . 001 , Mann–Whitney U-test ) . ( C ) Durations of rhythmic action potentials recorded from the DLMs of air-puff stimulated tethered flies . Bars represent the mean spike duration and diamonds represent the spike duration of an individual recording ( *p < 0 . 001 , Mann–Whitney U-test ) . ( D ) Representative traces of electrophysiological recordings from DLMs of individuals with dFz2 knockdown at the indicated developmental stages are shown . ( E ) Quantification of dFz2 transcript levels after knockdown by dFz2 RNAi in serotonergic ( TRHGAL4 ) and dopaminergic ( THGAL4 ) neurons . The Y-axis represents log2 fold changes calculated by the ΔΔCt method . Each value is the mean ± SEM of three independent experiments , obtained from three independent RNA samples ( *p < 0 . 05 , one-way ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 00310 . 7554/eLife . 07046 . 004Figure 1—figure supplement 1 . Normal flight in flies with knockdown of dFz2 in non-dopaminergic neurons . Percentage flight times of individual flies after knockdown of dFz2 in glutamatergic ( OK371GAL4 ) , mostly GABAergic ( GADGAL4 ) and a peptidergic neuron subset ( P386GAL4 ) . Knockdown of dFz2 in these neuronal domains ( red bars ) did not lead to flight deficits . Knockdown of dFz2 by specific GAL4s was compared to their respective GAL4 controls ( gray bars ) ( *p < 0 . 001 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 00410 . 7554/eLife . 07046 . 005Figure 1—figure supplement 2 . Expression of multiple dFz2-IR strains in dopaminergic neurons exhibits flight defects . ( A ) Percentage flight times of dFz2-IR ( BL27568 , BL31390 , BL31312 ) heterozygotes ( gray bars ) and flies with knockdown of dFz2 in dopaminergic neurons ( red bars ) . Knockdown was achieved by temperature controlled THGAL4; GAL80ts expression . Flies , with knockdown post-embryonic development ( PEL ) , exhibit significant flight defects as compared to controls ( *p < 0 . 001 , Mann–Whitney U-test ) . ( B ) Quantification of dFz2 transcript levels in THGAL4;GAL80ts;BL27568 ( 29°C PEL ) is shown . No significant reduction in transcript levels was observed . The Y-axis represents log2 fold changes calculated by the ΔΔCt method . Each value is the mean ± SEM of three independent experiments obtained from three independent RNA samples . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 00510 . 7554/eLife . 07046 . 006Video 1 . dFz2 knockdown in dopaminergic neurons result in flight defect . Real time video recording of air-puff induced flight in the following genotypes from left to right . ( 1 ) THGAL4/+ , ( 2 ) THGAL4;dFz2-IR , ( 3 ) dFz2-IR/+ . Following a gentle air-puff THGAL4;dFz2-IR flies were able to initiate but not maintain flight for as long as control flies of the genotypes THGAL4/+ and dFz2-IR/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 006 An earlier study demonstrated that pan-neuronal knockdown of dFz2 in pupae leads to flight deficits . However , dFz2 knockdown in adults did not affect flight ( Agrawal et al . , 2013 ) . Existing larval progenitor neurons undergo extensive re-modeling of their axonal and dendritic arbors during pupal stages to form synaptic connections of the mature adult flight circuit ( Fernandes and VijayRaghavan , 1993; Consoulas et al . , 2002 ) . We therefore determined the developmental stage at which dFz2 function is required in dopaminergic neurons for flight . For this purpose , we used the TARGET ( temporal and regional gene expression targeting ) system ( McGuire et al . , 2003 ) . TARGET regulates GAL4 expression by a temperature sensitive GAL80ts element , which can be expressed and repressed at 18°C and 30°C , respectively ( McGuire et al . , 2004 ) . Experimental animals of the genotype THGAL4 , GAL80ts;dFz2-IR were shifted to the permissive GAL4 expression temperature ( 30°C ) either during pupal or adult stages . This allowed stage-specific knockdown of dFz2 . Upon knockdown of dFz2 in TH-expressing neurons through pupal development a significant reduction of flight time was observed , whereas normal flight bouts , as measured for 30 s , were observed upon knockdown of dFz2 in adults ( Figure 1B ) . Flight deficits upon knockdown of dFz2 in pupae were equivalent to those observed upon knockdown throughout post-embryonic development confirming that dFz2 requirement for flight is primarily in pupal dopaminergic neurons during circuit maturation . A physiological correlate of flight is rhythmic patterns of action potentials recorded from the dorsal longitudinal muscles ( DLMs ) during tethered flight . A reduction in duration of flight patterns was observed upon knockdown of dFz2 in TH-expressing neurons during pupal development ( Figure 1C , D ) . dFz2 knockdown at pupal and adult stages in TH-expressing neurons was confirmed by qPCR . As a control , we also confirmed dFz2 knockdown in serotonergic neurons targeted by TRHGAL4 where no flight deficits were observed ( Figure 1E ) . The specificity of flight deficits obtained upon dFz2 knockdown was tested by expression of three additional RNAi strains for dFz2 ( BL27568 , BL31390 , and BL31312 ) . Significant flight defects ranging from 67% to 72% were obtained upon knockdown through post-embryonic development ( Figure 1—figure supplement 2 ) . The difference in flight deficits between dFz2-IR and the three other RNAi strains is very likely due to a difference in their efficacy of knockdown ( compare Figure 1E with Figure 1—figure supplement 2B ) . Therefore , the dFz2-IR strain was used for all subsequent analyses . Dopaminergic neurons marked by THGAL4 have been broadly classified into seven clusters in the brain ( Figure 2A; Table 1 ) . In addition TH-expressing neurons are present in each segment of the ventral ganglion ( Mao and Davis , 2009; Sadaf et al . , 2015 ) . In the brain , two neuronal clusters referred to as PAM and PAL ( Protocerebral Anterior Lateral ) are located in the anterior region , whereas five neuronal clusters , PPM1 , 2 , and 3 ( Protocerebral Posterior Medial ) , PPL1 and 2 ( Protocerebral Posterior Lateral ) are located in the posterior region ( Figure 2A ) . In order to identify TH-expressing neurons that require dFz2 function for flight , three independent GAL4 strains ( THC’ , THC1 , and THF2; Liu et al . , 2012 ) , with differential expression in central brain clusters and the ventral ganglion , were tested ( Table 1 ) . Significant flight deficits were observed upon expression of dFz2-IR under control of THC1GAL4 , but not with THC’GAL4 and THF2GAL4 ( Figure 2B ) . These data suggested that either all or some neurons in the PAM , PPM1 , and T3 regions , marked by THC1GAL4 , but poorly marked or not marked by THC’GAL4 and THF2GAL4 , form part of the flight circuit and require dFz2 signaling during pupal development . Next , we tested two strains ( NP6510GAL4 and R58E02GAL4 ) ( Riemensperger et al . , 2013 ) which drive expression uniquely in the PAM neurons ( Table 1 ) . Significant flight deficits were observed in flies with knockdown of dFz2 by either NP6510GAL4 or R58E02GAL4 ( Figure 2B , C , Figure 2—figure supplement 1 , Video 2 ) implicating these dopaminergic neurons as part of a central brain flight circuit . These data do not rule out a role for additional central neurons or ventral ganglion neurons in the regulation of flight . Expression of dFz2 was confirmed in adult PAM neurons by immunohistochemistry ( Figure 2—figure supplement 2 ) . Knockdown by dFz2-IR in PAM neurons resulted in significant loss of dFz2 immunostaining ( Figure 2—figure supplement 2 ) . Moreover , in support of the pupal requirement for dFz2 ( Agrawal et al . , 2013 ) ( Figure 1B ) , PAM neurons marked by R58E02GAL4 do not express TH in the larval stages ( Figure 2—figure supplement 3 ) , indicating that TH immunoreactivity in these neurons is acquired during pupal maturation , as observed by co-localization of TH immunostaining with R58E02GAL4-driven GFP in pupae ( Figure 2—figure supplement 3 ) . With this we concluded that PAM neurons require dFz2 signaling during functional maturation of the flight circuit in pupae . 10 . 7554/eLife . 07046 . 007Figure 2 . RNAi-mediated knockdown of dFz2 function in Protocerebral Anterior Medial ( PAM ) dopaminergic neurons causes flight deficits . ( A ) Expression pattern of THGAL4 in the anterior and posterior regions of the brain are shown . Dotted line markings show the neuronal clusters . PAM: protocerebral anterior medial; PAL: protocerebral anterior lateral; PPM1 , PPM2 , PPM3: protocerebral posterior medial 1 , 2 , and 3; PPL1 , PPL2: protocerebral posterior lateral 1 and 2 . ( B ) Percentage flight times of heterozygous GAL4 controls ( gray bars ) and GAL4-specific knockdown of dFz2 ( red bars ) . Knockdown of dFz2 in PAM-expressing GAL4 individuals ( THC1GAL4 , NP6510GAL4 , R58E02GAL4 ) resulted in significantly reduced flight times when compared to their respective GAL4 controls ( *p < 0 . 001 , Mann–Whitney U-test ) . ( C ) Durations of rhythmic action potentials recorded from the DLMs of air-puff stimulated tethered flies . Average Spike durations were reduced upon expression of dFz2 RNAi in NP6510GAL4 and R58E02GAL4 as compared to GAL4s controls ( *p < 0 . 001 , Mann–Whitney U-test ) . ( D ) Expression of THC1GAL4 , THF2GAL4 , NP6510GAL4 , and R58E02GAL4 in the PAM neuronal cluster is shown . Except THF2GAL4 , all other GAL4s express in dopaminergic PAM neurons . Expression was analyzed from 10 brain hemispheres . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 00710 . 7554/eLife . 07046 . 008Figure 2—figure supplement 1 . Electrophysiological traces showed reduced firing upon knockdown of dFz2 . Representative traces for electrophysiological recordings obtained from DLMs of flies of the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 00810 . 7554/eLife . 07046 . 009Figure 2—figure supplement 2 . Expression of dFz2 in PAM dopaminergic neurons . Expression of GFP ( Anti GFP; green ) and dFz2 ( Anti Fz2; red ) in PAM dopaminergic neurons of R58E02GAL4> mCD8GFP flies and R58E02GAL4> mCD8GFP; dFz2-IR flies . dFz2 immunostaining was absent upon expression of dFz2-IR . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 00910 . 7554/eLife . 07046 . 010Figure 2—figure supplement 3 . Expression of TH in PAM dopaminergic neurons during development . Expression of GFP ( Anti GFP; green ) and TH ( Anti TH; red ) in the larval and pupal brain of R58E02GAL4>mCD8GFP organisms are shown . Separate confocal stacks for the anterior and posterior regions of the brain are shown . TH immunoreactivity ( red ) co-localizes with R58E02GAL4-driven mGFP ( green ) in pupae but not in larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 01010 . 7554/eLife . 07046 . 011Table 1 . Summary of expression pattern of GAL4sDOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 011THC’THC1THF2NP6510R58E02PAM+++−+++++PAL++−−−PPM1+++−−−PPM2+++−−PPM3−−+−−PPL1−+++−−PPL2+++−−T1++−−−T2−++−−T3−+−−−Ab+++−−Table summarizing the expression pattern of THC’GAL4 , THC1GAL4 , THF2GAL4 , NP6510GAL4 , and R58E02GAL4 in specified dopaminergic neuronal clusters . Clusters shown in Figure 2A and thoracic ganglion ( T1 , T2 , T3 , Ab ) were examined for the expression . Plus ( + ) and minus ( − ) indicate the presence and absence of dopaminergic positive neurons , respectively . Double plus ( ++ ) and triple plus ( +++ ) indicate the presence of >5 and >50 dopaminergic positive neurons , respectively . 10 brain hemispheres were analyzed for the expression . 10 . 7554/eLife . 07046 . 012Video 2 . dFz2 knockdown in PAM neurons result in flight defect . Real time video recording of air-puff induced flight in the following genotypes from left to right . ( 1 ) R58E02GAL4/+ , ( 2 ) R58E02GAL4;dFz2-IR , ( 3 ) dFz2-IR/+ . Following a gentle air-puff R58E02GAL4;dFz2-IR flies were able to initiate but not maintain flight for as long as control flies of the genotypes THGAL4/+ and dFz2-IR/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 012 A role for dFz2 in flight circuit development was originally identified in a screen for GPCRs that signal through changes in intracellular Ca2+ ( Agrawal et al . , 2013 ) . In order to identify molecules that function downstream of Fz2 for development of the adult flight circuit , interactions of candidate genes were tested . Reports from vertebrates suggest that dFz2 activates downstream Ca2+ signaling through non-canonical mechanisms ( Slusarski et al . , 1997 ) . We tested flight in animals with dFz2 knockdown in combination with both canonical ( Figure 3—figure supplement 1 ) and non-canonical ( Figure 3A ) candidate genes . In the canonical signaling pathway dFz2 along with its co-receptor lipoprotein receptor related-protein 5/6 ( LRP5/6 , encoded by Arrow in Drosophila ) activates Dishevelled . Activated dishevelled functions to stabilize β catenin ( Drosophila armadillo ) and hence promote β catenin entry into the nucleus followed by enhanced transcription of downstream target genes . We over-expressed wild-type Dishevelled , a point mutant ( G64V ) in the DIX-domain of Dishevelled ( specifically activates the canonical pathway; Penton et al . , 2002 ) and a constitutively active form of Armadillo ( UAS-armS10; Morel and Arias , 2004 ) in the background of dFz2 down-regulation either in dopaminergic neurons or across all neurons . The resultant progeny were tested for flight . Up-regulation of canonical signaling molecules did not rescue flight deficits of dFz2 down-regulation ( Figure 3—figure supplement 1 ) . Moreover , normal flight times were observed in flies with RNAi knockdown of canonical dFz2 pathway components like LRP5/6 , Dishevelled , and GSK3β in either the pan-neuronal domain or in dopaminergic neurons ( Figure 3—figure supplement 1 ) . RNAi strains for dishevelled and GSK3β were validated by quantitative PCR ( qPCR; Figure 3—figure supplement 1 ) , whereas RNAi for LRP5/6 was validated by ubiquitous expression with Act5CGAL4 that resulted in embryonic lethality ( Dietzl et al . , 2007 ) . These results do not support a role for canonical dFz2 signaling in dopaminergic neurons for maturation of the flight circuit in pupae . 10 . 7554/eLife . 07046 . 013Figure 3 . dFz2 function is mediated through G-protein Go and IP3-mediated calcium signaling in dopaminergic neurons . ( A ) A schematic showing dFz2-mediated activation of Go followed by IP3R-mediated Ca2+ signaling pathway and Store-operated Ca2+ entry ( SOCE ) through dSTIM and dOrai . Red ( down-regulation ) and green ( over-expression ) arrows indicate the two strategies used for testing this signaling mechanism . ( B ) Percentage flight times of the indicated genotypes are shown . Knockdowns flight times were compared to their respective heterozygote controls , whereas AcGo rescue of dFz2 knockdown was compared to dFz2 knockdown ( *p < 0 . 001 , Mann–Whitney U-test ) . ( C ) Percentage flight times of heterozygous controls ( gray bars ) followed by over-expression of calcium signaling molecules ( itpr+ , dStim+ , dOrai+ ) in flies with knockdown of dFz2 ( green bars ) . Overexpression of calcium signaling molecules ( itpr+ , dStim+ , dOrai+ ) rescued flight defects significantly when compared to flies with dFz2 knockdown ( *p < 0 . 001 , Mann–Whitney U-test ) . ( D ) Durations of rhythmic action potentials recorded from the DLMs of air-puff stimulated tethered flies . Spike durations were reduced upon expression of Go RNAi or UAS-PTX . 16 in dopaminergic neurons and partially rescued upon over-expression of calcium signaling molecules ( itpr+ , dStim+ , dOrai+ ) when compared to knockdown of dFz2 ( *p < 0 . 001 , Mann–Whitney U-test ) . ( E ) Representative electrophysiological recordings from DLMs of the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 01310 . 7554/eLife . 07046 . 014Figure 3—figure supplement 1 . The canonical Fz2/β catenin signaling pathway does not function downstream of dFz2 in the context of flight circuit maturation . ( A ) A schematic of canonical Wnt/dFz2 signaling . Upon activation by Wnt , in collaboration with a co-receptor ( lipoprotein receptor-related protein or LRP5/6; encoded by gene Arrow in Drosophila ) , dFz2 activates the cytosolic protein Dishevelled , which is an inhibitor of a negative regulator , Axin of the pathway . Axin is required for formation of a β catenin destruction complex . Activated Dishevelled prevents formation of the β catenin destruction complex and hence stabilizes formation and accumulation of β catenin ( encoded by Armadillo ) . β catenin moves to the nucleus and activates downstream signaling . ( B ) Percentage flight times for heterozygotes ( gray bars ) ; dopaminergic ( THGAL4 ) and pan-neuronal ( ElavC155GAL4 ) over-expression of Dishevelled ( Dsh ) , point mutant in DIX-domain of Dishevelled ( DshG64V ) ( yellow bars ) or activated form of Armadillo ( Armact ) ( green bars ) in the background of dFz2 knockdown are shown . Over-expression of indicated transgenes was compared to the dFz2 knockdown and found not to be significantly different ( p < 0 . 001 , Mann–Whitney U-test ) . ( C ) Average percentage flight times ( bars ) of individuals ( diamonds ) with knockdown of LRP5/6 ( Arr-IR; blue bars ) , Dishevelled ( Dsh-IR; yellow bars ) or GSK-3 ( Sgg-IR; red bars ) in dopaminergic ( THGAL4 ) and pan-neuronal ( ElavC155GAL4 ) domains . ( D ) Quantification of transcripts in total RNA isolated from heads , upon expression of RNAi for Dishevelled ( Dsh-IR ) and Shaggy ( Sgg-IR ) is shown . The Y-axis represents log2 fold changes calculated by the ΔΔCt method . Each value is the mean ± SEM of three independent experiments , obtained from three independent RNA samples . Gene expression was reduced significantly as compared to the THGAL4 control ( *p < 0 . 05 , one-way ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 01410 . 7554/eLife . 07046 . 015Figure 3—figure supplement 2 . Go functions downstream of dFz2 in the context of flight circuit maturation . Percentage flight times are shown for flies with dopaminergic specific expression of UAS-PTX . 16 ( red bars ) at the indicated developmental stages obtained by temporal activation of THGAL4; GAL80ts . Flight deficits observed were significantly different from THGAL4; GAL80ts heterozygote controls ( gray bars; *p < 0 . 001 , Mann–Whitney U-test ) . Over-expression of AcGi , AcGq , or AcGs in the background of dopaminergic ( THGAL4 ) -driven dFz2 knockdown either did not rescue flight or affected viability . Flight in organisms with pan-neuronal ( ElavC155GAL4 ) knockdown of dFz2 ( red bars ) , compared with constitutively active forms of Gi ( AcGi ) , Gq ( AcGq ) , or Gs ( AcGs ) in the background of dFz2 knockdown ( green bars ) during pupal development , was not significantly different ( p < 0 . 001 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 01510 . 7554/eLife . 07046 . 016Figure 3—figure supplement 3 . Non-canonical dFz2/Ca2+ signaling functions downstream of dFz2 in the context of flight circuit maturation . ( A ) Schematic of putative non-canonical dFz2/Ca2+ signaling depicting the strategies used for testing this pathway . Red arrows represent knockdown and green arrows show over-expression of the indicated molecule . ( B ) Percentage flight times of individuals of the indicated genotypes . Over-expression of dFz2 partially rescues the flight defects shown by pan-neuronal knockdown of calcium signaling molecules ( itpr , dStim , dOrai ) ( *p < 0 . 001 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 016 Next , we tested flight deficits by expression of previously implicated non-canonical candidates that link dFz2 activation with Ca2+ signaling ( Figure 3A ) ( Sheldahl et al . , 2003 ) . From genetic studies , we know that the heterotrimeric G-protein , Gq , which links GPCR activation to intracellular store calcium release , does not function downstream of dFz2 signaling in the context of Drosophila flight ( Agrawal et al . , 2013 ) . Therefore , we tested the requirement of other heterotrimeric G-proteins from Drosophila . Constitutively , active forms of the α subunits of Gs ( UASAcGs ) , Go ( UASAcGo ) , and Gi ( UASAcGi ) were tested in flies with dFz2 knockdown in pupal stages . Pan-neuronal expression of AcGo in pupae with dFz2 knockdown , rescued flight defects to a significant extent ( Figure 3B ) , whereas constitutively active forms of Gi , Gq , or Gs did not ( Figure 3—figure supplement 2 ) . A partial rescue of flight defects was also observed upon AcGo expression in dopaminergic neurons in pupae ( Figure 3B ) . These data support Go activation by dFz2 in dopaminergic neurons of the maturing flight circuit during pupal development . To confirm the requirement of Go in dopaminergic neurons , we down-regulated Go function either by expression of an RNAi construct ( Vecsey et al . , 2014 ) THGAL4;GoRNAi or by expression of pertussis toxin which inhibits Go function ( THGAL4;UASPTX . 16 ) and tested the progeny for flight . In Drosophila , pertussis toxin is a selective inhibitor of Go signaling ( Hopkins et al . , 1988; Ferris et al . , 2006 ) . Expression of Go-IR and PTX . 16 in dopaminergic neurons reduced both flight times ( Figure 3B ) , and the maintenance time of flight patterns recorded from the DLMs ( Figure 3D , E ) . Moreover , down-regulation of Go signaling in PAM neurons with R58E02GAL4 , by expression of either Go-IR or PTX . 16 resulted in significant loss of flight ( Figure 3B ) . PTX induced flight deficits required expression in pupae and not in adults ( Figure 3B and Figure 3—figure supplement 2 ) . Flight deficits induced by pan-neuronal knockdown of dFz2 were rescued to a significant extent by over-expression of the ER Ca2+ depletion sensor , dSTIM+ ( Agrawal et al . , 2013 ) , suggesting that activation of Go by dFz2 evokes Ca2+ signals in Drosophila neurons . As in other organisms ( Feske et al . , 2006; Prakriya et al . , 2006; Vig et al . , 2006 ) , in Drosophila neurons as well Ca2+ release through the IP3R leads to clustering of dSTIM , which in turn promotes Store-operated Ca2+ entry ( SOCE ) through dOrai ( Venkiteswaran and Hasan , 2009; Agrawal et al . , 2010 ) ( Figure 3A ) . In a converse experiment , we tested the effect of over-expression of a dFz2+ transgene on flight deficits induced by pan-neuronal knockdown of the IP3R ( itpr-IR ) , dSTIM ( dSTIM-IR ) and dOrai ( dOrai-IR ) and observed a significant rescue in all three conditions tested ( Figure 3—figure supplement 3 , Video 3 , and Video 4 ) . These data suggest that , as in vertebrate neurons , dFz2 links to intracellular calcium signaling in Drosophila . Next , we tested the effect of over-expression of dSTIM+ , on flight deficits induced by dFz2 knockdown in dopaminergic neurons . A partial but significant rescue of flight was observed ( Figure 3C ) accompanied by a rescue of the duration of firing patterns from the DLMs ( Figure 3D , E ) . Over-expression of the IP3R and dOrai also rescued flight deficits observed by knockdown of dFz2 in dopaminergic neurons ( Figure 3C , D , E ) . Together these data support the idea that maturation of dopaminergic neurons in the flight circuit requires intracellular Ca2+ signaling by activation of dFz2 and Go . The mechanism by which Go activates Ca2+ signaling through IP3R/dSTIM requires further investigation ( Figure 3A ) . 10 . 7554/eLife . 07046 . 017Video 3 . Overexpression of IP3R in dopaminergic neurons rescues flight defects of dFz2 downregulation . Real time video recording of air-puff induced flight in the following genotypes from left to right . ( 1 ) THGAL4;dFz2-IR;itpr+ , ( 2 ) THGAL4;dFz2-IR , ( 3 ) dFz2-IR/+ . Following a gentle air-puff THGAL4; dFz2-IR; itpr+ flies were able to initiate and maintain flight for a longer duration as compared to THGAL4;dFz2-IR . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 01710 . 7554/eLife . 07046 . 018Video 4 . Flight defects in dFz2 knockdown individuals can be rescued by over-expression of dSTIM . Real time video recording of air-puff induced flight in the following genotypes from left to right . 1 ) THGAL4; dFz2-IR; dSTIM+ , 2 ) THGAL4; dFz2-IR , 3 ) dFz2-IR/+ . Following a gentle air-puff THGAL4; dFz2-IR; dSTIM+ flies were able to initiate and maintain flight for a longer duration as compared to THGAL4; dFz2-IR . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 018 Synaptic function of developing hippocampal neurons can be modulated by Ca2+ signaling downstream of Fz2 ( Varela-Nallar et al . , 2010 ) . In Drosophila , neuronal activity can be increased by over-expression of a voltage-gated sodium channel , NaChBac ( Nitabach et al . , 2006 ) . Therefore , we tested flight in organisms with dFz2 knockdown and increased neuronal activity by NaChBac expression . Flight was restored close to 100% upon expression of NaChBac in dopaminergic neurons ( THGAL4 ) and more specifically in PAM neurons ( NP6510GAL4 and R58E02GAL4; Figure 4A , Video 5 ) . Moreover , raising neuronal activity during pupal development in parallel with dFz2 knockdown compensated for loss of flight observed in the knockdown condition ( Figure 4B ) . These data suggest that Fz2/Ca2+ signaling can contribute to the synaptic activity of dopaminergic PAM neurons in pupae . The requirement for synaptic activity in maturing PAM neurons was tested directly by expression of a temperature sensitive mutant of the dynamin orthologue , shibirets ( Shits ) . Expression of Shits blocks vesicle endocytosis at 30°C ( Kitamoto , 2001 ) and its expression during pupal development , either in TH neurons ( THGAL4 ) or exclusively in PAM neurons ( NP6510GAL4 and R58E02GAL4 ) resulted in significant loss of flight ( Figure 4C ) . Temporal expression of Shits in adult PAM neurons also resulted in a flight deficit ( Figure 4D ) , supporting the requirement of active synaptic transmission in PAM neurons for adult flight . 10 . 7554/eLife . 07046 . 019Figure 4 . Knockdown of dFz2 affects neuronal activity of maturing flight circuit PAM neurons . ( A ) Percentage flight times of individual heterozygous controls ( gray bars ) , dFz2 knockdown ( dFz2-IR ) in dopaminergic neurons ( THGAL4 ) and PAM neurons ( NP6510GAL4 , R58E02GAL4 ) ( red bars ) followed by over-expression of NaChBac in presence of dFz2-IR ( green bars ) ; ( *p < 0 . 001 , Mann–Whitney U-test ) . ( B ) Percentage flight times for heterozygotes of THGAL4;GAL80ts ( gray bars ) followed by stage-specific knockdown of dFz2 ( red bars ) and over-expression of NaChBac in flies with dFz2 knockdown ( green bars ) as indicated . Over-expression of NaChBac during pupal development rescued flight as did over-expression post-egg laying ( PEL ) ( *p < 0 . 001 , Mann–Whitney U-test ) . ( C ) Percentage flight times upon expression of Shibirets ( Shits 30°C; red bars ) either in pupal or no expression ( Shits 22°C; gray bars ) . Expression was either in dopaminergic neurons ( THGAL4 ) or PAM neurons ( NP6510GAL4 , R58E02GAL4 ) . Flight was tested at 25°C . Expression of Shits in pupal resulted in reduced flight times . ( D ) Percentage flight times upon adult expression of Shibirets ( Shits 30°C; red bars ) or no expression ( Shits 22°C; gray bars ) , in PAM neurons with R58E02GAL4 . Flight was tested at 30°C . Expression of Shits resulted in reduced flight times ( *p < 0 . 001 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 01910 . 7554/eLife . 07046 . 020Video 5 . Increased neuronal activity in PAM neurons rescues flight in individuals with dFz2 knockdown . Real time video recording of air-puff induced flight in the following genotypes from left to right . ( 1 ) R58E02GAL4;dFz2-IR;NaChBac , ( 2 ) R58E02GAL4;dFz2-IR , ( 3 ) dFz2-IR/+ . Following a gentle air-puff R58E02GAL4; dFz2-IR;NaChBac flies were able to initiate and maintain flight for a longer duration as compared to R58E02GAL4;dFz2-IR . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 020 The cellular effect of reduced dFz2 expression in PAM neurons was investigated next . TH levels in PAM neurons marked by R58E02GAL4 , appear reduced upon dFz2 knockdown as judged by immunohistochemistry ( compare anti-TH panels in Figure 5A , B ) . Expression of NaChBac with dFz2-IR restored TH expression close to wild-type levels ( Figure 5A , B , C ) . Quantification of TH immunostaining across multiple samples revealed a significant reduction upon dFz2 knockdown which was restored by expression of NaChBac ( Figure 5D , E ) . Furthermore , TH transcript levels were significantly reduced by dFz2 knockdown and were restored upon expression of NaChBac ( Figure 5—figure supplement 1 ) . Thus , altered TH levels corroborated well with flight deficits and their rescue in various genotypes . Numbers of TH-positive neurons in the PAM cluster were not significantly different between the three genotypes as judged by anti-TH immunostaining ( Figure 5G ) . Surprisingly , R58E02GAL4-driven GFP expression was also reduced upon expression of dFz2-IR and was restored back upon expression of NaChBac ( Figure 5—figure supplement 1 ) . Consequently , there was an apparent reduction in the numbers of GFP positive cells upon dFz2-IR expression which was restored by NaChBac expression ( Figure 5F ) . Because the number of TH-expressing cells of the PAM cluster remained unchanged upon dFz2 knockdown and after NaChBac rescue ( Figure 5G ) , we hypothesized that dFz2/Ca2+ signaling regulates TH expression in PAM neurons during pupal development . Moreover , our data support a role for dFz2/Ca2+ signaling in regulating expression of the R58E02GAL4 transgene where GAL4 is under control of the fumin gene encoding a Dopamine Transporter , DAT ( Liu et al . , 2012 ) . 10 . 7554/eLife . 07046 . 021Figure 5 . Expression of TH is reduced in PAM neurons by dFz2 knockdown . ( A ) Expression of GFP ( Anti GFP; green ) and TH ( Anti TH; red ) is shown in PAM dopaminergic neurons marked by R58E02GAL4>mCD8GFP . ( B ) Significant reduction of GFP and TH immunoreactivity is observed in PAM neurons of R58E02GAL4>mCD8GFP; dFz2-IR individuals; which is ( C ) rescued by over-expression of NaChbac ( R58E02GAL4> mCD8GFP; dFz2-IR;NaChBac ) . ( D ) Scatter plot with the mean intensity of TH expression in individual PAM neurons ( N = 1280 ) in the indicated genotypes . Cells were obtained from 16 brain hemispheres; *p < 0 . 05 , one-way ANOVA . ( E ) A Kolmogorov-Smirnov ( K-S ) plot analyzing the distribution of the mean intensity of TH immunoreactivity in PAM neurons . The frequency distribution is significantly shifted to the left for R58E02GAL4>mCD8GFP;dFz2-IR as compared to R58E02GAL4>mCD8GFP indicating a significantly higher percentage of cells with lower mean intensity . Frequency distribution of R58E02GAL4> mCD8GFP; dFz2-IR; NaChBac is shifted back towards the control distribution R58E02GAL4>mCD8GFP , indicating a significant rescue ( *pK-S < 0 . 05 ) . ( F ) Total number of GFP positive cells and ( G ) TH positive cells were counted in the indicated genotypes . No difference in the number of TH cells was found; however GFP cells were reduced upon dFz2 knockdown ( *p < 0 . 05 , one-way ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02110 . 7554/eLife . 07046 . 022Figure 5—figure supplement 1 . Expression of GFP is altered upon expression of dFz2-IR in PAM neurons using R58E02GAL4 . ( A ) Scatter plot of the mean intensity of GFP expression in individual PAM neurons ( N = 1280 ) from 16 brain hemispheres of the indicated genotypes ( *p < 0 . 05 , one-way ANOVA ) . ( B ) Kolmogorov-Smirnov ( K-S ) plot analyzing the distribution of the cellular mean intensity shown in the scatter plot . The frequency distribution is shifted significantly to the left for R58E02GAL4>mCD8GFP;dFz2-IR as compared to R58E02GAL4>mCD8GFP indicating the presence of a higher percentage of cells with lower mean intensity . Frequency distribution of R58E02GAL4> mCD8GFP; dFz2-IR; NaChBac is shifted towards the right indicating rescue in the percentage of cells with higher mean intensity ( *pK-S < 0 . 05 ) . ( C ) Quantification of TH transcripts upon expression of RNAi for dFz2 with or without NaChBac in PAM neurons ( R58E02GAL4 ) . The Y-axis represents log2 fold changes calculated by the ΔΔCt method . Each value is the mean ± SEM of three independent experiments , obtained from three independent RNA samples . Gene expression of TH was rescued upon expression of NaChBac with dFz2-IR ( *p < 0 . 05 , one-way ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02210 . 7554/eLife . 07046 . 023Figure 5—figure supplement 2 . Altered GFP expression was seen upon expression of dFz2-IR in dopaminergic neurons . ( A ) mGFP expression in the PAM dopaminergic neurons observed in brains of animals of the indicated genotypes . The number of PAM dopaminergic neurons was significantly reduced upon expression of ( B ) dFz2-IR ( arrows ) . The number of GFP-positive cells was partially rescued by over-expression of either ( C ) dSTIM+ ( arrows ) or ( D ) itpr+ ( arrows ) or ( E ) NaChBac ( arrows ) . ( F ) Quantification of cells in specified clusters of dopaminergic neurons , for the indicated genotypes , is shown . Clusters that were examined: PAM , protocerebral anterior medial; PAL , protocerebral anterior lateral; PPM1 , PPM2 , PPM3 , protocerebral posterior medial; PPL1 , PPL2 , protocerebral posterior lateral and T1 , T2 , T3-Ab , thoracic ganglion . The bars represent mean number of cells and circles represent number of cells from ∼15 individual brain hemispheres . Cell number appeared reduced in the PAM cluster upon expression of dFz2-IR ( PAM; red bar ) ; this was partially rescued by over-expression of either dSTIM+ or itpr+ or NaChBac ( PAM; blue , green , magenta bars ) ( *p < 0 . 05 , one-way ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02310 . 7554/eLife . 07046 . 024Figure 5—figure supplement 3 . Knockdown of dFz2 in OK371GAL4-expressing neurons does not affect TH expression-positive PAM neurons . Expression of GFP ( Anti GFP; green ) and TH ( Anti TH; red ) is shown in PAM neurons of the indicated genotypes . No change in the intensity of either GFP or TH immunostaining was observed upon knockdown of dFz2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 024 The status of TH and GFP expression in PAM neurons was further investigated after dFz2 knockdown with THGAL4 . Unlike R58E02GAL4 , THGAL4-driven mGFP marks a small subset of PAM neurons . This consists of two clusters of 6–7 neurons each ( Figure 2A and Figure 5—figure supplement 2 ) ( Riemensperger et al . , 2013 ) . THGAL4-driven expression of dFz2-IR resulted in significant loss of GFP expression in PAM neurons ( Figure 5—figure supplement 2 ) . Interestingly , GFP expression in five TH-positive neurons of the PAL cluster remained unaffected by knockdown of dFz2 , suggesting that dFz2 regulation of TH expression maybe PAM specific ( data not shown ) . Over-expression of either dSTIM+ , IP3R or NaChBac in the background of dFz2 knockdown could partially rescue loss of GFP expression in the PAM neurons ( Figure 5—figure supplement 2 ) . Thus , THGAL4-driven GFP expression in PAM neurons correlated with flight deficits and their rescue ( Figures 1A , 3C , 4A ) . Quantification of TH immunoreactivity in PAM cells by THGAL4-driven dFz2 knockdown was technically not possible because the few THGAL4-positive cells of the PAM cluster could not be identified in the dFz2 knockdown condition . Taken together these data support the idea that dFz2/Ca2+ signaling in PAM neurons drives transcription of two key dopamine synthesis and uptake molecules , TH and DAT . The transcriptional regulation extends to GAL4 transgenic constructs containing TH and DAT regulatory sequences . As controls we tested TH immunoreactivity of PAM neurons in flies with dFz2 knockdown in glutamatergic neurons ( OK371GAL4 ) . Both TH immunoreactivity of PAM neurons ( Figure 5—figure supplement 3 ) and flight patterns ( Figure 1—figure supplement 1 ) were similar to controls . Based on our observation that dFz2 knockdown in PAM neurons leads to flight deficits accompanied by a significant reduction of TH expression , we tested the requirement of TH in PAM neurons for flight . Over-expression of a neuronal-specific TH cDNA transgene ( UASDTH1 ) ( Friggi-Grelin et al . , 2003 ) in flies with dFz2 knockdown by PAMGAL4 strains ( NP6510GAL4 and R58E02GAL4 ) could rescue flight deficits significantly ( Figure 6A , Video 6 , Figure 6—figure supplement 1 ) . Furthermore , knockdown of TH with an RNAi ( dTH-IR ) resulted in significant loss of flight and reduced TH expression ( Figure 6A–C , F , Figure 6—figure supplement 1 ) . Moreover , knockdown of TH in PAM neurons affected R58E02GAL4-driven GFP expression suggesting feedback regulation of dopamine transporter ( DAT ) by dopamine levels . Over-expression of the DTH1 neuronal cDNA could rescue TH immunoreactivity in the R58E02GAL4-expressing PAM neurons with dFz2 knockdown ( Figure 6B , C , E ) . However , GFP immunoreactivity remained low and unchanged between dFz2-IR- and dFz2-IR;DTH1-expressing PAM neurons ( Figure 6—figure supplement 1 ) . These data suggest that rescue of flight by over-expression of DTH1 by passes the transcriptional regulation of DAT by dopamine and of endogenous TH by dFz2/Ca2+ signaling . They confirm the requirement for TH expression in PAM neurons for flight . 10 . 7554/eLife . 07046 . 025Figure 6 . Expression of DTH1 in PAM neurons rescues flight defects shown by dFz2 knockdown . ( A ) Percentage flight times of individual heterozygous control flies ( gray bars ) , flies with expression of dFz2-IR and THRNAi ( dTH-IR ) in PAM neurons ( R58E02GAL4 ) ( red bars ) , and flies with over-expression of DTH1 in the presence of dFz2-IR ( green bars ) . Expression of DTH1 rescued the flight defect of dFz2 knockdown flies to a significant extent ( *p < 0 . 001 , Mann–Whitney U-test ) . ( B ) Scatter plot of the mean intensity of TH expression in individual PAM neurons ( N = 1280 ) from 16 brain hemispheres of the indicated genotypes ( *p < 0 . 05 , one-way ANOVA ) . ( C ) Kolmogorov–Smirnov ( K-S ) plot analyzing the distribution of the cellular mean intensity shown in B . The frequency distribution is significantly shifted to the left for R58E02GAL4> mCD8GFP;dTH-IR as compared to R58E02GAL4>mCD8GFP indicating a higher number of cells with lower mean intensity of TH . Frequency distribution of R58E02GAL4> mCD8GFP; dFz2-IR;DTH1 is shifted back to the right indicating fewer cells with lower mean intensity ( *pK-S < 0 . 05 ) . ( D ) Expression of GFP ( Anti GFP; green ) and TH ( Anti TH; red ) is shown in PAM dopaminergic neurons in R58E02GAL4>mCD8GFP , ( E ) R58E02GAL4>mCD8GFP; dFz2-IR; DTH1 and ( F ) R58E02GAL4> mCD8GFP;dTH-IR . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02510 . 7554/eLife . 07046 . 026Figure 6—figure supplement 1 . Expression of GFP is altered in PAM neurons upon knockdown of dFz2 in the presence of either DTH1 or dsDTH . ( A ) Scatter plot of the mean intensity of GFP expression in individual PAM neurons ( N = 1280 ) from 16 brain hemispheres of the indicated genotypes ( *p < 0 . 05 , one-way ANOVA ) . ( B ) Kolmogorov-Smirnov ( K-S ) plot analyzing the distribution of the cellular mean intensity shown in the scatter plot . The frequency distribution is significantly shifted to the left for R58E02GAL4>mCD8GFP; dFz2-IR; DTH1 and R58E02GAL4>mCD8GFP; dTH-IR individual cells as compared to R58E02GAL4>mCD8GFP expressing neurons indicating a higher percentage of cells with lower mean intensity ( *pK-S < 0 . 05 ) . ( C ) Percentage flight times of individual flies of the indicated genotypes . Expression of DTH1 rescued the flight defect of dFz2-IR expressing individuals to a significant extent ( *p < 0 . 001 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02610 . 7554/eLife . 07046 . 027Video 6 . DTH over-expression in PAM neurons rescues flight in individuals with dFz2 knockdown . Real time video recording of air-puff induced flight in the following genotypes from left to right . ( 1 ) R58E02GAL4;dFz2-IR;DTH1 , ( 2 ) R58E02GAL4;dFz2-IR , ( 3 ) dFz2-IR/+ . Following a gentle air-puff R58E02GAL4; dFz2-IR;DTH1 , flies were able to initiate and maintain flight for a longer duration as compared to R58E02GAL4;dFz2-IR . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 027 PAM neurons send projections to the horizontal lobes of the MB neuropil ( Aso et al . , 2012; Burke et al . , 2012; Liu et al . , 2012; Riemensperger et al . , 2013 ) . The MB is a paired brain structure that controls several higher brain functions in insects ranging from olfactory memory formation and reinforcement ( Kahsai and Zars , 2011; Waddell , 2013 ) to locomotor activity ( Helfrich-Forster et al . , 2002; Serway et al . , 2009; Riemensperger et al . , 2013 ) . Axons and dendrites of Kenyon cells , positioned in the calyx region form the MB neuropil which is subdivided into the α , β , α′ , β′ , and γ lobes ( Strausfeld et al . , 2003 ) . We reasoned that flight deficits observed due to reduced levels of TH in PAM neurons might derive from reduced dopamine release and signaling in postsynaptic MB neurons . This idea was tested by silencing specific MB neuropil lobes with mb186bGAL4 and mb247GAL4 drivers . mb186bGAL4 is a recently generated split GAL4 strain ( Aso et al . , 2014 ) whose expression is restricted to the α′ β′ lobes ( Vogt et al . , 2014 ) , whereas mb247GAL4 is expressed in the α , β , and γ lobes ( Zars , 2000; Krashes et al . , 2007; Pech et al . , 2013 ) . Synaptic release in MB neurons of adult flies was inhibited by expression of a temperature sensitive dominant negative dynamin transgene ( UAS-Shits ) under control of either mb186bGAL4 or mb247GAL4 drivers . Blocking synaptic release in α′ , β′ lobes resulted in a strong flight deficit , whereas silencing of the α , β , and γ lobes did not have a significant effect on flight , for 30 s ( Figure 7A ) . Reduced flight bouts were accompanied by loss of rhythmic spiking in the DLMs of flies with silenced α′ , β′ lobes neurons ( Figure 7B ) . These data support a requirement for post-synaptic dopamine receptors in MB neurons that function for maintenance of acute flight . We tested this requirement further by RNAi-mediated knockdown of the four dopamine receptors—DopECR ( CG18314 ) , Dop1R1 ( CG9652 ) , Dop1R2 ( CG18741 ) , and Dop2R ( CG33517 ) , in either the α′ , β′ neurons ( mb186bGAL4 ) or the α , β , γ neurons ( mb247GAL4; Figure 7—figure supplement 1 ) . A reduction in the length of flight bouts was observed specifically upon knockdown of Dop1R2 in the α′ , β ′ neurons ( Figure 7A ) . The role of α′ β′ lobes , in flight was supported by another GAL4 driver , c305GAL4 , which expresses in the α′ β′ lobes and faintly in the γ lobe ( Krashes et al . , 2007; Pech et al . , 2013 ) . Blocking synaptic activity or knockdown of Dop1R2 using c305aGAL4 resulted in significant flight deficits ( Figure 7—figure supplement 2 ) . 10 . 7554/eLife . 07046 . 028Figure 7 . Mushroom body α’/β’ neurons regulate flight through Dop1R2 . ( A ) Percentage flight times of individual flies of the indicated genotypes . Flight defects were seen by reducing the activity of α’/β’ neurons ( mb186bGAL4 , red bar ) and by knockdown of Dop1R2 in mushroom body α’/β’ neurons ( *p < 0 . 01 , Mann–Whitney U-test ) . ( B ) Electrophysiological responses from the DLMs showed similar responses as observed during flight . ( C ) Flight times during longer flight tests monitored over 15 min are shown . Over-expression of NaChBac rescued flight time partially when compared to knockdown to dFz2 ( *p < 0 . 001 , Mann–Whitney U-test ) . ( D ) Percentage of flies that either do not initiate flight ( 0 s ) or fly for time-periods within the binned intervals ( 20 s each ) is shown for the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02810 . 7554/eLife . 07046 . 029Figure 7—figure supplement 1 . Dopamine receptor knockdown in MB neurons . Percentage flight times of individuals upon knockdown of the indicated Dopamine receptors . The knockdowns were either in α’/β’ neurons ( mb186bGAL4 ) or in α/β and γ neurons ( mb247GAL4 ) of the mushroom body . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 02910 . 7554/eLife . 07046 . 030Figure 7—figure supplement 2 . Synaptic activity in α’/β’ lobes required for flight . ( A ) Percentage flight times of individual flies of the indicated genotypes . Flight defects were seen by reducing the activity ( 30°C ) of α’/β’ neurons using c305aGAL4 ( *p < 0 . 01 , Mann–Whitney U-test ) . ( B ) Percentage flight times of flies upon knockdown of Dopamine receptors . Knockdown of Dop1R2 using c305aGAL4 resulted in significant flight deficits ( *p < 0 . 01 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 03010 . 7554/eLife . 07046 . 031Figure 7—figure supplement 3 . Knockdown of dFz2 does not affect climbing ability of flies . Percentage climbers are shown for knockdown of dFz2 either in multiple dopaminergic neurons ( THGAL4 ) or in the PAM dopaminergic cluster ( R58E02GAL4 ) . Blocking synaptic vesicle recycling by expression of UAS-Shits in PAM neurons ( R58E02GAL4 ) in adults , by keeping the flies at 30°C for 15 min and immediately testing for climbing at 30°C , affected climbing times significantly ( *p < 0 . 001 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 03110 . 7554/eLife . 07046 . 032Figure 7—figure supplement 4 . Maintenance of flight requires dFz2/Ca2+ signaling in dopaminergic neurons . Long flight times monitored for 15 min are shown for the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 03210 . 7554/eLife . 07046 . 033Figure 7—figure supplement 5 . Maintenance of flight requires Fz2/Ca2+ signaling in dopaminergic neurons . Percentage of flies that do not initiate ( 0 s ) or fly for time periods within the binned intervals ( 20 s each ) is shown for the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 07046 . 033 Unlike dopaminergic neurons located in the ventral ganglion , which directly modulate flight motor neuron function as demonstrated recently ( Sadaf et al . , 2015 ) , the PAM–MB circuit described here is not known to project to flight motor neurons in the ventral ganglion ( Riemensperger et al . , 2013 ) . Rather , PAM-MB circuits function to reinforce both aversive and appetitive olfactory responses ( Waddell , 2013 ) . To test possible re-inforcement of flight time by the PAM-MB circuit , we monitored longer flight bouts in several genotypes . Knockdown of dFz2 in PAM neurons ( R58E02GAL4>dFz2-IR ) significantly reduced the duration of flight bouts , monitored up to 15 min , from an average of 13 . 04 ± 0 . 4 min in controls to 0 . 85 ± 0 . 3 min in the knockdowns ( Figure 7C ) . Maintenance of flight bouts was rescued significantly upon increasing neuronal activity by expression of NaChBac ( Figure 7C ) . These data were analyzed further by binning flight bouts in 20-s intervals ( Figure 7D ) . Flight time of dFz2 knockdown flies clustered towards the left among shorter flight bouts , whereas control flies clustered towards the right with longer flight bouts ( Figure 7D ) . Distribution of flight times in NaChBac-rescued flies appeared intermediate . All rescued flies flew for longer than 20 s and a small percentage flew for longer than 10 min . Interestingly , DTH1-rescued flies exhibit shorter flight bouts as compared with NaChBac-rescued organisms ( Figure 7C , D ) . This difference in rescue abilities may in part be due to the previous observation that NaChBac rescue restores TH immunoreactivity to PAM neurons ( Figure 5C ) , whereas TH rescue very likely cannot restore the excitability deficit of PAM neurons . Long flight bouts were also tested in flies with dFz2 knockdown by THGAL4 , followed by rescue with dSTIM+ , itpr+ , and NaChBac ( Figure 7—figure supplement 4 ) . The NaChBac rescue profile was very similar to that observed by NaChBac rescue of PAM-specific dFz2 knockdown , whereas , dSTIM+ and itpr+ rescue profiles resembled the DTH1 rescue in Figure 7C , D ( Figure 7—figure supplement 5 ) . These data support a role for the PAM-MB circuit in maintenance of long flight bouts through dopaminergic synapses on α' β' MB lobes .
Differentiation of neuronal subtypes , after genetic specification , is subject to multiple signals many of which generate and modify electrical activity of the cognate neurons ( Borodinsky et al . , 2014 ) . We demonstrate a requirement for dFz2/Ca2+ signaling for maintaining TH levels in a subset of central brain dopaminergic neurons—the PAM cluster . Our results support transcriptional regulation of TH and very likely the dopamine transporter ( DAT ) by dFz2/Ca2+ signaling in the PAM neurons . A significant compensation of the flight deficit was observed in flies with dFz2 knockdown in PAM neurons upon over-expression of the sodium channel NaChBac , indicating that dFz2/Ca2+ signaling also affects neural activity of PAM neurons . Moreover , flight deficits were observed upon expression of Shits in PAM neurons during the pupal stages , supporting a role for neural activity and synaptic transmission in their development . Increased TH transcripts and TH immunoreactivity after rescue by NaChBac suggests that dFz2/Ca2+ signaling can in part be compensated by raised neural activity , and possibly the two signaling mechanisms function in parallel for maintaining TH transcription in PAM neurons . It is likely that in addition to TH and DAT , dFz2/Ca2+ signals exert their influence on other transcripts in PAM neurons . Transcriptional profiling of the PAM neurons is necessary to address this possibility . Interestingly , despite an increase in the number of PAM neurons during pupal maturation ( Figure 2—figure supplement 3 ) , we do not observe an affect of dFz2/Ca2+ signaling on the number of PAM neurons . Knockdown of dFz2 by DdcGAL4 that marks >60 PAM neurons resulted in flight time of ∼65% ( Figure 1A ) , whereas a stronger flight deficit was observed with NP6510GAL4 which marks just 15 PAM neurons ( Figure 2B ) . Moreover , the flight deficit obtained with R58E02GAL4 which marks ∼100 PAM neurons was similar to the flight deficit obtained with NP6510GAL4 ( Figure 2B ) . Thus , the numbers of PAM neurons that express dFz2-IR do not correlate with the extent of observed flight deficits , suggesting that flight is regulated by a subset of PAM neurons and their projections to the MB . Further , analysis with GAL4 strains that mark PAM neuronal sub-domains would be helpful in identifying such flight specific PAM neurons . Down-regulation of GFP fluorescence in PAMGAL4 strains upon dFz2 knockdown prevented direct analysis of their projections to the MB . Based on a recent study demonstrating similar PAM-MB connections for negative geotaxis ( Riemensperger et al . , 2013 ) , we measured climbing in PAM > dFz2-IR flies . This appeared similar to controls ( Figure 7—figure supplement 3 ) , supporting the idea that connections of PAM neurons to the MB are maintained upon dFz2 knockdown . Thus as compared to climbing , flight appears more sensitive to the observed imbalance of TH . However , as expected inhibition of synaptic release from PAM neurons ( R58E02GAL4>Shits; at 30°C ) affected both flight ( Figure 4D ) and climbing ( Figure 7—figure supplement 3 ) . In vertebrates , βγ subunits of the trimeric Go protein activate phospholipase Cβ , which in turn enhances IP3 formation followed by IP3 receptor-mediated release of calcium from endoplasmic reticular stores ( Rebecchi and Pentyala , 2000 ) . In Xenopus embryos , non-canonical Wnt/Ca2+ signaling , acting through Fz receptors , activates the Nuclear Factor of Activated T cells ( NFAT ) which regulates transcription of genes required for dorsoventral axis formation ( Saneyoshi et al . , 2002 ) . Apart from NFAT , non-canonical Fz2/Ca2+ signaling can also activate calcium calmodulin-dependent protein kinase II ( CamKII ) and protein kinase C ( PKC ) which regulate activity of transcription factors , such as NFκB and CREB ( Sheldahl et al . , 1999; Kuhl et al . , 2000; Slusarski and Pelegri , 2007 ) . Non-canonical dFz2/Ca2+ signaling has been poorly characterized in Drosophila . dFz2 can be cleaved and imported into the nucleus in Drosophila neurons ( Mosca and Schwarz , 2010 ) . However , we do not favor direct transcriptional control of TH by cleaved dFz2 for the following reasons . Rescue of flight in dFz2 knockdown flies can be achieved by AcGo , itpr+ , and dSTIM+ ( Figure 2 ) . These data support a link between dFz2 activation of Go at the membrane followed by intracellular Ca2+ release through the IP3R and dSTIM-mediated calcium entry . This mechanism is broadly similar to what has been observed in vertebrates . Moreover , we did not detect nuclear dFz2 in PAM neurons ( Figure 2—figure supplement 2 ) . Our observation that knockdown of dFz2 reduced not only transcripts from the endogenous TH gene but also affected GFP expression from a DAT promoter transgene ( R58E02GAL4>GFP ) suggests co-ordinated transcriptional regulation of genes for maintenance of dopamine levels by Fz2/Ca2+ signaling in PAM neurons . However , the molecular mechanism by which reduced Fz2/Ca2+ signaling regulates transcription of TH and very likely other genes , in PAM neurons remains to be elucidated . Insect MBs are lobed structures located bilaterally in the protocerebrum of the central nervous system . Neuro-anatomical studies have demonstrated the presence of both efferent neurons arising from MB lobes as well as afferent connections supplying the MB lobes from protocerebral regions ( Ito et al . , 1998 ) including dopaminergic innervations from the PAM neurons ( Mao and Davis , 2009 ) . The Drosophila MB has been studied extensively as a central brain hub for olfactory associative memory and behavior ( Heisenberg , 2003; Fiala , 2007 ) . PAM-MB circuits are required for aversive as well as rewarding reinforcement of olfactory information ( Aso et al . , 2012; Burke et al . , 2012; Liu et al . , 2012 ) . Further analysis of our data revealed that majority ( ∼55% ) of flies with dFz2 knockdown in the PAM neurons fly for 1–20 s as compared to controls that can fly for 700–900 s or more . Rescue of flight deficits in flies with dFz2 knockdown in PAM neurons either by over-expression of a sodium channel ( NaChBac ) or a transgene encoding TH ( UAS-DTH1 ) supports a requirement for both neural activity and dopamine release from the PAM neurons for maintenance of longer flight bouts ( Figure 7 ) . Reduced flight times are very likely due to lack of dopaminergic reinforcement during flight arising from reduced strength of PAM-MB signaling . The role of higher brain centres in Drosophila flight has been investigated primarily in the context of visual cues , and these studies identified the central complex as a key area for visual associative learning ( Ofstad et al . , 2011 ) . The flight circuit identified here appears similar to the one identified recently for the startle induced climbing response which requires PAM dopaminergic inputs to the β′ lobe ( Riemensperger et al . , 2013 ) . Taken together , our findings support an emerging role for the Drosophila MB in coordinated motor behavior , previously considered unlikely ( Wolf et al . , 1998 ) . Dopaminergic inputs from the PAM to the MBs might help integrate olfactory sensory information with motor behavior essential in a natural environment . Further investigations should allow a better understanding of how MB centres for re-inforcement of olfactory memory interact with the flight motor system .
Drosophila was reared on corn flour/agar media supplemented with yeast , grown at 25°C , unless otherwise mentioned in the experimental design . The pan-neuronal GAL4 driver ( ElavC155GAL4 ) , aminergic GAL4 ( DdcGAL4 ) ( Li et al . , 2000 ) and mushroom body drivers c305aGAL4 and mb247GAL4 were obtained from Bloomington Stock Center , Bloomington , IN . mb186bGAL4 was obtained from Anja Beatrice Freidrich ( MPG , Germany ) . The dopaminergic GAL4 ( THGAL4 ) , serotonergic GAL4 ( TRHGAL4 ) , and two other GAL4s , NP6510GAL4 and R58E02GAL4 were generously provided by Serge Birman ( CNRS , ESPCI Paris Tech , France ) ( Riemensperger et al . , 2013 ) . The various dopaminergic subdomain GAL4 drivers used , THC’GAL4 , THC1GAL4 , and THF2GAL4 , were obtained from Mark N Wu ( Johns Hopkins University , Baltimore ) ( Liu et al . , 2012 ) . The peptidergic GAL4 ( P386GAL4 ) was obtained from Paul Taghert ( Washington University , St . Louis ) ( Taghert et al . , 2001 ) . UAS strains of Frizzled-2 RNAi ( 9739R-1 ( II ) , referred to as dFz2-IR in the text and figures ) and itpr RNAi ( 1063-R2 ) were obtained from National Institute of Genetics Fly Stocks Centre , Kyoto , Japan ( NIG ) . The UASRNAi strains for dSTIM ( 47073 ) , dOrai ( 12221 ) , Arrow ( 6707 and 36286 ) , Dishevelled ( 101525 ) , Shaggy ( 101538 ) , DopECR ( 103494 ) , Dop1R1 ( 107058 ) , Dop1R2 ( 105324 ) , Dop2R ( 11470 , 11471 ) , and TH ( 108879 ) were obtained from Vienna Drosophila RNAi center , Vienna , Austria ( VDRC ) . UASRNAi strains for Frizzled-2 ( BL27568 , BL31390 and BL31312 ) were also obtained from Bloomington Stock Center , Bloomington , IN . RNAi strains are referred to as IR indicating the presence of an interference RNA . We obtained UAS-DTH1 from Serge Birman ( CNRS , ESPCI Paris Tech , France ) , UASAcGo ( GoαQ205L ) from Yu Fengwei ( National University of Singapore , Singapore ) , UASAcGi ( GiαQ205L ) from Jurgen Knoblich ( Institute of Molecular Biotechnology , Austria ) , UAS-PTX . 16 from Gregg Roman ( University of Houston , Texas ) and UASFz2 from Stephen Cohen ( Institute of Molecular Cell Biology , Singapore ) . AcGs ( GαsQ215L ) BL6490; Go RNAi , BL34653; UASNaChBac , BL9468; UASDishevelled , BL9453 ( Dsh ) ; BL9522 ( DshG64V ) , UASArmadilloactive , BL4782 ( Armact ) , and UAS-Shits ( BL44222 ) were obtained from the Bloomington Stock Center , Bloomington , IN . UASdOrai+ ( Venkiteswaran and Hasan , 2009 ) , UASdSTIM+ ( Agrawal et al . , 2010 ) , and UASAcGq3 ( Ratnaparkhi et al . , 2002 ) have been published . The GAL80ts strain with two inserts of tubP-GAL80ts on the second chromosome was generated by Albert Chiang , NCBS , Bangalore , India . Progeny were collected upon eclosion and aged for 3–4 days . For flight tests , flies were anaesthetized on ice for 15 min and a thin metal wire was glued between the neck and thorax region with the help of nail polish . To test air-puff-stimulated flight responses , videos were recorded for 30 s after a gentle mouth-blown air puff was delivered to the tethered fly . These videos were analyzed and percentage flight times were calculated . For short flight assays 30 s was taken as 100% flight time . For the long flight assay air-puff-stimulated flight times were monitored for 15 min . For each genotype , a minimum of 30 flies were tethered and tested along with 30 control flies . Flight times of individual flies were noted , and data from a minimum of 30 flies were taken for calculation of the mean and standard error of mean ( SEM ) . Significance testing between the raw data of control and experimental genotypes was performed with the Mann–Whitney U-test using GraphPad Prism 6 ( GraphPad Software Inc , La Jolla , CA , USA ) . Data are represented as bar graphs of the mean percentage flight times . Diamonds inside each bar represent the flight time of individual flies . Electrophysiological recordings were obtained from the indirect dorsal longitudinal flight muscles ( DLMs ) as described previously ( Banerjee et al . , 2004 ) . Briefly , an un-insulated 0 . 127-mm tungsten electrode , sharpened by electrolysis to attain a 0 . 5 μm tip diameter , was inserted in the DLMs ( fiber a ) . A similar electrode was inserted in the abdomen for reference . Air-puff stimulated recordings were obtained for 30 s . All recordings were performed using an ISO-DAM8A amplifier ( World Precision Instruments , Sarasota , FL ) with filter set up of 30 Hz ( low pass ) to 10 kHz ( high pass ) . Gap free mode of pClamp8 ( Molecular Devices , Union City , CA ) was used to digitize the data ( 10 kHz ) on a Pentium 5 computer equipped with Digidata 1322A ( Molecular Devices ) . The duration of rhythmic action potential was analyzed using Clampfit ( Molecular Devices ) and the mean and standard error ( SEM ) were plotted using Origin 8 . 0 software ( MicroCal , Origin Lab , Northampton , MA , USA ) . Spike durations in individual flies have been represented as diamonds within the histograms . Progeny were collected upon eclosion and aged for 3–4 days . To test for climbing , flies in batches of 10 were transferred into cylinder of diameter 2 . 5 cm . Numbers of flies that crossed the 8 cm mark on the cylinder within 12 s , after three gentle taps , were recorded . This procedure was repeated three times with three independent batches of flies . Means and SEM were calculated using the Origin 8 . 0 software ( MicroCal , Origin Lab , Northampton , MA , USA ) . For isolation of RNA , the central nervous system ( CNS ) was dissected from adult flies . For each genotype , three independent sets of RNA were isolated each from eight dissected CNS preparations . Total RNA was isolated using TRIzol Reagent ( Invitrogen Life Technologies , Carlsbad , CA , USA ) according to the manufacturer's specifications . Integrity of RNA was confirmed by visualization on a 1% TAE ( 40 mM Tris pH 8 . 2 , 40 mM acetate , 1 mM EDTA ) agarose gel . Total RNA ( 500 ng ) was treated with DNase in a volume of 45 . 5 μl with 1 μl ( 1U ) DNase I ( Amplification grade , Invitrogen Life Technologies , Carlsbad , CA , USA ) with 1 mM dithiothreitol ( DTT ) ( Invitrogen Life Technologies , Carlsbad , CA , USA ) , 40U of RNase Inhibitor ( Promega , Madison , WI , USA ) in 5X First Strand Buffer ( Invitrogen Life Technologies , Carlsbad , CA , USA ) for 30 min at 37°C and heat inactivated for 10 min at 70°C . The reverse transcription reaction was performed in a final volume of 50 μl by addition of 1 μl ( 200U ) Moloney murine leukemia virus ( M-MLV ) reverse transcriptase ( Invitrogen Life Technologies , Carlsbad , CA , USA ) , 2 . 5 μl ( 500 ng ) random hexaprimers ( MBI Fermentas , Glen Burnie , MD , USA ) and 1 μl of a 25 mM dNTP mix ( GE Healthcare , Buckinghamshire , UK ) . Samples were incubated for 10 min at 25°C , then 60 min at 42°C and heat inactivated for 10 min at 70°C . The polymerase chain reactions ( PCRs ) were performed using 1 μl of cDNA as a template in a 25 μl reaction under appropriate conditions to check the integrity of cDNA prepared . Real time quantitative PCR ( qPCR ) was performed on an ABI 7500 Fast machine ( Applied Biosystems , Foster City , California , USA ) operated with ABI 7500 software version 2 ( Applied Biosystems , Foster City , California , USA ) using MESA GREEN qPCR MasterMIx Plus for SYBR Assay I dTTp ( Eurogentec , Belgium ) . Each qPCR experiment was repeated three times with independently isolated RNA samples . qPCRs were performed with rp49 primers as internal controls and primers specific to gene of interest using dilutions of 1:10 . Sequences of the primers used in the 5′ to 3′ directions are given below . The sequence of the forward primer is given first in each case: dfz2 GGTTACGGAGTGCCAGTCAT; CACAGGAAGAACTTGAGGTCC , rp49 CGGATCGATATGCTAAGCTGT; GCGCTTGTTCGATCCGTA , dsh CCAAATCCCAAGGGCTACTTC; ATAATACTGTCGTGCGATGTGAG sgg GCTGCTCGAGTATACGCCC; CACTAGGCTGGGCTGTATTGA th GTTGCAGCAGCCCAAAAGAAC; GAGACCGTAATCATTTGCCTTGC . The cycling parameters were 95°C for 5 min , 40 cycles of 95°C for 15 s , and 60°C for 1 min followed by 1 cycle of 72°C for 5 min . The fluorescent signal produced from the amplicon was acquired at the end of each polymerization step at 60°C . A melt curve was performed after the assay to check for specificity of the reaction . The fold change of gene expression in the genotype relative to wild-type was determined by the comparative ΔΔCt method ( Lorentzos et al . , 2003 ) . In this method , the fold change = 2−ΔΔCt where ΔΔCt = ( Ct ( target gene ) − Ct ( rp49 ) ) mutant2 − ( Ct ( target gene ) − Ct ( rp49 ) ) Wild type . Immunohistochemistry was performed on Drosophila adult brains expressing cytosolic GFP ( UASGFP ) with the specified GAL4 strains , after fixing the dissected tissue in 4% paraformaldehyde . The following primary antibodies were used: mouse monoclonal anti-TH antibody ( 1:50 , #22941 , ImmunoStar , Hudson , WI , USA ) , rabbit anti-GFP antibody ( 1:10 , 000; #A6455 , Molecular Probes , Eugene , OR , USA ) , mouse anti-Fz2 ( 1:20; #12A7 , DSHB , University of Iowa ) . 12A7 was deposited to the DSHB by Nusse , R ( DSHB Hybridoma Product 12A7 ) . Fluorescent secondary antibodies were used at a dilution of 1:400 as follows: anti-rabbit Alexa Fluor 488 ( #A1108 ) and anti-mouse Alexa Fluor 568 ( #A1104 , Molecular Probes , Eugene , OR , USA ) . After antibody staining , confocal analysis was performed on an Olympus Confocal FV1000 microscope and visualized using the FV10-ASW 1 . 3 viewer ( Olympus Corporation , Tokyo , Japan ) . Mean intensity of TH or GFP immunostaining was calculated using ImageJ Version 10 . 2 ( U . S . National Institutes of Health , Bethesda , Maryland , USA , http://imagej . nih . gov/ij/ , 1997–2014 ) . Region of interest was drawn around each neuron and mean intensities were obtained for TH and GFP for all the neurons . Median shown by the horizontal line and spread of 25–75% of cell intensities represented as big square was calculated and plotted using Origin 8 . 0 software ( MicroCal , Origin Lab , Northampton , MA , USA ) with the data from all the neurons . Significant difference between the different groups of cell intensities was calculated using One-way analysis of variance ( ANOVA ) for p < 0 . 05 . | The fruit fly Drosophila melanogaster is an aerial acrobat . These insects can suddenly change direction in less than one hundredth of a second , explaining why a moving fly can be so difficult to swat . To perform their aerial manoeuvres , the flies continually combine information from multiple senses , including vision , hearing and smell , and use these data to control the activity of the neural circuits that support flight . These flight circuits are established during the pupal stage of fly development , during which the fly transforms from a larva into its adult form . In 2013 , researchers showed that a protein called dFrizzled2 must be present in pupae for flight circuits to mature correctly . This protein forms part of a pathway that ultimately controls which specific chemicals—called neurotransmitters—are released by neurons to communicate with other cells . Agrawal and Hasan—who worked on the 2013 study—now extend their findings to investigate the role of dFrizzled2 in more detail . The new experiments show that for the flight circuits to mature , dFrizzled2 must be active in a cluster of neurons known collectively as PAM . Specifically , dFrizzled2 is needed to make an enzyme that helps to produce a neurotransmitter called dopamine . This in turn enables the PAM neurons to communicate with a region of the fruit fly brain called the mushroom body , which it thought to play an important role in complex behaviors such as reward-based learning . The absence of dFrizzled2 results in adult flies that rarely remain airborne for more than 20 s at a time , whereas normal flies can typically fly for over 700 s . Given that dopamine is known to signal reward , one possibility is that the dopamine signals from the PAM neurons to the mushroom body serve as a reward to encourage continuous flight . Mutant flies that lack dFrizzled2—and thus these dopamine signals—lose their motivation to fly after only a few seconds . Overall , Agrawal and Hasan's findings suggest that the mushroom body has an important role in coordinating a fly's movements with information from it senses . Future research will be needed to determine exactly how the mushroom body performs this role . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2015 | Maturation of a central brain flight circuit in Drosophila requires Fz2/Ca2+ signaling |
Understanding T cell function in vivo is of key importance for basic and translational immunology alike . To study T cells in vivo , we developed a new knock-in mouse line , which expresses a fusion protein of granzyme B , a key component of cytotoxic granules involved in T cell-mediated target cell-killing , and monomeric teal fluorescent protein from the endogenous Gzmb locus . Homozygous knock-ins , which are viable and fertile , have cytotoxic T lymphocytes with endogeneously fluorescent cytotoxic granules but wild-type-like killing capacity . Expression of the fluorescent fusion protein allows quantitative analyses of cytotoxic granule maturation , transport and fusion in vitro with super-resolution imaging techniques , and two-photon microscopy in living knock-ins enables the visualization of tissue rejection through individual target cell-killing events in vivo . Thus , the new mouse line is an ideal tool to study cytotoxic T lymphocyte biology and to optimize personalized immunotherapy in cancer treatment .
Cytotoxic T lymphocytes ( CTLs ) are not only essential for the removal of foreign agents such as viruses or bacteria , but also play a key role in modern personalized cancer immunotherapy ( Porter et al . , 2011; Sharma and Allison , 2015; Watanabe et al . , 2018; Minutolo et al . , 2019 ) . Accordingly , a detailed mechanistic understanding of the major CTL function , that is the killing of cells through release of cytotoxic substances from cytotoxic granules ( CGs ) at the immune synapse ( IS ) , is of utmost interest for basic and clinical science alike ( Dustin and Long , 2010; Griffiths et al . , 2010; Mukherjee et al . , 2017; Xiong et al . , 2018 ) . The release of perforin and granzymes from CGs has been studied in great detail using cultured primary CTLs ( Friedl et al . , 2005; Kupfer , 2006; Orange , 2008; Dustin and Long , 2010; Griffiths et al . , 2010; Voskoboinik et al . , 2015 ) , but corresponding approaches have neglected the fact that CTLs act in concert with other immune cells to perform their function in vivo . In essence , CTL function in isolation differs considerably from CTL function in vivo . However , in vivo studies on CTLs have been scarce due to their technical difficulties and the lack of suitable markers ( Mempel et al . , 2006; Breart et al . , 2008; Germain et al . , 2012; Halle et al . , 2017; Lodygin and Flügel , 2017; Torcellan et al . , 2017; Cazaux et al . , 2019; Malo and Hickman , 2019 ) . We report here the generation of a GzmB-mTFP knock-in ( KI ) mouse that expresses a fluorescent fusion protein consisting of granzyme B ( GzmB ) , a CG-resident serine protease , and monomeric teal fluorescent protein ( mTFP ) from the endogenous Gzmb locus . The new GzmB-mTFP-KI allows the observation of individual CTLs and even CGs in living mice at any time point of interest . We show that GzmB-mTFP-KIs are viable , fertile and free of any obvious defects , that their T cell-specific functions are wild-type-identical , and that their CTLs can be imaged with all major super-resolution techniques in vitro and in vivo . We expect that the GzmB-mTFP-KI will be a highly valuable tool to investigate CTL function in vitro and in vivo - in the context of both , basic CTL biology and clinical aspects of CTL function , such as CTL-based personalized cancer immunotherapy .
To create a specific , endogenous fluorescent label for cytotoxic granules ( CG ) we chose GzmB ( Young et al . , 1986; Masson and Tschopp , 1987; Krahenbuhl et al . , 1988 ) , which belongs to a family of serine proteases that induce apoptosis of target cells and which is present in CGs of natural killer cells and CD4+ and CD8+ T lymphocytes ( Peters et al . , 1991 ) . In contrast to perforin , a CG-specific pore-forming protein , GzmB deletion does not lead to a killing defect in CTLs ( Simon et al . , 1997 ) . Using CRISPR-Cas9 technology and a corresponding HDR fragment , we replaced the Stop codon in exon 5 of the mouse Gzmb gene with a sequence encoding a flexible GGSGGSGGS linker , which has a high probability to be cleaved in the acidic environment of the lysosome ( Huang et al . , 2014 ) , the coding sequence of monomeric teal fluorescent protein ( mTFP ) , and a Stop codon ( Figure 1A and Figure 1—figure supplement 1 ) . We generated homozygous GzmB-mTFP-KIs , which were viable and fertile and showed no obvious phenotypic changes . PCR analyses of CTL lysates derived from wild-type , heterozygous and homozygous GzmB-mTFP-KI mice verified the expected genotypes ( Figure 1B ) . As envisioned by our design , Western blot analyses of lysates of CTLs four and five days after activation showed that the fusion protein is efficiently cleaved into GzmB and mTFP ( Figure 1C ) , ensuring a correct function of GzmB in the killing process . As expected , Western blot ( days 0–5; Figure 1D ) and FACS analyses ( days 0–10; Figure 1E ) demonstrated a continuous up-regulation of GzmB expression following CTL activation . The expression levels of the fusion protein varied between different preparations ( 59 . 1% ( day 4 , Figure 1C ) , 53 . 6% ( day 5 , Figure 1C ) and 183 . 9% ( day 5 , Figure 1D ) of wt level for GzmB ) as expected , but we always observed a robust fluorescence without the requirement to change the intensity of the excitation lasers for the experiments shown in the following figures . To verify the usefulness of the GzmB-mTFP-KI in vitro and in vivo imaging , we next assessed the correct localization of GzmB and mTFP to CGs . We isolated CTLs from the KI mice and fixed them five days after activation . We then performed immunofuorescence labeling with anti-GzmB and anti-perforin antibodies and detected by structured illumination microscopy ( SIM ) an excellent co-localization of mTFP , GzmB , and perforin signals ( Pearson's coefficient of correlation 0 . 84 ± 0 . 02 for GzmB-mTFP vs . GzmB , 0 . 69 ± 0 . 03 for GzmB-mTFP vs . perforin , n = 10; Manders' coefficient of correlation 0 . 79 ± 0 . 02 for GzmB-mTFP vs . GzmB , 0 . 80 ± 0 . 01 for GzmB vs . GzmB-mTFP , 0 . 70 ± 0 . 02 for GzmB-mTFP vs . perforin , 0 . 71 ± 0 . 03 for perforin vs . GzmB-mTFP , n = 10; Figure 2A–C ) . Further , we cross-bred the GzmB-mTFP-KI with a Synaptobrevin-2 ( Syb2 ) -mRFP-KI . Using the Syb2-mRFP-KI , we had shown previously that Syb2 is almost exclusively localized to CGs in CTLs and acts as the vSNARE in CG fusion with the plasma membrane ( Matti et al . , 2013 ) . Comparison of endogenous GzmB-mTFP and Syb2-mRFP fluorescence in CTLs on poly-ornithine or anti-CD3 coated coverslips again revealed excellent co-localization ( Pearson's coefficient of correlation 0 . 88 ± 0 . 10 for poly-ornithine , and 0 . 86 ± 0 . 08 for anti-CD3 , n = 25; Manders' coefficient of correlation 0 . 86 ± 0 . 08 for GzmB-mTFP vs . Syb2-mRFP and 0 . 79 ± 0 . 17 for Syb2-mRFP vs . GzmB-mTFP on poly-ornithine , 0 . 79 ± 0 . 13 for GzmB-mTFP vs . Syb2-mRFP and 0 . 81 ± 0 . 13 for Syb2-mRFP vs . GzmB-mTFP on anti-CD3 , n = 25; Figure 2D–F ) . These data show that the cleavage of GzmB-mTFP we observed ( Figure 1C–D ) occurs inside CGs , so that the mTFP in the GzmB-mTFP-KI represents an excellent endogenous marker for studies of CG maturation , transport and fusion , without affecting GzmB localization and function . To visualize CG transport along the microtubular network , we incubated primary CTLs from GzmB-mTFP-KIs with silicon rhodamine ( SiR ) -tubulin , seeded them onto anti-CD3 coated coverslips to induce immune synapse ( IS ) formation and followed transport of CGs along microtubuli by live stimulated emission depletion ( STED ) microscopy ( Figure 2G ) . Within five minutes after seeding , microtubules with the characteristic microtubule organizing center ( MTOC ) had formed and all CGs were being transported toward the IS at the bottom of the cells . During transport , the average distance between CGs and microtubuli was 137 . 5 ± 19 . 6 nm ( 19 cells , 80 granules; Figure 2G–I ) , comparable to that of wild-type control cells . Thus , CTLs of the GzmB-mTFP-KI mouse are ideally suited to follow maturation and trafficking of endogenously labeled CGs in vitro with super-resolution microscopy . We next tested whether CG fusion at the IS can be observed and quantified in real time using the GzmB-mTFP-KI . For this purpose , we seeded primary GzmB-mTFP-KI CTLs on anti-CD3 coated coverslips and performed total internal reflection microscopy ( TIRFM ) . The presence of essential components of pSMAC and cSMAC allows the CTLs to form an IS on the area facing the coverslip . Within two minutes CGs arrived at the IS and frequently fused , as indicated by a sudden drop in fluorescence due to the diffusion of the fluorophore ( Figure 3A ) . Quantification showed that the average number of CTLs showing CG fusion ( 59 . 0 ± 2 . 9% for wild-type control and 57 . 7 ± 4 . 6% for KI , N = 3 , n = 70; Figure 3B ) and the number of CG fusion events per cell ( 7 . 09 ± 0 . 48 for wild-type control and 7 . 23 ± 0 . 64 for KI , N = 3 , n = 70; Figure 3C ) were indistinguishable between wild-type and GzmB-mTFP-KI CTLs , in accordance with our recent findings on fusion kinetics and TCR stimulation ( Estl et al . , 2020 ) . Furthermore , there was no difference in the kinetics of CG movement into the TIRF field between wild-type and GzmB-mTFP-KI ( Figure 3D ) . We also used CTLs isolated from GzmB-mTFP/Syb2-mRFP double KI mice to assess simultaneous fluorescence signals in mTFP and mRFP channels . Again , we found a perfect co-localization of both fluorophores , with fusion events detected in both channels . Following fusion , the mRFP fluorescence remained visible at the plasma membrane , as expected for a membrane protein like synaptobrevin2 ( Figure 3E ) . Our TIRFM data show that CG fusion events at the IS can be reliably observed with CTLs derived from the GzmB-mTFP-KI mouse , making the new KI an ideal tool to study molecular mechanisms of CG fusion in primary CTLs without the need for transfection of CG markers . To confirm that primary GzmB-mTFP-KI CTLs maintain their ability to kill target cells , we performed two independent killing assays and compared the killing efficiency of CTLs from GzmB-mTFP-KIs to that of wild-type CTLs . First , we stably transfected P815 target cells with Casper3-GR , a FRET based sensor consisting of green and red fluorescent proteins TagGFP and TagRFP connected by a linker containing the caspase-3 cleavage sequence DEVD ( Shcherbo et al . , 2009 ) . The activation of caspase-3 by GzmB during apoptosis leads to cleavage of the DEVD sequence and elimination of FRET , which can be detected as a decrease in the red emission of TagRFP and a simultaneous increase in green emission of TagGFP . We found that wild-type and GzmB-mTFP-KI CTLs killed with a similar frequency ( 59 . 0 ± 2 . 9% for wild-type and 57 . 7 ± 4 . 6% for KI , N = 3 , n = 70 ) and efficiency , as indicated by the same fold ratio change between the green and the red channel ( 2 . 40 ± 0 . 33 for wild-type , n = 8; 2 . 34 ± 0 . 22 for KI , n = 7; Figure 4A ) . Second , we performed a flow cytometry-based degranulation assay ( Betts et al . , 2003 ) . At all three days after activation tested ( days 3 , 5 and 7 after activation ) , the degranulation measured by LAMP1 ( CD107a ) expression on the surface was comparable between wild-type and GzmB-mTFP-KI CTLs [day 3: 84 . 0 ± 2 . 5% ( wild-type ) vs . 82 . 7 ± 0 . 3% ( KI ) ; day 5: 84 . 2 ± 2 . 5% ( wild-type ) vs . 82 . 6 ± 0 . 3% ( KI ) ; day 7: 88 . 3 ± 0 . 6% ( wild-type ) vs . 85 . 4 ± 0 . 2% ( KI ) , N = 3; Figure 4B ) . Thus , the killing capacity of GzmB-mTFP-KI CTLs is indistinguishable from that of wild-type CTLs in vitro . CTLs fight infections and cancer in vivo in concert with other immune cells , like CD4+ T cells and macrophages . Therefore , we tested the ability of GzmB-mTFP-KI CTLs to reject foreign tissue transplanted into the anterior chamber of the eye ( ACE ) ( Speier et al . , 2008a; Speier et al . , 2008b; Abdulreda et al . , 2013 ) . Originally developed to follow insulin secretion from Langerhans islets in Diabetes treatment ( Speier et al . , 2008a; Rodriguez-Diaz et al . , 2012 ) , the ACE model allows longitudinal , non-invasive in vivo imaging of anti-islet immune responses with single-cell resolution in real time ( Abdulreda et al . , 2013; Abdulreda et al . , 2019 ) . We transplanted Langerhans islets from allogenic DBA/2 donor mice into the anterior chamber of the eye of GzmB-mTFP-KIs recipients , which had been bred in the C57BL/6 background ( Figure 5A ) . At post operational day 4 ( POD4 ) following transplantation , we started weekly in vivo observations with a confocal laser scanning or a two-photon microscope . Between POD4 and POD7 , transplanted islets became capillarized and blood-perfused , indicating successful engraftment ( Figure 5—figure supplement 1A , Video 1 ) . Starting between POD7 and POD14 , GzmB-mTFP positive cells invaded the islets , reached a maximum in number within two weeks ( 26 ± 8 . 3 cells/100 µm³ , N = 4 animals , n = 11 islets ) and eventually declined with the disappearance of the island allografts ( Figure 5—figure supplement 1B–D ) . To exclude that the islet shrinking resulted from islet necrosis and/or unspecific rejection , we performed control experiments with transplanted syngenic islets from C57BL/6 donors . As expected , we observed engraftment of the syngenic islets after 4–7 days , but we did not detect any GzmB-mTFP positive cells nor any islet rejection during three months of observation post-transplantation , demonstrating the high specificity of the ACE model . To verify the identity of GzmB-mTFP positive cells inside the anterior chamber , we fixed transplanted eyes at POD14-18 , produced histochemical samples and counterstained the tissue with specific antibodies against CD4+ T helper cells and CD8+ CTLs ( Figure 5B ) . We found almost no overlap between GzmB-mTFP positive cells and CD4+ cells , while the majority of CD8+ cells also were GzmB-mTFP positive . These data again demonstrate the high specificity of the GzmB-mTFP-KI as a marker for CTLs . The small number of GzmB-mTFP positive , but CD8 negative cells might represent natural killer ( NK ) cells . Next , we used the ACE model for long-term in vivo observations of CTL migration . We imaged islets in 40 µm thick layers ( 11 planes ) and analyzed the migration of invading CTLs in 4D ( volume and time ) as previously described ( Abdulreda et al . , 2011 ) . Figure 5C shows a representative , confocal overview of an islet with green spots indicating the positions of single CTLs . Figure 5—video 1 shows the respective long-term observation over 32 min . Analysis of CTL velocity from these data yielded an average migration velocity of 5 . 52 ± 1 . 5 µm/min ( mean ±s . d . ; median 5 . 34 ) at POD14 ( Figure 5D ) . At later PODs , the velocity increased due to the advanced tissue clearance ( data not shown ) . Ultimately , the specific , endogenous labeling of individual granules should enable us to visualize single fusion events of CG in vivo . A prerequisite for this goal is to achieve sufficient resolution and acquisition frequencies to discriminate single vesicles and track them . As shown in Figure 5—videos 1 and 2 and in Figure 5E , we clearly identified individual granules in migrating CTLs . The nominal lateral resolution in the raw data of the videos and the image in Figure 5E was ~0 . 7 µm ( pixel size 340 × 340 nm ) . The apparent resolution as determined by 'full width at half maximum ( FWHM ) ' analysis was dependent on the recording conditions and the position of the cell within an islet . Typically within the first 40 µm from the islet surface , we reached a resolution of ~1 µm . While using two-photon instead of confocal microscopy ( Figure 5E ) did not increase the maximal resolution , the signal to noise ratio and the light-penetration depth were substantially increased . New technological developments , like a combination of two-photon microscopy with STED microscopy ( Takasaki and Sabatini , 2014 ) will further improve 3D resolution in in vivo studies and will , in combination with the GzmB-mTFP-KI , allow the observation of individual CG fusion events in vivo .
The elimination of infected or malignant cells through CTL-mediated target cell killing is essential to maintain health . While in vitro studies indicated that CTL-mediated target cell killing is highly efficient , two-photon imaging of lymph nodes in living mice demonstrated that CTLs kill on average only 4 . 5 target cells per day and require the cooperativity of several CTLs acting in concert ( Halle et al . , 2016; Nolz and Hill , 2016; Halle et al . , 2017 ) . This surprising inefficiency in vivo is most likely caused by rate-limiting steps like CTL priming , CTL trafficking , CTL properties and target cell properties ( Halle et al . , 2017 ) . For the optimization of vaccines and CTL-based immunotherapies it is therefore of utmost importance to study CTL behavior in the physiological setting of a living organism . We report the generation of a GzmB-mTFP-KI , in which all CGs of T cells are endogenously fluorescent due to an mTFP-tag at the CG-resident enzyme GzmB . We chose the mTFP tag because of its ideal suitability for two-photon imaging in vivo and for Förster resonance energy transfer ( FRET ) ( Day et al . , 2008; Drobizhev et al . , 2011; Gossa et al . , 2015 ) . mTFP is a derivative of cyan fluorescent protein ( CFP ) from Clavularia coral ( Ai et al . , 2006 ) , is monomeric ( 26 . 9 kDa ) unlike the tetrameric CFP , has a high brightness , photostability and quantum yield , is only weakly acid sensitive and has a narrow emission spectrum ( Ai et al . , 2006; Day et al . , 2008 ) . The brightness , photostability and high quantum yield of mTFP are ideally suited for longitudinal , non-invasive imaging over extended time periods . In addition , mTFP is tolerated by the immune system upon adoptive transfer , which favors its use for imaging immune cells in vivo ( Gossa et al . , 2015 ) . The biggest concern when using fusion protein tags is that the attached tag might interfere with the expression , localization or function of the tagged protein , in our case GzmB . GzmB is synthesized as a zymogen and cleaved by cathepsin C in lysosomes ( Jenne and Tschopp , 1988; Krahenbuhl et al . , 1988; Caputo et al . , 1993; Perišić Nanut et al . , 2014; Voskoboinik et al . , 2015 ) , which removes an N-terminal 18-residue signal peptide and the downstream activation dipeptide Gly-Glu ( Caputo et al . , 1993 ) . To avoid interference with proper GzmB localization we therefore chose the C-terminus for fusion of mTFP . Further , we added a flexible GGSGGSGGS linker between GzmB and mTFP , which is cleaved in CGs ( Figure 1C; Huang et al . , 2014 ) . Our co-localization analyses with perforin and the CG v-SNARE synaptobrevin2 ( Figure 2 ) and TIRF imaging of individual CG fusion events ( Figure 3 ) demonstrate that the mTFP and GzmB in the GzmB-mTFP-KI are correctly and specifically localized to CG , so that GzmB is not functionally compromised ( see below ) and mTFP can be used as a very faithful CG marker . Beyond subcellular protein trafficking and localization , tags can affect protein function . Indeed , a recently published GzmB-tdTomato-KI exhibits reductions of 75% in GzmB expression , of 30% in CTL degranulation and of 25% in GzmB protease activity ( Mouchacca et al . , 2013; Mouchacca et al . , 2015 ) . In striking contrast to this , our analyses of GzmB-mTFP-KI CTLs did not reveal any changes in GzmB expression ( Figure 1C–D ) , CG transport ( Figure 2G–H ) , CG fusion efficiency ( Figure 3C–D ) , degranulation capacity ( Figure 4B ) or killing efficiency ( Figure 4A ) . This may be partly due to the smaller size of mTFP as compared to tdTomato ( 26 . 9 vs . 54 . 2 kDa ) . More importantly , though , the fact that GzmB-mTFP appears to be efficiently cleaved into GzmB and mTFP in the acidic environment of the CG lumen ( Figure 1C–D ) contributes substantially to the almost perfect functionality of GzmB in the GzmB-mTFP-KI , so that GzmB can operate in an unperturbed manner while mTFP can still be used to follow CG localization , movement and fusion . The advantages and potential of the novel GzmB-mTFP-KI are manifold . While many current in vivo studies require the removal of CTLs from lymphatic tissue and their exogenous labeling and re-injection via the tail vein , no manipulations are necessary for confocal or two-photon imaging in the GzmB-mTFP-KI ( Figure 5 ) . Further , the subcellular targeting of GzmB-mTFP to CGs allows the investigation of individual CGs without compromising analyses of single CTLs in tissue . In combination with FRET-based apoptosis reporters ( Breart et al . , 2008; Shcherbo et al . , 2009; Garrod et al . , 2012; Cazaux et al . , 2019 ) expressed in the islets the GzmB-mTFP-KI might shed new light on the degranulation process in vivo . In conclusion , the new GzmB-mTFP-KI mouse now allows the investigation of rate-limiting processes in CTL function like priming , differentiation , migration and killing in order to optimize vaccines and immunotherapies for virus infections and cancer .
GzmB-mTFP-KI mice were generated by CRISPR-Cas9 technology with an HDR fragment designed to replace the endogenous Stop codon of the Gzmb gene by sequences encoding a GGSGGSGGS-linker , mTFP and a Stop codon ( Figure 1A , Figure 1—figure supplement 1 ) . Mouse Gzmb gene sequences required for the HDR fragment were PCR amplified from genomic C57BL/6J ( Janvier ) DNA , subcloned into pMiniT ( NEB ) and sequence verified , and the sequence encoding GGSGGSGGS-mTFP-Stop was obtained from a previously generated plasmid ( pMAX-Synaptobrevin2-mTFP ) . Using these sequences , the HDR fragment was generated by Gibson assembly ( Gibson , 2009 ) . For zygote injections , we used a single sgRNA ( 5’-GTC CAG GAT TGC TCT AGG AC-3’; PAM , AGG ) , in vitro transcribed capped Cas9-D10A mRNA ( 'nickase' version of Cas9 [Gasiunas et al . , 2012; Jinek et al . , 2012; Cong et al . , 2013] ) , and the purified HDR fragment . The strategy to use a single sgRNA in combination with Cas9-D10A mRNA was chosen to minimize the likelihood of off-target cleavage events . The CRISPR-Cas9 reagents ( 10 ng/µl sgRNA , 10 ng/µl Cas9-D10A mRNA , 20 ng/µl HDR fragment ) were injected into the pronucleus and cytosol of zygotes ( C57BL/6J , Janvier ) , which were then re-implanted into seven recipient mice . We obtained 32 offspring , which were genotyped for the proper insertion of the HDR fragment by PCR , using primer 36016 ( ATCAAAGAACAGGAGAAGACCCAG , Exon 3 ) in combination with primer 33615 ( GGTGTTGGTGCCGTCGTAGGG , mTFP ) ( 1433 bp fragment ) and primer 19524 ( ACCGCATCGAGATCCTGAACC , mTFP ) in combination with primer 36017 ( AATGGCTAAGCAATCCCATCAGG , downstream of HDR1 ) ( 1565 bp fragment ) . Of the 32 offspring , one carried the desired mutation . This mouse was subsequently bred to C57BL/6J mice ( Janvier ) to obtain germline transmission of the GzmB-mTFP-KI mutation . Genotyping of later generations was done by short-range PCR ( Figure 1A , Figure 1—figure supplement 1 ) . As soon as the paper is published , the GzmB-mTFP-KI mouse line will be deposited at the European Mouse Mutant Archive ( EMMA ) . Until the line is established at EMMA , mice will be provided by the authors upon request . C57BL/6 mice for other experiments were purchased from Charles River Laboratories ( Sulzfeld , Germany ) . Donor mice for islet of Langerhans isolation were DBA/2 mice ( 12–30 weeks old ) . All animal experiments were performed according to German law and European directives , and with permission of the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit ( LAVES animal license number 33 . 9-42502-04-13/1359 ) and the state of Saarland ( Landesamt für Gesundheit und Verbraucherschutz; animal license number 41–2016 ) . Casper3-GR ( FP971; Evrogen ) vector was digested with BamH1 and Not1 to remove TagRFP-linker-TagGFP . The same digestion was performed for pMAX vector and both vector and insert were ligated . The following primary antibodies were used: anti-GAPDH ( RRID:AB_561053 ) , anti-tRFP ( RRID:AB_2571743 ) and anti-GzmB ( RRID:AB_2114432 ) . Secondary antibodies were HRP-conjugated donkey anti-rabbit and Alexa Fluor 647 goat anti-mouse IgG ( H+L ) ( Thermo Fisher Scientific ) . For structured illumination microscopy ( SIM ) , Alexa647-coupled anti-granzymeB ( RRID:AB_2294995 ) and Alexa647-coupled anti-perforin ( RRID:AB_493255 ) antibodies were used . For total internal reflection fluorescence microscopy ( TIRFM ) a hamster anti-mouse CD3ε ( RRID:AB_394591 ) antibody was used for coating coverslips and stimulating cells . For FACS , a rat anti-mouse CD107a-PE ( LAMP-1-PE ) ( RRID:AB_1732051 ) antibody was used . Splenocytes were isolated from 8 to 18 week-old C57BL6/N , Syb2-mRFP/GzmB-mTFP double knock-in mice or Granzyme B-mTFP-KI mice , as described before ( Chang et al . , 2016 ) . Briefly , naive CD8+ T cells were isolated from splenocytes using Dynabeads FlowComp Mouse CD8+ kit ( Thermo Fisher Scientific ) . The isolated naive CD8+ T cells were cultured for up to 12 to 14 d ( 37°C , 5% CO2 ) with anti-CD3/anti-CD28 activator beads with 100 U/mL IL2 ( 1:0 . 8 ratio ) at a density of 1 × 106 cells/mL in AIMV ( Thermo Fisher Scientific ) containing 10% FCS and 50 µM 2-ME to generate effector CTLs . P815 target cells were cultured in RPMI 1640 medium ( Thermo Fisher Scientific ) containing 10% FCS , 1% pen/strep , and 10 mM HEPES . For functional effector:target cell conjugations P815 target cells were incubated with 10 µg/ml anti-CD3ε antibody for 15–30 min at 37°C and washed with 1xPBS . 0 . 2 × 106 P815 cells were plated on 12 mm glass coverslips . Carefully , 0 . 2 × 106 CTLs were added on top of settled P815 target cells and imaged for 45 min to 1 hr . Cells were fixed in ice-cold 4% PFA in Dulbecco’s 1xPBS ( Thermo Fisher Scientific ) for 20 min . For staining , cells were permeabilized with 0 . 1% TritonX-100 in Dulbecco’s 1xPBS ( permeabilizing solution ) for another 20 min followed by 30 min blocking in solution containing 2% BSA prepared in permeabilizing solution . Cells were stained with either Alexa 647 conjugated anti-GzmB or Alexa 647 conjugated anti-perforin antibodies ( BD Biosciences ) and observed at an Elyra PS . 1 microscope ( Zeiss , Jena , Germany ) . Images were acquired with a 63x Plan-Apochromat ( NA 1 . 4 ) with laser excitation at 488 , 561 , and 635 nm and then processed to obtain higher resolutions ( RRID:SCR_013672 ) . For analysis of co-localization , Pearson’s and Manders’ coefficients of correlation ( Pearson , 1909; Manders et al . , 1993 ) were determined using the JACoP plugin of ImageJ v1 . 46 . The TIRFM setup ( Visitron Systems GmbH , Puchheim , Germany ) was based on an IX83 ( Olympus ) equipped with the Olympus autofocus module , a UAPON100XOTIRF NA 1 . 49 objective ( Olympus ) , a 488 nm 100 mW laser and a solid-state laser 100 mW emitting at 561 nm , the iLAS2 illumination control system ( Roper Scientific SAS , France ) , the evolve-EM 515 camera ( Photometrics ) and a filter cube containing Semrock ( Rochester , NY , USA ) FF444/520/590/Di01 dichroic and FF01-465/537/623 emission filter . The setup was controlled by Visiview software ( version 4 . 0 . 0 . 11 , Visitron GmbH ) . For TIRFM , day seven bead activated CTLs isolated from GzmB-mTFP-KI or GzmB-mTFP/Syb2-mRFP double knock-in mice were used . 2 hr before experiment , beads were removed from CTLs and roughly ( 0 . 2–0 . 3 × 106 cells ) were resuspended in 30 μL of extracellular buffer ( 2 mM HEPES , 140 mM NaCl , 4 . 5 mM KCl , and 2 mM MgCl2 ) containing no Ca2+ and allowed to settle for 1–2 min on anti-CD3ε antibody ( 30 μg/mL ) coated coverslips . Cells were then perfused with extracellular buffer containing 10 mM Ca2+ for visualizing CG fusion . Cells were imaged for 7 min at room temperature at 488 nm or alternating between 488 nm and 561 nm with acquisition frequency of 10 Hz and exposure time of 100 ms . Images and time-lapse series were analyzed using ImageJ ( RRID:SCR_003070 ) or the FIJI package of ImageJ ( RRID:SCR_002285 ) . CG fusion analysis was performed using ImageJ with the plugin Time Series Analyzer ( RRID:SCR_014269 ) . A sudden drop in GzmB-mTFP or GzmB-mCherry fluorescence occurring within 300 ms ( three acquisition frames ) was defined as fusion ( Ming et al . , 2015 ) . The number of vesicles was shown according to the corresponding fluorescence intensity of the vesicles . Mouse CTLs were homogenized with a syringe in lysis buffer ( 50 mM Tris ( pH 7 . 4 ) , 1 mM EDTA , 1% Triton X-100 , 150 mM NaCl , 1 mM DTT , 1 mM deoxycholate , protease inhibitors , and PhosSTOP; Roche ) on ice and insoluble material was removed by centrifugation at 11 , 300 g for 10 min . The protein concentration was determined using Quick Start Bradford 1x Dye Reagent ( 5000205; Bio-Rad ) . Proteins were separated by SDS-PAGE ( NuPAGE; Thermo Fisher Scientific ) , transferred to nitrocellulose membranes ( Amersham ) , and blocked by incubation with 5% skim milk powder in 20 mM Tris , 0 . 15 M NaCl , pH 7 . 4 , and TBS for 1–2 hr and blotted with specific antibodies . Blots were developed using enhanced chemiluminescence reagents ( SuperSignal West Dura Chemoluminescent Substrate; Thermo Fisher Scientific ) and scanned . CTLs isolated from wt and GzmB-mTFP-KI mouse were stimulated with anti-CD3/CD28 beads . Day 3 , 5 , 7 and 10 activated CTLs from both wt and GzmB-mTFP knock-in mouse were used for degranulation assay . 2 hr before degranulation , anti-CD3/CD28 beads were removed from cells and allowed the cells to rest in the 37°C incubator . 0 . 2 × 106 CTLs were used for each condition . Cells were plated in duplicates in 96-well plates either coated with or without anti-CD3/CD28 Ab , along with antibiotic Monensin ( BD Biosciences ) to block endocytosis of CD107a ( LAMP1a ) from the CTL membrane . Cells were allowed to degranulate for 2 hr at 37°C . Cell were washed and resuspended in isolation buffer ( 0 . 1% BSA + 2 mM EDTA in D-PBS , pH 7 . 4 ) for acquisition . Degranulation was measured with anti-CD107a conjugated with PE ( RRID:AB_1732051 ) . Data were acquired in FACSAria3 ( BD Pharmingen ) . Data were analyzed using FlowJo software ( RRID:SCR_008520 ) . Gates were set based on unstained controls . Briefly , 0 . 2 × 104 P815 target cells stably expressing Caspr3-GR ( FRET construct containing Tag-GFP and Tag-RFP with a target cleavage site DEVD of Caspase 3 ( activated via GzmB ) were washed once with PBS and resuspended in 100 µl AIMV medium . P815 cells were incubated with 30 µg/ml anti-CD3 antibody at 37°C for 20 min and plated into 8-well strips ( BD Falcon ) . After 2–3 min , 0 . 2 × 106 day 5 CTLs from either wild type or granzyme B-mTFP knock-in mouse were added gently to the target cells and imaged for 45 min to 1 hr . The cytotoxicity was calculated from the decrease in Tag-RFP and increase in Tag-GFP fluorescence . Fold ratio change of Tag-GFP/Tag-RFP values were calculated using ImageJ ( RRID:SCR_003070 ) software . Islets of Langerhans were isolated as described ( Stull et al . , 2012 ) . In short , mice were killed by cervical dislocation . After opening of the abdominal cavity , the bile duct was ligated by use of surgical suture near to the liver . A cut ( 0 . 5 mm ) was made into the intestine where the bile duct ended ( papilla ) . Through that cut a 0 . 40 mm wide blunt cannula was inserted into the bile duct half way toward the ligation . Through that cannula 3 mL of a Ringer solution containing 50 µg Liberase TL ( Merck ) was slowly injected into the bile duct , and inflated the pancreatic tissue . After injection , the pancreas was removed from the mouse abdomen . For digestion the pancreas was heated up to 37°C in 5 mL of Ringer solution and was gently agitated in a prewarmed water bath for 12 min . Digestion was stopped by adding 25 mL of cold ( 4°C ) Ringer solution . The pancreas then was smashed mechanically . Undigested debris was separated from Islets and exocrine material by filtration through a steel mesh ( 200 µm mesh size ) . Finally islets were separated from exocrine material by a density centrifugation ( 900 g , 18 min ) using a Ringer/Histopaque gradient ( Histopaque 1077 , Merck ) . Islets are supposed to stay at the Ringer/Histopaque interphase , while exocrine material goes to the pellet . After isolation islets were collected by filtering the supernatant through a 70 µm strainer , washed and resuspended in RPMI 1740 cell culture medium and kept in an incubator ( 37°C , 5% CO2 ) for up to 4 days before further use . Transplantation was performed as described ( Abdulreda et al . , 2013 ) . In short , mice ( 12–20 weeks old ) were anesthetized with isoflurane . An opening ( 500 µm ) was cut into the cornea with a sharp injection needle ( 30 G ) . Through this opening a blunt cannula ( 0 . 4 mm ) preloaded with 10–20 islets was inserted . The cannula was connected to a foot panel-steered injector through which the islets were injected into the eye under optical control ( binocular microscope ) . After transplantation mice were kept for 4 days under pain relief medication ( Caprofen , 5 mg/kg bodyweight ) before any further experimental manipulation . Imaging was performed as described previously ( Abdulreda et al . , 2011 ) . In short , transplanted mice were pre-anesthetized with isoflurane in a box . Under anesthetization animals were then mounted to a three point head holder with a respirator mask for continuing anesthesia . The mouse laid on a heating plate and the head was tilted in a way that the corneal surface of the transplanted eye was in perpendicular orientation to an objective of an upright microscope . Two different microscopes ( LSM880 confocal microscope , LSM880 two-photon microscope , both Carl Zeiss Jena ) were used for live imaging . Both systems were equipped with a 20x Plan-Apochromat water immersion objective ( Carl Zeiss Jena , NA 1 . 0 ) . If not otherwise stated data were acquired as volume ( xyz ) and time ( t ) 4D stacks ( xyzt; 11 planes over 40 µm , 30–60 s interval ) and are displayed as maximum intensity projections . For data acquisition Zen software ( RRID:SCR_013672 ) was used . Data analyses and 3D reconstructions were performed by using Imaris9 . 3 ( RRID:SCR_007370 ) . Deconvolution was done with Autoquant X3 ( RRID:SCR_002465 ) . For immunhistochemistry mice were deeply anesthetized by a intraperitoneal injection of a mixture of ketamine ( 280 mg/kg bodyweight ) and xylacine ( 20 mg/kg body weight ) . Access to the heart was obtained by removing part of the rips together with the sternum and lung displacement . Subsequently an injection needle connected to a perfusion pump punctured the left ventricle , and the right atrium was opened with small scissors . Perfusion was started immediately by a perfusion pump . Approximately 200 mL of fixative ( Ringer solution + 4% paraformaldehyde ) was pumped through the vascular network of the mouse . After perfusion the transplanted eye was isolated and postfixed in the same fixative for another 4 hr . After washing ( 3x , 2 hr each ) the eye was incubated in 30% sucrose solution over night ( 4°C ) before embedding in Tissue-Tek O . C . T . compound ( Sakura Finetek Europe B . V . ) . For embedding the eye was placed in a small beaker and surrounded by Tissue-Tek O . C . T . compound , dipped for 1 min into 2-methylbutane at −80°C and stored at −80°C until cutting . A cryostat ( Thermo Scientific HM525 ) was used to produce 10–20 µm thick eye slices . Slices were collected on Superfrost+ slides ( VWR ) . Staining with primary and secondary antibodies followed standard protocols . Samples were mounted using ProLong Gold Antifade ( RRID:SCR_015961 ) as a mounting medium . Statistical differences in data were calculated with paired or unpaired Student’s t-test . Unless mentioned otherwise , all data are presented as average ± standard average of the mean ( s . e . m . ) . Data were analyzed with ImageJ v1 . 46 ( RRID:SCR_003070 ) ( Schneider et al . , 2012 ) , Excel ( RRID:SCR_016137 ) , SigmaPlot 13 ( RRID:SCR_003210 ) and Imaris 9 . 3 ( RRID:SCR_007370 ) and graphed using Affinity Designer Software ( RRID:SCR_016952 ) . | Cytotoxic , or killer , T cells are a key part of the immune system . They carry a lethal mixture of toxic chemicals , stored in packages called cytotoxic granules . Killer T cells inject the contents of these granules into infected , cancerous or otherwise foreign cells , forcing them to safely self-destruct . In test tubes , T cells are highly efficient serial killers , moving from one infected cell to the next at high speed . But , inside the body , their killing rate slows down . Researchers think that this has something to do with how killer T cells interact with other immune cells , but the details remain unclear . To get to grips with how killer T cells work in their natural environment , researchers need a way to follow them inside the body . One approach could be to use genetic engineering to attach a fluorescent tag to a protein found inside killer T cells . That tag then acts as a beacon , lighting the cells up and allowing researchers to track their movements . Tagging a protein inside the cytotoxic granules would allow close monitoring of T cells as they encounter , recognize and kill their targets . But fluorescent tags are bulky , and they can stop certain proteins from working as they should . To find out whether it is possible to track killer T cells with fluorescent tags , Chitirala , Chang et al . developed a new type of genetically modified mouse . The modification added a teal-colored tag to a protein inside the granules of the killer T cells . Chitirala , Chang et al . then used a combination of microscopy techniques inside and outside of the body to find out if the T cells still worked . This analysis showed that , not only were the tagged T cells able to kill diseased cells as normal , the tags made it possible to watch it happening in real time . Super-resolution microscopy outside of the body allowed Chitirala , Chang et al . to watch the killer T cells release their toxic granule content . It was also possible to follow individual T cells as they moved into , and destroyed , foreign tissue that had been transplanted inside the mice . These new mice provide a tool to understand how killer T cells really work . They could allow study not only of the cells themselves , but also their interactions with other immune cells inside the body . This could help to answer open questions in T cell research , such as why T cells seem to be so much more efficient at killing in test tubes than they are inside the body . Understanding this better could support the development of new treatments for viruses and cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"immunology",
"and",
"inflammation"
] | 2020 | Studying the biology of cytotoxic T lymphocytes in vivo with a fluorescent granzyme B-mTFP knock-in mouse |
Unrelated genes establish head-to-tail polarity in embryos of different fly species , raising the question of how they evolve this function . We show that in moth flies ( Clogmia , Lutzomyia ) , a maternal transcript isoform of odd-paired ( Zic ) is localized in the anterior egg and adopted the role of anterior determinant without essential protein change . Additionally , Clogmia lost maternal germ plasm , which contributes to embryo polarity in fruit flies ( Drosophila ) . In culicine ( Culex , Aedes ) and anopheline mosquitoes ( Anopheles ) , embryo polarity rests on a previously unnamed zinc finger gene ( cucoid ) , or pangolin ( dTcf ) , respectively . These genes also localize an alternative transcript isoform at the anterior egg pole . Basal-branching crane flies ( Nephrotoma ) also enrich maternal pangolin transcript at the anterior egg pole , suggesting that pangolin functioned as ancestral axis determinant in flies . In conclusion , flies evolved an unexpected diversity of anterior determinants , and alternative transcript isoforms with distinct expression can adopt fundamentally distinct developmental roles .
The specification of the primary axis ( head-to-tail ) in embryos of flies ( Diptera ) offers important advantages for studying how new essential gene functions evolve in early development . This process rests on lineage-specific maternal mRNAs that are localized at the anterior egg pole ( ‘anterior determinants’ ) , which , surprisingly , have changed during the evolution of flies . While the anterior determinants of most flies remain unknown , they can be identified by comparing the transcriptomes of anterior and posterior egg halves ( Klomp et al . , 2015 ) . Furthermore , their function can be analyzed in the syncytial early embryos of a broad range of species via microinjection , considering timing and subcellular localization . It is therefore possible to conduct phylogenetic comparisons at the functional level . Finally , when the function of anterior determinants is suppressed , embryos develop into an unambiguous , predictable phenotype: these embryos lack all anterior structures and develop as two outward facing tail ends ( ‘double abdomen’ ) . Anterior determinants can be encoded by new genes with a dedicated function in establishing embryonic polarity . One example is bicoid in the fruit fly Drosophila melanogaster . Maternal mRNA of bicoid is localized in the anterior pole of the egg and Bicoid protein is expressed in a gradient in the early embryo ( Berleth et al . , 1988 ) . Bicoid-deficient embryos fail to develop anterior structures and instead form a second tail end , or a symmetrical double abdomen when the maternal activity gradient of another gene , hunchback , is disrupted simultaneously ( Driever , 1993 ) . The bicoid gene originated in the lineage of cyclorrhaphan flies more than 140 million years ago by duplication of zerknüllt ( zen; aka Hox3 ) , which , in insects , plays an important role in extraembryonic tissue development ( Schmidt-Ott et al . , 2010 ) . The expression and function of cyclorrhaphan bicoid orthologs are conserved but bicoid has not been found outside this group , and has been lost in some lineages within the Cyclorrhapha . Another example is panish , which encodes the anterior determinant of a midge , Chironomus riparius . This gene evolved by gene duplication of the Tcf homolog pangolin ( pan ) and capture of the maternal promoter of a nucleoside kinase gene , and has been called panish ( for pan“ish’ ) ( Klomp et al . , 2015 ) . Pangolin functions as the effector of ß-catenin-dependent Wnt signaling pathway ( ‘canonical’ Wnt signaling ) but Panish lacks the ß-catenin domain of Pangolin , and sequence similarity between Pangolin and Panish is limited to the cysteine-clamp domain ( 30 amino acids ) . panish has not been found outside the family Chironomidae , suggesting that lower dipterans use different anterior determinants . Here , we have used embryos of a wider range of dipteran species that lack bicoid and panish to address the question of how anterior determinants evolve . We started our analysis with moth flies ( Psychodidae: Clogmia albipunctata , Lutzomyia longipalpis ) and subsequently extended it to mosquitoes ( Culicidae: Culex quinquefasciatus , Aedes aegypti , Anopheles gambiae , Anopheles coluzzii ) , and to crane flies ( Tipulidae: Nephrotoma suturalis ) ( Figure 1A ) . Our results reveal three distinct old genes that evolved anterior determinants by localizing an alternative maternal transcript isoform at the anterior egg pole of the respective species . Therefore , alternative transcription might have played an important role in the evolution of this gene function and gene regulatory networks in fly embryos .
We annotated 5602 transcripts from the anterior and posterior transcriptomes of 1 hr-old bisected Clogmia embryos and ranked them according to the magnitude of their differential expression scores and P values ( Figure 1B ) . In the anterior embryo , the most enriched transcript was homologous to odd-paired , the Drosophila homolog of mammalian Zic ( zinc finger of the cerebellum ) genes . ZIC proteins are known to function as transcription factors or co-factors ( Houtmeyers et al . , 2013 ) . odd-paired was discovered in a screen for early Drosophila segmentation genes and subsequently classified as a ‘pair-rule’ gene , since odd-paired mutants fail to develop alternating segments ( Jürgens et al . , 1984 ) . During the Drosophila segmentation process , odd-paired is expressed in a single broad domain and controls the 'frequency-doubling' of other pair-rule genes ( Clark and Akam , 2016 ) . The Clogmia genome contains a single odd-paired locus ( Cal-opa ) ( Vicoso and Bachtrog , 2015 ) . Using RNA-seq data from preblastoderm and blastoderm embryos and Rapid Amplication of cDNA Ends ( RACE ) , we identified maternal and zygotic Cal-opa transcripts with alternative first exons that we mapped onto a 54 kb genomic scaffold ( Figure 1C ) . The maternal transcript ( Cal-opaMat ) was detected in preblastoderm embryos ( 0 . 5 hr-old ) and syncytial blastoderm embryos ( 4 hr-old ) . The zygotic transcript ( Cal-opaZyg ) was found in cellularized blastoderm embryos ( 7 hr-old ) and gastrulating embryos ( 9 hr-old ) . Protein alignments with homologs from other flies suggest that Cal-opaZyg encodes the full-length Cal-Opa protein ( 655 amino acids ) , while Cal-opaMat encodes a truncated protein variant ( 635 amino acids ) , lacking the N-terminal 20 amino acids of Cal-OpaZyg ( Figure 1—figure supplement 1 ) . To confirm the alternative Cal-opaMat and Cal-opaZyg transcripts and their non-overlapping expression patterns , we performed whole mount RNA in situ hybridization experiments with transcript-specific probes . The Cal-opaMat transcript was anteriorly localized in preblastoderm embryos but absent at the cellular blastoderm stage . Conversely , the Cal-opaZyg transcript was absent in preblastoderm embryos but expressed broadly in the trunk region of 7 hr-old blastoderm embryos ( Figure 1D ) , like odd-paired in Drosophila . These observations suggest that Cal-opa produces transcript isoforms with spatially and temporally distinct expression patterns . To determine the function of Cal-opaMat and Cal-opaZyg , we established a protocol for microinjecting early Clogmia embryos and conducted transcript-specific RNA interference ( RNAi ) experiments . Injection of Cal-opaMat double-stranded RNA ( dsRNA ) led to mirror-image duplications of the tail end ( double abdomen; Figure 2A and Figure 2—figure supplement 1 ) . In contrast , injection of dsRNA targeting Cal-opaZyg resulted in half the number of segmental expression domains of Cal-slp ( the ortholog of pair-rule gene sloppy-paired ) and caused defects in segmentation , dorsal closure , and head development but did not alter embryo polarity ( Figure 2B ) . Finally , injection of dsRNA targeting both Cal-opaMat and Cal-opaZyg resulted in double abdomens with missing segments ( Figure 2—figure supplement 2 ) . These observations indicate distinct roles of Cal-opaMat and Cal-opaZyg in specifying embryo polarity and in segmentation , respectively . We noticed that maternal transcripts of Cal-slp and Cal-mira , a homolog of miranda , which encodes an adaptor protein for cell fate determinants in Drosophila ( Ikeshima-Kataoka et al . , 1997; Adams et al . , 2000 ) , were also slightly enriched in the anterior portion of embryo ( Figure 1B ) . This observation was confirmed by RNA in situ hybridizations ( Figure 2—figure supplement 3 ) . Injection of Cal-slp dsRNA resulted in head and dorsal closure defects while Cal-mira dsRNA caused labrum and antennal defects , but in both cases embryo polarity was retained ( Figure 2—figure supplement 3 ) . To test whether Cal-opaMat can induce head development ectopically , we injected Cal-opaMat mRNA into the posterior pole of 1 hr-old embryos . These embryos expressed a head marker , Cal-otd ( ortholog of ocelliless/orthodenticle ) , on both ends of the embryo and developed a symmetrical double head , including some duplicated thoracic elements ( Figure 2C and Figure 2—figure supplement 4 and Video 1 ) . These observations suggest that anterior enrichment of maternal transcripts other than Cal-opaMat mRNA is not essential for head development , and that Cal-opaMat localization is sufficient for establishing embryo polarity . Next , we asked whether the early timing of odd-paired expression is critical for its function as anterior determinant in moth flies . To test this hypothesis , we conducted posterior injections of Cal-opaMat mRNA during the syncytial blastoderm stage ( 4 hr ) and examined Cal-otd expression after gastrulation . These embryos developed with normal head-to-tail polarity ( Figure 2D ) . This result places the requirement of odd-paired for axis specification prior to the syncytial blastoderm stage and suggests that early timing of odd-paired activity is essential for its function as anterior determinant . Cal-opaMat and Cal-opaZyg mRNAs not only differ in the timing of expression , but also differ in their 5’UTRs and predicted N-terminal protein sequences , as mentioned above ( Figure 1C and Figure 1—figure supplement 1 ) . To test whether the open reading frame difference is required for the anterior determinant function , we injected Cal-opaZyg mRNA at the posterior pole of preblastoderm embryos . Embryos from this experiment also developed as double heads ( Figure 2D ) . Because the translation start site of Cal-OpaMat is located downstream of the Cal-OpaZyg translation start site , we also tested Cal-opaZyg mRNA in which the putative start codon for Cal-OpaMat was mutated to encode leucine ( Cal-opaZyg-Met21Leu ) . Posterior injection of Cal-opaZyg-Met21Leu mRNA also resulted in double heads ( Figure 2D ) . These findings indicate that the protein difference between Cal-OpaMat and Cal-OpaZyg is not essential for the anterior determinant function of odd-paired in Clogmia . To test whether only a small portion of the Cal-opa open reading frame is required for its function as anterior determinant , we examined the ability of various truncated variants of Cal-Opa mRNAs ( Figure 1—figure supplement 1 ) to induce head development at the posterior egg pole ( Figure 2D ) . mRNAs of protein variants with large N-terminal truncation ( Cal-Opa115-655 and Cal-Opa182-655 ) retained the ability to induce double heads . However , mRNAs of protein variants with C-terminal truncation ( Cal-Opa182-445 and Cal-Opa1-337+22 , a hypothetical splice variant ) failed to induce double heads . These results indicate that the ability of Cal-Opa to specify embryo polarity requires the C-terminal portion of the protein but is largely insensitive to N-terminal truncation , corroborating our above conclusion that N-terminal differences between Cal-OpaMat and Cal-OpaZyg were not essential for evolving the anterior determinant function of Cal-opa . In Drosophila and other dipterans , maternal germ plasm in the posterior embryo not only specifies primordial germ cells but also contributes to and stabilizes embryo polarity via nanos , which suppresses the translation of anterior determinants in the posterior embryo ( Tautz , 1988; Gavis and Lehmann , 1992; Struhl et al . , 1992; Lemke and Schmidt-Ott , 2009 ) . The activity of nanos in the posterior preblastoderm is dependent on oskar ( Lehmann , 2016 ) , which is conserved in many insects ( Ewen-Campen et al . , 2010 ) . However , in Clogmia , expression profiling of anterior and posterior egg halves did not reveal any posteriorly localized maternal transcripts ( Figure 1B , alpha = 0 . 001 , unadj . ) , and no oskar homolog was found in our Clogmia transcriptomes or the Clogmia genome ( Vicoso and Bachtrog , 2015 ) . To test whether Clogmia lacks maternal germ plasm , we examined the expression of candidate germ cell markers , including Clogmia homologs of nanos ( Cal-nos1 , Cal-nos2 , Cal-nos3 , and Cal-nos4 ) , vasa ( Cal-vas ) , tudor ( Cal-tud ) , and germ cell-less ( Cal-gcl ) . Cal-nos1 , Cal-nos3 , and Cal-nos4 were not localized in the posterior of preblastoderm embryos but were expressed in a small set of cells at the posterior pole of cellular blastoderm and gastrulating embryos along with Cal-vas , Cal-tud , and Cal-gcl that were expressed more broadly ( Figure 2—figure supplement 5 ) . These observations suggest that Clogmia lacks maternal germ plasm and that Clogmia may induce the germ cell fate zygotically . To test this hypothesis , we examined Cal-nos expression in Cal-opaMat RNAi embryos . Cal-nos positive cells were duplicated in double abdomens ( Figure 2E ) , indicating that Clogmia uses an inductive mechanism for germ cell specification , which is repressed in the anterior embryo by Cal-opaMat . Therefore , axis specification in the Clogmia embryo is independent from germ cell specification . To our knowledge , Clogmia is also the first example of inductive germ cell specification in flies . Maternal odd-paired transcript is absent in freshly deposited eggs of chironomids ( Klomp et al . , 2015 ) and mosquitoes ( Akbari et al . , 2013 ) , both of which belong to the Culicomorpha lineage ( Figure 1A ) . To test whether localized maternal odd-paired transcript is broadly conserved in the Psychodomorpha lineage , we examined maternal transcript localization in the eggs of the sand fly Lutzomyia longipalpis , a moth fly species of public health concern due to its role in the transmission of visceral leishmaniasis . Of 5392 annotated transcripts , the most enriched maternal transcript in the anterior half of 1–2 hr old embryos was homologous to odd-paired and was therefore named Llo-opaMat ( Figure 3A ) . In the posterior Lutzomyia embryo , the most enriched transcript was homologous to oskar , indicating that Lutzomyia eggs contain maternal germ plasm at the posterior pole , unlike the Clogmia eggs . These findings suggest that a broad range of moth flies use odd-paired transcript as anterior determinant , and that maternal germ plasm was lost only in the Clogmia lineage . Close examination of Lutzomyia transcriptomes from 1 hr-old and 24 hr-old embryos also revealed zygotic odd-paired transcript ( Llo-opaZyg ) ( Figure 3B ) . Llo-opaMat and Llo-opaZyg share the same open reading frame but differ at their untranslated 5’ and 3’ ends . Since the N-terminal ends of Llo-OpaMat/Llo-OpaZyg and Cal-OpaZyg proteins are homologous ( Figure 1—figure supplement 1 ) , we infer that the N-terminal truncation of Cal-OpaMat occurred after the transcript had evolved maternal expression and anterior localization . The detection of Llo-opaZyg transcript in 24 hr-old embryos coincided with that of gap and pair-rule segmentation gene homologs ( Figure 3—figure supplement 1 ) , indicating that Llo-opaZyg functions during segmentation . The odd-paired gene of ancestral moth flies could have evolved the ability to establish the embryo polarity via specific amino acid substitutions . In this case , odd-paired homologs from species with a different anterior determinant , such as Drosophila or Chironomus , should not induce ectopic head development in Clogmia embryos . Alternatively , odd-paired could have evolved its role as axis determinant in moth flies independent of any amino acid substitution via co-option . In this case , odd-paired homologs from Drosophila or Chironomus could have the ability to induce head development in Clogmia embryos when appropriately expressed . To test this possibility , we injected odd-paired mRNA from Lutzomyia , Chironomus , or Drosophila into the posterior pole of early Clogima embryos . All of these odd-paired homologs induced double heads in Clogmia , provided that the endogenous kozak sequence of Cal-opaMat was used for optimal translation efficiency ( Figure 3C ) . Since neither Drosophila nor Chironomus uses odd-paired for specifying embryo polarity , these results suggest that amino acid substitutions were not essential for the evolution of the anterior determinant function of odd-paired in moth flies . We therefore propose that this gene function evolved via co-option when alternative maternal transcript of moth fly odd-paired became enriched at the anterior egg pole . Given that freshly deposited mosquito eggs lack maternal odd-paired transcript orthologs ( Akbari et al . , 2013 ) , we extended our search for anterior determinants to mosquitoes . Initially , we focused on the Southern House Mosquito Culex quinquefasciatus , a vector of West Nile virus ( the leading cause of mosquito-borne disease in the continental United States ) and of Wuchereria bancrofti ( the major cause of lymphatic filariasis ) . This species was chosen because their eggs are large and have clearly distinguishable anterior and posterior egg poles . We annotated 8239 Culex transcripts from the pooled anterior and posterior transcriptomes of 1 hr-old preblastoderm embryos and ranked them according to the magnitude of their differential expression scores and P values ( Figure 4A ) . In the posterior embryo , the most enriched transcript was related to nanos , consistent with the presence of maternal germ plasm in this species ( Figure 4D ) ( Juhn et al . , 2008 ) . The most enriched transcript in the anterior embryo was closely related to an uncharacterized gene of Drosophila ( CG9215 ) . However , reciprocal BLAST searches suggest that CG9215 belongs to a poorly defined larger gene family in Drosophila that might be represented by a single gene in mosquitoes . We named this gene cucoid to reflect its bicoid-like function in a culicine mosquito . cucoid encodes a protein with five C2H2 zinc finger domains ( Figure 4—figure supplement 1 ) . RACE experiments with cDNA from 0 to 7 hr-old embryos revealed three alternative cucoid transcripts with distinct 3’ ends ( cucoidA , cucoidB , and cucoidC ) ( Figure 4B ) , but only cucoidB and cucoidC were recovered from cDNA of 0–2 hr-old preblastoderm embryos , suggesting that one or both these transcripts might be maternally localized at the anterior pole . To test this hypothesis , we performed RNA in situ hybridizations with specific probes for cucoidA ( probe A ) or cucoidB ( probe B ) and , due to the very short sequence unique to cucoidC ( 121 nucleotides ) , a probe against all three isoforms ( probe C ) . cucoidA and cucoidB expression was detected in the fore and hind gut of extended germbands but not in 1 hr-old preblastoderm embryos . In contrast , the probe against all three isoforms detected maternally localized cucoid transcript at the anterior pole in addition to the zygotic expression pattern ( Figure 4C ) . Taken together , these results suggest that only the cucoidC isoform is maternally localized at the anterior pole and could function as anterior determinant . To test this hypothesis , we injected cucoid dsRNA from the shared 5’ region and examined the expression of a posterior marker ( Cqu-cad ) in gastrulating embryos . Many of these embryos expressed Cqu-cad in the anterior and underwent ectopic gastrulation at the anterior pole , suggesting that normal head-to-tail polarity was lost ( Figure 4E ) . Taken together , our results suggest that cucoid acquired the anterior determinant function via the localization of a maternal transcript isoform with an alternative 3’ end . We obtained similar results in another culicine mosquito , the Yellow Fever Mosquito Aedes aegypti , which transmits Dengue , Chikungunya , and Zika viruses . In this species , expression profiling of 5802 transcripts from the anterior and posterior transcriptomes of 1 hr-old preblastoderm embryos also identified cucoid ( Aae-cucoid ) as the gene with the most significantly enriched transcript in the anterior embryo ( Figure 5A ) . RACE experiments with cDNA from 1 hr to 6 hr-old embryos revealed three similar Aae-cucoid transcripts with alternative 3’ ends ( Aae-cucoidA , Aae-cucoidB , and Aae-cucoidC ) ( Figure 5B ) , and RNA in situ hybridization experiments with a probe against all three isoforms confirmed the anterior localization of Aae-cucoid transcript in preblastoderm embryos ( Figure 5C ) . Aae-cucoid expression in Aedes germbands could not be examined for technical reasons . In the posterior embryo , no highly enriched transcripts were observed . This was unexpected given that whole mount in situ hybridizations revealed posterior localized transcript of Aedes nanos in 1 hr-old embryos ( Figure 5C ) and that maternal transcript of Aedes oskar is also localized at the posterior pole ( Juhn and James , 2006 ) . Low statistical power of our differential expression analysis in Aedes might explain this discrepancy , since we could have confounded the anterior and posterior pole in some of the bisected Aedes eggs ( see Materials and methods ) . Alternatively , only a small portion of these transcripts might be localized at the posterior pole . Injection of Aae-cucoid dsRNA against the shared region of all transcripts resulted in double abdomens ( Figure 5D ) , suggesting that cucoid evolved its function as anterior determinant prior to the divergence of the Culex and Aedes lineages . The Anopheles gambiae species complex constitutes an outgroup to the Culex-Aedes clade . It includes eight or more sub-Saharan species that are difficult to distinguish due to widespread genealogical heterogeneity across the genome , incomplete lineage sorting and introgression ( Thawornwattana et al . , 2018 ) . We interchangeably used A . gambiae and A . coluzzii , two sibling species within this species complex that are responsible for the majority of malaria transmission in Africa , to identify the anterior determinant of this mosquito lineage . Whole mount RNA in situ hybridizations with a probe against the Anopheles gambiae ortholog of cucoid did not detect any anterior localized transcript in 1 hr-old embryos , suggesting that Anopheles uses a different anterior determinant than Culex and Aedes . To further test this possibility , we sequenced the anterior and posterior transcriptomes 1 hr-old preblastoderm embryos of A . gambiae and ranked 9353 transcripts according to the magnitude of their differential expression scores and P values . In the posterior embryo , the most enriched transcript was homologous to nanos . In the anterior embryo , the most enriched transcript was homologous to pangolin ( also known as Tcf ) ( Figure 6A ) . To test for potential alternative maternal and zygotic isoforms of pangolin in Anopheles , we mapped the assembled transcripts and 5’ and 3’ RACE products from 1 to 6 hr-old embryos onto an available A . gambiae genome assembly ( AgamP4 ) . We identified two alternative transcript variants with non-overlapping 3’UTRs but nearly identical open reading frames that we named Aga-panMat and Aga-panZyg , respectively ( Figure 6B and Figure 6—figure supplement 1 ) . Aga-panMat was tightly localized at the anterior pole of 1–2 hr-old preblastoderm embryos and only weakly expressed in elongated germbands , whereas Aga-panZyg was expressed segmentally in elongated germbands but not in embryos younger than 2 hr-old preblastoderm embryos ( Figure 6C ) . Both pangolin isoforms were conserved in Anopheles coluzzii with sequence identity above 99% and the maternal isoform ( Aco-panMat ) was localized at the anterior pole ( Figure 6—figure supplement 2 ) . The stage-specific expression of both pangolin isoforms was also conserved outside the Anopheles gambiae species complex in Anopheles stephensi ( Figure 6—figure supplement 3 ) . Alignments of dipteran Pangolin proteins suggest that , in Anopheles , the maternal variant includes an additional seven amino acids at C-terminal end due to alternative polyadenylation and splicing ( Figure 6—figure supplement 1 ) . To investigate the function of localized maternal pangolin expression in Anopheles , we specifically targeted this isoform in Anopheles coluzzii . Injection of Aco-panMat-specific dsRNA into several hundred 1 hr-old A . coluzzii embryos resulted in only 37 cuticles with variable phenotypes , including anterior truncations and , in extreme cases , double abdomens ( Figure 6D ) . We noticed perturbed segmentation boundaries in the double abdomens , suggesting that Aga-panMat may also function in segmentation , as suggested by its weak zygotic expression pattern . Injection of Aco-panZyg-specific dsRNA into 1 hr-old A . coluzzii embryos resulted in severe segmentation defects that were difficult to characterize , but no double abdomens or anterior-specific truncation defects were found ( Figure 6—figure supplement 4 ) . Taken together with the isoform-specific transcript localization data presented above , these RNAi results support the hypothesis that pangolin acquired the anterior determinant function via the localization of a maternal transcript isoform with an alternative 3’ end . Anterior-localized maternal pangolin ( Tc-pan ) transcript has also been observed in the eggs of a beetle ( Tribolium castaneum ) ( Bucher et al . , 2005 ) , but the function of this transcript remains unknown . Previous Tc-pan RNAi experiments targeted both maternal and zygotic transcripts and only revealed a function in posterior development , due to the role of zygotic Tc-pan in canonical Wnt signaling in the posterior growth zone ( Bolognesi et al . , 2008; Fu et al . , 2012; Prühs et al . , 2017; Ansari et al . , 2018 ) . To test whether ancestral dipterans localized maternal pangolin transcript at the anterior pole of the egg , we established a culture of the crane fly Nephrotoma suturalis ( Tipulidae ) , which belongs to the Tipulomorpha , one of the oldest branches of dipterans ( Grimaldi and Engel , 2005; Wiegmann et al . , 2011 ) . We sequenced the anterior and posterior transcriptomes of freshly deposited Nephrotoma egg halves and ranked 5371 transcripts according to the magnitude of their differential expression scores and P values ( Figure 6E ) . The most enriched transcript in the posterior embryo was related to oskar , suggesting that crane fly eggs contain maternal germ plasm at the posterior pole . The most enriched transcript in the anterior embryo was homologous to pangolin and therefore named Nsu-pan . The anterior localization of this transcript was confirmed by RNA in situ hybridization ( Figure 6F ) . RACE experiments with cDNA from 1 hr-old embryos identified multiple isoforms with slightly variable 5’ ends but the same open reading frame . An alignment of the predicted Nsu-Pan protein from this open reading frame with other dipteran Pangolin homologs revealed conserved N-terminal and C-terminal ends in Nsu-Pan . Taken together with our Anopheles data , our results in Nephrotoma suggest that ancestral dipteran insects localized maternal pangolin transcript in the anterior egg pole , where this transcript may have functioned as anterior determinant . In the midge Chironomus , the ortholog of pangolin ( Cri-pan ) is not expressed maternally but its diverged paralog panish functions maternally as anterior determinant ( Klomp et al . , 2015 ) . Given that panish evolved from pangolin via gene duplication , panish probably inherited its role from pangolin . Therefore , it is possible that Cri-pan and panish are still functionally equivalent when expressed at the anterior pole of preblastoderm Chironomus embryos . Alternatively , Panish may have co-evolved with the targets required for anterior patterning and Pangolin can no longer interact with those targets . In this case , Cri-pan should no longer be able to fulfill the function of panish . To distinguish between these possibilities , we examined the ability of panish and Cri-pan mRNAs to rescue the RNAi phenotype of panish . We have previously shown that dsRNA of the panish 3’UTR can induce the double abdomen phenotype with a penetrance of nearly 100% , and that this phenotype can be rescued in roughly half of the embryos by injecting panish mRNA with heterologous UTRs at the anterior pole , shortly after the injection of dsRNA ( Klomp et al . , 2015 ) . We used this assay to compare the functions of panish mRNA ( positive control ) , frame shifted panish mRNA ( negative control ) , Cri-pan mRNA , and a modified Cri-pan mRNA designed to better resemble panish mRNA ( Cri-pantrunc . mRNA ) . Cri-pantrunc . mRNA encodes a N-terminal truncated Cri-Pan variant , lacking the ß-Catenin binding and HMG box domains , with two mutations in the cysteine-clamp domain to mimic conserved changes of the Panish cysteine-clamp ( Figure 6—figure supplement 5A ) . Only panish mRNA rescued panish RNAi embryos ( Figure 6—figure supplement 5B ) , suggesting that panish co-evolved with its targets and functionally diverged after its origin via gene duplication from pangolin .
In this study , we have identified three unrelated old genes that encode the anterior determinant in moth flies , culicine mosquitoes , and anopheline mosquitoes , respectively ( Figure 7 ) . All three genes not only localize their maternal transcript at the anterior egg pole; they also are subject to alternative transcription , which allows a single gene to generate multiple transcript isoforms with distinct 5’ and 3’ ends through the use of alternative promoters ( alternative transcription initiation ) and polyadenylation signals ( alternative transcription termination ) . In moth flies , the localized maternal odd-paired transcript that functions as anterior determinant has an alternative first exon compared to the canonical isoform ( Figure 1B–D , Figure 2 , and Figure 3A–B ) . In the case of mosquitoes , maternal transcript isoforms of cucoid ( in culicine mosquitoes ) or pangolin ( in anopheline mosquitoes ) with alternative last exons are localized at the anterior pole of the egg and function as anterior determinant ( Figure 4 , Figure 5 , and Figure 6 ) . Since anterior determinants are localized in the anterior egg , and signals for the subcellular localization of transcripts are typically found in UTRs ( Holt and Bullock , 2009 ) , it is possible that alternative transcription facilitates the evolution of anterior determinants by providing the UTR sequence for isoform-specific localization signals that do not interfere with other gene functions . For example , it has been shown that alternative last exons of transcript isoforms confer isoform-specific localization in neurons ( Taliaferro et al . , 2016; Ciolli Mattioli et al . , 2019 ) . Additional experiments will be needed to test whether the unique UTR sequences of anterior determinants are essential for their localization at the anterior egg pole . In addition to changes in UTR sequences , alternative transcription also can result in the truncation or elongation of the open reading frame . For example , the anterior determinant of Clogmia ( Cal-OpaMat ) lacks the N-terminal 20 amino acids ( Figure 1—figure supplement 1 ) , and the anterior determinant of Anopheles ( Aga-panMat ) encodes protein that includes additional seven amino acids at the C-terminal end ( Figure 6—figure supplement 1 ) . However , these changes to the protein may not have been important for adopting a function as anterior determinant . The truncation in the maternal Odd-paired protein that we observed in Clogmia is not conserved in Lutzomyia , in which Llo-opaMat and Llo-opaZyg encode the same protein , and full-length Odd-paired homologs from these and other species can function as anterior determinant in Clogmia ( Figure 3B–C ) . Also , the elongation of Aga-PanMat protein is not conserved in Nephrotoma , in which the localized Nsu-panMat transcript encodes a Pangolin protein with a conserved C-terminal end ( Figure 6—figure supplement 1 ) . Therefore , modifications in the open reading frame of these genes may reflect secondary changes . Unlike the anterior determinants identified in this study , the previously described anterior determinants of Drosophila and Chironomus are encoded by newly evolved genes , bicoid and panish . These genes seem to be dispensable outside the context of axis specification ( Driever , 1993; Klomp et al . , 2015 ) , suggesting that they evolved specifically for this function . They could have acquired their function de novo via protein evolution or via inheritance from the progenitor gene . Our findings suggest that the role of pangolin in axis specification was already present in ancestral dipterans ( Figure 7 ) . We therefore propose that panish , which evolved from pangolin via gene duplication in the Chironominae lineage ( Figure 7—figure supplement 1 ) , inherited its function from pangolin . Future examinations of pangolin isoforms and their expression in the eggs of chironomids that lack panish orthologs ( species representing basal chironomid lineages ) could reveal intermediate steps in this process , such as a localized truncated pangolin isoform . Similarly , bicoid could have acquired its function de novo via protein evolution or via inheritance from its progenitor gene , zerknüllt . Several previous studies have hypothesized that Bicoid replaced Orthodenticle , a conserved homeodomain protein with similar DNA-binding affinity that functions in animal head development ( Wimmer et al . , 2000; Schröder , 2003; Lynch et al . , 2006; Datta et al . , 2018 ) . Ancestrally reconstructed homeodomains confirmed that a single amino acid change in the homeodomain of Bicoid ( Q50K ) , which is shared by Bicoid and Orthodenticle , caused a dramatic shift of Bicoid’s DNA-binding affinity in vitro and target recognition in vivo ( Liu et al . , 2018 ) . We cannot rule out that orthodenticle functioned as anterior determinant in ancestral brachyceran flies . However , in analogy with our findings , it is also possible that a maternal zerknüllt isoform became localized at the anterior pole of the egg and acquired the role of anterior determinant via co-option , prior to the origin of bicoid via gene duplication ( Figure 7—figure supplement 2 ) . Maternal zerknüllt expression is common in lower Diptera but was lost in Cyclorrhapha ( Stauber et al . , 2002 ) . If bicoid inherited its function from zerknüllt , the Q50K mutation in the homeodomain of Bicoid must have been a secondary , potentially maladaptive change . In this case , it may have been fixed in the cyclorrhaphan stem lineage via a compensatory or balancing mechanism and would have driven co-evolution of its targets . It may be objected that , in Drosophila , reverting the K50 residue of the Bicoid homeodomain to Q50 is lethal and results in a bicoid null phenotype ( Liu et al . , 2018 ) . However , how the ancestral gene network responded to the Q50K mutation of Bicoid cannot be inferred from observations in Drosophila . Moreover , published biochemical data suggest that the Q50K mutation increases interaction with the consensus Bicoid binding DNA motif much stronger than it reduces interaction with the consensus Zerknüllt binding DNA motif . It is therefore conceivable that the Q50K mutation had a less dramatic effect in ancestral flies , in which the target genes of the anterior determinant were activated via Zerknüllt binding sites , than in Drosophila , in which the target genes of the anterior determinant are activated via Bicoid binding sites . Examination of the mechanisms that determine embryo polarity in non-cyclorrhaphan Brachycera flies might help to test this hypothesis . If panish and bicoid inherited their functions , their evolutionary origin and divergence could have served the purpose of reducing the pleiotropy of their progenitor genes , pangolin and zerknüllt , rather than allowing them to take on an entirely new function in development . Recent genome-wide analyses have shown that alternative transcription is a widespread phenomenon . For example , there are on average four alternative transcription start sites per gene in humans ( Forrest et al . , 2014 ) and at least 50–70% of mammalian genes are subject to alternative polyadenylation ( Shepard et al . , 2011; Adams et al . , 2000; Derti et al . , 2012 ) . Alternative transcript isoforms can be tightly regulated in a cell or tissue specific manner and can affect transcription and translation efficiency as well as splicing ( An et al . , 2008; Davuluri et al . , 2008; Lau et al . , 2010; Pinto et al . , 2011; Smibert et al . , 2012; Anvar et al . , 2018; Tushev et al . , 2018; Taliaferro et al . , 2016; Ciolli Mattioli et al . , 2019; Shabalina et al . , 2010 ) . Functional studies in model organisms have shown that alternative transcription can generate dominant negative and alternatively localized protein isoforms ( Davuluri et al . , 2008; Bharti et al . , 2008; Vacik et al . , 2011; Berkovits and Mayr , 2015; Schrankel et al . , 2016 ) , while misregulation of alternative transcript isoforms has been associated with human diseases including cancer ( Mayr and Bartel , 2009; Wiesner et al . , 2015; Pal et al . , 2012; Shapiro et al . , 2011 ) . However , the contribution of alternative promoters ( alternative transcription initiation ) and polyadenylation signals ( alternative transcription termination ) to the evolution of new gene functions and regulatory networks remains poorly understood ( Carroll et al . , 2005; Peter and Davidson , 2011; Wittkopp and Kalay , 2011; de Klerk and 't Hoen , 2015 ) . The results of several previous studies suggest that alternative transcription may underlie the evolutionary diversification of gene functions . For example , a large fraction of alternative promoter sequences is conserved between human and mice , but those with cell or tissue restricted expression have frequently changed during mammalian evolution ( Baek et al . , 2007; Forrest et al . , 2014 ) , suggesting that alternative promoters may have played a significant role in cell type evolution . Another likely substrate for evolutionary diversification are the protein terminal ends generated by alternative transcription in conjunction with alternative splicing . These terminal ends commonly contain intrinsically disordered regions which are enriched in sites that mediate protein-protein interactions ( Buljan et al . , 2013; Shabalina et al . , 2014 ) . A recent case study found that a light fur color variant of beach mice evolved repeatedly via selection for an alternative agouti isoform with increased translation efficiency ( Linnen et al . , 2013; Mallarino et al . , 2017 ) . Our in vivo study revealed three old genes that evolved the anterior determinant function by localizing an alternative transcript isoform at the anterior pole of the egg . Therefore , we propose that differential expression of alternative transcript isoforms can result in the evolution of new gene functions , independent of , and prior to gene duplication and sub-functionalization . Given that alternative transcription is a widespread phenomenon , it could play an important role in the evolution of gene regulatory networks .
Coding sequences from Clogmia , Lutzomyia , Chironomus , Anopheles , Nephrotoma , Culex , and Aedes were amplified from embryonic cDNA with primers constructed from RNA-seq data . Coding sequence of odd-paired was amplified from cDNA ( FI01113 ) that was obtained from the BDGP Gold collection of the Drosophila Genomics Resource Center . Amplified cDNA was cloned into the expression vector pSP35T ( Amaya et al . , 1991 ) , using In-Fusion HD Cloning Kit ( Clontech ) , and PstI- or EcoRI-linearized vector was used for mRNA synthesis using mMESSAGE mMACHINE SP6 . panish , Cri-pan , and Cri-pantrunc mRNAs were synthesized from PCR template ( containing T7 polymerase binding site ) using mMESSAGE mMACHINE T7ultra . Mutations in the open reading frame ( Cal-opaZyg-Met21Leu , panish FS , Cri-pantrunc ) and in the kozak sequences of Dme-opa and Cri-opa were generated using QuikChange Lightning Site-Directed Mutagenesis Kit ( Agilent ) . Double-stranded RNA ( dsRNA ) was generated from PCR-amplified templates using embryonic cDNA and primers containing T7 polymerase binding sites as described ( Klomp et al . , 2015 ) . Forward and reverse primer sequences for generating templates for mRNA synthesis were: Forward and reverse primer sequences for dsRNA ( lengths of dsRNAs in brackets; gene specific sequence of primers underlined ) : Chironomus embryo injection was done as previously described ( Klomp et al . , 2015 ) . Clogmia eggs were dissected from ovaries and activated under water . Eggs of Aedes , Culex , and Anopheles were collected in a dark chamber on a moist filter paper for about 30 min ( Fisher Scientific , Cat . No 09–795C ) . These eggs were transferred to another filter paper cut into 4 cm x 2 cm pieces and aligned perpendicularly to the edge of a cover glass ( Fisher Scientific , Cat . No 12-648-5C ) with the prospective injection side pointing towards the glass edge . We noticed that injecting eggs near the anterior or posterior pole was critical for survival of the procedure . During the alignment procedure , water was applied to the filter paper as needed to prevent eggs from desiccation . After aligning the eggs , the cover glass was removed and excess water on the filter paper was absorbed using filter paper . A second cover glass with a layer of double-sided tape ( Scotch 3M ) was slightly pressed against the aligned eggs to transfer the eggs to the double-sided tape . The embryos were then immediately covered with halocarbon oil to prevent desiccation . For cuticle preparations , the embryos were injected under halocarbon oil 27 ( Sigma , MKBZ7202V ) . The oil was washed off under a gentle stream of water immediately after injection . In the case of Clogmia , Aedes , and Culex the cover glass was transferred to a moist chamber ( petri dish with wet kimwipe paper ) and kept at 28 °C , and water was added every day to prevent desiccation . In the case of Anopheles , the eggs were allowed to develop under water . Removal of the halocarbon oil was critical to ensure embryo survival until late developmental stages and hatching . For eggs to be fixed within a day following injection , we used a 1:1 mixture of halocarbon oil 27 and halocarbon oil 700 ( Sigma , MKCB5817 ) and left the injected eggs immersed in the oil until fixation . Embryos were injected with quartz needles using a Narishige IM-300 microinjector . Quartz capillaries ( Sutter Instruments Q100-70-10 ) were pulled with a Sutter instrument P-2000 laser-based micropipette puller . Our settings for the needle puller were: Heat 645 , Fil 4 , Vel 40 , Del 125 , Pul 130 . Needles were back-filled and the tip was broken open at the time of injection by slightly touching the first egg . Clogmia Embryos were dechorionated using a 10% dilution of commercial bleach ( 8 . 25% sodium hypochloride ) for 3 min . For Nephrotoma embryos , a 25% dilution was used for 3 min until the chorion became slightly transparent . Embryos of Aedes , Culex , and Anopheles were dechorionated as described ( Juhn and James , 2012 ) . Dechorionated embryos were fixed in a 50 mL falcon tube , using 20 mL of boiling salt/detergent-solution ( 100 μL 10% triton-X , 500 μL 28% NaCl , up to 20 mL of water ) . After 10 s , water was applied to the tube to cool down the embryos . If needed , the embryos were devitellinized in a 1:1 mixture of n-heptane and methanol by gentle shaking . Embryos with vitelline membrane attached were further devitellinized using sharp tungsten needles in an agar plate covered with methanol . Devitellized embryos were stored in 100% methanol at −20°C . RNA in situ hybridizations were conducted as described ( Klomp et al . , 2015 ) , using digoxigenin ( DIG ) -labeled probes and Fab fragments from anti-DIG antibodies conjugated with alkaline phosphatase ( AP ) ( Roche , IN , USA ) . Probes were prepared from PCR templates , using sequence-specific forward primers and reverse primers with T7 promoter sequence ( see above for Cal-opaMat , Cal-opaZyg , Cal-cad , Cal-slp , Cal-mira , Aga-panMat , Aga-panZyg , and Aae-cucoid; gene specific sequence underlined ) . Total RNA was phenol/chloroform extracted from Clogmia ( 1 hr-old and 9 hr-old embryos ) , Anopheles ( 1–6 hr-old embryos ) , Culex ( 0–7 hr old embryos ) , and Nephrotoma ( 1–29 hr-old embryos ) fixed in TRIzol Reagent ( Invitrogen ) and precipitated with isopropanol . 5’/3’ RACE was performed using SMARTer RACE 5'/3' Kit ( Clontech ) with the custom-made primers ( including at the 5’ end 15 nucleotides of pRACE vector sequence ) . Gene specific sequences are underlined . Cuticles were prepped four to five days after injection . Eggshells was removed with tungsten needles and the embryos were transferred to a glass block dish with a drop of 1:4 glycerol/acetic acid . Following incubation in 1:4 glycerol/acetic acid overnight at room temperature , the cuticles were transferred onto a glass slide , oriented , mounted in 1:1 Hoyer’s medium/lactic acid ( Stern and Sucena , 2000 ) , covered with a cover glass , and dried overnight at 65 °C . Bisection of anterior and posterior embryo halves , RNA extraction , and sequencing were conducted as described ( Klomp et al . , 2015 ) . In the case of Clogmia , anterior or posterior embryo halves from three 1 hr-old embryos were pooled and RNA-seq data were obtained from two replicates . In case of Lutzomyia , embryo halves from ten 1–2 hr-old embryos were pooled four replicates were generated . In case of Anopheles ( G-3 strain ) , embryo halves from five 1 hr-old embryos were pooled and three replicates were generated , In the case of Culex , embryo halves from seven 1 hr-old embryos were pooled and three replicates were generated . In case of Aedes ( Liverpool ‘black eye’ strain ) , embryo halves from five 1 hr-old embryos were pooled and four replicates were generated . In case of Nephrotoma , embryo halves from nine 1 hr-old embryos were pooled and three replicates were generated . Stage-specific Clogmia transcriptomes were generated from the offspring of a single mother and total RNA from five embryos was used for each stage . In the case of Lutzomyia , about 100 staged embryos were pooled for RNA extraction , and two independent RNA extractions from each time point were combined and submitted for sequencing . Prior to library construction , RNA integrity , purity , and concentration were assessed using an Agilent 2100 Bioanalyzer with an RNA 6000 Nano Chip ( Agilent Technologies , USA ) . Purification of messenger RNA ( mRNA ) was performed using the oligo-dT beads provided in the Illumina TruSEQ mRNA RNA-SEQ kit ( Illumina , USA ) . Complementary DNA ( cDNA ) libraries for Illumina sequencing were constructed using the Illumina TruSEQ mRNA RNA-SEQ kit ( Illumina , USA ) , using the manufacturer-specified protocol . Briefly , the mRNA was chemically fragmented and primed with random oligos for first strand cDNA synthesis . Second strand cDNA synthesis was then carried out with dUTPs to preserve strand orientation information . The double-stranded cDNA was then purified , end repaired , and ‘a-tailed’ for adaptor ligation . Following ligation , the samples were selected a final library size ( adapters included ) of 400–550 bp using sequential AMPure XP bead isolation ( Beckman Coulter , USA ) . The libraries were sequenced in an Illumina HiSeq 4000 DNA sequencer , utilizing a pair end sequencing flow cell with a HiSeq Reagent Kit v4 ( Illumina , USA ) . The TrimGalore ( Krueger , 2012 ) wrapper for Cutadapt ( Martin , 2011 ) and FastQC ( Andrews , 2010 ) was used to remove adapters and low quality sequences from raw fastq files . Overlapping reads were combined with Flash ( Magoč and Salzberg , 2011 ) prior to assembly . Trinity 2 . 4 . 0 ( Grabherr et al . , 2011 ) on the Indiana University Karst high-performance computing cluster was used for assembling contiguous sequences ( contigs ) from the paired end ( PE ) sequence data of Clogmia , Lutzomyia , Anopheles , and Nephrotoma . ABySS 2 . 0 ( Jackman et al . , 2017 ) was used for assembling contigs from Culex and Aedes data . Only contigs of 200 nucleotides or greater were retained . BLAST+ tools ( Camacho et al . , 2009 ) were used to annotate contigs by conducting best-reciprocal-blast first against the Drosophila melanogaster transcriptome ( BDGP6 ) peptide sequences ( blastx/tblastn ) and then the coding sequence ( tblastx ) with a maximum threshold evalue of 1e-10 . Biomart and AnnotationDbi packages were used for gene ids and names . The longest open reading frames ( ORFs ) of unannotated transcripts were compared to the RefSeq invertebrate protein database ( downloaded 4-1-2017 ) using blastp ( max evalue 1e-10 ) followed by a similar comparison to remove transcripts with ORFs matching RefSeq plant , protozoan , archaea , bacteria , fungi , plasmid , or viral sequences ( downloaded 6-1-2017 ) . Remaining transcripts were designated by the top BLAST hit in D . melanogaster . Cleaned paired-end read data was aligned and analyzed using R base ( Ihaka and Gentleman , 1996 ) and Bioconductor ( Gentleman et al . , 2004 ) software packages . Sequence alignment was conducted with the seed-and-vote aligner , Subread , as implemented in the Rsubread package ( Liao et al . , 2013 ) with up to five multi-mapping locations , six mismatches , and 20 subreads/seeds per read . Sequence file manipulation , including sorting and indexing of ‘ . bam’ files , was done using Rsamtools ( Morgan et al . , 2013 ) . To avoid potential biases in transcript localization unrelated to anterior-posterior axis formation , transcripts annotated with mitochondrial , ribosomal , or ambiguous status ( e . g . , predicted , hypothetical , or uncharacterized ) were filtered out prior to the differential expression comparisons . Transcripts with 20 or fewer counts in any of the A-P pairs were also excluded from the analysis prior to library normalization . Lower scoring , potentially related transcripts matching a given gene from the D . melanogaster transcriptome were retained for initial differential expression comparisons but removed for clarity of presentation in subsequent analyses and volcano plots . Trimmed mean of M-values ( TMM ) ( Robinson and Oshlack , 2010 ) was used for normalization and EdgeR ( Robinson et al . , 2010 ) was used to perform quasi-likelihood F-tests between A-P samples , corrected for multiple testing using FDR ( Benjamini-Hochberg ) . Following filtering based on annotation and detection of >20 counts per paired samples , we used the following number of transcripts for differential expression comparisons: 5602 for Clogmia; 5392 for Lutzomyia; 8239 for Culex; 5802 for Aedes; 9353 for Anopheles; 5371 for Nephrotoma . RNA-seq reads from stage-specific transcriptomes were mapped to genomic scaffolds containing a gene of interest using TopHat RNA-seq aligner ( Trapnell et al . , 2009 ) . Publicly available Anopheles stephensi transcriptomes used in this paper were: SRR515316 , SRR515341 , SRR514863 , and SRR515304 . This project was deposited at the National Center for Biotechnology Information under Bioproject ID PRJNA454000 and the reads were deposited in the Short Reads Archives under accessions SRR7132661 , SRR7132662 , SRR7132659 , SRR7132660 , SRR7132665 , SRR7132666 , SRR7132663 and SRR7132664 for Clogmia , SRR7134470 , SRR7134469 , SRR7134472 , SRR7134471 , SRR7134468 , and SRR7134467 for Lutzomiya , SRR8729860 , SRR8729859 , SRR8729858 , SRR8729857 , SRR8729856 and SRR8729855 for Anopheles , SRR8729854 , SRR8729853 , SRR8729852 and SRR8729851 for Aedes , SRR8729868 , SRR8729867 , SRR8729870 , SRR8729869 , SRR8729864 and SRR8729863 for Culex and SRR8729866 , SRR8729865 , SRR8729872 , SRR8729871 , SRR8729861 and SRR8729862 for Nephrotoma . Transcript sequences are listed on the Key Resources Table . | With very few exceptions , animals have ‘head’ and ‘tail’ ends that develop when they are an embryo . The genes involved in specifying these ends vary between species and even closely-related animals may use different genes for the same roles . For example , the products of two unrelated genes called bicoid in fruit flies and panish in common midges accumulate at one end of their respective eggs to distinguish head from tail ends . It remained unclear how other fly species , which have neither a bicoid nor a panish gene , distinguish the head from the tail end , or how genes can evolve the specific function of bicoid and panish . Cells express genes by producing gene templates called messenger ribonucleic acids ( or mRNAs for short ) . The central portions of mRNAs , known as protein-coding sequences , are then used to produce the protein . Proteins can play several distinct roles , which they acquire through evolution . This can happen in different ways , for example , genetic mutations in the part of a gene that codes for protein may alter the resulting protein , giving it a new activity . Alternatively , sequences at the beginning and the end of an mRNA molecule that do not code for protein , but regulate when and where proteins are made , can influence a protein’s role by changing its environment . Many genes produce mRNAs with alternative sequences at the beginning or the end , a process known as alternative transcription . Here , Yoon et al . identified three unrelated genes that perform similar roles to bicoid and panish in the embryos of several different moth flies and mosquitoes . These genes appear to have acquired their activity because one of their alternative transcripts accumulated at the future head end , rather than through mutations in the protein-coding sequences . Studying multiple species also made it clear that panish inherited its function from a localized alternative transcript of an old gene that duplicated and diverged . These findings suggest that alternative transcription may provide opportunities for genes to evolve new roles in fundamental processes in flies . Most animal genes use alternative start and stop sites for transcription , but the reasons for this remain largely obscure . This is especially the case in the human brain . The findings of Yoon et al . , therefore , raise the question of whether alternative transcription has played an important role in the evolution of the human brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"evolutionary",
"biology"
] | 2019 | Embryo polarity in moth flies and mosquitoes relies on distinct old genes with localized transcript isoforms |
Changes in cancer cell identity can alter malignant potential and therapeutic response . Loss of the pulmonary lineage specifier NKX2-1 augments the growth of KRAS-driven lung adenocarcinoma and causes pulmonary to gastric transdifferentiation . Here , we show that the transcription factors FoxA1 and FoxA2 are required for initiation of mucinous NKX2-1-negative lung adenocarcinomas in the mouse and for activation of their gastric differentiation program . Foxa1/2 deletion severely impairs tumor initiation and causes a proximal shift in cellular identity , yielding tumors expressing markers of the squamocolumnar junction of the gastrointestinal tract . In contrast , we observe downregulation of FoxA1/2 expression in the squamous component of both murine and human lung adenosquamous carcinoma . Using sequential in vivo recombination , we find that FoxA1/2 loss in established KRAS-driven neoplasia originating from SPC-positive alveolar cells induces keratinizing squamous cell carcinomas . Thus , NKX2-1 , FoxA1 and FoxA2 coordinately regulate the growth and identity of lung cancer in a context-specific manner .
Cancer progression is often accompanied by profound changes in cellular identity . Cellular identity , or differentiation state , influences not only intrinsic malignant potential , but also response to therapy , even in tumors harboring the same targetable mutations ( Cohen and Settleman , 2014 ) . Although tissue of origin is a major determinant of cancer cell identity , cancer cells can also undergo lineage switching in the course of their natural history and in response to the selective pressure of targeted therapy . In lung adenocarcinoma , absence of the pulmonary lineage specifier NKX2-1/TTF1 correlates with non-pulmonary cellular identities and poor prognosis compared with NKX2-1-positive tumors ( Barletta et al . , 2009; Cardnell et al . , 2015 ) . Moreover , lung adenocarcinomas can undergo lineage switching during the evolution of drug resistance that reduces their dependence on the oncogenic signaling pathway being targeted ( Rotow and Bivona , 2017 ) . Taken together , these observations indicate that there is a need to understand the critical regulators of cancer cell identity . In previous work , we and others have shown that loss of NKX2-1 is sufficient to cause lineage switching in a mouse model of KRASG12D-driven lung adenocarcinoma ( Maeda et al . , 2011; Snyder et al . , 2013; Tata et al . , 2018 ) . Nkx2-1 deletion in established tumors causes cancer cells to shed their pulmonary identity and adopt a gastric-like differentiation state characterized by extensive mucin production and expression of multiple gastrointestinal markers , including HNF4α and Gastrokine 1 . These tumors morphologically resemble a subtype of human lung cancer called invasive mucinous adenocarcinoma ( IMA ) , which also expresses gastrointestinal markers and is predominantly driven by KRAS mutations ( Guo et al . , 2017 ) . Approximately 10–15% of human lung adenocarcinomas express HNF4α with no detectable NKX2-1 ( 9 ) , including both IMAs and more moderately differentiated tumors . In many of these tumors , the NKX2-1 gene appears to be silenced by genetic and/or epigenetic mechanisms ( Hwang et al . , 2016; Matsubara et al . , 2017 ) . Aside from NKX2-1 itself , the Polycomb Repressive Complex 2 ( PRC2 ) appears to play a role in suppressing mucinous differentiation in KRAS-driven , p53-deficient lung adenocarcinoma ( Serresi et al . , 2016 ) . However , the precise mechanisms by which a gastric gene expression program is activated in NKX2-1-deficient tumors remain to be fully elucidated . Many of the gastrointestinal transcripts expressed in IMA are known targets of the forkhead box transcription factors FoxA1 and FoxA2 ( FoxA1/2 ) . These transcription factors govern the development of a variety of tissues and are expressed in both the adult lung and GI tract ( reviewed in Golson and Kaestner , 2016 ) . FoxA1/2 are also expressed in both murine and human IMA ( Figure 1A and Figure 1—figure supplement 1A–B ) . We previously found that Nkx2-1 deletion in autochthonous lung tumors caused FoxA1/2 to re-localize from the regulatory elements of pulmonary-specific genes ( such as Sftpa1 ) to those of genes ( such as Hnf4a ) that are expressed in both the GI tract and IMA ( Snyder et al . , 2013 ) . Given that NKX2-1 physically interacts with FoxA1/2 ( Snyder et al . , 2013; Minoo et al . , 2007 ) , we hypothesized that NKX2-1 promotes FoxA1/2 interaction with regulatory elements of the pulmonary differentiation program at the expense of those governing gastric identity . However , these data did not demonstrate a functional role for FoxA1/2 in the activation of the gastric program in these tumors . To address this question directly , we used conditional alleles of Foxa1 ( Gao et al . , 2008 ) and Foxa2 ( Sund et al . , 2000 ) to abrogate their function in an autochthonous mouse model of NKX2-1-negative lung adenocarcinoma . We found that FoxA1/2 are critical and redundant regulators of both the gastric differentiation program and growth of NKX2-1-negative tumors . Moreover , we found that the cellular identity adopted by tumors was highly dependent on the context in which FoxA1/2 activity is lost , suggesting that a cell’s baseline epigenetic state can influence the identity it adopts in response to changes in lineage specifier expression .
To test the hypothesis that FoxA1/2 are required for lung adenocarcinoma cells to undergo a pulmonary to gastric lineage switch upon loss of NKX2-1 expression , we incorporated conditional alleles of Foxa1 and Foxa2 into a mouse model of NKX2-1-deficient lung adenocarcinoma ( Snyder et al . , 2013 ) . In this model , intratracheal delivery of virus expressing Cre recombinase simultaneously activates a conditional allele of oncogenic Kras ( KrasLSL-G12D/+ ) and silences conditional alleles of Nkx2-1 ( Nkx2-1F/F ) alone or in addition to Foxa1 ( Foxa1F/F ) and/or Foxa2 ( Foxa2F/F ) . Initial evaluation by morphology ( H and E ) and immunohistochemistry ( IHC ) showed that tumors lacking either FoxA1 or FoxA2 were indistinguishable from control tumors ( Figure 1A ) . In sharp contrast , concomitant deletion of Foxa1 and Foxa2 led to the emergence of small neoplastic lesions ( Figure 1A , right column ) in the alveoli that were completely devoid of the glandular architecture and mucin production that characterizes NKX2-1-deficient tumors . Absence of mucin production was apparent by H and E staining and further demonstrated by Alcian Blue staining ( Figure 1B ) and IHC for Muc5AC ( Figure 1—figure supplement 1C ) . Given the dramatic change in the morphology of lung neoplasia lacking NKX2-1 , FoxA1 and FoxA2 , we used IHC to assess the differentiation state of these lesions . Cytokeratin 8 ( CK8 ) was expressed in lesions arising in mice of all genotypes ( Figure 1—figure supplement 1C ) , showing that cells lacking all three transcription factors retained an epithelial identity and did not undergo a complete epithelial to mesenchymal transition . HNF4α and PDX1 are transcription factors that regulate gastrointestinal differentiation and are expressed in human invasive mucinous adenocarcinoma and mouse models of this disease ( Snyder et al . , 2013; Skoulidis et al . , 2015 ) . Both transcription factors , as well as the HNF4α target PK-L , were undetectable in FoxA1/2-deficient neoplasia ( Figure 1C ) . Additional markers of gastrointestinal differentiation , including Gastrokine 1 , Cathepsin E and Galectin 4 , were also not expressed in these lesions ( Figure 1—figure supplement 1C ) . All these markers were retained in lesions lacking either FoxA1 or FoxA2 alone ( Figure 1C and Figure 1—figure supplement 1C ) . Taken together , these data show that FoxA1 and FoxA2 are required for mucin production and key elements of the gastrointestinal differentiation program in NKX2-1-negative lung tumors in a functionally redundant manner . Most lesions in KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice exhibited complete loss of FoxA1/2 expression when analyzed at 11 weeks post-infection . However , these mice also harbored a variable but substantial quantity of tumors ( ‘incomplete recombinants’ ) that retained FoxA1 or FoxA2 as well as targets such as HNF4α ( Figure 2—figure supplement 1A–B ) . Since incomplete recombinants were often larger than the lesions lacking NKX2-1 , FoxA1 and FoxA2 ( i . e . ‘complete recombinants’ ) ( Figure 2—figure supplement 1B ) , we speculated that they might have gradually outgrown the complete recombinants over time . Consistent with this possibility , we found that 5 weeks after tumor initiation , incomplete recombinants comprised a much smaller proportion of overall tumor burden than at 11 weeks ( Figure 2—figure supplement 1A–B ) . Based on these data , we chose the 5-week timepoint to quantitate tumor burden and proliferation rates among the different genotypes . We found that concomitant deletion of both Foxa1 and Foxa2 led to an approximately 10-fold reduction in tumor burden when measured at 5 weeks post-initiation ( Figure 2A ) . This was accompanied by reduced lesion size and , to a lesser extent , fewer lesions/mm2 ( Figure 2—figure supplement 1C–D ) . In contrast , deletion of either Foxa1 or Foxa2 alone had little to no effect on tumor burden . To determine why loss of FoxA1/2 activity caused such a severe inhibition of tumorigenesis , we analyzed proliferation and apoptosis in tumors of each genotype . BrdU incorporation was reduced by ~50% in FoxA1/2-deficient lesions in comparison with control lesions ( Figure 2B–C ) . IHC for the proliferation markers MCM2 and KI67 also demonstrated that FoxA1/2-deficient lesions proliferate at a significantly lower rate than controls ( Figure 2—figure supplement 1E–G ) . In contrast , the apoptotic rate of FoxA1/2-deficient lesions was no different than controls as measured by IHC for cleaved caspase-3 ( Figure 2—figure supplement 1H ) . In addition to these short-term measurements , we assessed the long-term impact of Foxa1/2 deletion in a survival analysis ( Figure 2D ) . Mice in the three control groups survived for a similar duration after tumor initiation ( median survival 143–160 days ) . In contrast , deletion of both Foxa1 and Foxa2 led to a dramatic increase in survival ( median survival 293 days ) . Histopathologic analysis showed that approximately 80% of the tumor burden in KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice consisted of mucinous HNF4α-positive adenocarcinomas ( Figure 2—figure supplement 1A ) . This suggests that these mice ultimately succumbed to growth of incomplete recombinants and that the complete recombinants likely had little impact on overall survival . We also noted extensive extracellular mucin secretion in the tumors of KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F mice ( Figure 2—figure supplement 1I ) . This phenomenon was rarely observed in tumors from other control groups , which predominantly produced intracellular mucin , suggesting that FoxA1 and FoxA2 likely have some specific functions in the regulation of the differentiation state of NKX2-1-negative adenocarcinoma . Taken together , these data show that lack of FoxA1/2 activity at tumor initiation severely impairs the proliferation and long-term growth potential of NKX2-1-negative lung adenocarcinoma . We next sought to analyze the changes in gene expression induced by deletion of Foxa1 and Foxa2 in NKX2-1-deficient tumors . Our mice harbor a Cre-dependent tdTomato reporter allele ( Madisen et al . , 2010 ) that enables tumor cell isolation by fluorescence-activated cell sorting ( FACS ) . For sorting experiments , we initiated tumors with the Ad5-SPC-Cre adenovirus ( Sutherland et al . , 2011 ) , which restricts Cre activity to SPC-positive lung epithelial cells , obviating the need to exclude stromal cells from the sorted population . ( SPC-Cre induces lesions identical to lentiviral-driven Cre ( Figure 6 and data not shown ) ) . However , we lacked a cell surface marker that would enable us to differentially isolate complete from incomplete recombinants in KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice during sorting . Single-cell RNA-Seq can be used to deconvolute gene expression profiles of mixed cell populations from the murine lung bioinformatically and thereby assign an identity to each cell ( Treutlein et al . , 2014 ) . We therefore proceeded with single-cell RNA-Seq analysis on FACS-sorted lung tumor cells via the Fluidigm C1 Autoprep microfluidic system . We sorted tumor cells from one KrasLSL-G12D/+ mouse , one KrasLSL-G12D/+; Nkx2-1F/F mouse , and two KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice . After Illumina sequencing and transcript quantitation , we used the SC3 clustering package ( Kiselev et al . , 2017 ) for quality control , filtering and clustering . A total of 134 cells were considered to be of sufficient quality for further analysis ( Supplementary files 1–2 ) , which yielded three distinct clusters ( tSNE plot , Figure 3A ) . Cluster 1 ( C1 , n = 62 cells ) contained cells from mice of all three genotypes . Using the SC3 package , we identified marker genes for this cluster ( defined as 'genes that are highly expressed in only one of the clusters and are able to distinguish one cluster from all the remaining ones’ ) . These included canonical NKX2-1 target genes Sftpa1 and Sftpb ( Supplementary file 3 ) . From these data , we infer that C1 represents tumor cells that are phenotypically NKX2-1-positive . In contrast , cluster 2 ( C2 , n = 31 cells ) only contained cells from KrasLSL-G12D/+; Nkx2-1F/F and KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice . Numerous gastrointestinal transcripts were identified as marker genes for this cluster , including Hnf4a , Gkn1 , Lgals4 and Ctse . Thus , C2 appears to include incomplete recombinants from KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice that express sufficient levels of FoxA1 and/or FoxA2 to maintain a gastric differentiation state . In contrast , cluster 3 ( C3 , n = 41 cells ) contained only cells from KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice and expressed marker genes not characteristic of either a pulmonary or gastric differentiation state , suggesting that C3 likely contains cells completely deficient for NKX2-1 , FoxA1 and FoxA2 ( i . e . complete recombinants ) . Several different analyses further validated our classification of C1 and C2 as NKX2-1-positive and NKX2-1-negative cells , respectively . First , we identified differentially expressed genes between C1 and C2 using an independent software package ( SCDE ) and found that many pulmonary and gastric transcripts were differentially expressed between the two clusters ( Supplementary file 3 ) . We then performed RNA-Seq on sorted bulk tumor cells from KrasLSL-G12D/+ and KrasLSL-G12D/+; Nkx2-1F/F mice ( n = 3 each , Supplementary file 4 ) and found a strong correlation ( Pearson correlation coefficient = 0 . 62 ) between the differentially expressed genes identified in single cell and bulk analyses ( Figure 3—figure supplement 1A ) . We also compared our single-cell datasets with published data from other groups . We found that normal murine type two pneumocytes ( Treutlein et al . , 2014 ) , which are the presumed cell of origin for NKX2-1-positive tumor cells , clustered with presumptive NKX2-1-positive C1 cells ( Figure 3—figure supplement 1B ) . We also used principal component analysis ( PCA ) to compare our single-cell data with a gene signature of human IMA ( Guo et al . , 2017 ) . In this analysis , the IMA signature caused C2 cells to cluster separately from the other cells ( Figure 3—figure supplement 1C ) . This shows that C2 cells are more similar to IMA than C1 and C3 , as would be expected if they represent the NKX2-1-negative tumor cell population . To characterize the identity of our tumor cells in a global manner , we compared our single-cell RNA-Seq data with total RNA-Seq data from a panel of mouse tissues ( Supplementary file 5 ) . The 50 genes in each tissue with the highest expression compared to the other tissues in the panel were identified . Expression data for this set of genes was extracted from the single cell and tissue datasets and evaluated using two approaches: tSNE ( Figure 3B ) and hierarchical clustering on principal components ( HCPC , Figure 3—figure supplement 1D ) , which combines PCA , hierarchical clustering and k-means clustering . In both approaches , we found that C1 was most similar to normal lung , and that C2 was most similar to glandular stomach . C3 cells clustered near the upper GI tract , in particular the forestomach and esophagus . However , cosine similarity analysis ( Supplementary file 6 ) showed that C3 cells are not as closely related to esophagus/forestomach as C1 and C2 are to lung and glandular stomach , respectively . This bioinformatic analysis is in consonance with microscopic evaluation of complete recombinants , which lack morphological features of a multi-layered , keratinizing squamous epithelium ( Figure 1 ) that is found in the normal esophagus and forestomach . Moreover , the vast majority of complete recombinants cells do not express ΔNp63 , a master regulator of squamous differentiation , or the squamous marker cytokeratin 5 ( CK5 ) ( Figure 3—figure supplement 1E ) . Thus , complete ablation of NKX2-1 , FoxA1 and FoxA2 causes lung tumor cells to adopt an identity that is neither pulmonary nor gastric , but also is not fully squamous . Indeed , it appears that the exact differentiation state adopted by these cells is not well represented in the panel of tissues evaluated . Recent studies have described a small but discrete transitional zone at the squamocolumnar junction ( SCJ ) of the gastrointestinal ( GI ) tract , just proximal to the glandular stomach , which is not included as a discrete entity in our tissue panel . This transitional zone consists of a bilayered epithelium expressing high levels of cytokeratin 7 ( CK7 ) ( Jiang et al . , 2017; Wang et al . , 2011 ) , including a ΔNp63/CK5-positive basal layer and a ΔNp63/CK5-negative luminal layer . Intriguingly , complete recombinants in KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice have uniformly high levels of CK7 protein that are comparable to the SCJ ( Figure 3C ) . Manual inspection of genes specifically expressed in C3 vs . both C1 and C2 ( using both SC3 and SCDE ) revealed that several of these genes are expressed at high levels at the SCJ of the GI tract and/or the cervix . These genes include Chil4 ( Nio et al . , 2004 ) , Gda and Mmp7 ( Herfs et al . , 2012 ) , and Vcam1 ( Figure 3C ) . Other C3-specific genes are expressed at higher levels throughout the forestomach and esophagus than glandular stomach , including Cav1 , Cdh13 , Hilpda , Fbln2 and Rbp7 ( Uhlén et al . , 2015 ) . Using IHC , we found that protein levels of several these genes are much higher in complete recombinants than in NKX2-1-negative lesions ( Figure 3C and Figure 3—figure supplement 1E ) . These data led us to evaluate FoxA1/2 levels at the SCJ of the murine GI tract ( Figure 3—figure supplement 1F ) . Both FoxA1 and FoxA2 are expressed in the glandular stomach . Interestingly , FoxA2 expression ends at the SCJ and is absent throughout the squamous forestomach and esophagus . In contrast , FoxA1 levels are very low but detectable at the SCJ then increase in the proximal forestomach and esophagus . Thus , overall FoxA1/2 levels appear to reach their nadir at the SCJ and distal forestomach of the normal murine GI tract . Taken together , our data show that FoxA1/2 are required for NKX2-1-deficient lung tumor cells to adopt a gastric identity . Moreover , concomitant loss of NKX2-1 and FoxA1/2 activity at tumor initiation leads to a distinct differentiation state characterized by expression of multiple markers of the transitional epithelium normally found at the SCJ of the GI tract . Although KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F and KrasLSL-G12D/+; Nkx2-1F/F; Foxa2F/F mice exhibited minimal obvious phenotypes at early timepoints ( Figures 1–2 ) , we found that a subset of these mice developed macroscopic adenosquamous carcinomas ( AdSCCs ) at 20 weeks post-initiation ( Figure 4 and Figure 4—figure supplement 1A ) . In contrast , we did not find AdSCCs in any of the KrasLSL-G12D/+; Nkx2-1F/F mice aged to 20 weeks post-initiation . Human AdSCC is an uncommon but aggressive lung cancer subtype that contains a mix of clonally related adenocarcinoma and squamous cell components ( Shu et al . , 2013; Tochigi et al . , 2011 ) . In our mice , AdSCCs consisted of a mucinous adenocarcinoma component that was continuous with , and typically circumscribed , a well-differentiated , keratinizing squamous cell carcinoma component ( Figure 4A–C ) . Both components were tdTomato-positive , indicating that these tumors had arisen through Cre-mediated recombination ( Figure 4B ) . Although both components were NKX2-1-negative , markers of gastric differentiation were restricted to the adenocarcinoma component , and markers of squamous differentiation ( including ΔNp63 and cytokeratins 5 and 14 ( CK5 and CK14 ) , but not SOX2 ) were selectively expressed in the SCC component ( Figure 4B and Figure 4—figure supplement 1C ) . Given that genetic deletion of Foxa1 and Foxa2 at initiation completely suppressed mucinous gastric differentiation , we evaluated expression of both transcription factors in AdSCCs . In KrasLSL-G12D/+; Nkx2-1F/F; Foxa2F/F mice , we found that FoxA2 was absent in both components , whereas FoxA1 was expressed in the adenocarcinoma components but absent in the SCC ( Figure 4B ) . AdSCC in the KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F mouse exhibited the opposite pattern , that is , FoxA1 loss in both components and FoxA2 expression only in the adenocarcinoma component ( Figure 4—figure supplement 1B ) . Thus , the SCC component is always associated with stochastic loss of FoxA1/2 expression . Given that KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice contain incomplete recombinants that retain either FoxA1 or FoxA2 , we also observed AdSCCs in a subset of these mice at 20 weeks ( Figure 4—figure supplement 1A ) . As expected , AdSCCs in KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice always expressed either FoxA1 or FoxA2 in the adenocarcinoma component and stochastic loss of the other paralogue in the squamous component . These data suggest that when only one FoxA paralogue is expressed in mucinous lung adenocarcinoma , stochastic loss of the other FoxA paralogue can occur as the tumors progress . This stochastic loss of FoxA activity is associated with a profound change in differentiation state , with FoxA1/2-negative cells upregulating a keratinizing squamous differentiation program . This is in sharp contrast to the differentiation state of tumor cells in which FoxA1/2 loss was engineered at the time of tumor initiation ( Figure 3 ) , which led to an SCJ-like phenotype . These results raise the possibility that the genetic and/or epigenetic context in which FoxA1/2 activity is lost may have a significant influence on the cellular identity adopted by lung tumor cells . We next analyzed FoxA1/2 expression by IHC in human AdSCC ( n = 12 ) to determine whether these transcription factors are differentially expressed between adenocarcinoma and squamous components ( Figure 4D–E ) . FoxA1 and FoxA2 were expressed in the adenocarcinoma component of all cases . In half of the cases , FoxA1 and FoxA2 were both downregulated in the squamous component ( n = 5 cases with complete loss of expression and n = 1 case with detectable but diminished expression ) . In the other half , either FoxA1 ( n = 5 ) or FoxA2 ( n = 1 ) exhibited downregulation in the squamous component . Thus , half of the human AdSCC examined exhibit the same pattern of FoxA1/2 downregulation that we observe in our mouse model . Moreover , all cases exhibit at least partial reduction in expression of FoxA1 or FoxA2 in the squamous component . Taken together , these data suggest that reduced FoxA activity is commonly associated with adenosquamous transdifferentiation in human lung cancer . To test the hypothesis that loss of FoxA1/2 activity might promote squamous differentiation only in specific contexts , we generated KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice as well as controls wild type for either Foxa1 or both Foxa1 and Foxa2 . In these mice , delivery of the FlpO recombinase ( via Ad5CMV-FlpO adenovirus ) to the lung epithelium activates transcription of the KrasG12D oncogene from its endogenous locus ( Young et al . , 2011 ) and transcription of CreERT2 from the Rosa26 locus ( Schönhuber et al . , 2014 ) . Tamoxifen is then used to activate the CreERT2 protein and drive recombination of lineage specifiers in KRASG12D-expressing lung neoplasia . To determine whether loss of NKX2-1 , FoxA1 and FoxA2 in established neoplasia was sufficient to induce full squamous differentiation , we administered tamoxifen 1 week after tumor initiation with Ad5CMV-FlpO , then analyzed tumors 4 weeks later ( outline in Figure 5A ) . Histopathologic analysis of controls showed that the lungs contained mucinous adenocarcinoma that expressed HNF4α and the expected pattern of FoxA1/2 ( Figure 5B and Figure 5—figure supplement 1A ) . Almost all lesions in Nkx2-1F/F and Nkx2-1F/F; Foxa2F/F mice were ΔNp63-negative ( Figure 5B and Figure 5—figure supplement 1C ) . Indeed , only one mouse in each control group exhibited a single lesion with ΔNp63-positive cells . In contrast , all KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice harbored numerous non-mucinous lesions lacking FoxA1/2 and HNF4α . Most of these lesions were morphologically similar to the SCJ-like lesions generated with Cre-mediated recombination at tumor initiation ( Figure 1 ) . Strikingly , four out of eight KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice harbored well-differentiated squamous cell carcinomas ( SCCs ) characterized by a stratified squamous epithelium with extensive keratinization ( Figure 5B and Figure 5—figure supplement 1B ) . In contrast to the AdSCCs that arise stochastically from mucinous adenocarcinomas ( Figure 4 ) , these SCCs appeared to be discrete lesions and were not surrounded by HNF4α-positive mucinous adenocarcinoma ( Figure 5B ) . As expected , all SCCs in this model expressed ΔNp63 ( Figure 5B ) . Interestingly , we even detected ΔNp63 in a significant minority of non-keratinizing lesions in these mice ( Figure 5B ) , which contrasts with the lack of ΔNp63 expression in the complete recombinants of KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice ( Figure 3—figure supplement 1E ) . Most of the microscopic analysis of KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice ( Figures 1–4 ) was performed on lesions generated with lentivirus expressing Cre under the control of the Pgk promoter . To control for the possibility that the use of adenovirus and/or the CMV promoter might have played a role in the phenotypes observed with sequential recombination , we infected KrasLSL-G12D/+; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice ( n = 6 ) with Ad5CMV-Cre and harvested tumors 5 weeks after infection . Importantly , none of the mice harbored SCCs , despite the presence of multiple complete recombinants ( Figure 5—figure supplement 1C ) . Interestingly , ΔNp63 expression was slightly higher in these lesions than in lesions from mice of the same genotype infected with Pgk-Cre lentivirus ( data not shown ) . Taken together , these data identify a specific context in which loss of FoxA1/2 activity is sufficient to induce full squamous differentiation in the lung . Since FoxA1/2 loss was induced only 1 week after KRASG12D expression in this experiment , it seems likely that enhanced competence for squamous differentiation is a direct result of KRASG12D expression rather than stochastic genetic alterations accruing over time . Moreover , the fact that only a subset of neoplastic lesions are keratinizing SCC raises the possibility that a specific subpopulation of lung epithelial cells may exhibit enhanced competence for squamous differentiation in this system . To define more precisely the cell type from which SCCs arise in the sequential recombination model , we generated an adenovirus in which expression of the FlpO recombinase is driven by the murine SPC promoter . This promoter has been extensively validated to drive Cre expression primarily in type 2 pneumocytes of the alveoli ( Sutherland et al . , 2011 ) . To validate this promoter in our sequential recombination system , we generated KrasFSF-G12D/+; RosaFSF-CreERT2 harboring a CAG-LSL-HA-UPRT transgene ( Gay et al . , 2013 ) , in which the HA-tagged UPRT enzyme is only expressed after Cre-based recombination of the STOP cassette . These mice were infected with Ad5-SPC-FlpO or Ad5-CMV-FlpO , treated with tamoxifen 1-week post-infection , and subjected to histopathologic analysis 3 weeks post-infection . Whereas recombination was readily detectable in the bronchioles and alveoli of Ad5-CMV-FlpO infected mice , recombination was restricted to the alveoli of Ad5-SPC-FlpO infected mice ( Figure 6—figure supplement 1A ) . Next , we infected a cohort of KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice , along with KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F controls with Ad5-SPC-FlpO . As an additional control , we infected a group of KrasLSL-G12D/+; RosaLSL-tdTomato; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice with Ad5-SPC-Cre . All mice were treated with tamoxifen 1 week after infection and analyzed 5 weeks after infection . As expected , the lungs of KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F controls harbored numerous mucinous lesions in the alveoli that expressed FoxA1/2 and lacked squamous markers such as ΔNp63 and CK5 ( Figure 6 , left panel ) . KrasFSF-G12D/+; RosaFSF-CreERT2; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice harbored FoxA1/2-negative lesions of two distinct morphologies ( Figure 6 , central panels ) . All mice harbored SCJ-like lesions that were predominantly CK7-positive/CK5-negative and expressed ΔNp63 in a minority of cells . Moreover , 63% of these mice ( n = 5 out of 8 ) harbored well-differentiated , keratinizing SCCs that were CK7-negative/CK5-positive and expressed ΔNp63 ( Figure 6 , central panels and Figure 6—figure supplement 1B ) . Overall , these phenotypes were very similar to those observed when tumors were initiated with Ad5-CMV-FlpO in these mice ( Figure 5 ) . Importantly , we did not identify SCC in any of the KrasLSL-G12D/+; RosaLSL-tdTomato; Nkx2-1F/F; Foxa1F/F; Foxa2F/F mice infected with Ad5-SPC-Cre . These mice harbored CK7-positive/CK5-negative SCJ-like lesions that were essentially identical to lesions initiated by lentivirus in previous experiments ( Figures 1–3 ) . Taken together , these data show that loss of NKX2-1 , FoxA1 , and FoxA2 in SPC-positive alveolar cells has distinct outcomes depending on the state of oncogenic signaling in these cells . When these lineage specifiers are lost in normal SPC-positive cells ( concomitant with KRASG12D activation ) , the resulting neoplasia equilibrates to a uniform SCJ-like state marked by a CK7-positive/CK5-negative immunophenotype . In contrast , SPC-positive cells that have already been subjected to oncogenic signaling from KRASG12D for ~1 week have the potential to undergo full squamous transdifferentiation ( CK7-negative/CK5-positive ) and become well-differentiated keratinizing SCCs .
Lung adenocarcinomas can adopt a variety of differentiation states , and changes in cellular identity can have both prognostic and therapeutic implications for patients with this disease . We have previously shown that engineered loss of the pulmonary lineage specifier NKX2-1 causes lung adenocarcinoma cells to shed their pulmonary identity and adopt a gastric differentiation state that is also observed in human IMA ( Snyder et al . , 2013 ) . Here , we show that FoxA1 and FoxA2 are required for lung adenocarcinomas to adopt a mucinous , gastric differentiation state in the absence of NKX2-1 . Although FoxA1/2 can regulate lung adenocarcinoma biology individually in some contexts ( Li et al . , 2015 ) , their functional redundancy in IMA is consistent with their frequently redundant role in endodermal tissue specification ( reviewed in Golson and Kaestner , 2016 ) . The precise mechanisms by which FoxA1/2 specifically activate a gastric program in NKX2-1 negative lung cancer , as opposed to other potential endodermal differentiation states ( e . g . hepatic , pancreatic , lower GI tract etc . ) , remain to be determined . However , it appears likely that FoxA1/2 regulate gastrointestinal differentiation programs in other types of cancer . For example , pancreatic ductal adenocarcinoma ( PDAC ) and its precursors often express many of the same foregut markers as NKX2-1-negative lung adenocarcinoma ( Tata et al . , 2018; Bailey et al . , 2016; Prasad et al . , 2005 ) . FoxA1/2 levels are much higher in the subset of PDAC expressing a foregut differentiation program than in those tumors with a more mesenchymal/squamous differentiation state ( Bailey et al . , 2016 ) . In addition , aberrant activation of a gastrointestinal differentiation program in prostate cancer , which can mediate castration resistance , is driven by HNF4γ in cooperation with FoxA1 ( Shukla et al . , 2017 ) . Interestingly , the precise consequences of FoxA1/2 loss in lung cancer are highly dependent on the specific context in which it occurs ( model , Figure 7 ) . When Nkx2-1 , Foxa1 and Foxa2 are deleted at tumor initiation , the resulting lung lesions lacked evidence of either pulmonary or gastric differentiation ( Figure 3 ) . Instead , complete recombinants expressed several genes enriched at the SCJ of the GI tract , which contains a small but distinct non-keratinized transitional columnar epithelium marked by high levels of CK7 ( Jiang et al . , 2017 ) . The transitional epithelium consists of a ΔNp63-positive basal progenitor layer , which can give rise to Barrett’s metaplasia , and a differentiated ΔNp63-negative luminal layer . Thus , in the absence of FoxA1/2 activity , Nkx2-1 deletion causes normal lung epithelial cells to adopt a cell fate resembling the transitional epithelium that localizes immediately proximal to the glandular stomach . Given the lack of ΔNp63 expression in complete recombinants , we speculate that these tumor cells more closely resemble the luminal cells of the transitional epithelium , which may account in part for their limited proliferative capacity . In contrast , both stochastic ( Figure 4 ) and engineered ( Figures 5–6 ) loss of FoxA1/2 in established KRASG12D-driven lesions initiated in SPC-positive cells was accompanied by activation of a robust squamous differentiation program , as evidenced by a stratified multi-layered epithelium with extensive keratinization and expression of ΔNp63 and CK5 . These data suggest that signaling from KRASG12D enhances the capacity of SPC-positive cells to activate a squamous differentiation program in the absence of NKX2-1 and FoxA1/2 . Additional studies will be needed to determine the mechanism ( s ) that account for this enhanced propensity for squamous differentiation . ΔNp63 is an activator of squamous differentiation and is generally thought to function as an oncogene in SCC ( Watanabe et al . , 2014 ) , so the increased levels of ΔNp63 when Nkx2-1;Foxa1/2 deletion occurs in established lesions ( Figures 5–6 ) vs . at tumor initiation ( Figure 3 ) are likely to be one major factor that dictates whether cells adopt an SCJ-like vs . SCC fate . We speculate that signaling from KRASG12D either alters the epigenetic state of elements regulating ΔNp63 directly , or influences the activity of its numerous upstream regulators ( Yoh and Prywes , 2015 ) . It is also unclear why SPC-positive cells adopt two distinct fates in our sequential mutagenesis experiments ( SCJ-like vs SCC , Figure 6 ) . Intrinsic heterogeneity of the SPC-positive population ( prior to KRASG12D expression ) could account for this observation ( Kim et al . , 2005; Nabhan et al . , 2018; Zacharias et al . , 2018 ) . Alternatively , heterogeneous response to KRASG12D signaling could also play a role . Proliferation rate can influence changes in cell identity ( Soufi and Dalton , 2016 ) , and it is possible that only a subset of SPC-positive cells are actively cycling one week after KRASG12D expression . Regardless , the fact that SPC-positive cells readily give rise to SCC contrasts with other investigations of cell type specificity in mouse models of SCC . In KrasG12D; Lkb1 conditional mice CC10-positive lung epithelial cells are the predominant cell of origin for adenosquamous carcinomas , whereas SPC-positive cells mainly give rise to adenocarcinomas ( Nagaraj et al . , 2017; Zhang et al . , 2017 ) . In other murine models driven by SOX2 expression and either deletion of Pten and Cdkn2ab ( Ferone et al . , 2016 ) or Nkx2-1 ( 7 ) , both cell types can give rise to SCCs , although CC10-positive cells appeared to do so more efficiently . Context also appears to be critical for the effect of FoxA1/2 loss on tumor growth . We have previously shown that Nkx2-1 deletion augments KRASG12D-driven tumorigenesis ( Snyder et al . , 2013 ) . Concomitant Foxa1/2 deletion at initiation reverses this phenotype ( Figure 2 ) , showing that when FoxA1/2 are absent at tumor initiation , NKX2-1-negative lesions equilibrate to a low proliferation state that never progresses to macroscopic disease . However , the stochastic emergence of macroscopic FoxA1/2-negative AdSCCs ( Figure 4 ) argues that there is a selective advantage to loss of FoxA1/2 in some established lung neoplasia . This is further reinforced by the observation that a subset of human AdSCCs downregulate FoxA1/2 in their squamous component ( Figure 4 ) . An apparently dichotomous and context-specific contribution of FoxA1/2 to malignant potential has been observed in tumors from other tissues ( reviewed in Golson and Kaestner , 2016 ) . For example , one study of human lung SCC reported that 43% of cases lacked FoxA1 expression by IHC , and that FoxA1 positivity was significantly correlated with unfavorable survival ( Deutsch et al . , 2012 ) . In PDAC , FoxA1 can promote metastasis ( Roe et al . , 2017 ) , despite the fact low levels of FoxA1/2 ( as well as other lineage specifiers associated with endodermal differentiation ) are found in the subtype of pancreatic ductal adenocarcinoma that confers the worst prognosis ( Bailey et al . , 2016 ) . A comprehensive evaluation of FoxA1/2 loss at distinct stages of tumorigenesis will be needed to delineate fully its context-specific role in lung tumor growth . In summary , this work expands our understanding of the lineage specifiers that coordinately regulate the growth and identity of lung adenocarcinoma . We show that FoxA1 and FoxA2 regulate the growth and gastric identity of NKX2-1-negative lung adenocarcinoma . In the absence of FoxA1/2 activity , NKX2-1-negative tumor cells adopt a more proximal cell fate with features of either the transitional epithelium of the SCJ or the squamous epithelium of the forestomach/esophagus , depending on the context of FoxA1/2 loss . Squamous transdifferentiation has been linked to drug resistance in human lung adenocarcinomas ( Hou et al . , 2017 ) , and it will be interesting to determine whether FoxA1/2 downregulation plays a role in this process . More broadly , our results show that the effects of lineage specifier inactivation in cancer can be highly context-dependent , and provide an experimental system for future work to elucidate the mechanistic basis for this specificity .
Mice harboring KrasLSL-G12D ( Jackson et al . , 2001 ) , KrasFSF-G12D ( Young et al . , 2011 ) , Rosa26LSL-tdTomato ( Madisen et al . , 2010 ) , Rosa26FSF-CreERT2 ( 30 ) , Nkx2-1F/F ( Kusakabe et al . , 2006 ) , Foxa1F/F ( Gao et al . , 2008 ) , Foxa2F/F ( Sund et al . , 2000 ) and CAG-LSL-HA-UPRT ( Gay et al . , 2013 ) alleles have been previously described . Rosa26LSL-tdTomato and CAG-LSL-HA-UPRT mice were obtained from the Jackson Laboratories ( Bar Harbor , Maine ) . All animals were maintained on a mixed C57BL/6J × 129SvJ background . Mice were infected intratracheally with adenovirus ( University of Iowa , Gene Transfer Vector Core ) or lentivirus as described ( DuPage et al . , 2009 ) . Animal studies were approved by the University of Utah IACUC , and conducted in compliance with the Animal Welfare Act Regulations and other federal statutes relating to animals and experiments involving animals and adhere to the principles set forth in the Guide for the Care and Use of Laboratory Animals , National Research Council ( PHS assurance registration number A-3031–01 ) . Tamoxifen ( Sigma , St . Louis , MO ) was dissolved in corn oil to a concentration of 20 mg/ml and administered at a dose of 120 mg/kg per day for 6 doses over 9 days . This was followed by ad libitum feeding with tamoxifen-supplemented chow ( 500 mg/ kg; Envigo , Indianopolis , IN ) in place of standard chow for the duration of experiment . Lentivirus was produced by transfection of 293 T cells with TransIT-293 ( Mirus Bio , Madison , WI ) , lentiviral backbone as well as packaging vectors Δ8 . 9 ( gag/pol ) and CMV-VSV-G ( DuPage et al . , 2009 ) . Supernatant was collected at 36 , 48 , 60 and 72 hr after transfection . For in vivo infection , virus was concentrated by ultracentrifugation at 25 , 000 r . p . m . for 105 min and re-suspended in an appropriate volume of 1X PBS . Cell line identity was authenticated using STR analysis at the University of Utah DNA Sequencing Core . Cells tested negative for mycoplasma . We first generated a pCDH-SPC-Flpo lentiviral vector by PCR amplifying the murine SPC promoter ( Sutherland et al . , 2011 ) and cloning into SpeI-XbaI sites of pCDH-CMV-Flpo plasmid . The pCDH-mSPC-Flpo vector was then digested with ClaI-PacI and blunt ended with Klenow to clone into EcoRV site of the adenovirus shuttle plasmid G0687 pacAd5mcsSV40pA ( University of Iowa , Viral Vector Core Facility ) . Correct identity and orientation of the construct was confirmed via Sanger sequencing . Further recombination and adenovirus production and purification was carried out by University of Iowa Viral Vector Core ( cat . # VVC-Snyder-6695 ) . All tissues were fixed in 10% formalin overnight , and lungs were perfused with formalin via the trachea . Tissues were transferred to 70% ethanol , embedded in paraffin , and four-micrometer sections were cut . To detect mucin , sections were stained with 1% Alcian Blue pH 2 . 5 at the HCI Research Histology Shared Resource . Immunohistochemistry ( IHC ) was performed manually on Sequenza slide staining racks ( ThermoFisher Scientific , Waltham , MA ) . Sections were treated with Bloxall ( Vector labs ) followed by Horse serum ( Vector Labs , Burlingame , CA ) or Rodent Block M ( Biocare Medical , Pacheco , CA ) , primary antibody , and HRP-polymer-conjugated secondary antibody ( anti-Rabbit , Goat and Rat from Vector Labs; anti-Mouse from Biocare . The slides were developed with Impact DAB or VIP ( Vector ) and counterstained with hematoxylin . Slides were stained with antibodies to BrdU ( BU1/75 , Abcam , Cambridge , MA ) , Cadherin 13 ( EPR9621 , Abcam ) , Cathepsin E ( LS-B523 , Lifespan Biosciences , Seattle , WA ) , Caveolin 1 ( EPR15554 , Abcam ) , CHIL3/4 ( EPR15263 , Abcam ) , Cleaved caspase-3 ( 5A13 , CST , Danvers , MA ) , Cytokeratin 5 ( EP1691Y , Abcam ) , Cytokeratin 7 ( EP17078 , Abcam ) , Cytokeratin-8 ( TROMA-I , DSHB , Iowa City , Iowa ) , Cytokeratin 14 ( EPR17350 , Abcam ) , FoxA1 ( EPR10881-14 , Abcam ) , FoxA2 ( EPR4466 , Abcam ) , Galectin 4 ( AF2128 , R and D Systems , Minneapolis , MN ) , Gastrokine 1 ( 2E5 , Abnova , Taipei City , Taiwan ) , GDA ( EPR18751 , Abcam ) , HNF4α ( C11F12 , CST ) , KI67 ( SP6 , Abcam ) , MCM2 ( ab31159 , Abcam ) , MMP7 ( AF2967 , R and D Systems ) Muc5AC ( SPM488 , Abnova ) , NKX2-1 ( EP1584Y , Abcam ) , p40 ( ΔNp63 ) ( BC28 , Biocare ) , PDX1 ( F109-D12 , DSHB ) , PIGR ( 7C1 , Abcam ) , PK-LR ( EPR11093P , Abcam ) , RFP ( Rockland Immunochemicals , Limerick , PA ) , SOX2 ( C70B1 , CST ) and VCAM1 ( EPR5047 , Abcam ) . Pictures were taken on a Nikon Eclipse Ni-U microscope with a DS-Ri2 camera and NIS-Elements software . For double immunostaining , slides were blocked sequentially with Bloxall , horse serum and Rodent Block M , then incubated with antibodies of interest from different species ( Rabbit and Mouse ) simultaneously . Slides were incubated with a mouse secondary followed by DAB ( brown ) . This was followed by incubation with a rabbit secondary antibody and ImPACT VIP ( purple , Vector lab ) . Tumor quantitation was performed on hematoxylin and eosin-stained or IHC-stained slides using NIS-Elements software . All histopathologic analysis was performed by a board-certified anatomic pathologist ( E . L . S . ) . 7–20 weeks after tumor initiation with Ad5-SPC-Cre ( Sutherland et al . , 2011 ) , tumor-bearing mice were euthanized using carbon dioxide and the rib-cage was dissected to reveal trachea and heart . Cadiac perfusion of the pulmonary vasculature was performed using PBS until the lungs turned pale . The lungs were inflated with an enzymatic digest solution ( Collagenase type I ( Thermo Fisher Scientific ) ; Elastase ( Worthington Biochemical , Lakewood , NJ ) , Dispase ( Corning CB-40235 , VWR , Radnor , PA ) and Dnase I ( DN25 , Sigma ) ) and then minced and digested with the enzyme digest solution at 37 C for 45 min . The digested tissue was then passed through an 18-gauge syringe needle followed by 100 , 70 and 40 micron filters to generate a single-cell suspension . The suspension was treated with Red Blood Cell Lysis Buffer ( eBioscience , ThermoFisher Scientific , ) and then reconstituted in 1X PBS supplemented with 2% fetal bovine serum , 2% BSA and DAPI ( Sigma ) . Cells were sorted using BD FACSAria for tdTomato-positive and DAPI-negative cells into PBS + 10% serum . Sorted tumor cells ( 200–300 cells/ul ) were mixed with C1 Suspension Reagent ( Fluidigm , South San Francisco , CA ) and loaded on a 5–10 μm C1 Single-cell Auto Prep IFC for mRNA Seq ( Fluidigm cat# 100–5760 ) . Captured cells were visualized and scored by microscopy . Amplified cDNA products derived from captured cells were harvested and concentrations were measured using the Qubit dsDNA HS Assay Kit . Amplified products were normalized to a concentration of 0 . 2 ng/ul and sequencing libraries were prepared using the Nextera XT DNA Library Preparation Kit ( cat# FC131-1096 , Illumina , San Diego , CA ) and dual indexed adapters ( FC-131–2001 , FC-131–2002 ) according to the modified protocol described by Fluidigm . Purified libraries were qualified on an Agilent Technologies 2200 TapeStation using a D1000 ScreenTape assay ( cat# 5067–5582 and 5067–5583 ) . The molarity of adapter-modified molecules was defined by quantitative PCR using the Kapa Biosystems ( Wilmington , MA ) Kapa Library Quant Kit ( cat# KK4824 ) . Individual libraries were normalized to 10 nM and equal volumes were pooled in preparation for Illumina sequence analysis . Sequencing libraries ( 25 pM ) were chemically denatured and applied to an Illumina HiSeq v4 single read flow cell using an Illumina cBot . Hybridized molecules were clonally amplified and annealed to sequencing primers with reagents from an Illumina HiSeq SR Cluster Kit v4-cBot ( GD-401–4001 ) . Following transfer of the flowcell to an Illumina HiSeq 2500 instrument ( HCSv2 . 2 . 38 and RTA v1 . 18 . 61 ) , a 50-cycle single-read sequence run was performed using HiSeq SBS Kit v4 sequencing reagents ( FC-401–4002 ) . RNA was isolated by trizol-chloroform extraction followed by column-based purification . Sorted cells were lysed in 1 ml Trizol ( ThermoFisher Scientific ) , followed by phenol-chloroform extraction . The aqueous phase was brought to a final concentration of 50% ethanol , and RNA was purified using the PureLink RNA Mini kit according to the manufacturer’s instructions ( ThermoFisher Scientific ) . Library preparation was performed using the TruSeq Stranded RNA kit with Ribo-Zero Gold ( Illumina ) . Libraries were sequenced on an Illumina HiSeq 2500 ( 50 cycle single-read sequencing ) . Mouse FASTA and GTF files were downloaded from Ensembl release 82 and a reference database was created using RSEM version 1 . 2 . 12 ( Li and Dewey , 2011 ) . RSEM and the Bowtie 1 . 0 . 1 aligner were used to map reads and estimate transcripts and gene counts using rsem-calculate-expression with the forward-prob 0 option for reversely stranded Illumina reads . The expected gene counts were filtered to remove 12371 features with zero counts and 10100 features with fewer than 10 reads in any sample . Differentially expressed genes were identified using a 5% false discovery rate with DESeq2 version 1 . 16 . 0 ( 59 ) . Formalin fixed , paraffin-embedded ( FFPE ) tumors were obtained in accordance with protocols approved by the Institutional Review Boards of the University of Utah and Intermountain Healthcare . Additional lung adenocarcinomas were evaluated on commercially available tissue microarrays ( US BioMax , Rockville , MD ) . The patient IDs and cluster names from 68 KRAS-mutants listed in supplementary figure 2A in Skoulidis et al . ( 2015 ) were saved to a sample table with 23 KL , 30 KP and 15 KC samples corresponding to genetic alterations in STK11/LKB1 ( KL ) , TP53 ( KP ) , and CDKN2A/B inactivation coupled with low expression of NKX2-1 ( KC ) . The patient IDs were matched to a count matrix from the TCGA Lung Adenocarcinoma project ( LUAD ) using the TCGAbiolinks package and HTSeq counts in the GDC harmonized dataset ( Colaprico et al . , 2016 ) . Eleven patients with a matched normal sample were also included as a fourth group for comparison . The count matrix was filtered to remove 5789 features with zero counts and 19 , 546 features with fewer than 10 reads in any sample . The sample table and filtered count matrix were loaded into DESeq2 version 1 . 16 . 0 ( 59 ) to estimate normalized counts and identify differentially expressed genes using a 5% false discovery rate . p-Values were calculated using the unpaired two-tailed Mann-Whitney ( non-parametric ) U test , Chi-squared test or Fisher’s Exact Test . RNA-Seq statistics are described above . | Among all cancers , lung cancers cause the most deaths worldwide . There are many different types of lung cancer , each of which contain lung cancer cells that look different . As a general rule , lung cancer cells that look the most like healthy lung cells are the least aggressive . Cancer cells that take on the appearance of other tissues in the body are more aggressive and often respond poorly to treatment . In one uncommon type of lung cancer called invasive mucinous adenocarcinoma ( IMA , for short ) , the cancer cells start to resemble the cells that line the inside of the stomach . For example , these lung cancer cells activate genes more typically active in stomach cells , and they start to make a lot of mucus . Previous studies with mice showed that losing a single protein called NKX2-1 can cause this switch from lung to stomach cell identity . However , it is not clear exactly how this switch happens and which other proteins are involved . Camolotto et al . have now addressed these issues by studying two DNA-binding proteins called FoxA1 and FoxA2 . There were two main reasons for choosing these specific proteins . First , they can physically interact with the NKX2-1 protein , so losing NKX2-1 affects how FoxA1 and FoxA2 interact with DNA . Second , the two proteins switch on many of the stomach-related genes that are also activated in IMA . Camolotto et al . activated a gene that commonly drives lung cancer and deleted the gene for NKX2-1 in the lungs of mice , mimicking IMA . As expected , these mice developed lung tumors that resembled stomach tissue . When the genes for FoxA1 and FoxA2 were deleted at the same time , the tumors stopped producing the mucus-related proteins . Further experiments showed that these cancer cells adopt a different cell identity also found in the digestive tract . Mice with tumors lacking both FoxA1 and FoxA2 survived for longer than those still containing these proteins . Lastly , when the genes for NKX2-1 , FoxA1 and FoxA2 were deleted later , in lung tumors that had already formed , the outcome was a more aggressive type of lung cancer that also occurs in human patients . These experiments demonstrate that losing FoxA1 and FoxA2 at different times affects what kind of lung tumor can grow . Future studies will need to examine how these different lung cancer types respond to therapy and whether lung cancer cells switch identities to evade therapy . This knowledge may eventually lead to new treatments for lung cancer patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2018 | FoxA1 and FoxA2 drive gastric differentiation and suppress squamous identity in NKX2-1-negative lung cancer |
Reliably detecting unexpected sounds is important for environmental awareness and survival . By selectively reducing responses to frequently , but not rarely , occurring sounds , auditory cortical neurons are thought to enhance the brain's ability to detect unexpected events through stimulus-specific adaptation ( SSA ) . The majority of neurons in the primary auditory cortex exhibit SSA , yet little is known about the underlying cortical circuits . We found that two types of cortical interneurons differentially amplify SSA in putative excitatory neurons . Parvalbumin-positive interneurons ( PVs ) amplify SSA by providing non-specific inhibition: optogenetic suppression of PVs led to an equal increase in responses to frequent and rare tones . In contrast , somatostatin-positive interneurons ( SOMs ) selectively reduce excitatory responses to frequent tones: suppression of SOMs led to an increase in responses to frequent , but not to rare tones . A mutually coupled excitatory-inhibitory network model accounts for distinct mechanisms by which cortical inhibitory neurons enhance the brain's sensitivity to unexpected sounds .
Across sensory modalities , cortical neurons exhibit adaptation , attenuating their responses to redundant stimuli ( Das and Gilbert , 1999; Ulanovsky et al . , 2003; Garcia-Lazaro et al . , 2007; Asari and Zador , 2009; Khatri et al . , 2009 ) . Adaptation to stimulus context is thought to increase efficiency of sensory coding under the constraints of limited resources ( Barlow , 1961 ) . Yet , the neuronal-circuit mechanisms that facilitate adaptation in the cortex remain poorly understood . In the primary auditory cortex ( A1 ) , the vast majority of neurons exhibit stimulus-specific adaptation ( SSA , Figure 1 ) . When presented with a sequence of two tones , one of which occurs frequently ( termed ‘standard’ ) and another rarely ( termed ‘deviant’ ) , the neuron's response to the standard tone becomes weaker , but the response to the deviant tone remains strong ( Ulanovsky et al . , 2003; Szymanski et al . , 2009; Farley et al . , 2010; Fishman and Steinschneider , 2012 ) . Whereas SSA has also been found in sub-cortical structures e . g . in the auditory midbrain ( Malmierca et al . , 2009; Zhao et al . , 2011; Thomas et al . , 2012 ) and the auditory thalamus ( Kraus et al . , 1994; Anderson et al . , 2009; Antunes et al . , 2010; Bauerle et al . , 2011 ) , it is weak in the lemniscal areas of the auditory pathway , which project to A1 , and stronger in those non-lemniscal areas which receive feedback from A1 ( Ulanovsky et al . , 2004 , Perez-Gonzalez et al . , 2005 , Duque et al . , 2012 ) . Therefore , cortical circuits are proposed to contribute to and amplify SSA in A1 ( Ulanovsky et al . , 2003 , Szymanski et al . , 2009 , Bauerle et al . , 2011 , Fishman and Steinschneider , 2012 , Escera and Malmierca , 2014 ) , through a combination of plastic modulation of thalamocortical inputs and intra-cortical inhibitory circuits , which would allow for selective suppression of neuronal responses to specific stimuli ( Nelken , 2014 ) . Our study tests whether and how inhibitory neurons contribute to cortical SSA . 10 . 7554/eLife . 09868 . 003Figure 1 . Nearly all recorded A1 neurons exhibit stimulus-specific adaptation . ( A ) Diagrams of oddball stimuli; oddball stimuli are composed of a 2 . 5-Hz train of 100-ms long sine-wave tone pips separated by 300 ms of silence ( gray and red dots ) . Each tone pip is at one of two frequencies , tone A or B . In oddball stimulus 1 , 10% of all pips are tone A and 90% of pips are tone B . In oddball stimulus 2 , the tone probabilities are reversed . The less frequent tone is referred to as the deviant tone ( red dots ) . The more frequent tone is referred to as the standard ( gray dots ) . ( B ) Left: diagram of recording . Electrode was lowered perpendicular to the brain surface . Virus was injected in A1 . Right: the frequencies of tones A and B ( dashed black and gray lines ) are selected based on the frequency response functions of neurons of interest . Mean firing rate ( FR ) of five co-tuned neurons ( colored lines ) recorded simultaneously in a single session in response to 65 dB tone pips at 50 frequencies logarithmically spaced from 1 to 80 kHz . FR is normalized to the peak response of each neuron . ( C ) A representative neuron exhibited suppressed responses to a tone presented as a standard ( gray raster and PSTH ) compared to the same tone presented as a deviant ( red raster and PSTH ) . Left: responses to tone A , presented as a deviant in oddball stimulus 1 , and a standard in oddball stimulus 2 . Right: responses to tone B . Shaded regions indicate standard ( gray ) and deviant ( red ) tones trials . Gray dashed lines indicate tone onset and offset times . ( D ) Population histogram of stimulus-specific adaptation ( SSA ) index exhibited by all neurons included in the analysis . Gray and white bars indicate neurons expressing significant and non-significant SSA , respectively . Spike count for response to deviant tones was significantly greater than for response to standard tones ( Wilcoxon rank sum test , one tail , p < 0 . 05 ) . The black marker indicates the population average SSA index . ( E ) Left: diagram of electrode spanning A1 . Right: representative peri-stimulus current source density ( CSD ) . Top: mean response to deviant tones . Bottom: mean response to standard tones . Gray dashed lines indicate tone onset and offset . Green dashed lines indicate the location of the granular layer . Negative CSD values ( blue ) indicate current sinks , while positive CSD values ( red ) indicate current sources . ( F ) Mean CSD collected from the thalamo-recipient layer , in response to standard ( gray ) and deviant ( red ) tones . Gray dashed lines indicate tone onset and offset . ( G ) Mean SSA index across sessions measured from thalamo-recipient granular layer CSD , infra- and supra-granular layer cortical CSD and mean neuronal spiking activity SSA index averaged over sessions . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 00310 . 7554/eLife . 09868 . 004Figure 1—figure supplement 1 . Local field potentials recorded in A1 exhibit SSA . ( A ) Representative peri-stimulus local field potentials ( LFPs ) across cortical layers . Top: mean response to deviant tones . Bottom: mean response to standard tones . Gray dashed lines indicate tone onset and offset . Green dashed lines indicate the margins between cortical layers . ( B ) Mean LFP collected from the thalamo-recipient granular layer , in response to standard ( gray ) and deviant ( red ) tones . Gray dashed lines indicate tone onset and offset . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 004 Auditory cortex , like other sensory cortices , contains morphologically and physiologically diverse inhibitory interneurons , which form dense interconnected networks with excitatory neurons ( DeFelipe , 2002; Douglas and Martin , 2004 ) . While different interneuron types have been hypothesized to carry out specialized complementary functions in sensory processing ( DeFelipe , 2002; Markram et al . , 2004; Isaacson and Scanziani , 2011; Kepecs and Fishell , 2014; Marlin et al . , 2015 ) , their function in driving changes in dynamic auditory processing has not been previously established . We hypothesized that the two most common types of interneurons in the cortex , parvalbumin- ( PVs ) and somatostatin-positive interneurons ( SOMs ) ( Xu et al . , 2010; Rudy et al . , 2011 ) , facilitate SSA in excitatory neurons of A1 in a complementary fashion . PVs , a subset of which receive direct thalamic inputs ( Staiger et al . , 1996 ) , may amplify SSA in excitatory neurons by providing a constant inhibitory drive; equally strong inhibitory drive would attenuate the weak response to standard tones relatively more than the strong response to deviant tones , leading to a greater differential between standard vs deviant tone spiking response . SOMs , which target distal dendrites of pyramidal cells ( McGarry et al . , 2010; Gentet et al . , 2012 ) , have excitatory synapses that exhibit facilitation upon repetitive stimulation ( Reyes et al . , 1998; Silberberg and Markram , 2007 ) . Therefore , inputs from SOMs may exert a stimulus-specific increase in suppression of excitatory neurons that is selective to the standard tone and does not generalize to the deviant tone . Alternatively , they may contribute to selective adaptation in excitatory neurons through differential post-synaptic integration . To tease apart the function of different inhibitory types in SSA , we tested whether optogenetic suppression of either PV or SOM interneurons during sound presentation reduced SSA in putative excitatory neurons in the auditory cortex ( Hamilton et al . , 2013; Pi et al . , 2013; Weible et al . , 2014 ) . We found that both types of interneurons contribute to SSA in the cortex , with PVs providing constant inhibition , and SOMs increasing their effect with repeated tones .
We recorded spiking activity of neurons as well as local field potentials ( LFPs ) in A1 in head-fixed mice under light isoflurane anesthesia . SSA was measured from the firing rate ( FR ) of neurons in response to tones presented as a series of ‘oddball’ stimuli . Each oddball stimulus consisted of a sequence of tone pips at one of two frequencies ( tones A and B ) . In each oddball stimulus , one tone was presented as the rare ( deviant ) tone , while the other was presented as the frequent ( standard ) tone ( A to B ratio of 90:10 or 10:90 , Figure 1A ) . A third stimulus was also presented ( equal stimulus ) , with tones A and B being presented equally often ( 50:50 ) . The frequencies of tone A and B were selected at 0 . 39 octave intervals , narrower than the typical tuning bandwidth of A1 neurons ( Hackett et al . , 2011; Guo et al . , 2012; Kanold et al . , 2014 ) , such that they activated the majority of recorded neurons on each session ( Figure 1B ) . As expected , for a representative neuron recorded in A1 , the mean FR in response to a tone was lower when the tone was presented as the standard than as the deviant ( Figure 1C ) , exhibiting SSA . To quantify the level of adaptation for each neuron , we computed the index of the change in FR to the same tone when it was presented as the deviant vs the standard ( SSA index ) . SSA index is 1 when adaptation is complete ( i . e . , no response to the standard , and significant response to the deviant ) , and 0 when there is no adaptation ( i . e . , the response to the standard and deviant is equal ) . Almost all neurons recorded in A1 exhibited significant SSA ( Figure 1D , standard tone-evoked FR significantly lower than the deviant tone-evoked FR in N = 138 out of 147 neurons , Wilcoxon rank sum test p < 0 . 05 ) . We first tested whether SSA is present in inputs from the thalamus . Current source density ( CSD ) analysis has been extensively used to quantify inputs from the thalamus ( Metherate and Cruikshank , 1999; Kaur et al . , 2005; Szymanski et al . , 2009; Happel et al . , 2014 ) . We used a linear probe to record LFPs using electrodes spaced 50 microns apart inserted perpendicularly to brain surface in the primary auditory cortex . The multi-electrode probe is 775-µm long , spanning layers 1–6 of mouse A1 . CSD is computed as the second spatial derivative of the LFPs across the depth of the cortex ( Figure 1E , Figure 1—figure supplement 1A , 20 sessions , 15 mice ) . Typically , in response to tones , CSD exhibits a negative basin , termed sink , within a short delay of tone onset , localized to electrodes in thalamo-recipient layer ( Figure 1F , Figure 1—figure supplement 1B ) ( Kaur et al . , 2005; Szymanski et al . , 2009 ) . The amplitude of current in the sink was taken as a measure of the combined strength of post-synaptic inputs onto layer 4 neurons , which should reflect the strength of the thalamic inputs to the cortex ( Metherate and Cruikshank , 1999; Kaur et al . , 2005; Szymanski et al . , 2009; Happel et al . , 2014 ) . We compared the amplitude of the CSD sink for each tone when presented as a deviant or standard , and computed their ratio ( Figure 1F ) . The sink amplitude was lower for the standard as compared to the deviant tones ( Figure 1F , G ) , suggesting that excitatory signals produced by thalamo-cortical inputs exhibit SSA , consistent with previous findings ( Szymanski et al . , 2009 ) . This finding supports the ‘adaptation in narrowly tuned inputs’ model , which postulates that SSA in broadly tuned neurons in A1 reflects adaptation in either thalamocortical inputs or at the stage of integration of thalamocortical inputs , specific to inputs tuned to the standard tone ( Mill et al . , 2011; Taaseh et al . , 2011; Nelken , 2014 ) . Importantly , across sessions , the SSA index of the granular layer CSD sinks was significantly lower than that of either the non-thalamo-recipient layers ( Δ = −28% , p-value from one-sided test after correction ( p1 ) = 6e−4 , z = −3 . 4 , Bonferroni corrected for two tests ( C = 2 ) ) or the SSA index of the mean spiking activity of A1 neurons ( Δ = 23% , p1 = 0 . 029 , z = −2 . 1 , C = 2 ) in each session ( N = 20 sessions in 15 mice , Figure 1G ) , suggesting that additional intra-cortical mechanisms may contribute to SSA in the cortex . We next tested whether cortical inhibitory interneurons may contribute to SSA . Since different inhibitory neuronal subtypes can differentially affect sensory responses of putative excitatory neurons ( Lee et al . , 2012; Wilson et al . , 2012; Cottam et al . , 2013 ) , we separately tested the role of PVs and SOMs . We used targeted viral delivery in the auditory cortex of mice to drive Archaerhodopsin ( Arch ) expression , which hyperpolarizes neurons when stimulated by light , in either PVs or SOMs ( Chow et al . , 2010 ) . A modified adeno-associated virus ( AAV ) encoding anti-sense code for Arch and a fluorescent reporter , under the FLEX cassette , was injected into PV-Cre or SOM-Cre mice ( Boyden et al . , 2005; Sohal et al . , 2009; Cardin et al . , 2010; Zhang et al . , 2010; Deisseroth , 2011 ) ( Figure 2A ) . 2 weeks following virus injection , Arch was expressed selectively in PVs or SOMs in auditory cortex at expected levels ( Kvitsiani et al . , 2013 ) ( Figure 2B , C PV-Cre: N = 250 neurons in 4 mice , specificity = 92 ± 1% , efficiency = 73 ± 5% . SOM-Cre: N = 149 neurons in 5 mice , specificity = 95 ± 2% , efficiency = 86 ± 5% ) . To activate Arch , a light guide was positioned to cast 180 mW/mm 532-nm light onto A1 surface , perpendiular to cortical layers . In vitro intracellular recordings from optically identified PVs or SOMs ( Figure 2—figure supplement 1 , Figure 2—figure supplement 2 ) demonstrate that light cast over the auditory cortex in vitro drives a strong suppressive current ( Figure 2D , Figure 2—figure supplements 1C , D , 2C , D ) and hyperpolarizes the membrane potential in these neurons ( Figure 2—figure supplements 1B , 2B ) . Assuming a 100-fold attenuation of light over 1 mm of brain tissue ( Aravanis et al . , 2007 ) , the estimated irradiance in the deepest cortical layer ( 1 . 8 mW/mm2 ) was strong enough to induce hyperpolarizing current in neurons in vitro ( Figure 2D ) . In vivo , in both PV-Cre and SOM-Cre mice , illuminating the auditory cortex suppressed spiking activity in a small subset of recorded neurons ( Figure 2E , F , left , putative inhibitory neurons ) and increased activity in a great majority of recorded neurons ( Figure 2E , F right , putative excitatory neurons ) . Shining light over A1 increased spontaneous neuronal activity in the majority of the recorded neurons in both PV-Cre mice ( N = 115 neurons , 102 increased , 0 decreased , in 10 mice ) ( Figure 2G ) and SOM-Cre mice ( N = 104 neurons , 61 increased , 3 decreased , in 9 mice ) ( Figure 2H ) . These measurements demonstrate that casting light over A1 selectively and effectively suppresses the activity of either PVs or SOMs . 10 . 7554/eLife . 09868 . 005Figure 2 . Cell type-specific optogenetic suppression of parvalbumin-positive and somatostatin-positive neurons . ( A ) Optogenetic methods diagram . Top: A1 was injected with AAV-FLEX-Arch-GFP . During experiments , an optic fiber was positioned to target A1 and neuronal activity was recorded using a multichannel silicon probe in A1 . Bottom: green light ( 532 nm ) suppresses PVs in PV-Cre mice or SOMs in SOM-Cre mice . ( B ) Transfection of interneurons with Archaerhodopsin ( Arch ) . Immunohistochemistry demonstrating co-expression of the Arch and an interneuron-type reporter in A1 . Top: PV-Cre mouse A1 . Red: anti-body stain for parvalbumin . Green: Arch-GFP . Merge; co-expression of Arch and parvalbumin . Bottom: SOM-Cre mouse A1 . Red: anti-body stain for somatostatin . Green: Arch-GFP . Merge; co-expression of Arch and somatostatin . Scale Bar = 25 µm . ( C ) Efficiency and specificity of transfection of interneurons with Arch . Bar Plots: efficiency ( Ef ) and specificity ( Sp ) of visual transfection of PVs ( top ) and SOMs ( bottom ) with Arch . Ef , percent of labeled interneurons expressing Arch . Sp , percent of Arch-expressing cells , which are also labeled interneurons . ( D ) Mean Arch-mediated outward current evoked in response to increasing photostimulation power , recorded in vitro by whole-cell patch recording in putative excitatory neurons from PV-Cre ( blue , N = 5 ) and Som-Cre ( orange , N = 5 ) mice . The gray dashed line indicates the level of irradiance expected in in vivo experiments at the deepest recording sites , in cortical layer 6 . ( E , F ) Tone responses of representative neurons , which are suppressed ( left ) or activated ( right ) by photostimulation , from PV-Cre ( E ) and SOM-Cre ( F ) mice . Raster plot of spike times ( bottom ) and PSTH ( top ) of a single neuron response to a 100-ms long tone ( gray dashed lines , shaded region ) on light-on ( overlapping 250-ms light pulse , green shading ) and light-off trials . Light-on trials: green . Light-off trials: black . ( G , H ) Modulation of spontaneous FR by interneuron photosuppression recorded in PV-Cre ( G ) and SOM-Cre ( H ) mice . Each neuron is represented by a circle that is filled for those with significantly increased ( green ) or decreased ( red ) FR or unfilled for those without significant modulation . Gray dashed line , identity line . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 00510 . 7554/eLife . 09868 . 006Figure 2—figure supplement 1 . Optogenetic control of PVs in mouse primary auditory cortex via photostimulation of Arch in acute slices . ( A ) Sustained high-frequency firing pattern typical of a PV-positive FS cell ( top ) in response to rectangular current injection ( bottom; 600 pA ) recorded in vitro via whole-cell patch clamp . Inset , epifluorescence ( i ) and corresponding IR-DIC image ( ii ) of the depicted cell . Scale bar , 20 µm . ( B ) Membrane hyperpolarization mediated by 532-nm light . ( C ) Outward current mediated by photoactivation of Arch . ( D ) Plot of light-induced outward current vs illuminance ( mW/mm2 ) . Error bars , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 00610 . 7554/eLife . 09868 . 007Figure 2—figure supplement 2 . Optogenetic control of SOMs in mouse primary auditory cortex via photostimulation of Arch in acute slices . ( A ) Adapting discharge pattern typical of a somatostatin-positive cell ( top ) in response to rectangular current injection ( bottom; 200 pA ) recorded in vitro via whole-cell patch clamp . Inset , endogenous GFP fluorescence of the recorded cell illustrating AAV9 . Arch . GFP expression ( i ) filled with Alexa 594 ( ii ) and imaged using a two-photon microscope . Scale bar , 20 µm . ( B ) Membrane hyperpolarization mediated by 532-nm light . ( C ) Outward current mediated by photoactivation of Arch . ( D ) Plot of light-induced outward current vs illuminance ( mW/mm2 ) . Error bars , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 007 To test the function of PVs and SOMs in SSA , their activity was suppressed during every fifth tone of the oddball stimulus by illuminating A1 ( Figure 3A ) . To directly test the effect of interneuron suppression , we computed the SSA index separately on light-on and light-off trials for neurons responsive to both tones A and B ( SSA was found in 63 out of 67 tone-responsive neurons in PV-Cre mice , 42 out of 43 tone-responsive neurons in SOM-Cre mice ) . Photosuppression of either PVs or SOMs affected the responses of neurons to the tones ( Figure 3B , C ) , resulting in a significant reduction in SSA index across the population ( Figure 3E , F , PV-Cre: Δ = −41% , p1 = 1e−12 , t ( 66 ) = 8 . 6 . SOM-Cre: Δ = −25% , p1 = 2e−6 , t ( 42 ) = 5 . 4 ) . Photo-manipulation-affected responses only to the tone during which it was presented , but not to subsequent tones ( Figure 3—figure supplement 1 ) . Additionally , photo-manipulation was limited to cortex since it did not affect thalamo-recipient layer CSD tone responses and SSA ( Figure 3—figure supplement 2 ) . In a control group of PV-Cre or SOM-Cre mice ( 6 mice ) , we injected a modified AAV , which encoded anti-sense fluorescent reporter alone under the FLEX cassette , and computed the effect of casting light on SSA ( SSA was found in 33 out of 37 tone-responsive neurons in control mice ) . In this control group , SSA was not affected by light ( Figure 3D , G , p > 0 . 05 , t ( 36 ) = −2 . 0 ) , confirming that Arch expression was required for the effect of the light . Therefore , the effects of interneurons are specific to intra-cortical mechanisms . These results demonstrate that both types of interneurons contribute to the reduction of the response of the neuron to the stimulus during SSA . 10 . 7554/eLife . 09868 . 008Figure 3 . Optogenetic suppression of either PVs or SOMs reduces SSA in putative excitatory neurons in the auditory cortex . ( A ) Diagram of oddball stimuli with light; two oddball stimuli are presented ( as in Figure 1A ) , with 250-ms light pulses ( green bars ) delivered during every fifth tone , starting 100 ms before tone onset . ( B–D ) Representative neuron PSTH in response to tone A ( left ) and B ( right ) as a standard ( gray ) or deviant ( red ) on light-on ( light colors ) and light-off trials ( dark colors ) . Neurons recorded in PV-Cre ( B , E ) , SOM-Cre ( C , F ) , and control ( D , G ) mice . ( E–G ) Effect of interneuron photosuppression on SSA . Left: SSA index on light-on vs light-off trials . Each neuron is represented by a circle that is filled if the neuron exhibits significant SSA , that is , its FR in response to deviant tones is greater than that to standard tones . The respective representative neuron in B , C , and D is indicated by a red circle . Gray dashed line , identity line . Right: mean SSA index on light-on ( green ) and light-off ( gray ) trials over neuronal population . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 00810 . 7554/eLife . 09868 . 009Figure 3—figure supplement 1 . Photostimulation during standard tone does not affect SSA during subsequent tones on light-off trials . ( A ) Diagram of oddball stimuli illustrating post-photostimulation tone number: tones and light pulses indicated as in Figure 3A . Numbers indicate each tone position relative to light pulses as included in the analysis below . Any tones following deviant tones were excluded from the analysis . ( B ) The mean population FR in response to standard ( gray ) and deviant ( red ) tones subsequent to light-on trials is not affected by light presentation ( dark bars: light-off . light bars: light-on ) . For each neuron , responses are normalized by the response to the third post-laser standard tone ( T3 , indicated by blue dashed line ) . In PV-Cre mice , the standard tone-evoked FR with light-on ( T0 ) and the tone preceding it ( T−1 ) were significantly higher than that of standard T3 ( N = 159 , T−1: Δ = 10% , p2 = 0 . 037 , t ( 158 ) = 2 . 9 , C = 9 . T0: Δ = 170% , p2 = 2e−7 , t ( 158 ) = 5 . 9 , C = 9 ) , while the two post-light tones ( T1 and T2 ) were not significantly different ( N = 159 , T1 and T2: p2 > 0 . 05 , t ( 158 ) < 2 . 6 , C = 9 ) . In SOM-Cre mice , the light-on standard T0-evoked FR was greater than that of T3 ( N = 114 , Δ = 54% , p2 = 4e−8 , t ( 113 ) = 6 . 4 , C = 9 ) , while all light-off tones were not significantly different ( T−1 , T1 , and T2: p2 > 0 . 05 , t ( 113 ) < 0 . 9 , C = 9 ) . In control mice , no standard tones evoked greater FR than T3 , ( N = 107 , T−1 through T2: p2 < 0 . 05 , t ( 106 ) < 2 . 7 ) . In all three groups , deviant tones in all positions evoked greater FRs than standard T3 , ( Δ > 209% , p2 < 5e−5 , t ( 106 ) > 4 . 7 , C = 9 ) . ( C ) Mean SSA index for each sequential tone position ( for T−1 , 0 , 1 , 2 , 3 ) calculated based on the pair of standard and deviant tones at each respective position . Each tone response , tone A or B , was used to calculate a separate SSA index: SSA Index= DA – SADA+ SA or DB – SBDB+ SB Where S and D indicate mean FR evoked by standard and deviant tone probabilities , respectively , and their subscripts indicate the tone frequency condition . Compared to T3 , SSA index was significantly reduced only for T0 , the only light-on trial , in both PV-Cre and SOM-Cre mice ( PV-Cre: Δ = −40% , p2 = 4e−10 , t ( 158 ) = −6 . 9 , C = 4 . SOM-Cre: Δ = −29% , p2 = 2e−7 , t ( 158 ) = −5 . 8 , C = 4 ) , as expected from Figure 3E–G . In both PV-Cre and SOM-Cre mice , the SSA index at all of the other sequential tone positions , T−1 through T2 was not significantly different than that of T3 ( p2 > 0 . 05 , t ( 113 ) < 1 . 9 , C = 4 ) , indicating that the effects of photosuppression were not detectable beyond T0 . In control mice , the SSA index was not different compared to T3 for any tone position , even T0 ( p2 > 0 . 05 , t ( 106 ) < 1 . 8 , C = 4 ) . Together , this analysis demonstrates that the optogenetic effects are acute to illumination periods and unlikely to confound interpretation of effects observed during light-off trials . In all panels , single , and triple stars indicate p < 0 . 05 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 00910 . 7554/eLife . 09868 . 010Figure 3—figure supplement 2 . Interneuron photosuppression does not affect thalamocortical responses to standard or deviant . ( A ) In PV-Cre and SOM-Cre mice , the mean granular layer CSD SSA index was not significantly different between the light-off and light-on conditions for standard or deviant tones ( p2 > 0 . 05 , for each condition; left , PV-Cre: N = 16 . Center , SOM-Cre: N = 12 ) . ( B ) In both experimental groups , the mean granular layer CSD amplitude was not significantly different between the light-off and light-on conditions for standard or deviant tones ( p2 > 0 . 05 , for each condition; left , PV-Cre: N = 8 . Center , SOM-Cre: N = 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 010 A decrease in the SSA index may be due to several factors: ( 1 ) an increase in response to the standard only , ( 2 ) a decrease in response to the deviant , or ( 3 ) an increase in response both to the standard and the deviant , but with a relatively greater increase for the standard . Therefore , we next investigated the effect of interneuron photosuppression on FR of putative excitatory neurons evoked by the standard and deviant tones separately . The effects of PVs and SOMs diverged; in addition to increasing spontaneous activity ( Δ = 185% , p-value from one-sided t-test after correction ( p2 ) = 3e−11 , t ( 159 ) = −7 . 2 ) , suppressing PVs led to increased FR to both the standard ( Δ = 102% , p2 = 3e−11 , t ( 159 ) = −7 . 2 ) and deviant ( Δ = 56% , p2 = 9e−12 , t ( 159 ) = −7 . 4 ) tones ( N = 160 , Figure 4A–C , Figure 4—figure supplement 1A ) . 83% of neurons exhibited greater FR to the standard and 46% to the deviant during PV suppression ( Figure 4—figure supplement 1B ) . The difference in FR due to suppression of PVs was not significantly different between the standard and deviant tones ( p2 > 0 . 05 , t ( 159 ) = −0 . 1 , C = 2 ) but both were greater than the difference in the spontaneous FR ( standard: Δ = 25% , p2 = 0 . 001 , t ( 159 ) = −3 . 6 , C = 2 . Deviant: Δ = 26% , p2 = 0 . 039 , t ( 159 ) = −2 . 4 , C = 2 ) , indicating that the change in tone-evoked FR was similar regardless of tone probability ( Figure 4B , bottom panel ) . Because an equal increase in the FR produces a weaker relative effect on the response to the deviant ( which is higher than to the standard ) , PV inactivation decreases SSA index ( Figure 3E ) . 10 . 7554/eLife . 09868 . 011Figure 4 . PVs and SOMs differentially affect response to standard and deviant tones . ( A , D ) Top: mean response to deviant ( left , red ) and standard ( right , black ) tones , during light-on ( light colors ) and light-off trials ( dark colors ) . Bottom: mean of the difference between responses on light-on and light-off trials for each neuron for deviant ( left , red ) and standard ( right , black ) tone . Each trace is a population average of putative excitatory neuron PSTHs normalized to each neuron's maximum deviant tone-evoked FR on light-off trials . Shaded regions around traces indicate standard error ( SE ) . Dashed lines indicate light onset ( green ) and tone onset and offset ( gray ) . Neurons recorded in PV-Cre ( A ) , SOM-Cre ( D ) mice . ( B , E ) ( Top ) Mean population FR on light-on and light-off trials; ( bottom ) mean population FR difference between light-on and light-off conditions for deviant ( red ) and standard ( gray ) tones and spontaneous activity ( blue ) . Normalization as in A . Neurons recorded in PV-Cre ( B ) , SOM-Cre ( E ) mice . ( C , F ) Modulation of PV-Cre mouse putative excitatory neuron FR response to tones by interneuron photosuppression . Neuronal responses to each tone are represented by two circles , one for standard ( black ) and one for deviant ( red ) tone responses . Filled circles represent significantly increased ( gray , pink ) or decreased ( black , red ) response; unfilled circles: responses without significant modulation . Gray dashed line , identity line . Neurons recorded in PV-Cre ( C ) , SOM-Cre ( F ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01110 . 7554/eLife . 09868 . 012Figure 4—figure supplement 1 . PVs and SOMs differentially affect response to standard and deviant tones . ( A , C ) Correlation between standard and deviant tone response change by photostimulation . Each neuron's response to each tone , A and B , is represented by one circle . Gray dashed line , identity line . Green dashed line , regression line . Neurons recorded in PV-Cre ( A ) and SOM-Cre ( C ) mice . ( B , D ) Proportion of putative excitatory population exhibiting significantly increased ( gray , pink ) , decreased ( black , red ) , or unchanged ( unfilled ) FR to standard and deviant tones due to photosuppression . Neurons recorded in PV-Cre ( A ) , SOM-Cre ( C ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01210 . 7554/eLife . 09868 . 013Figure 4—figure supplement 2 . Consistent effects of PV and SOM suppression in response to equal probability tones . ( A ) Diagram of equal probability tone stimulus; an equal number of pseudorandom tones A and B are presented with 250-ms light pulses ( green bars ) delivered during every fifth tone , starting 100 ms before tone onset . ( B ) Effect of interneuron photosuppression on putative excitatory neuron responses to standard and deviant and equal probability tones . Mean FR of single neuron responses to standard ( gray ) , equal ( green ) , and deviant ( red ) tones on laser-off ( dark colors ) vs laser-on ( light colors ) trials . Top: responses to tone A . Bottom: responses to tone B . Left: neuron from PV-Cre mouse . Center: neuron from SOM-Cre mice . Right: neuron from control mouse . ( C , E ) PSTH of FR to equal probability tones , during light-on ( light green ) and light-off trials ( dark green ) . Each trace is a population average of putative excitatory neuron PSTHs normalized to each neuron's maximum deviant tone-evoked FR on light-off trials . Shaded regions around traces indicate standard error ( SE ) . Dashed lines indicate light onset ( green ) and tone onset and offset ( gray ) . Neurons recorded in PV-Cre ( C , N = 160 ) and SOM-Cre ( E , N = 114 ) mice . ( D , F ) Population mean spontaneous FR ( 50 ms prior to tone onset , yellow ) and equal-tone evoked FR ( 50 ms from tone onset , green ) for light-off ( dark colors ) and light-on ( light colors ) trials . Normalized as in C . Neurons display an increase in spontaneous FR and equal tone-evoked FR with light-on for both PV-Cre ( D—Spn: Δ105% , p2 = 3e−6 , t ( 159 ) = −8 . 1 . Equ: Δ = 41% , p2 = 1e−13 , t ( 159 ) = 4 . 8 ) and SOM-Cre ( F , Spn: Δ = 17% , p2 = 0 . 002 , t ( 113 ) = −3 . 1 . Equ: Δ = 17% , p2 = 0 . 012 , t ( 113 ) = −2 . 54 ) mice . ( G , H ) Modulation of PV-Cre mouse putative excitatory neuron FR response to tones by interneuron photosuppression . Left: circle: Response of each neuron to tone A and/or B . Filled: significantly increased ( light green ) or decreased ( dark green ) response; Unfilled: non-significant modulation . Gray dashed line , identity line . Right: fraction of neuronal tone responses in the population that increased ( light green ) , decreased ( dark green ) , or did not significantly change with light . Neurons recorded in PV-Cre ( G ) , SOM-Cre ( H ) mice . ( I , K ) Mean of the difference between light-on and light-off trials for each neuron for equal probability tones FR response PSTHs . Normalization and dashed lines as in C . Neurons recorded in PV-Cre ( I ) , SOM-Cre ( K ) mice . ( J , L ) Mean population FR difference between light-on and light-off conditions for spontaneous activity ( yellow ) and equal probability tones ( green ) . Measured and normalized as in D and F . Neurons display a larger increase in equal-tone evoked FR than spontaneous FR with light-on for those recorded in both PV-Cre ( J , Δ = 32% , p2 = 0 . 029 , t ( 159 ) = 2 . 2 ) , SOM-Cre ( L , Δ = 118% , p2 = 0 . 047 , t ( 113 ) = 2 . 0 ) mice . In all panels , single , double and triple stars indicate p < 0 . 05 , 0 . 01 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01310 . 7554/eLife . 09868 . 014Figure 4—figure supplement 3 . PVs and SOMs have differential effects on SSA across different layers of cortex . ( A ) Diagram of multi-electrode recording across the supra-granular , granular , and infra-granular layers of A1 . ( B ) SSA index for cortical supra-granular ( Sup , cyan ) , granular ( Grn , yellow ) , and infra-granular ( Inf , magenta ) layers on light-off ( dark colors ) and light-on ( light colors ) trials . ( C ) Difference in SSA index between responses on light-on and light-off trials for each layer as shown in B . Suppressing PVs reduced SSA throughout all cortical layers ( B left , Sup: N = 15 , Δ = −31 , p2 = 0 . 002 . Grn: N = 27 , p2 = 2e−4 , z = 3 . 8 . Inf: N = 79 , Δ = −39% , p2 = 1e−8 , z = 5 . 7 ) . Notably , the effect of PVs was significantly stronger in the granular than in the infra-granular layers ( C left—Δ = 194% , p = 0 . 014 , C = 2 ) but was not different between the supra-granular and the granular or infra-granular layers ( p > 0 . 05 , z < 1 . 8 , C = 2 ) . In the controls , SSA index was not significantly reduced between light-on and light-off trials in any layer ( B right , Sup: N = 3 . Grn: N = 21 . Inf: N = 75 . For each layer: p2 > 0 . 05 , z < 1 . 4 ) , demonstrating that the light-induced effects required Arch . In contrast , suppressing SOMs reduced SSA in the granular ( N = 7 , Δ = −42% , p2 = 0 . 031 ) and infra-granular ( N = 63 , Δ = −24% , p2 = 6e−7 , z = 5 . 0 ) layers but did not have a significant effect on SSA in the supra-granular layers ( N = 3 , p2 > 0 . 05 ) ( B , center ) . In SOM-Cre mice and controls , there was no difference between effects of photosuppression on SSA index in different layers ( C , center and right , p > 0 . 05 , z < 1 . 1 ) . Signed rank test for B and ranked sum test used for C In all panels , double and triple stars indicate p < 0 . 01 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01410 . 7554/eLife . 09868 . 015Figure 4—figure supplement 4 . Differences between PV and SOM effects on standard and deviant tones are preserved for subsets of neurons matched for FR . ( A ) Left: two subsets of neurons recorded in PV-Cre mice with matched FR response magnitude to standard ( gray , above x-axis ) and deviant ( red , below x-axis ) tones on light-off trials . Right: difference between light-on and light-off FR in response to standard ( gray ) and deviant ( red ) tones for the respective subsets of neurons . ( B ) Same as A for neurons recorded in SOM-Cre mice . In all panels , triple stars indicate p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01510 . 7554/eLife . 09868 . 016Figure 4—figure supplement 5 . Effects of PV suppression are identical for tones that evoke strong or weak responses in putative excitatory neurons . Each neuron's response to oddball tones A and B is pooled according to their response strength . The tone which evokes a higher peak FR as a deviant is pooled across neurons as the ‘strong tone’ response , while the tone which evoked a lower peak FR is pooled as the ‘weak tone’ response . ( A–J ) Data are presented as in Figure 4 . Strong tone response data are presented on the left ( A , C , D , G , I ) with solid lines and filled bars , and weak tone response data are presented on the right ( B , E , F , H , J ) with dashed lines and unfilled bars . All data are from PV-Cre mice . ( K ) Mean population FR difference between light-on and light-off conditions for deviant ( red ) and standard ( gray ) tones and spontaneous activity ( blue ) for strong ( filled ) and weak ( unfilled ) tones . Measured and normalized as in D and F . Photosuppression of PVs led to increased spontaneous FR ( Spn ) and standard ( Stn ) and deviant ( Dev ) tone-evoked FR for both strong ( D—Spn: Δ = 187% , p2 = 4e−7 , t ( 50 ) = −5 . 8 . Stn: Δ = 71% , p2 = 3e−10 , t ( 50 ) = −7 . 8 . Dev: Δ = 24% , p2 = 0 . 002 , t ( 50 ) = −3 . 3 ) and weak tones ( F—Spn: Δ = 171% , p2 = 3e−7 , t ( 50 ) = −5 . 9 . Stn: Δ = 89% , p2 = 2e−8 , t ( 50 ) = −6 . 5 . Dev: Δ = 58% , p2 = 2e−7 , t ( 50 ) = −6 . 0 ) ( N = 51 ) . There were no significant differences between strong and weak tones for the change in spontaneous FR and standard and deviant tone-evoked FR ( K , Spn , Stn and Dev: p > 0 . 05 , t ( 50 ) < 2 . 0 ) . In all panels , double and triple stars indicate p < 0 . 05 , 0 . 01 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01610 . 7554/eLife . 09868 . 017Figure 4—figure supplement 6 . Effects of SOM suppression are identical for tones that evoke strong or weak responses in putative excitatory neurons . ( A–K ) Data are presented as in Figure 4—figure supplement 5 . All data are from SOM-Cre mice . Photosuppression of SOMs lead to increased spontaneous FR and standard tone-evoked FR and did not change deviant tone-evoked FR for both strong ( D—Spn: Δ = 45% , p2 = 7e−7 , t ( 33 ) = −6 . 1 . Stn: Δ = 27% , p2 = 4e−5 , t ( 33 ) = −4 . 7 . Dev: p2 > 0 . 05 , t ( 33 ) = −0 . 2 ) and weak tones ( F—Spn: Δ = 45% , p2 = 0 . 001 , t ( 33 ) = −3 . 5 . Stn: Δ = 32% , p2 = 0 . 003 , t ( 33 ) = −3 . 2 . Dev: p2 > 0 . 05 , t ( 33 ) = −0 . 1 ) ( N = 34 ) . There were no significant differences between strong and weak tones for the change in spontaneous FR and standard and deviant tone-evoked FR ( Spn , Stn , and Dev: p > 0 . 05 , 0 . 28 and 0 . 95 , t ( 33 ) < 1 . 2 ) . In all panels , double and triple stars indicate p < 0 . 05 , 0 . 01 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01710 . 7554/eLife . 09868 . 018Figure 4—figure supplement 7 . Differences between PV and SOM effects on standard and deviant tones are preserved for subsets of neurons matched for strength of laser effects on standard tones . ( A ) Two subsets of tone responses ( N = 66 ) matched across PV-Cre ( above x-axis ) and SOM-Cre ( below x-axis ) mice for standard tone-evoked FR difference between light-on and light-off conditions . ( B ) Difference between light-on and light-off FR for spontaneous FR ( blue ) and standard ( gray ) and deviant ( red ) tone-evoked FR and for the PV-Cre ( left ) and SOM-Cre ( right ) subsets . With PV photosuppression , spontaneous FR , standard , and deviant tone-evoked FR increased ( Spn: 20% , p2 = 1e−12 , t ( 65 ) = 8 . 8 , Stn: 19% , p2 = 1e−11 , t ( 65 ) = 8 . 2 , Dev: 21% , p2 = 0 . 001 , t ( 65 ) = 3 . 3 ) , and there were no significant differences between spontaneous and tone-evoked FR changes ( Spn vs Stn: p2 > 0 . 05 , t ( 65 ) = 0 . 1 , C = 3 , Spn vs Dev: p2 > 0 . 05 , t ( 65 ) = −0 . 3 , C = 3 , Stn vs Dev: p2 > 0 . 05 , t ( 65 ) = −0 . 3 , C = 3 ) . With SOM photosuppression , spontaneous FR and standard tone-evoked FR increased ( Spn: 17% , p2 = 1e−8 , t ( 56 ) = 6 . 6 , Stn: 19% , p2 = 2e−11 , t ( 65 ) = 8 . 1 ) , while deviant tone-evoked FR did not change ( p > 0 . 05 , t ( 65 ) = 0 . 9 ) . These changes were not significantly different between spontaneous FR and standard tone-evoked FR ( Spn vs Stn: p > 0 . 05 , t ( 65 ) = −1 . 2 ) , but both were greater than the change in deviant tone-evoked FR ( Spn vs Dev: 309% , p2 = 0 . 022 , t ( 65 ) = 2 . 8 , C = 3 , Stn vs Dev: 360% , p2 = 0 . 003 , t ( 3 . 5 ) , C = 3 ) . By design , the change in standard tone-evoked FR was nearly identical between PV-Cre and SOM-Cre mice ( p1 > 0 . 05 , t ( 65 ) = −0 . 1 , C = 3 ) . Spontaneous FR was also similarly modulated by PV and SOM photosuppression ( p1 > 0 . 05 , t ( 65 ) = 0 . 8 , C = 3 ) . However , deviant tone-evoked FR was more strongly modulated by PV photosuppression than by SOM photosuppression ( 405% , p1 = 0 . 029 , t ( 65 ) = 2 . 4 , C = 3 ) . In all panels , single , double and triple stars indicate p < 0 . 05 , 0 . 01 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 01810 . 7554/eLife . 09868 . 019Figure 4—figure supplement 8 . Differences between PV and SOM effects on standard and deviant tone responses are preserved when FRs are normalized by the mean onset response . A , B , C , and D as in Figure 4A , B , D , E , respectively . ( A , B ) In PV-Cre mice , spontaneous FR and standard and deviant-tone evoked FR increased with light ( B top—Spn: Δ = 210% , p2 = 2e−9 , t ( 159 ) = −6 . 4 . Stn: Δ = 116% , p2 = 9e−10 , t ( 159 ) = −6 . 5 . Dev: Δ = 56% , p2 = 5e−11 , t ( 159 ) = −7 . 1 ) . For FR changes between light-on and light-off conditions , there was no difference significant difference between standard and deviant-tone evoked FRs ( B , bottom , Stn vs Dev: p2 >0 . 05 , t ( 159 ) = 0 . 7 , C = 2 ) , but both were great than the difference in spontaneous FR ( Spn vs Stn: Δ = 34% , p2 = 1e−4 , t ( 159 ) = −4 . 1 , C = 2 . Spn vs Dev: Δ = 26% , p2 = 0 . 029 , t ( 159 ) = −2 . 5 ) . ( C , D ) In PV-Cre mice , spontaneous and standard tone-evoked FRs increased with light ( D top—Spn: Δ = 46% , p2 = 2e−10 , t ( 113 ) = −7 . 0 . Stn: Δ = 26% , p2 = 2e−7 , t ( 113 ) = −5 . 5 ) , but deviant tone-evoked FRs did not ( Dev: p2 > 0 . 05 , t ( 113 ) = −1 . 0 ) . For FR changes between light-on and light-off conditions , there was no difference between spontaneous and standard tone-evoked FR ( D , bottom , Spn vs Stn: p2 > 0 . 05 , t ( 113 ) = −0 . 3 , C = 2 ) , but both were significantly great than deviant tone-evoked FR differences ( Spn vs Dev: Δ = 298% , p2 = 0 . 011 , t ( 113 ) = 2 . 8 Stn vs Dev: Δ = 282% , p2 = 0 . 016 , t ( 113 ) = 2 . 7 , C = 2 ) . In all panels , single , and triple stars indicate p < 0 . 05 and 0 . 001 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 019 By contrast , suppressing SOMs led to an increase in FR for spontaneous activity ( Δ = 46% , p2 = 2e−9 , t ( 113 ) = −6 . 5 ) and during the standard ( Δ = 29% , p2 = 2e−8 , t ( 113 ) = −6 . 1 ) but not deviant ( p2 > 0 . 05 , t ( 113 ) = −0 . 8 ) tone ( N = 114 , Figure 4D–F , Figure 4—figure supplement 1C ) . 52% of neurons exhibited greater FR to the standard and only 11% to the deviant during PV suppression ( Figure 4—figure supplement 1D ) . The increase in FR for spontaneous activity was not different than that during the standard tone ( p2 > 0 . 05 , t ( 113 ) = 0 . 2 , C = 2 ) and the differences in FR due to suppression of SOMs were stronger for spontaneous activity and the standard tone than the deviant tone ( spontaneous: Δ = 390% , p2 = 0 . 004 , t ( 113 ) = 3 . 1 . Standard: Δ = 378% , p2 = 0 . 005 , t ( 113 ) = 3 . 1 ) ( Figure 4E , bottom panel ) , thereby accounting for the change in SSA with SOM inactivation ( Figure 3F ) . Responses to the equal stimulus evoked consistent , yet weaker effects ( Figure 4—figure supplement 2 ) . PVs and SOMs differ in their density among different layers of the cortex and in laminar sources and targets of their inputs and outputs ( Markram et al . , 2004; Xu and Callaway , 2009; Fino et al . , 2013 ) . The effects of PV and SOM suppression on SSA had differential laminar distribution ( Figure 4—figure supplement 3 ) . The effect of PVs on SSA was equally strong in the supra-granular and infra-granular layers , but stronger in the granular layer , that is , the thalamo-recipient layer . This differential effect is consistent with the relative proportion of cortical interneurons that are PVs , which is higher in granular than either in infra- or supra-granular layers ( Lee et al . , 2010; Xu et al . , 2010; Ouellet and de Villers-Sidani , 2014 ) . In contrast , suppressing SOMs reduced SSA in the granular and infra-granular , but not supra-granular layers . The relative proportion of cortical interneurons that are SOMs is greatest in the granular and infra-granular layers , but still present in supra-granular layers ( Lee et al . , 2010; Xu et al . , 2010; Ouellet and de Villers-Sidani , 2014 ) . As some SOMs predominantly target the distal dendrites of pyramidal neurons ( Markram et al . , 2004 ) , the effect of suppressing SOMs in supra-granular layers may be evident in recordings of pyramidal neurons with cell bodies in deeper layers , supporting our results . In addition , cortical extracellular recordings may be biased toward neurons in granular and infra-granular layers , precluding adequate sampling of activity in superficial layers . In controls , we did not observe a difference in the effect of light on SSA across layers , demonstrating that the differences are not due to differential artifact of light stimulation . Our results indicate that both PVs and SOMs affect SSA , but in different ways: ( 1 ) the increase in the FR of putative excitatory neurons due to PV suppression is constant , either during presentation of the standard or the deviant , and greater than changes in spontaneous activity . Thus , PVs amplify SSA in excitatory neurons by exerting a relatively stronger inhibitory drive for the standard than for the deviant . ( 2 ) Suppression of SOMs leads to increased putative excitatory neuron activity only during the spontaneous firing or the presentation of the standard , but not for the deviant . This suggests that the strength of SOM-mediated inhibitory drive is not significant in response to the deviant but increases with repeated presentations of the standard . In neurons exhibiting SSA , responses to the deviant are stronger than to the standard . This difference might lead to a ‘ceiling’ effect , reducing the effect of PV photosuppression on FR to the deviant , but not standard ( Olsen et al . , 2012 ) . However , restricting the analysis to two subpopulations of neurons , which have matched mean and standard deviation of FR to the standard vs the deviant tones ( Ulanovsky et al . , 2004; Rust and Dicarlo , 2010 ) , preserved the observed effects of photosuppression ( Figure 4—figure supplement 4 ) . Suppressing PVs led to an equal increase in FR to both the standard and the deviant tone ( N = 54—standard: Δ = 62% , p2 = 6e−8 , t ( 53 ) = 6 . 3 . Deviant: Δ = 55% , p2 = 3e−5 , t ( 53 ) = 4 . 5 . Standard vs deviant: p2 > 0 . 05 , t ( 53 ) = 0 . 5 ) . In contrast , suppressing SOMs led to a significant increase in FR to the standard , but no change in FR to the deviant ( N = 44—standard: Δ = 30% , p2 = 7e−6 , t ( 43 ) = 5 . 1 . Deviant: p2 > 0 . 19 , t ( 43 ) = 1 . 3 . Standard vs deviant: Δ = 382% , p2 = 6e−4 , t ( 43 ) = 3 . 7 ) . For neurons that responded more strongly to one of the tones ( ‘strong’ vs ‘weak’ tone ) , a ceiling effect would predict that the effect of interneuron suppression would be stronger for the weak than the strong tone . However , PV and SOM suppression exhibited a similar effect on responses to the strong and the weak tones in neurons that exhibited differential responses to two tones ( Figure 4—figure supplements 5 , 6 ) . Suppressing PVs led to similar increases in tone-evoked FR between weak and strong tones for both deviant ( N = 51 , p2 > 0 . 05 , t ( 50 ) = 1 . 0 ) and standard tones ( p2 > 0 . 05 , t ( 50 ) = −1 . 9 ) . Suppressing SOMs also led to similar differential effects between strong and weak tones; standard tone-evoked FR increased equally ( N = 34 , p2 > 0 . 05 , t ( 33 ) = 1 . 1 ) and deviant tone-evoked FR was equally unchanged ( p2 = 0 . 05 , t ( 33 ) = −0 . 1 ) . Combined , these analyses demonstrate that the effect of PV photosuppression on SSA cannot be explained by the ceiling effect for either PVs or SOMs . Although Arch drove strong currents in both SOM and PV neurons ( Figure 2D , Figure 2—figure supplements 1 , 2 ) , there might be a difference in expression level or efficacy of Arch between SOM-Cre and PV-Cre mice , leading to a stronger effect of photosuppression in PV-Cre than in SOM-Cre mice on tone-evoked FRs ( Figure 4B , E ) . Alternatively , the difference might be attributable to the morphological or functional differences between SOMs and PVs . To address this confound , we selected tone responses that exhibited matched difference in standard tone-evoked FR between light-on and light-off trials ( N = 66 , Figure 4—figure supplement 7 ) . Within these matched subpopulations , PV and SOM photosuppression exhibited differential effects similar to those of the whole population . The change in FR due to PV suppression was not significantly different between responses to the standard and deviant ( p2 > 0 . 05 , t ( 65 ) = −0 . 3 , C = 3 ) . By contrast , the change in deviant tone-evoked FR due to SOM suppression was significantly weaker than that for the standard tone ( Δ = −78% , p2 = 0 . 003 , t ( 3 . 5 ) , C = 3 ) . By the design of the analysis , the effect of PV or SOM suppression on standard tone-evoked FR was nearly identical ( p1 > 0 . 05 , t ( 65 ) = −0 . 1 , C = 3 ) . However , the change in deviant tone-evoked FR was greater for PV photosuppression than SOM photosuppression ( Δ = 404% , p1 = 0 . 029 , t ( 65 ) = 2 . 4 , C = 3 ) . Since the observed differential effects of PV and SOM suppression persisted in subsets of neurons that were matched for photosuppression-induced change in standard tone-evoked FR , these differences are unlikely due to differential expression or efficacy of Arch in the PV-Cre and SOM-Cre mice , but rather reflect functional differences between the two types of interneurons . Within the oddball sequence , after the presentation of the deviant tone , SSA takes several repeats of the standard tone to reach an adapted state ( Ulanovsky et al . , 2004 ) . Consistent with previous findings ( Ulanovsky et al . , 2004 ) , presentation of the deviant tone temporarily reduced SSA without photosuppression ( Figure 5A–C , dark color bars ) ; following the deviant tone ( T0 ) , the first two standard tones ( T1 and T2 ) evoked elevated FRs compared to the fourth standard tone ( T4 ) ( PV-Cre , Figure 5B—N = 148 , T1: Δ = 60% , p2 = 3e−8 , t ( 146 ) = 6 . 3 , C = 11 , T2: Δ = 26% , p2 = 0 . 043 , t ( 146 ) = 2 . 9 , C = 11 . SOM-Cre , Figure 5C—N = 102 , T1: Δ = 72% , p2 = 1e−5 , t ( 101 ) = 5 . 2 , C = 11 , T2: Δ = 31% , p2 = 0 . 013 , t ( 101 ) = 3 . 3 , C = 11 ) . The third standard tone ( T3 ) and the tone prior to the deviant tone ( T−1 ) evoked responses similar to T4 ( PV-Cre , Figure 5B—T−1 and T3: p2 > 0 . 05 , t ( 146 ) < 2 . 5 , C = 11 . SOM-Cre , Figure 5C—T−1 and T3: p2 > 0 . 05 , t ( 101 ) < 2 . 9 , C = 11 ) . Neurons in which response to T0 did not produce spikes were excluded . Suppressing PVs led to a significant and equal increase in FR to four consecutive presentations of the standard following the deviant ( Figure 5B , left , for each tone , T−1 through T4 , with light-on compared to T4 with light-off: Δ > 132% , p2 < 2e−9 , t ( 146 ) ≥ 6 . 8 , C = 11 . Figure 5B , right , change in FR between light-on and light-off responses to each T−1 through T3 as compared to T4: p > 0 . 05 , t ( 146 ) < 1 . 8 , C = 5 ) . In contrast with PVs , suppressing SOMs led to a progressively increasing effect on FR to consecutive presentations of the standard tone following the deviant ( Figure 5C , left , for each standard tone , T−1 through T4 , with light-on compared to T4 with light-off: Δ > 64% , p2 < 9e−4 , t ( 101 ) ≥ 4 . 1 , C = 11 . Figure 5C , right , difference between FR change in T1 and T4 with light-on: p = 0 . 008 , t ( 101 ) = −3 . 2 , C = 5 . Repeated measures ANOVA with tone number ( T1 through T4 ) as a factor: F ( 3 , 300 ) = 4 . 30 , p = 0 . 0054 ) . These results are consistent with the interpretation that the inhibitory drive from PVs is constant throughout the stimulus regardless of tone history , whereas the effect of SOM modulation increases with repeated presentations of the standard tone . 10 . 7554/eLife . 09868 . 020Figure 5 . Post-deviant time course of interneuron-mediated effect on SSA . ( A ) Diagram of oddball stimuli illustrating post-deviant tone number used in subsequent analysis; Tones and light pulses are as indicated in Figure 3A . Numbers indicate each tone position relative to deviant tones . Responses to any standard tones following light-on standards were excluded from the analysis . ( B , C ) Left: mean population FR in response to standard tones ( gray ) subsequent to deviant tones ( red ) within the oddball sequence on light-off ( dark colors ) and light-on ( light colors ) trials . All responses are normalized to the response to the fourth post-deviant standard tone on light-off trials ( green dashed line ) . Right: difference between FR on light-on and light-off trials in response to standard ( gray ) and deviant ( red ) tones . ( B ) : PV-Cre mice . ( C ) : SOM-Cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 02010 . 7554/eLife . 09868 . 021Figure 5—figure supplement 1 . Initial time course of interneuron-mediated effect on SSA . The inhibitory influence of PV+ interneurons is persistent while that of SOM+ interneurons builds up over the first 40 tones . ( A , B , C , D ) Top: mean population FR in response to consecutive tones of the oddball sequence . Lines represent FR to standard tones on light-off ( dark gray ) and light-on ( light gray ) trials , interpolated to continuous lines . Dots represent FR to deviant tones on light-off ( red ) and light-on ( pink ) trials . Bottom: difference between FR on light-on and light-off trials to standard tones of the oddball sequence . Left: whole-oddball sequence . Right: first 50 tones of each sequence . A , B: PV-Cre mice . C , D: SOM-Cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 021 The time course of the effect of interneuron photosuppression on FR of the putative excitatory neurons at the beginning of each oddball sequence exhibited similar differences between PVs and SOMs . After the onset of each oddball sequence , SSA develops over the course of several standard tone presentations ( Ulanovsky , 2004 ) . As expected , on light-off trials , FR decreased in response to the standard tone over the first 20 repetitions of the tone ( Figure 5—figure supplement 1 ) . For PV-Cre mice , the difference in FR to the standard tone between light-on and light-off trials did not change over this time and stayed positive for the remainder of the oddball stimulus ( Figure 5—figure supplement 1A ) . Over the first 20 trials , FR adapted with a similar time course for both the light-on and light-off trials , so the change due to PV photosuppression in FR to standard stayed constant ( Figure 5—figure supplement 1A , B ) . In contrast , for SOM-Cre mice , FR on light-on trials increased over the first 40 trials , whereas on light-off trials , it decreased ( Figure 5—figure supplement 1A ) . As a result , the difference due to photo-manipulation in FR to the standard tone increased over the first 40 trials and then stayed consistently positive throughout the stimulus presentation ( Figure 5—figure supplement 1C ) . These results demonstrate that the PV-mediated effect on putative excitatory neuronal responses did not change with repeated presentations of the standard tone , whereas the SOM-mediated effect increased with the repeated stimulus . In order to understand how PVs and SOMs exert differential control of SSA in putative excitatory neurons , we used optogenetic tagging to identify the specific interneurons and to quantify whether PVs and SOMs exhibited SSA ( Lima et al . , 2009 ) . Through targeted viral delivery to AC , we drove Channelrhodopsin-2 ( ChR2 ) expression , which depolarizes neurons when stimulated by light , in either PVs or SOMs ( Chow et al . , 2010 ) ( Figure 6A , D , Figure 6—figure supplement 1A ) . A modified AAV encoding anti-sense code for ChR2 and a fluorescent reporter , under the FLEX cassette , was injected into PV-Cre or SOM-Cre mice ( Boyden et al . , 2005; Sohal et al . , 2009; Cardin et al . , 2010; Zhang et al . , 2010; Deisseroth , 2011 ) and resulted in specific expression of ChR2 , localized to PVs or SOMs ( Figure 6—figure supplement 1B , c PV-Cre; N = 183 neurons in 3 mice , specificity = 67 ± 1% , efficiency = 76 ± 5% . SOM-Cre: N = 202 neurons in 4 mice , specificity = 90 ± 3% , efficiency = 81 ± 4% ) . Neurons were identified as PVs or SOMs if they responded to brief ( 5 ms ) flashes of light with spikes within 1 . 5–4 . 5 ms of laser pulse onset ( Figure 6A , D ) . 10 . 7554/eLife . 09868 . 022Figure 6 . PV and SOM interneurons exhibit SSA . ( A , D ) Optogenetic methods . A1 was injected with AAV-FLEX-ChR2-tdTomato . During experiments , an optic fiber was positioned to target A1 and neuronal activity was recorded using a multichannel silicon probe in A1 . Top diagram: blue light ( 473 nm ) excites PVs in PV-Cre mice or SOMs in SOM-Cre mice . Bottom: peri-stimulus spike raster of a representative optogenetically identified PV ( top ) or SOM ( bottom ) . Shaded region , blue light on . ( A ) PV-Cre . ( D ) SOM-Cre . ( B , E ) PSTH of PVs ( B ) or SOMs ( E ) FR response to deviant ( red ) and standard ( black ) tones . Normalization and dashed lines as in Figure 4A , B . ( C , F ) Mean PVs ( C ) or SOMs ( F ) FR response over the 100 ms of deviant ( red ) and standard tones ( gray ) , and 100 ms of spontaneous activity prior to tone onset ( blue ) . Each line represents a single neuron's response to each conditions , and its color indicates the magnitude of significant differences between two conditions; pink , gray , blue , and dashed black lines indicate a greater response to deviant tone , standard tone , silence and no significant change , respectively . ( G ) Mean SSA index of putative excitatory neurons , PVs , and SOMs . Circles represent SSA index values of individual neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 02210 . 7554/eLife . 09868 . 023Figure 6—figure supplement 1 . Optical tagging of PVs and SOMs . ( A ) Diagram of optogenetic methods . A1 was injected with AAV-FLEX-ChR2-tdTomato . During experiments , an optic fiber was positioned to target A1 and neuronal activity was recorded using a multichannel silicon probe in A1 . ( B , C ) Transfection of interneurons with Channelrhodopsin-2 ( ChR2 ) . Images: immunohistochemistry demonstrating co-expression of ChR2 and an interneuron-type reporter in A1 . Bar plots: efficiency ( Ef ) and specificity ( Sp ) of visual transfection of PVs ( top ) and SOMs ( bottom ) with ChR2 . Ef , percent of labeled interneurons expressing ChR2 . Sp , percent of ChR2-expressing cells , which are also labeled interneurons . ( B ) PV-Cre mouse A1 . Green; anti-body stain for parvalbumin . Red; ChR2-tdTomato . Merge; co-expression of ChR2 and PVs . ( C ) SOM-Cre mouse A1 . Green; anti-body stain for somatostatin . Red; ChR2-tdTomato . Merge; co-expression of ChR2 and SOMs . Scale Bar = 25 µm . ( D , E ) Fraction of PVs ( D ) or SOMs ( E ) exhibiting a greater response to deviants than standards ( pink ) , the reverse ( gray ) , or neither ( white ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 02310 . 7554/eLife . 09868 . 024Figure 6—figure supplement 2 . PVs and SOMs have different adaptation profiles for equal probability tones . ( A , C ) PSTH of PVs ( A ) or SOMs ( C ) FR response to standard ( black ) , equal probability ( green ) , and deviant ( red ) tones . Normalization and dashed lines as in Figure 4A , B . ( B , D ) Mean PV ( B ) or SOM ( D ) population FR response to standard ( gray ) , equal probability ( green ) , or deviant ( red ) tones over 100-ms tone duration . The mean spontaneous FR ( during the 100 ms prior to all tones ) of oddball and equal probability stimuli was subtracted from respective tone-evoked mean FRs . In PVs , equal probability tones evoked FRs greater than standard tones ( B—N = 16 , Δ = 110% , p2 = 0 . 030 , z = −2 . 4 , C = 2 ) and not significantly different that deviant tones ( p2 > 0 . 05 , z = −1 . 7 , C = 2 ) . In SOMs , equal tones evoked higher FRs than standard tones ( C—N = 28 , Δ = 95% , p2 = 0 . 022 , z = −2 . 6 , C = 2 ) , and lower FRs than deviant tones ( Δ = −36% , p2 = 0 . 049 , z = −2 . 3 , C = 2 ) . In both types of interneuron , deviant tones evoked higher FRs than standard tones ( B , PV—Δ = 188% , p2 = 0 . 010 , z = −2 . 8 , C = 2 . C , SOM—Δ = 205% , p2 = 0 . 002 , z = −3 . 3 , C = 2 ) . In all panels , single and double stars indicate p < 0 . 05 and 0 . 01 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 024 Both PVs and SOMs exhibited SSA , evidenced by a significant reduction in standard tone-evoked FR compared to the deviant tone response ( Figure 6B , C , E , F , PV: N = 16 , Δ = −32% , p2 = 0 . 023 , z = −2 . 5 , C = 2 . SOM: N = 28 , Δ = −41% , p2 = 0 . 002 , z = −3 . 3 , C = 2 . Signed-rank test ) . The SSA index was not significantly different between PVs and SOMs ( Figure 6G , neurons responsive to both tones A and B—PV: N = 5 , SOM: N = 12 . PV and SOM: p2 > 0 . 05 , C = 2 . Rank sum test ) and both were similar to the mean SSA index in putative excitatory neurons ( Figure 6G—Exc: N = 67 . Exc vs PV: p > 0 . 05 , z = 0 . 7 , C = 2 . Exc vs SOM: p > 0 . 05 , z = 0 . 4 , C = 2 ) . PVs and SOMs exhibited some differences in relative response changes between the deviant , the standard , and the equal tones ( Figure 6 , Figure 6—figure supplement 2B , D ) ; PVs' response to the equal tones did not decrease significantly as compared to deviant tones ( N = 16 p2 > 0 . 05 , z = −1 . 7 , C = 2 ) , whereas SOMs adapted in their response to equal tones ( Δ = −36% , p2 = 0 . 049 , z = −2 . 3 , C = 2 ) , and then further to standard tones ( N = 28 , Δ = −49% , p2 = 0 . 022 , z = −2 . 6 , C = 2 ) . These results suggest that SOMs may adapt at a faster time scale than PVs with repeated presentation of tones . Our results of recordings from PVs and SOMs present a surprising finding that PVs and SOMs adapt in response to repeated tones , countering our initial hypothesis that SOMs saturate in responses to the deviant , or facilitate with repeated presentation of a tone . How can an adapting interneuron contribute to added adaptation in excitatory neurons ? To address this question , we next developed a model of coupled excitatory–inhibitory neuronal populations . Excitatory and inhibitory neurons form tight mutually coupled networks in A1 , and we hypothesized that through differential post-synaptic integration by excitatory neurons , interneurons can amplify adaptation in excitatory neurons . As a proof-of-principle that would account for our findings that PVs and SOMs exhibit similar magnitude of SSA , yet have a differential effect on SSA in putative excitatory neurons , we constructed a simplified model of mutually coupled inhibitory–excitatory neuronal populations . We tested how responses of the model putative excitatory neurons are affected by manipulation of activity of PVs or SOMs ( Figure 7A ) . Thalamocortical tone-evoked inputs were modeled including an adaptation term and resulted in reduced responses of excitatory , PV , and SOM populations to repeated tones ( Figure 7—figure supplement 1A , B ) . The model replicated the differential effects of manipulation of PV and SOM activity on responses to standard and deviant tones in putative excitatory neurons ( Figure 7B–E ) : when PVs were suppressed optogenetically , the responses to both the standard and the deviant tones increased ( Figure 7B , C ) . By contrast , when SOMs were suppressed , although the spontaneous FR and standard tone-evoked FR were elevated , the responses to the deviant tone remained constant , whereas the responses to the standard tone increased ( Figure 7D , E ) . SOMs have been shown to inhibit PVs ( Cottam et al . , 2013; Pfeffer et al . , 2013; Sturgill and Isaacson , 2015 ) . Including inhibition between SOMs and PVs did not affect the model outcome , with suppression of PVs resulting in suppression of excitatory responses to both the standard and the deviant , and suppression of SOMs driving specific suppression of excitatory responses to the standard , but not the deviant ( Figure 7—figure supplement 2 ) . 10 . 7554/eLife . 09868 . 025Figure 7 . Mutually coupled excitatory-PV-SOM neuronal model accounts for differential effects of PVs and SOMs on SSA in putative excitatory neurons . ( A ) Center: diagram of coupled network model . Excitatory ( Exc ) and two types of inhibitory interneurons ( PV and SOM ) receive tone-evoked inputs . They make reciprocal connections on each other; Exc makes excitatory synapses on PV or SOM; PV and SOM inhibit Exc . Closed circles: excitatory synapses . Open circles: inhibitory synapses . Orange outlines: excitatory input–output pathway . Purple outlines: PV input–output pathway . Green outlines: SOM input–output pathway . The effect of optogenetic modulation was modeled as an additional input current delivered to inhibitory neuronal populations . Adaptation was modeled as decaying synaptic coefficient with slow adaptation . Left and right inset plots: combined input–output transfer function that represents the transformation between synaptic inputs and the activity of excitatory neurons . The values of inputs are depicted by arrows for the spontaneous and tone-evoked activity in response to deviant and standard tones under light-off ( dark color ) and light-on ( light color ) conditions , with change due to light highlighted by light green arrows . ( B , D ) Tone-evoked responses of model neuronal excitatory population to deviant ( red ) and standard tones ( gray ) , that is , the first and fourth consecutive tone presented , under light-off ( dark colors ) and light-on ( light colors ) conditions . Dashed lines indicate light onset and offset ( green ) and tone onset and offset ( gray ) . ( B ) Light suppresses PVs . ( D ) Light suppresses SOMs . ( C , E ) Left: spontaneous FR ( blue ) and standard ( black ) and deviant ( red ) tone-evoked FRs on light-off ( dark colors ) and light-on ( light colors ) conditions . Right: mean difference between responses on light-on and light-off conditions . ( C ) Light suppresses PVs . ( E ) Light suppresses SOMs . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 02510 . 7554/eLife . 09868 . 026Figure 7—figure supplement 1 . Adaptation to repeated tones in model excitatory and inhibitory neurons . Responses evoked by four consecutive tones Exc ( purple ) , PVs ( orange , A ) , and SOMs ( green , B ) . Note adaptation in the responses of both excitatory and inhibitory neurons . During fourth tone , there is light-evoked suppression of interneuron activity . Ligh-on: solid; light-off: dashed lines . ( A ) Light suppresses PVs . ( B ) Light suppresses SOMs . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 02610 . 7554/eLife . 09868 . 027Figure 7—figure supplement 2 . Excitatory–inhibitory model with inhibitory inputs from SOM to PV population accounts for differential effects of PVs and SOMs on SSA in putative excitatory neurons . ( A ) Center: diagram of coupled network model . Model is as in Figure 7 , with additional inhibitory inputs from SOM to Exc population . ( B , E ) Tone-evoked responses of model neuronal excitatory population to deviant ( red ) and standard tones ( gray ) , that is , the first and fourth consecutive tone presented , under light-off ( dark colors ) and light-on ( light colors ) conditions . Dashed lines indicate light onset and offset ( green ) and tone onset and offset ( gray ) . ( B ) Light suppresses PVs . ( E ) Light suppresses SOMs . ( C , F ) Left: spontaneous FR ( blue ) and standard ( black ) and deviant ( red ) tone-evoked FRs on light-off ( dark colors ) and light-on ( light colors ) conditions . Right: mean difference between responses on light-on and light-off conditions . ( C ) Light suppresses PVs . ( F ) Light suppresses SOMs . ( D , G ) Responses evoked by four consecutive tones Exc ( purple ) , PVs ( orange ) , and SOMs ( green ) . Note adaptation in the responses of both excitatory and inhibitory neurons . During the fourth tone , there is light-evoked suppression of interneuron activity . Dark traces: light-off . Light traces: light-on . ( D ) Light suppresses PVs . ( G ) Light suppresses SOMs . DOI: http://dx . doi . org/10 . 7554/eLife . 09868 . 027 An explanation for the difference of the effects of PVs and SOMs can be provided by examining the combined transfer function between pre-synaptic inputs and post-synaptic activity of excitatory neurons separately for PVs and SOM suppression ( Figure 7A , insets ) : light-driven modulation of PV activity has the same effect on excitatory neuron responses at spontaneous , standard tone-evoked , and deviant tone-evoked activity ( Figure 7A , left inset ) . Spontaneous , standard , and deviant input levels all fall within the linear portion of the transfer function between inputs and change in the excitatory neuron activity . On the other hand , for SOMs , modulation of their activity in the deviant tone-evoked regime drives small to no changes in excitatory neuronal activity , whereas modulation of SOM activity in the spontaneous and standard tone-evoked regime drives significant changes in excitatory neuronal activity ( Figure 7A , right inset ) . The deviant tone-evoked activity falls on the saturating part of the input–output transfer function , whereas the standard tone-evoked and spontaneous inputs fall on the linear part of the transfer function . Then , shifts in SOM inputs due to photosuppression evoke small changes during deviant tone responses , but larger changes during either standard or spontaneous activity . Either PV or SOM manipulation would result in reduction of combined SSA of excitatory neurons .
The majority of neurons in the auditory cortex selectively reduce their responses to frequent , but not rare sounds , exhibiting SSA . However , the cortical mechanisms involved in the production and stimulus-specificity of SSA within the auditory cortex are not well understood . Here , we found that , in addition to adaptation at the level of thalamocortical inputs , two distinct types of interneurons , PVs and SOMs , differentially contributed to SSA in the primary auditory cortex . Optogenetic suppression of either PVs or SOMs led to a reduction in SSA in putative excitatory neurons ( Figure 3 ) . Suppression of PVs led to an equal increase in the FR of the putative excitatory neurons in response to the standard tone and the deviant tone ( Figure 4 ) . By contrast , suppression of SOMs significantly increased the response to the standard tone but lacked a significant effect on the response to the deviant tone ( Figure 4 ) . This series of findings expands on the ‘adaptation in narrowly tuned units’ model , which proposes that repeated presentation of the standard stimulus drives adaptation within more narrowly tuned inputs , such as thalamocortical inputs ( Mill et al . , 2011; Taaseh et al . , 2011; Nelken , 2014 ) . Our data indicate dual effects of cortical inhibition on SSA: ( 1 ) PVs contribute to SSA by providing a constant level of inhibition , resulting in a relatively higher inhibitory drive during the presentation of the standard , as compared to the deviant . Taking into account the non-linear synaptic input to FR output function of a typical pyramidal neuron , the constant inhibition amplifies the effect of thalamocortical depression in suppressing the response of the neuron to repeated stimulus ( Figure 7A ) . ( 2 ) The selective increase of the inhibitory drive from SOMs for standard stimulus as compared to the deviant stimulus responses might be explained by a shift in the non-linear transfer function between inputs to SOMs and their outputs to excitatory neurons , possibly due to facilitation of SOM-to-excitatory neuron synapses ( Beierlein et al . , 2003; Silberberg and Markram , 2007 ) ( Figure 7A ) . Surprisingly , we found that , despite the differential effect of PV and SOM suppression on tone-evoked responses in putative excitatory neurons , both PVs and SOMs exhibit SSA . This finding is consistent with previous results that found that thalamocortical synapses onto inhibitory neurons and synapses from inhibitory neurons to excitatory cells can be depressing ( Tan et al . , 2008; Ma et al . , 2012 ) . How does suppression of these interneurons result in differential reduction in SSA in excitatory neurons ? Our model provides an intuition for this effect: the mutually coupled excitatory–inhibitory network model demonstrates that the observed differential effects of PV and SOM suppression may be due to their differential action on excitatory neuronal responses in the unadapted and adapted state ( Figure 7 ) . Tone-evoked responses of PVs would fall on the linear portion of the transfer function between PV activity and excitatory neuron depolarization , while the same tones maximally affect inputs from SOMs onto excitatory neurons , with SSA shifting the inputs to the linear , more sensitive range of inputs from SOMs . Thus , SSA may serve an additional function: to adjust the responses of neurons in a range that is more sensitive to small changes in the inputs from both excitatory and inhibitory neuronal populations . More generally , the simulation demonstrates that a circuit element , such as PVs or SOMs , that itself adapts may further amplify adaptation in the excitatory neurons . To estimate the differential contribution of PVs or SOMs inputs to the excitatory neurons , we measured the difference in the FR of neurons due to optogenetic partial suppression of their firing . This measurement provides an estimate of the change in the FR of the putative excitatory neurons with the change in combined inputs to the inhibitory neurons , thereby allowing estimation of the synaptic transfer function ( Figure 7A , insets ) . A simple biologically plausible network incorporating these transfer functions can reproduce the observed responses ( Figure 7 ) . There are several caveats to this interpretation . First , the FR may not linearly translate onto synaptic input strength because of the spiking non-linear rectification between the inputs and outputs of the putative excitatory neuron: a small change in FR in the low-FR regime might correspond to a greater change in the synaptic drive than a similar-sized change in FR in the high-FR regime . However , our findings would still hold were this the case: in examining the effect of SOM suppression on response to the deviant , the actual difference in the synaptic drive between the deviant and the standard would then be even greater than observed . At the other end of the non-linearity , the analysis of neuronal responses sorted based on their FR to the standard tone and the deviant tone revealed that the ‘ceiling effect’ would not contribute to a decreased effect of photostimulation on the response to the deviant in SOM-Cre mice ( Figure 4—figure supplements 4–7 ) . Second , PVs and SOMs may inhibit not only the excitatory neurons , but also each other . SOMs make synapses onto PVs ( Isaacson and Scanziani , 2011; Ma et al . , 2012; Cottam et al . , 2013 ) , thereby potentially suppressing them with repeated presentation of the standard . Therefore , when SOMs are suppressed , some PVs may be disinhibited and provide a stronger suppression of excitatory neurons . The null effect on responses to the deviant tone during SOM suppression could result from a combination of increase in inhibition from disinhibited PVs in addition to reduced inhibition of SOMs onto excitatory neurons . Including inhibition from SOMs to PVs in the proof-of-principle model supported experimental findings ( Figure 7—figure supplement 2 ) . Third , other interneuron types , such as vasopressin-positive interneuron may be involved in the circuit ( Pi et al . , 2013 ) , and the changes that we observe may reflect several inhibitory stages of processing . One must be cautious in translating the data from our experiments as a strict description of neuronal activity in awake animals , as our results were based on recordings from mice under light isoflurane anesthesia . Other forms of anesthesia , such as pentobarbital-based ( Cheung et al . , 2001; Gaese and Ostwald , 2001 ) , ketamine ( Otazu et al . , 2009 ) and high concentrations of isoflurane ( Cheung et al . , 2001; Ter-Mikaelian et al . , 2007 ) , can affect multiple aspects of sound-evoked responses in the auditory cortex . Nonetheless , our results are likely to extend for awake mice , since isoflurane anesthesia-induced effects on neuronal activity decrease as the concentration of isoflurane is reduced to the levels used in our recordings ( Land et al . , 2012 ) . In addition , all recordings and manipulations were performed under identical anesthetic conditions , and our conclusions are based on the relative comparison of the effects of suppressing PVs and SOMs , which are expected to hold under awake conditions ( Centanni et al . , 2013 ) . While not demonstrated directly , SSA has been linked to detection of deviant sounds ( Ulanovsky et al . , 2003 ) , which may be facilitated by a relatively enhanced neuronal response to a change in the ongoing sound ( Nelken et al . , 2003; Winkler et al . , 2009; Grimm and Escera , 2012 ) . By suppressing the responses to a frequently presented tone , the responses of neurons to a rare stimulus become relatively enhanced . However , whether and how modulating SSA in the auditory cortex affects auditory behavior has not yet been tested . Inhibitory interneurons may prove to have a complementary role in shaping auditory perception in addition to receptive field reorganization driven by synaptic plasticity ( Froemke et al . , 2013 ) . The use of optogenetic methods to test the function of inhibitory interneurons in SSA overcomes the limitations of lesion or pharmacological studies ( Elliott and Trahiotis , 1972; Duque et al . , 2014 ) , which only allow for prolonged , non-selective inactivation ( Moore et al . , 2001 ) . By combining optogenetic manipulation of interneuron activity with behavioral measurements , future experiments will explore whether interneuron-mediated SSA indeed affects the auditory behavior of the subject , such as enhanced ability to detect unexpected events . | In everyday life , we are often exposed to a mix of different sounds . An essential task for our brain is to separate the important sounds from the unimportant ones . For example , stepping out onto a busy street , you may at first be very aware of the noise of traffic . Later , you may start to ignore the din and instead only notice sounds that break the monotony: a honking car horn or maybe a stranger's voice . This is because the neurons in the auditory pathway respond differently to common and rare sounds . In particular , excitatory neurons in the region termed the ‘auditory cortex’ send fewer nerve impulses in response to frequent sounds , but respond vigorously to rare sounds . This phenomenon is called ‘stimulus-specific adaptation’ , but it is not known exactly which neurons in this brain region enable this process to occur . Now , Natan et al . have combined different cutting-edge neuroscience techniques to identify the circuit of brain cells that drives this stimulus specific adaptation . A technique called optogenetics was used to effectively ‘turn off’ each of two kinds of inhibitory neuron in the auditory cortex of mice , by exposing the brain to colored light from a laser . Natan et al . found that both kinds of inhibitory neuron amplified stimulus-specific adaptation , but via different mechanisms . One of these neuron types , called ‘parvalbumin-positive interneurons’ , exerted a general effect on excitatory neurons and suppressed responses to both frequent and rare sounds As the responses to rare sounds started off greater than the responses to frequent sounds , suppressing both by an equal amount actually led to an increase in the relative difference between them . On the other hand , the second kind of inhibitory neuron , called ‘somatostatin-positive interneurons’ , only reduced the excitatory neurons' responses to frequent sounds; these neurons had no effect on responses to rare noises . Future studies will test how specific adaptation in different contexts can help us to behaviorally detect rare sounds while ignoring common ones , and search for the circuits beyond the auditory cortex that support hearing in complex sound environments . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"neuroscience"
] | 2015 | Complementary control of sensory adaptation by two types of cortical interneurons |
Eukaryotic cells deploy autophagy to eliminate invading microbes . In turn , pathogens have evolved effector proteins to counteract antimicrobial autophagy . How adapted pathogens co-opt autophagy for their own benefit is poorly understood . The Irish famine pathogen Phytophthora infestans secretes the effector protein PexRD54 that selectively activates an unknown plant autophagy pathway that antagonizes antimicrobial autophagy at the pathogen interface . Here , we show that PexRD54 induces autophagosome formation by bridging vesicles decorated by the small GTPase Rab8a with autophagic compartments labeled by the core autophagy protein ATG8CL . Rab8a is required for pathogen-triggered and starvation-induced but not antimicrobial autophagy , revealing specific trafficking pathways underpin selective autophagy . By subverting Rab8a-mediated vesicle trafficking , PexRD54 utilizes lipid droplets to facilitate biogenesis of autophagosomes diverted to pathogen feeding sites . Altogether , we show that PexRD54 mimics starvation-induced autophagy to subvert endomembrane trafficking at the host-pathogen interface , revealing how effectors bridge distinct host compartments to expedite colonization .
Autophagy is a conserved eukaryotic cellular process that mediates the lysosomal degradation and relocation of cellular cargoes within double-membraned vesicles called autophagosomes ( He and Klionsky , 2009; Leidal et al . , 2020 ) . Although previously considered to be a bulk catabolic pathway tasked with maintaining cellular homeostasis under normal or stress conditions , it is now clear that autophagy can be highly selective ( Zaffagnini and Martens , 2016 ) . Autophagic cargoes are typically captured during autophagosome formation , a complex process regulated by a set of conserved autophagy-related proteins ( ATG ) as well as specialized autophagy adaptors and cargo receptors ( Mizushima et al . , 2011 ) . These captured cargoes are sorted within the autophagosome during maturation of the isolation membrane ( also known as the phagophore ) via the specific interactions between cargo receptors and ATG8 , which decorates the phagophore to serve as a docking platform for cargo receptors ( Slobodkin and Elazar , 2013; Ryabovol and Minibayeva , 2016 ) . The source of the phagophore is still under debate , but its primary source is considered to be the endoplasmic reticulum ( ER ) ( Bernard and Klionsky , 2013; Hamasaki et al . , 2013 ) . As the cargo is captured , the phagophore undergoes massive expansion and is finally sealed to form a mature autophagosome . Therefore , formation of the autophagosome requires additional lipid supplies that are needed for elongation and final sealing of the phagophore . Supporting this view , the essential autophagy protein ATG2 was recently discovered to have lipid transfer activity ( Valverde et al . , 2019; Maeda et al . , 2019; Osawa et al . , 2019 ) . To cope with cellular starvation , cells can rapidly generate hundreds of autophagosomes , conceivably requiring an available supply of lipids . Remarkably , in yeast , lipids mobilized from lipid droplets ( LDs ) were found to fuel autophagosome biogenesis during starvation-induced autophagy , which employs relatively larger autophagosomes ( Shpilka et al . , 2015 ) . In contrast , smaller autophagosomes of the cytosol-to-vacuole targeting ( Cvt ) pathway do not rely on LDs , suggesting that LDs are specifically recruited for starvation-induced autophagy in order to meet the increased demand of lipids required for the biogenesis of larger sized autophagosomes ( Shpilka et al . , 2015; Dupont et al . , 2014 ) . Although poorly characterized , there is accumulating evidence for autophagosome maturation relying on vesicle transport and membrane expansion events which are regulated by secretory pathways involving Rab GTPases ( Ao et al . , 2014 ) . For instance , the mammalian Rab8a that regulates polarized secretion and lipid droplet fusion events is also implicated in autophagy ( Bansal et al . , 2018; Vaibhava et al . , 2012 ) . The model plant Arabidopsis genome contains five closely related Rab8a members organized into the RabE group ( RabE1a-d ) , but whether any of the RabE1 members contribute to autophagy is unknown ( Rutherford and Moore , 2002 ) . Molecular mechanisms governing autophagosome biogenesis , including the sources of membrane precursors required for autophagosome elongation and the transport routes to position these lipid supplies at autophagosome assembly sites remain to be comprehensively described in plants . The discovery of various autophagy cargo receptors uncovered a multitude of selective autophagy pathways implicated in crucial cellular functions ranging from development to immunity in both plants and animals ( Zaffagnini and Martens , 2016 ) . For instance , the plant selective autophagy cargo receptor Joka2/NBR1 mediates antiviral immunity by eliminating viral components ( Hafrén et al . , 2017; Hafrén et al . , 2018; Jung et al . , 2020 ) . Joka2/NBR1 is also required for immunity against bacteria and oomycete pathogens , however , the extent to which defense-related autophagy acts against these pathogens is unknown ( Dagdas et al . , 2018; Üstün et al . , 2018 ) . Consistent with the important role of autophagy in plant immunity , adapted pathogens have evolved strategies to manipulate the host autophagy machinery ( Leary et al . , 2018; Üstün et al . , 2018; Leary et al . , 2019 ) . Plants detect pathogens via immune sensors consisting of surface localized pattern recognition receptors ( PRRs ) and intracellular nucleotide-binding leucine-rich repeat-containing proteins ( NLRs ) . In response , pathogens secrete an arsenal of effector proteins to modulate host immunity and processes to support their virulence . Interestingly , effectors function not only to evade and suppress host immunity but also to mediate nutrient uptake ( Win et al . , 2012 ) . Some filamentous plant pathogens , including the Irish potato famine pathogen Phytophthora infestans , produce hyphal extensions called haustoria that grow into the host cells to facilitate effector delivery and gain access to host nutrients ( Panstruga and Dodds , 2009 ) . A haustorium is a specialized infection structure that remains enveloped by an enigmatic host-derived membrane known as the extra-haustorial membrane ( EHM ) , whose functions and biogenesis are poorly understood ( Whisson et al . , 2016 ) . Notably , we previously showed that Joka2-mediated defense-related autophagy is diverted to the EHM during P . infestans infection ( Dagdas et al . , 2018 ) . The pathogen counteracts this by deploying PexRD54 , a host-translocated RXLR class of effector with five consecutive WY motifs , that targets plant autophagy ( Dagdas et al . , 2016 ) . PexRD54 carries a canonical C-terminal ATG8 interacting motif ( AIM ) that is typically found on autophagy cargo receptors to bind ATG8 ( Maqbool et al . , 2016 ) . Among the diverse set of potato ATG8 members ( Kellner et al . , 2017; Zess et al . , 2019 ) , PexRD54 preferentially binds the ATG8CL isoform and outcompetes Joka2/NBR1 from ATG8CL complexes , thereby disarming defense-related autophagy at the pathogen interface ( Dagdas et al . , 2016; Dagdas et al . , 2018 ) . Intriguingly , PexRD54 does not fully shutdown autophagy as has been shown for animal pathogens that suppress autophagy ( Choy et al . , 2012; Kimmey and Stallings , 2016; Real et al . , 2018; Xu et al . , 2019 ) . Instead , it stimulates formation of autophagosomes that accumulate around the pathogen interface ( Dagdas et al . , 2018 ) . How PexRD54 stimulates autophagy and in what way the pathogen benefits from this remains unknown . Here , we show that PexRD54 mimics carbon starvation-induced autophagy by coupling the host vesicle transport regulator Rab8a ( Speth et al . , 2009; Zheng et al . , 2005; Pfeffer , 2017 ) to autophagosome biogenesis at the pathogen interface . Unlike PexRD54 which activates autophagy by recruiting Rab8a to ATG8CL compartments , PexRD54’s AIM peptide fails to associate with Rab8a , and instead functions as an autophagy inhibitor . Thus , by using an effector protein and its peptide derivative as molecular tools to perturb host-autophagy , we provide insights into not only how vesicle transport processes selectively support autophagosome formation , but also how the pathogen exploits these pathways to undermine plant immunity .
We have previously shown that PexRD54 stimulates formation of ATG8CL-autophagosomes in a macroautophagy dependent manner , and PexRD54 itself is a substrate of host autophagy when expressed in plant cells ( Dagdas et al . , 2016; Maqbool et al . , 2016 ) . During infection , these pathogen-induced autophagosomes are diverted to the EHM in a process that relies on the core autophagy machinery ( Dagdas et al . , 2018 ) . Because AIM-mediated binding of PexRD54 to ATG8CL is essential for the activation of autophagosome formation by the effector protein ( Dagdas et al . , 2016; Maqbool et al . , 2016 ) , we reasoned that PexRD54 could stimulate autophagosome formation by either negative regulation of host autophagy suppressors or through positive regulation of host components to the autophagosome biogenesis sites . To understand how PexRD54 stimulates autophagy , we used N . benthamiana , because: it is widely accepted as a solanaceous model plant in the field of plant-pathogen interactions; enables rapid functional , and cell biology assays using Agrobacterium; and can be infected by P . infestans . To first address whether ATG8 binding is sufficient to trigger autophagosome induction , we generated a PexRD54 truncate comprising only the C terminal AIM peptide ( amino acids 350–381 , hereafter AIMp ) , and compared its potency to stimulate autophagosome formation to the full-length protein . Strikingly , instead of stimulating autophagosome formation , AIMp fused to RFP ( RFP:AIMp ) significantly reduced the number of ATG8CL-autophagosomes in leaf epidermal cells ( Figure 1A–B ) . Compared to RFP:GUS control , expression of RFP:AIMp reduced the number of GFP:ATG8CL-autophagosomes by ~sixfold , whereas cells expressing RFP:PexRD54 had a ~fourfold increase in GFP:ATG8CL-autophagosome numbers as has been shown before ( Figure 1A–B; Dagdas et al . , 2016 ) . The AIMp interacted with ATG8CL in planta ( Figure 1C–D ) , as was previously shown through in vitro studies ( Dagdas et al . , 2016; Maqbool et al . , 2016 ) . However , this association appears to take place mainly in the cytoplasm as the suppression of autophagosome formation by AIMp is such that we hardly observe any GFP:ATG8CL autophagosomes ( Figure 1A–B , E ) . In contrast , RFP:PexRD54 and GFP:ATG8CL produced strong overlapping fluorescence signals which peak at mobile , ring-like ATG8CL-autophagosome clusters that are induced by PexRD54 as described previously ( Dagdas et al . , 2016 ) . However , in cells expressing RFP:GUS control , we did not detect any RFP signal that peaks at ATG8CL-puncta ( Figure 1A–B , E ) . Taken together these results show that binding of PexRD54 to ATG8CL , although necessary , is not sufficient to activate autophagosome biogenesis . This suggests while the full-length protein stimulates autophagy , PexRD54’s AIM peptide functions as an autophagy suppressor . To determine whether AIMp negatively regulates autophagy , we investigated its impact on autophagic flux by monitoring GFP:ATG8CL depletion over time . As there are no specific antibodies available for individual plant ATG8 isoforms , GFP:ATG8 depletion is used to measure autophagic flux in plants rather than quantifying the ratio of lipidated to unlipidated ATG8 isoforms as for other model organisms . Consistent with the AIMp triggered decrease in ATG8CL-autophagosome numbers ( Figure 1A–B ) , RFP:AIMp stabilized GFP:ATG8CL compared to RFP:PexRD54 or the controls RFP:EV and RFP:GUS ( Figure 2A , Figure 2—figure supplement 1 ) . Western blotting showed that in the presence of RFP:AIMp , GFP:ATG8CL was still able to produce a strong protein signal even after 6 days of transient expression . On the other hand , GFP:ATG8CL protein signal was hardly detectable after just four days in the presence of RFP:PexRD54 or RFP:EV , indicating that the AIMp hampers autophagic flux ( Figure 2A ) . Consistent with our previous report that PexRD54 acts as a cargo receptor that activates autophagy ( Dagdas et al . , 2016 ) , we saw rapid decline of RFP:PexRD54 protein levels after 2 days post infiltration ( dpi ) with only free RFP detectable at six dpi , indicating vacuolar processing of RFP:PexRD54 ( Figure 2—figure supplement 1 ) . This was not due lack of PexRD54 expression at later time points as we could still detect mRNA expression of all constructs which are driven by constitutive 35S promoter ( Figure 2—figure supplement 2 ) . In line with this view , confocal microscopy revealed that while RFP:PexRD54 localizes to the cytoplasm at two dpi , RFP signal becomes vacuolar at 4 and 6 dpi ( Figure 2—figure supplement 3 ) , providing further evidence that PexRD54 itself is a substrate of autophagy and is degraded inside the plant vacuole , consistent with the flux assays we reported earlier ( Dagdas et al . , 2016; Maqbool et al . , 2016 ) . In contrast , RFP:AIMp remains mostly cytoplasmic at 2–6 dpi consistent with its function as a suppressor of ATG8 autophagy ( Figure 2—figure supplement 3 ) . However , we did not observe any stabilisation of the GFP control by RFP:AIMp , RFP:PexRD54 or RFP:GUS ( Figure 2—figure supplement 1 ) , indicating that reduced turnover of GFP:ATG8CL by RFP:AIMp is specific and is not due to altered Agrobacterium-mediated expression efficiency . Moreover , the AIM peptide mutated in the conserved ATG8-interacting region was unable to prevent ATG8CL depletion ( Figure 2—figure supplement 4 ) . Altogether , these data indicate that inhibition of autophagy by AIMp relies on ATG8 binding and AIMp can potentially block ATG8 dependent autophagy pathways . We next investigated the extent to which AIMp acts on other potato ATG8 isoforms . RFP:AIMp showed a robust stabilization of all six potato ATG8 isoforms ( Figure 2B ) , demonstrating that AIMp acts as a broad spectrum autophagy suppressor . To support these results , we measured the endogenous NBR1/Joka2 and ATG8 protein levels in N . benthamiana in the presence or absence of the AIMp . Consistently , ectopic expression of RFP:AIMp led to marked increase in both native NbJoka2 and NbATG8 ( s ) levels compared to RFP:GUS expression , whereas RFP:PexRD54 mildly increased levels of native NbJoka2 but not NbATG8s ( Figure 2—figure supplement 5A ) . These results further support the view that PexRD54’s AIMp suppresses autophagy non-selectively , whereas PexRD54 activates ATG8CL-autophagy while neutralizing Joka2-mediated autophagy as shown before ( Dagdas et al . , 2016 ) . Furthermore , AIMp enhanced protein levels of the recombinantly tagged derivatives of Joka2 and ATG8CL that are expressed under constitutive 35S promoter ( Figure 2—figure supplement 5B ) , indicating that AIMp-triggered Joka2 and ATG8CL stabilization is not due to gene expression differences caused by the AIMp . We then explored the potency of the AIMp in autophagy suppression when applied exogenously . For this we custom synthesized PexRD54’s AIM peptide ( AIMpsyn , 10 amino acids at the C terminus ) along with an AIM peptide mutant ( mAIMpsyn ) that has a two amino acid substitution in the conserved residues of the AIM ( Dagdas et al . , 2016 ) , fused to cell penetrating peptides . We first tested their activities in roots of a transgenic N . benthamiana line that stably express GFP:ATG8CL to test the uptake and efficacy of the peptide in different plant tissues and as many plant autophagy studies are performed in roots . Although both AIMpsyn and mAIMpsyn fused to 5-Carboxyfluorescein ( CF-AIMpsyn and CF-mAIMpsyn ) were effectively taken up by the root cells ( Figure 2—figure supplement 6 ) , only the wild-type AIMpsyn reduced the frequency of GFP:ATG8CL-puncta ( by ~10-fold ) compared to mAIMpsyn or buffer control ( Figure 2C–D ) . We also repeated these assays in leaf epidermal cells successfully , however , peptide translocation efficiency and thus autophagosome reduction by AIMpsyn were much lower in leaves compared to root cells ( Figure 2E–F , Figure 2—figure supplement 7 ) . These findings demonstrate that PexRD54’s AIM peptide suppresses autophagy in a variety of tissues , likely through binding to plant ATG8 isoforms with a high affinity and limiting their access to the autophagy adaptors that are essential for induction of autophagy . This implies that full-length PexRD54 carries additional features to stimulate autophagosome formation , by for instance , recruiting and/or manipulating other host components . We next set out to investigate the mechanism of autophagy activation by PexRD54 . Although the underlying molecular mechanisms are largely unknown , autophagosome biogenesis relies on vesicle trafficking and fusion events in yeast and animals ( Singh et al . , 2019; Nair et al . , 2011 ) . We therefore reasoned that in addition to binding ATG8CL , PexRD54 could possibly hijack host vesicle transport machinery to stimulate autophagosome biogenesis . Interestingly , our previous proteomics survey identified Rab8a , which shows the most sequence similarity to Arabidopsis RabE1a , a member of the small Ras-related GTPases that mediate vesicle transport and fusion events , as a candidate PexRD54 interactor ( Dagdas et al . , 2016 ) . We first validated PexRD54-Rab8a association through co-immunoprecipitation assays by co-expressing the potato Rab8a ( herein Rab8a ) or the N . benthamiana Rab8a ( NbRab8a ) with PexRD54 in planta ( Figure 3A , Figure 3—figure supplement 1 ) . Notably , the AIM mutant of PexRD54 ( PexRD54AIM ) that cannot bind ATG8CL still interacted with Rab8a to a similar degree as PexRD54 ( Figure 3A ) , indicating that PexRD54 associates with Rab8a independent of its ATG8CL-binding activity . Consistent with this , the AIMp failed to associate with Rab8a in pull down assays , although it still strongly interacted with ATG8CL ( Figure 3B ) . These results suggest that PexRD54’s N-terminal region preceding the C-terminal AIM mediates Rab8a association . We then investigated the subcellular distribution of PexRD54 and Rab8a through confocal microscopy in leaf epidermal cells . Both stably and transiently expressed Rab8a fused to GFP ( GFP:Rab8a ) produced fluorescent signals at both the plasma membrane and the vacuolar membrane ( tonoplast ) ( Figure 3—figure supplement 2 ) . In addition , GFP:Rab8a localized to mobile puncta ( 0 . 2–0 . 5 μm in diameter ) as well as to larger ring-shaped structures ( Figure 3—figure supplement 2 , Video 1 ) , indicating that Rab8a could be involved in multiple cellular trafficking events . To determine the subcellular compartment ( s ) where PexRD54 associates with Rab8a , we performed co-localisation experiments of GFP:Rab8a with RFP:PexRD54 , RFP:AIMp or RFP:GUS . In line with the pull-down assays ( Figure 3A–B ) , a subset of punctate structures labeled by GFP:Rab8a showed a clear overlap with RFP:PexRD54-puncta , whereas we did not detect any RFP signal peaking at GFP:Rab8a puncta in cells expressing RFP:AIMp or RFP:GUS ( Figure 3C ) . Together , these results demonstrate that PexRD54 associates with Rab8a in an AIM independent manner and raise the possibility that Rab8a could be an important component of PexRD54 driven autophagy . Because Rab GTPases function by converting between GTP and GDP bound states , we decided to generate Rab mutants that mimic the active ( GTP ) and inactive ( GDP bound ) conformations , which are helpful for characterization of the Rab GTPase functions . Although earlier work challenged the applicability of these mutations ( Langemeyer et al . , 2014; Nottingham and Pfeffer , 2014 ) , we reasoned that Rab8a mutants could still be useful to dissect the role of Rab8a in PexRD54 activated autophagy . To determine whether PexRD54 favors a particular form of Rab8a , we produced Rab8a point mutants that we presume to mimic the GTP ( Rab8aQ74L ) or GDP ( Rab8aS29N ) bound states and investigated their subcellular distribution ( Figure 3—figure supplement 3 ) . Unlike GFP:Rab8a , which predominantly labeled the plasma membrane , GFP:Rab8aS29N mutant showed an even distribution at the plasma membrane and the tonoplast ( Figure 3—figure supplement 2B–C ) . In addition , both GFP:Rab8aS29N and GFP:Rab8a marked punctate structures with varying size and shape ( Figure 3—figure supplement 3B–C ) . In contrast , GFP:Rab8aQ74L was mainly trapped in the tonoplast and showed reduced punctate distribution compared to GFP:Rab8a or GFP:Rab8aS29N ( Figure 3—figure supplement 3B–D ) , indicating that the Q74L mutant may not be representing the fully active form of Rab8a as previously reported for other Rab GTPases ( Langemeyer et al . , 2014; Nottingham and Pfeffer , 2014 ) . We next examined the extent to which Rab8a mutants colocalize with PexRD54 . When co-expressed with BFP:PexRD54 , both GFP:Rab8a and GFP:Rab8aS29N consistently produced sharp fluorescence signals that overlap with the typical ring-like autophagosomes marked by PexRD54 ( Figure 3—figure supplement 4A–B ) . However , GFP:Rab8aQ74L showed a similar localization pattern to the GFP control , and mostly did not produce fluorescence signals that peak at BFP:PexRD54-puncta ( Figure 3—figure supplement 4C–D ) . We quantified these observations in multiple independent experiments where GFP:Rab8a and GFP:Rab8aS29N frequently ( 68% , N = 23 ) labeled BFP:PexRD54-puncta , whereas GFP:Rab8aQ74L only did so significantly less often ( 25% , N = 20 ) ( Figure 3—figure supplement 4E ) . As an additional control , we also checked for colocalization between Rab8a mutants and PexRD54’s AIM peptide . However , we did not observe any puncta co-labeled by RFP:AIMp and GFP:Rab8a or any of the Rab8a mutants we tested ( Figure 3—figure supplement 5 ) . These observations are consistent with the results that PexRD54’s AIM peptide fails to associate with Rab8a ( Figure 3B ) and suppresses autophagosome formation ( Figures 1–2 ) . We then compared the binding affinity of PexRD54 to Rab8a and its mutants via in planta co-immunoprecipitation . Rab8aS29N pulled-down PexRD54 more than wild type GFP:Rab8a or GFP:Rab8aQ74L in planta ( Figure 3D ) , suggesting that PexRD54 preferentially associates with the GDP bound state of Rab8a ( S29N ) . co-immunoprecipitation of PexRD54 with Rab8a , Rab8aQ74L , Rab8aS29N , or GFP . FLAG:PexRD54 was transiently co-expressed with GFP:Rab8a , GFP:Rab8aQ74L , GFP:Rab8aS29N , or GFP:EV . IPs were obtained with anti-GFP antiserum and total protein extracts were immunoblotted with GFP and FLAG antisera . Red asterisks indicate expected band sizes . We next explored the potential role of Rab8a in PexRD54-triggered autophagy . To this end , we first investigated the extent to which PexRD54 associates with its two host interactors , Rab8a and ATG8CL . We achieved this through live-cell imaging of Rab8a and ATG8CL co-expressed in combination with either PexRD54 , AIMp , or BFP control . This revealed that BFP:PexRD54 , but not free BFP or BFP:AIMp , localizes to puncta co-labeled by RFP:ATG8CL and GFP:Rab8a ( Figure 4A–C ) . Furthermore , GFP:Rab8a localized to ring-shaped RFP:ATG8CL clusters triggered by BFP:PexRD54 , whereas no such structures occurred in cells expressing BFP control or BFP:AIMp ( Figure 4A–C ) . Notably , our quantitative imaging revealed that , even in the absence of PexRD54 , more than half of RFP:ATG8CL-puncta ( 60% , N = 18 images ) are positively labeled by GFP:Rab8a ( Figure 4A , D ) . However , BFP:PexRD54 expression significantly increased the frequency of GFP:Rab8a-positive RFP:ATG8CL-puncta ( 85% , N = 18 images ) ( Figure 4B , D ) . Conversely , we rarely detected any fluorescent puncta that were co-labeled by GFP:Rab8a and RFP:ATG8CL in the presence of BFP:AIMp ( 6% , N = 18 ) , which strongly suppresses autophagosome formation ( Figure 4C–D ) . These results indicate that a subset of Rab8a localizes to autophagy compartments marked by ATG8CL and this is enhanced by PexRD54 . Consistently , in plants stably expressing GFP:Rab8a , we observed a similar degree of PexRD54-triggered increase in ATG8CL-Rab8a colocalization in ring-shaped ATG8CL-clusters ( Figure 4—figure supplement 1 ) , suggesting that PexRD54 might boost Rab8a recruitment to autophagic compartments . To then gain biochemical evidence for PexRD54-mediated recruitment of Rab8a to ATG8CL compartments , we conducted in planta co-immunoprecipitation assays between Rab8a and ATG8CL in presence of PexRD54 , AIMp , or a control . We extracted proteins at an early time point ( 2 dpi ) to minimize differences in ATG8CL levels , especially because we express proteins that alter autophagic activity ( PexRD54 and AIMp ) . We observed noticeably more RFP:ATG8CL pulled down with GFP:Rab8a in presence of 3xHA:PexRD54 but not 3xHA:AIMp or 3xHA:EV ( Figure 4E ) . As the 3xHA tag is only 27aa and 3 . 3 kDa , expression of smaller constructs such as 3xHA:EV and 3xHA:AIMp are not visible on western blots . Therefore , we validated expression of the HA-tagged constructs by RT-PCR using the RNA extracts from the Agroinfiltrated leaf patches ( Figure 4—figure supplement 2 ) . Altogether , these results show that PexRD54 enhances Rab8a accumulation at ATG8CL-autophagosomes . To further ascertain the functional relationship between PexRD54 and Rab8a , we investigated the degree to which Rab8a associates with autophagy machinery . We monitored the association of RFP:Rab8a with the early autophagosome biogenesis marker protein ATG9:GFP in combination with BFP:PexRD54 , BFP:AIMp , or BFP . Confocal microscopy analyses revealed that ATG9:GFP puncta frequently associate with RFP:Rab8a-labeled vesicles . However , we detected an increased incidence of RFP:Rab8a puncta that are in contact with the mobile ATG9:GFP compartments ( Video 2 ) in the presence of BFP:PexRD54 ( 68% , N = 44 ) compared to free BFP ( 38% , N = 31 ) or BFP:AIMp ( 31% , N = 58 ) ( Figure 4—figure supplement 3 ) , indicating that PexRD54 stimulates association of Rab8a with the autophagosome biogenesis machinery . Notably , in the presence of BFP:PexRD54 , but not BFP:AIMp or BFP:EV , ATG9:GFP puncta showed more proximity to RFP:Rab8a puncta ( Figure 4—figure supplement 3 ) . Furthermore , time-lapse microscopy revealed that these mobile ATG9:GFP compartments co-migrate with BFP:PexRD54/RFP:Rab8a-positive puncta ( Video 2 ) . These results implicate Rab8a in autophagy and indicate that PexRD54 promotes Rab8a recruitment to autophagosome biogenesis sites . We next investigated whether Rab8a is required for PexRD54-mediated autophagy . We measured the impact of NbRab8a silencing on autophagy by quantifying the RFP:ATG8CL-autophagosome numbers . In the N . benthamiana genome , we identified at least four genes encoding full-length Rab8a like proteins ( NbRab8a1-4 ) . We first decided to generate a RNA interference ( RNAi ) construct ( RNAi:NbRab8a1-2 hereafter ) that can target the three prime untranslated regions ( UTR ) of NbRab8a1 and NbRab8a2 , the two closest homologs of the potato Rab8a found in N . benthamiana . RNAi:NbRab8a1-2 showed specific silencing of the NbRab8a1-2 but not NbRab8a3 and NbRab8a4 ( Figure 5—figure supplement 1 ) . In the absence of PexRD54 , silencing of NbRab8a1-2 did not alter the number of RFP:ATG8CL puncta ( Figure 5—figure supplement 2 ) . However , following stimulation of autophagy by transient expression of GFP:PexRD54 , the number of RFP:ATG8CL-puncta/cell in RNAi:NbRab8a1-2 background reduced by half compared to a RNAi:GUS control ( Figure 5—figure supplement 2 ) . This suggests that simultaneous knockdown of NbRab8a1 and NbRab8a2 does not affect basal autophagy , but negatively impacts PexRD54-triggered autophagy . To validate these results , we set up a complementation assay in which we silenced NbRab8a1-2 in transgenic N . benthamiana lines stably expressing the GFP tagged potato Rab8a , which evades RNA silencing because it lacks the three prime UTR targeted by the RNAi:NbRab8a1-2 construct . Consistent with the results obtained in Figure 5—figure supplement 2 , we detected greater than twofold decrease in the number of HA:PexRD54 triggered RFP:ATG8CL-puncta upon delivery of RNAi:NbRab8a1-2 construct in wild-type plants compared to RNAi:GUS ( Figure 5A–B , Figure 5—figure supplement 3 ) . On the other hand , the frequency of RFP:ATG8CL-puncta is not altered by RNAi:NbRab8a1-2 in cells expressing the HA vector control ( Figure 5A–B , Figure 5—figure supplement 3 ) . We then set up a genetic complementation assay using stable transgenic N . benthamiana lines expressing the potato GFP:Rab8a protein that is resistant to silencing by RNAi:NbRab8a1-2 construct . In stable transgenic plants expressing the silencing resistant potato GFP:Rab8a protein , RNAi:NbRab8a1-2 did not change the number of RFP:ATG8CL puncta with or without HA:PexRD54 , compared to cells that express RNAi:GUS control ( Figure 5C–D , Figure 5—figure supplement 3 ) . These results suggest that Rab8a positively regulates PexRD54-mediated autophagy . We next generated a hairpin-silencing construct ( RNAi:NbRab8a1-4 ) that targets all four Rab8a members in N . benthamiana . The RNAi:NbRab8a1-4 construct substantially silenced NbRab8a1 and NbRab8a3 while silencing NbRab8a2 and NbRab8a4 to a lesser extent ( Figure 5—figure supplement 4 ) . RNAi:NbRab8a1-4 did not however silence an unrelated Rab GTPase family member Rab11 ( Figure 5—figure supplement 4 ) . Intriguingly , knockdown of the four Rab8a isoforms using the RNAi:NbRab8a1-4 construct significantly reduced basal ATG8CL ( Figure 5—figure supplement 5 ) and PexRD54-triggered ATG8CL autophagosome numbers ( Figure 5—figure supplement 6 ) . This suggests a potential redundancy in Rab8a function in basal autophagy . Alternatively , NbRab8a3-4 might be involved in basal autophagy , whereas NbRab8a1-2 are not , which needs to be tested in future . To gain further genetic evidence for Rab8a’s positive role in PexRD54-triggered autophagosome formation , we used the dominant negative Rab8a mutant ( N128I ) ( Essid et al . , 2012 ) and measured its impact on formation of RFP:ATG8CL-autophagosomes in the presence or absence of HA:PexRD54 . Consistent with the silencing assays ( Figure 5A–D ) , GFP:Rab8aN128I led to 50% reduction in PexRD54 triggered ATG8CL-autophagosome numbers compared to wild-type GFP:Rab8a ( Figure 5—figure supplement 7 ) . We further validated the dominant negative role of GFP:Rab8aN128I , which significantly reduced PexRD54-triggered ATG8CL puncta compared to GFP control in two independent biological replicates ( Figure 5—figure supplement 8 ) . Since we found that PexRD54 associates with Rab8a and its mutants with varying affinities ( Figure 3D ) , we next checked whether ectopic expression of Rab8a and its mutants ( S29N and Q74L ) have any effect on the formation of PexRD54-autophagosomes . Compared to GFP control , GFP:Rab8a expression led to a slight increase ( ~1 . 5 fold ) in the number of BFP:PexRD54 puncta ( Figure 5—figure supplement 9 ) , suggesting that Rab8a could positively regulate autophagosome formation . Expression of GDP-bound GFP:Rab8aS29N substantially enhanced ( ~three fold ) the frequency of BFP:PexRD54 puncta compared to a GFP control , whereas GTP bound GFP:Rab8aQ74L did not lead to any significant changes in the number of BFP:PexRD54 puncta compared to the GFP control ( Figure 5—figure supplement 9 ) . These results are consistent with the pulldown assays , which revealed stronger interaction between PexRD54 and Rab8aS29N . These findings were initially surprising given the general view that Rab S-to-N mutations in this position lead to less active ( or inactive ) forms that mimic the GDP-bound state , whereas the Q-to-L mutations in this site are assumed to be locked in GTP-bound state that is more active . However , there are reports which revealed that these mutations cannot be generalized ( Langemeyer et al . , 2014; Nottingham and Pfeffer , 2014 ) . Consistent with this view , the GDP-bound form of the mammalian Rab8a ( T22N mutant ) was found to promote lipid droplet ( LD ) fusions , indicating that S-to-N mutation in this Rab protein functions differently . Therefore , further biochemical evidence is required to determine whether these mutations show perturbed GTPase activities . Nevertheless , together with the data presented in Figure 3D , these findings demonstrate that PexRD54-driven autophagy requires Rab8a . To better characterize the autophagy pathway stimulated by PexRD54 , we further investigated the interplay between Rab8a and ATG8CL . The weak interaction of ATG8CL and Rab8a in the absence of PexRD54 ( Figure 4E ) suggests for an indirect association potentially mediated through a host autophagy adaptor . Therefore , we explored whether increased ATG8CL and Rab8a association triggered by PexRD54 is a general hallmark of autophagy activation or is a process that is stimulated through plant selective autophagy adaptors . Because the plant autophagy cargo receptor Joka2 that mediates aggrephagy also binds ATG8CL and stimulates autophagosome formation ( Dagdas et al . , 2016; Jung et al . , 2020 ) , we tested if Joka2 could interact with Rab8a and enhance Rab8a-ATG8CL association . Unlike PexRD54 , Joka2 did not colocalize or interact with Rab8a ( Figure 6A–B ) . Moreover , our quantitative imaging revealed that Joka2 overexpression leads to a reduction of RFP:ATG8CL puncta positively labeled by GFP:Rab8a ( Figure 6C , D ) . This sharply contrasts with the positive impact of PexRD54 on ATG8CL-Rab8a association ( Figure 6C , D ) , indicating that the autophagy pathway mediated by Joka2 is different from the PexRD54 triggered autophagy , and possibly does not require Rab8a function . Supporting this , we did not detect any difference in formation of Joka2-triggered autophagosomes upon NbRab8a1-2 silencing compared to GUS silencing ( Figure 6E–F , Figure 6—figure supplement 1 ) . Collectively , these results indicate that Joka2-mediated autophagy pathway does not involve Rab8a , and the weak association between ATG8CL and Rab8a observed in the absence of PexRD54 is not mediated by Joka2 but potentially through an unknown autophagy adaptor . Since we found that Joka2-mediated aggrephagy pathway does not necessarily rely on Rab8a ( Figure 6 ) , we decided to test whether other plant autophagy pathways employ Rab8a . Because autophagy can be induced through carbon starvation ( Huang et al . , 2019 ) and recent studies revealed a link between Rab8a , lipid droplets ( LDs ) and autophagy induced by carbon starvation in different systems ( Shpilka et al . , 2015; Fan et al . , 2019; Wu et al . , 2014 ) , we tested whether Rab8a-ATG8CL association is altered during autophagy activation following light restriction . We detected a slight yet significant increase in the number of RFP:ATG8CL puncta upon incubation of plants for 24 hr in the dark compared to normal light conditions ( Figure 7A–B ) . Consistently , we detected slightly lower levels of endogenous ATG8s and transiently expressed RFP:ATG8CL following 24 hr dark treatment ( Figure 7—figure supplement 1 ) . However , when GFP:PexRD54 is present , we did not measure any further enhancement of RFP:ATG8CL puncta following 24 hr in the dark , suggesting that PexRD54-mediated autophagy can override or mask starvation-induced autophagy ( Figure 7A–B ) . Furthermore , in plants stably expressing GFP:Rab8a that are light restricted , we detected an increased degree of colocalization between RFP:ATG8CL and GFP:Rab8a ( Figure 7C–D , Figure 7—figure supplement 2 ) , in a similar fashion to enhanced ATG8CL-Rab8a association mediated by PexRD54 ( Figure 4A–D ) . Collectively , these data suggest that PexRD54 mimics carbon-starvation-induced autophagy . Recent studies have revealed that lipid droplets ( LDs ) contribute to carbon-starvation-induced autophagy during light restriction ( Shpilka et al . , 2015; Fan et al . , 2019 ) . In addition , the GDP-bound mutant form of the mammalian Rab8a is enriched at LD contact sites to regulate their fusion ( Wu et al . , 2014 ) . Therefore , we investigated Rab8a association with LDs under normal or starvation conditions . We first checked co-localization between GFP:Rab8a and LDs marked by the orange-red fluorescent fatty acid ( FA ) , BODIPY 558/568 C12 ( herein BODIPY-C12 ) . Confocal microscopy revealed that a small fraction of GFP:Rab8a puncta are labeled by the LD marker BODIPY-C12 under normal light conditions , whereas the frequency of this colocalization increased by ~1 . 4-fold when plants are light restricted ( Figure 7E–F , Figure 7—figure supplement 3 ) . Furthermore , we observed that GFP:Rab8a puncta positive for BODIPY-C12 are also labeled by BFP:ATG8CL ( Figure 7—figure supplement 4 ) . The stronger association of Rab8a and LDs upon light restriction , combined with the finding that LDs are recruited toward autophagosomes during carbon starvation ( Fan et al . , 2019 ) , suggest that Rab8a-LD association could be a hallmark of starvation-induced autophagy . Additionally , we observed that ring-shaped PexRD54 clusters labeled with Rab8a also tightly associate with BODIPY-C12 labeled puncta ( Figure 7G , Figure 7—figure supplements 5–6 ) . However , we did not detect BODIPY-C12 fluorescence signal filling the lumen of the PexRD54-labeled compartments ( Figure 7G , Figure 7—figure supplements 5–6 ) , indicating that fatty acids are likely not the autophagic cargoes of PexRD54 . Rather , we detected a BODIPY-C12 signal at the periphery of autophagosomes marked by PexRD54 , which overlaps with GFP:Rab8a fluorescence signal ( Figure 7G , Figure 7—figure supplements 5–6 ) . Moreover , these PexRD54/Rab8a-clusters are also accompanied by LDs densely labeled with only BODIPY-C12 as they navigate through the cytoplasm ( Video 3 ) . These findings suggest that fatty acids could be one of the potential membrane sources of the autophagosomes stimulated by PexRD54 as observed during carbon-starvation-induced autophagy in other systems ( Shpilka et al . , 2015 ) . To gain further evidence for this , we investigated the colocalization of PexRD54 and Rab8a with the LD structural membrane protein Oleosin ( Siloto et al . , 2006; Fan et al . , 2019; Singh et al . , 2009 ) . Similar to BODIPY-C12-labeled puncta ( Figure 7—figure supplement 5 ) , Oleosin labeled LDs clustered around PexRD54/Rab8a-positive ring-like autophagosomes ( Figure 7—figure supplement 6 ) . Although Oleosin-positive LDs were adjacent to PexRD54 autophagosomes , in contrast to BODIPY-C12 , Oleosin-YFP did not produce fluorescent signal that overlaps with PexRD54/Rab8a ring-like autophagosomes ( Figure 7—figure supplement 6 ) , suggesting that FAs but not LD surface proteins are transferred to PexRD54-triggered autophagosomes . Strikingly , stimulation of autophagy by BFP:PexRD54 , but not Joka2:BFP , led to enhanced association of BODIPY-C12 and GFP:Rab8a-labeled puncta , supporting the hypothesis that PexRD54 mimics autophagy induced during carbon starvation ( Figure 7H , Figure 7—figure supplement 7 ) . Together these results show that unlike the aggrephagy receptor Joka2 , PexRD54 triggers responses similar to carbon-starvation-induced autophagy including induction of autophagosomes and enhanced association of ATG8CL-autophagosomes with Rab8a and LDs ( Figure 6 and Figure 7 ) . However , it is possible that PexRD54 could activate other host autophagy pathways that favor pathogen virulence . To investigate Rab8a’s role in recruiting lipid droplets to PexRD54 puncta , we knocked down Rab8a with the RNAi:NbRab8a1-4 construct and quantified the proportion of PexRD54 autophagosomes associated with Oleosin-labeled LDs . We observed that Rab8a knockdown not only reduces the total amount of PexRD54 autophagosomes but also the proportion of those associated with LDs compared to a silencing control ( Figure 7—figure supplement 8 ) . Finally to gain further evidence of the positive contribution of LDs to PexRD54-triggered autophagosomes , we designed a RNAi silencing construct to knockdown Seipin that has been shown to contribute to LD biogenesis in yeast , animals , and plants ( Cai et al . , 2015; Yang et al . , 2012; Taurino et al . , 2018; Greer et al . , 2020 ) . RNAi-mediated knockdown of the NbSeipin-A significantly reduced the amount of Oleosin-labeled LDs compared to a silencing control ( Figure 7—figure supplements 9–10 ) consistent with the established role of Seipin in LD biogenesis . NbSeipin-A knockdown significantly reduced the amount of PexRD54 puncta in the cell compared to the control ( Figure 7—figure supplement 11 ) , supporting the notion that PexRD54 relies on host LD resources to stimulate autophagosome formation . Together , these data show that PexRD54 requires Rab8a to recruit lipid droplets and that LDs contribute to its induction of autophagosome formation . These findings are in agreement with the findings in yeast and mammalian cells that LDs supply membrane sources required for autophagosome formation during starvation-induced autophagy autophagosomes ( Shpilka et al . , 2015; Dupont et al . , 2014 ) . Our recent work revealed that the perihaustorial niche is a hot spot for the formation of ATG8CL autophagosomes stimulated by PexRD54 ( Dagdas et al . , 2016; Dagdas et al . , 2018 ) . Therefore , we next examined whether Rab8a-PexRD54 association occurs at perihaustorial ATG8CL-autophagosomes . We first checked GFP:Rab8a localization alone in the haustoriated cells . In infected leaf epidermal cells transiently or stably expressing GFP:Rab8a , we detected varying sizes of GFP:Rab8a puncta around the P . infestans haustoria ( Figure 8—figure supplement 1 ) . These structures included ring shaped compartments that are reminiscent of PexRD54-autophagosomes as well as smaller densely packed GFP-positive puncta and large vacuole like structures , indicating that Rab8a could regulate diverse trafficking pathways during infection ( Figure 8—figure supplement 1 , Videos 4–5 ) . To verify that the perihaustorial Rab8a puncta represent the PexRD54-autophagosomes , we imaged infected plant cells which co-express GFP:Rab8a and the autophagosome marker protein RFP:ATG8CL in combination with BFP:PexRD54 , BFP:AIMp , BFP , or Joka2:BFP . Confocal micrographs of haustoriated plant cells showed accumulation of RFP:ATG8CL-autophagosomes around the haustoria which are co-labeled with GFP:Rab8a , and are positive for BFP:PexRD54 but not BFP control ( Figure 8A–C ) . However , formation of perihaustorial puncta co-labeled by RFP:ATG8CL and GFP:Rab8a was suppressed by arresting autophagosome formation through expression of BFP:AIMp ( Figure 8B ) . Notably , we detected a subset of perihaustorial GFP:Rab8a puncta that are not labeled by RFP:ATG8CL , further supporting an ATG8CL-independent haustorial trafficking role for Rab8a ( Figure 8A–D , Figure 8—figure supplement 2 ) . In agreement with these findings , other Rab8 members including RabE1 family have been reported to localize to distinct subcellar compartments including Golgi and Peroxisomes in Arabidopsis ( Zheng et al . , 2005; Cui et al . , 2013 ) . Therefore , we next investigated the potential colocalization of Rab8a and PexRD54 with markers that label organelles such as Golgi , peroxisomes and mitochondria . Co-expression of Rab8a and PexRD54 with GmMan11-49-mCherry ( Golgi marker ) , ScCOX41-29-mCherry ( mitochondria marker ) , or CFP-GLOX ( peroxisome marker ) revealed extensive co-localization of Rab8a with the Golgi marker but not with the markers that label mitochondria or peroxisomes ( Figure 8—figure supplement 3A; Nelson et al . , 2007 ) . However , PexRD54-Rab8a puncta did not show any labeling by any of the organelle markers tested ( Figure 8—figure supplement 3A ) . Consistent with these findings , Rab8a distinctly localized at either the Golgi or PexRD54 puncta , but not at peroxisomes or mitochondria , at the haustorium interface ( Figure 8—figure supplement 3B ) . Therefore , we conclude that Rab8a localization to PexRD54 autophagosomes cannot be explained by Rab8a’s association with these organelles . Intriguingly , we noticed that peroxisomes frequently appear around the PexRD54-Rab8a puncta , suggesting a potential involvement of this organelle in PexRD54-mediated autophagy ( Figure 8—figure supplement 3 ) . We also observed that Rab8a and PexRD54 co-localize with both Bodipy C-12 and oleosin in clusters of vesicles around haustoria , linking Rab8a’s emerging role in lipid trafficking with the pathogen effector and haustorial interface ( Figure 8—figure supplement 4 ) . On the other hand , in line with our findings that the Joka2 pathway does not employ Rab8a ( Figure 6 ) , perihaustorial Joka2:BFP/RFP:ATG8CL puncta and GFP:Rab8a puncta were exclusive to each other ( Figure 8D ) . Consistent with our pull down assays ( Figure 3D ) , we did not detect any sharp GFP:Rab8aQ74L signal at the perihaustorial BFP:PexRD54 puncta ( Figure 8—figure supplement 5 ) . In contrast , GFP:Rab8a , and particularly GFP:Rab8aS29N , produced strong fluorescence signals peaking at perihaustorial BFP:PexRD54 puncta ( Figure 8—figure supplement 5 ) , indicating that both wild-type Rab8a and Rab8aS29N are enriched at the perihaustorial PexRD54 autophagosomes . Taken together , these results demonstrate that Rab8a localizes to distinct compartments that accumulate around the haustorium and PexRD54 stimulates diversion of Rab8a positive LDs to perihaustorial autophagosomes . Localization of Rab8a to Golgi around the haustorium is in agreement with the conserved role of Rab8 in yeast and metazoans to mediate polarized secretion of proteins ( Nielsen et al . , 2008 ) . These findings , combined with our data that PexRD54-mediated subversion of Rab8a to autophagosomes ( Figure 8 ) around the haustoria , imply a potential function of Rab8a in polarized defense-responses . Therefore , we next investigated the possible role of Rab8a in immunity against P . infestans . Simultaneous RNAi-mediated knockdown of all four NbRab8a members led to a consistent increase in disease symptoms and P . infestans hyphal growth ( Figure 8—figure supplement 6 ) . To further determine the role of the Rab8a family in immunity , we conducted a silencing complementation assays using a codon shuffled NbRab8a-1 construct fused to GFP ( GFP:Rab8asyn ) that can evade RNAi ( Figure 8—figure supplement 7 ) . The enhanced susceptibility phenotype caused by RNAi:NbRab8a1-4 silencing was rescued by simultaneous overexpression of GFP:Rab8asyn but not the GFP control , providing further evidence that Rab8a is required for basal resistance against P . infestans ( Figure 8—figure supplement 7 ) . Consistent with knockdown assays , overexpression of dominant negative GFP:Rab8aN128I enhanced plant susceptibility to P . infestans compared to a GFP control ( Figure 8—figure supplement 8 ) . Collectively , these results indicate that Rab8a contributes to plant immunity , whereas PexRD54 could possibly interfere with the defense-related role of Rab8a by subverting a subpopulation of it to autophagic compartments . The observed induction of haustoria targeted autophagosomes prompted the hypothesis that P . infestans could benefit by co-opting host autophagy to support its own growth . Therefore , we next explored how activation of autophagy by PexRD54 affects P . infestans host colonization . Our discovery of the AIMp as an ATG8-specific autophagy inhibitor that can be used to spatiotemporally arrest plant autophagy allowed us to test the impact of autophagy on P . infestans virulence . To this end , we decided to transiently interfere with pathogen-induced autophagy by expressing the AIMp . Our quantitative image analysis revealed that compared to RFP:PexRD54 expression , transient expression of RFP:AIMp led to ~threefold decrease in the number of haustoria that are associated with autophagosomes marked by GFP:ATG8CL ( Figure 9A–B ) . We then measured how autophagy suppression by the AIMp affects P . infestans infection . In multiple independent experiments ( six biological replicates ) , N . benthamiana leaf patches expressing RFP:AIMp showed a consistent reduction of quantitative disease symptoms compared to an RFP vector control ( Figure 9C–D ) . This indicated that AIMp-mediated arrest of host autophagy negatively impacts P . infestans virulence , supporting the hypothesis that PexRD54-triggered autophagy is beneficial to the pathogen . Collectively , these results suggest that P . infestans relies on host autophagy function to support its virulence . This could explain why the pathogen deploys full-length PexRD54 that can activate specific host autophagy pathways while subverting defense-related autophagy , instead of just the AIM peptide .
Effectors are excellent tools to dissect complex biological processes such as autophagy as they often display high target specificity ( Schardon et al . , 2016 ) . In this study , we not only expanded our knowledge of pathogen-triggered autophagy but also discovered an effector derived peptide ( AIMp ) that can specifically block autophagy . Here , the AIMp served as an excellent negative control to understand PexRd54 activities and to interpret the impact of autophagy manipulation by the pathogen . We used it as a tool to perturb PexRD54 activities without directly interfering with Rab8a functions . Most ATG mutants show pleiotropic affects and have non-autophagy-related roles , making it difficult to interpret the outcomes of autophagy inhibition . However , the AIMp blocks autophagy specifically , as it directly acts on ATG8 . Chemical inhibitors are also often used to measure autophagy flux . However , these inhibitors are mostly inefficient and lack the required specificity . We discovered that PexRD54’s AIM peptide is a strong autophagy suppressor effective against all potato ATG8 isoforms ( Figure 2 ) . Our data suggest that AIMp blocks the autophagosome biogenesis step , as both autophagosome formation and autophagic flux are suppressed by the AIMp ( Figures 1–2 ) . Conceivably , the AIM peptide competitively inhibits autophagosome biogenesis by occupying the AIM docking site of ATG8 that accommodates host autophagy regulators ( Noda et al . , 2008 ) . In agreement with our finding , autophagy cargo receptors that bind ATG8 via AIMs were recently shown to stimulate ATG8 ( Chang et al . , 2021 ) . The AIM peptide is a genetically encodable tool , which can enable spatio-temporal arrest of autophagy when expressed under inducible or tissue specific promoters . Thus , this should be of great interest for autophagy studies in plants and other systems , which can overcome the limitations of chemical autophagy inhibitors and autophagy mutants . We also developed AIM peptide derivatives with cell penetrating features , which allow studying the tissue-specific functions of autophagy . The cell penetrating AIM peptide can be used to study autophagy in plants and other eukaryotic systems that are not amenable to genetic manipulation . Autophagosome biogenesis is a complex multi-step process . But how could an effector activate such an intricate process ? We uncovered that PexRD54 either directly or indirectly recruits Rab8a to autophagosome biogenesis sites ( Figures 3–4 ) . Similar channeling of Rab8a to ATG8CL-autophagosomes occurs during carbon starvation ( Figure 7 ) , suggesting that PexRD54 mimics autophagy induction via nutrient deprivation . Interestingly , the mammalian autophagy cargo receptor Optineurin also interacts with both LC3 ( mammalian ATG8 isoform ) and Rab8a in mice ( Bansal et al . , 2018; Vaibhava et al . , 2012 ) . Although the functional implications of these interactions are unclear , a proposed model suggests that Optineurin mediates pre-autophagosomal membrane elongation through anchoring Rab8a to autophagosome assembly sites . Our data is consistent with this model and suggests that PexRD54 recruits Rab8a to facilitate autophagosome formation . The membrane elongation step of autophagosome formation relies on direct transport of lipids from various donor compartments . LDs provide a membrane source for autophagosome biogenesis specifically during starvation-induced autophagy ( Shpilka et al . , 2015 ) . More recently , a conserved acyl-CoA synthetase ( ACS ) from yeast was shown to be mobilized on nucleated phagophores where it locally mediates transport of fatty acids required for phagophore elongation ( Schütter et al . , 2020 ) . But how are these lipid sources mobilized to autophagosome biogenesis sites at first ? Intriguingly , the GDP bound form of the mammalian Rab8a mediates LD fusion events ( Wu et al . , 2014 ) . In line with these reports , we found that PexRD54 associated more strongly with the Rab8aS29N mutant ( presumably the GDP bound form ) and recruits Rab8a to autophagosome biogenesis sites to enhance LD localization around the ATG8CL-foci and stimulate autophagosome formation ( Figures 4 and 7 , Figure 4—figure supplement 3 , Figure 7—figure supplements 4–10 ) . However , we did not find any luminal signal with the FA marker , indicating that FAs are not the likely the cargoes of PexRD54 . Furthermore , we detected FAs labeled with Bodipy but not the LD membrane marker protein Oleosin localized to periphery of the PexRD54 compartments ( Figure 7G , Figure 7—figure supplements 5–6 ) . This suggests that PexRD54 could employ Rab8a to position LDs at ATG8CL nucleation sites to facilitate lipid transfer for autophagosome biogenesis . Interestingly , PexRD54-triggered association of Rab8a and LDs is also enhanced by carbon starvation but not by Joka2 ( Figure 7G–H , Figure 7—figure supplement 6 ) . We propose that upon carbon starvation , the plant uses LDs as an additional membrane source to accommodate for an increase in autophagosome biogenesis , and PexRD54 exploits this process to stimulate autophagy . Nevertheless , Rab8a may not be engaged in all autophagy routes , as it is dispensable for Joka2-mediated autophagy ( Figure 6 ) . This is consistent with the finding that Arabidopsis NBR1 ( Joka2 ) mutants are sensitive to a variety of abiotic stress conditions but not to carbon starvation ( Zhou et al . , 2013 ) . Our results are consistent with the recent findings in yeast , where LDs specifically contribute to starvation-induced autophagy ( Shpilka et al . , 2015 ) . Nevertheless , further assays are needed to determine whether Rab8a is implicated in other autophagy pathways . Altogether , combined with previous findings , we conclude that distinct cellular transport pathways feed autophagosome formation during diverse selective autophagy pathways in plants . Our data revealed that the Rab8a family contributes to basal resistance against P . infestans ( Figure 8—figure supplements 6–8 ) . Rab8a belongs to the Rab8 family of small GTPases that are implicated in polarized secretion events in eukaryotes ( Pfeffer , 2017 ) . Another Rab8 member known as RabE1d was found to contribute to bacterial resistance and regulate membrane trafficking in the model plant Arabidopsis ( Speth et al . , 2009; Zheng et al . , 2005 ) . However , the extent to which Rab8 family members function in immunity remains unknown . Our findings revealed that Rab8a is most likely involved in a diverse range of cellular transport pathways including autophagy . Consistent with the evolutionarily conserved role of Rab8 family members in polarized secretion , we detected Rab8a localization to Golgi around the haustorium . These findings suggest that Rab8a has a function in defense-related secretion that possibly contributes to plant focal immune responses . Further research is required to determine the cargoes and the potential defense-related functions of Rab8a during infection . Autophagy suppression by the AIM peptide leads to reduced pathogen virulence ( Figure 9 ) , indicating that host autophagy inhibition is not favorable to P . infestans . This could explain why P . infestans deploys PexRD54 , which can neutralize defense-related autophagy mediated by Joka2 , while enabling other autophagy pathways that are somewhat beneficial . Intriguingly , the autophagy pathway primed by PexRD54 resembles autophagy induced by carbon starvation , but not to autophagy induced by the aggrephagy cargo receptor Joka2 . In contrast to Joka2 , both PexRD54 and light restriction stimulated enhanced Rab8a-ATG8CL and Rab8a-LD associations ( Figure 6 and Figure 7 ) . Furthermore , their effect on autophagosome formation were not additive ( Figure 7A–B ) . This hints at the possibility that PexRD54 could facilitate nutrient uptake from the host cells by mimicking starvation conditions to stimulate autophagy . This view is further supported by our earlier finding that PexRD54-autophagosomes are diverted to the haustorial interface ( Dagdas et al . , 2016 ) . However , our data suggests that the fate of PexRD54 autophagosomes in infected and uninfected cells could be different , since in uninfected cells PexRD54 is degraded inside the plant vacuole following 4–6 days of ectopic expression ( Figure 2—figure supplements 1–3 ) . Therefore , PexRD54 may or may not be degraded in infected plant cells . PexRD54 is a member of RXLR effectors which typically show high level of expression during 2–5 days of infection . Consequently , even if PexRD54 is depleted via host autophagy machinery during infection , there would be more of it secreted via the haustorium to maintain PexRD54 virulence functions as long as the haustorium is accommodated in the host cells . PexRD54 likely remodels the cargoes engulfed in ATG8CL-autophagosomes targeted to pathogen interface , which are presumably assimilated by the parasite . Whether these cargoes are hydrolyzed in infected plant cells or directly absorbed by the pathogen remains to be determined . Alternatively , but not mutually exclusively , PexRD54 could also help neutralize defense-related host components by engulfing them in secure membrane-bound autophagy compartments . For instance , PexRD54 could promote diversion of Rab8a to autophagy to intervene with non-autophagy related immune functions of Rab8a ( Figure 8 ) . Therefore , we cannot exclude the possibility that PexRD54 could activate other host autophagy pathways that are beneficial to the pathogen , and further research is needed to determine whether PexRD54 could mimic other autophagy inducing conditions . In summary , our findings demonstrate that to support its virulence , P . infestans manipulates plant cellular degradative and transport systems by deploying an effector protein that imitates carbon starvation conditions . It also demonstrates effectors can act as adaptors to bridge multiple host components to modulate complex cellular processes for the benefit the pathogen . Further research is needed ( i ) to determine the cargo of these autophagosomes , ( ii ) whether they are secreted to the pathogen interface and subsequently absorbed by the pathogen , and ( iii ) the molecular players involved in pathogen subverted autophagy .
N . benthamiana WT and transgenic plants ( 35S::GFP:Rab8a and 35S::GFP:ATG8CL ) were grown and maintained in a greenhouse with high light intensity ( 16 hr light/8 hr dark photoperiod ) at 22–24°C . To apply carbon stress , plants were kept under a dark period of 24 hr before images were acquired . Images were acquired 3 days after infiltration ( dpi ) . 35S::GFP:Rab8a and 35S::GFP:ATG8CL lines were produced as described elsewhere ( A Simple and General Method for Transferring Genes into Plants , 1985 ) with the pK7WGF2::Rab8a and pK7WGF2::ATG8CL constructs , respectively . Experiments were conducted in N . benthamiana leaf epidermal cells unless stated otherwise . P . infestans 88069 strain was used in this study . Cultures were grown and maintained by routine passing on rye sucrose agar medium at 18°C in the dark ( van West et al . , 1998 ) . Zoospores were collected from 10 to 14 days old culture by flooding with cold water and incubation at 4°C for 90–120 min . Infection of agroinfiltrated leaves was carried out by addition of 10 mL droplets of zoospore solution at 50 , 000 spores/ml on detached N . benthamiana leaves ( Chaparro-Garcia et al . , 2011 ) . Infection for microscopic experiments carried out on attached leaves . Inoculated detached leaves or plants were kept in humid conditions . Day light/UV images were taken at 7 days post infection and lesion areas were measured in ImageJ . Various constructs used in this study were published previously . GFP:ATG8CL , GFP:PexRD54 , GFP:PexRD54aim , RFP:Rem1 . 3 constructs were previously described in Bozkurt et al . , 2015 . JOKA2:BFP , BFP:EV , BFP:PexD54 , BFP:ATG8CL , ATG9:GFP , 3xHA:EV , JOKA2:3xHA constructs were described in Dagdas et al . , 2016 . GFP:ATG8 1 . 1 , GFP:ATG81 . 2 , GFP:ATG8 2 . 2 , GFP:ATG8 3 . 1 , GFP:ATG8 3 . 2 , GFP:ATG8 four were described in Zess et al . , 2019 . RFP:PexRD54 , RFP:AIMp , RFP:ATG8CL , RFP:Rab8a , Joka2:RFP , BFP:AIMp , 3xHA:PexRD54 , 3xHA:PexRD54aim , and 3xHA:Rd54AIMp constructs were generated by Gibson assembly of each gene PCR fragment into EcoRV digested RFP/tagBFP/HA vectors ( N-terminal fusion for PexRD54 , PexRD54aim , PexRD54AIMp and ATG8CL , C-terminal fusion for Joka2 ) . For YFP:Oleosin , the eYFP fluorophore was split into N-terminal ( residue M1- A155 ) and C-terminal half ( residue D156 - K239 ) . The N-terminal split YFP half was used via a linker peptide RPACKIPNDLKQKVMNH and the C-terminal split YFP half via a linker peptide HNMVKQKLDNPIKCAPR . EcoRV restriction site was added at the end of each linker to allow linearization of the vector and provide an insertion site for subsequent cloning . The DNA fragment encoding Oleosin , together with the linker peptides and restriction sites were amplified from N . benthamiana gDNA using primers NYFP-Oleosin F and Oleosin-CYFP R then assembled into pK7WGF2 vector backbone by Gibson assembly . GFP:Rab8a and RFP:Rab8a constructs were generated by PCR amplification from Solanum tuberosum cDNA using primers GW_StRab8a-1_F GW_StRab8a-1_R followed by Gateway cloning into the entry vector pENTR/D/TOPO ( Invitrogen ) then into the pK7WGF2 ( GFP ) and pH7WGR2 ( RFP ) vectors , respectively . RFP:GUS was created from the pENTR-GUS control plasmid provided in the GATEWAY cloning kit and inserted into pH7WGR2 ( RFP ) via LR reaction . Single residue mutations of Rab8aS29N , Rab8aQ74L and Rab8aN128I were obtained by inverse polymerase chain reaction ( PCR ) amplification of the StRab8a entry clone with the primer pairs ( phosphorylated at five prime ends ) carrying desired mutations; ( i ) Rab8aS29N_F and Rab8aS29N_R; ( ii ) Rab8aQ74L_F and Rab8aQ74L_R; ( iii ) Rab8aN128I_F and Rab8aN128I_R . Templates were then eliminated by one-hour Dpn-I ( New England Biolabs ) restriction digestion at 37°C and the PCR products of mutants were ligated using standard protocols to obtain circular Gateway entry clones carrying desired mutations . Next , the entry clones of Rab8a mutants were recombined into destination vectors pK7WGF2 or pB7RWG2 by Gateway LR reaction . All remaining constructs were amplified from existing constructs previously described ( Bozkurt et al . , 2015; Dagdas et al . , 2016; Dagdas et al . , 2018 ) , using primer pairs GA_RD54_F with GA_RD54_R for PexRD54 , GA_RD54_F with GA_LIR2_R for PexRD54aim , GA_AIMp_F with GA_LIR2_R for PexRD54AIMp , GA_ATG8C_F with GA_ATG8C_R for ATG8CL and GA_NbJoka2_1_Fr with GA_NbJoka2_1_Rv . Silencing constructs for Rab8a were amplified using primer combinations NbRab8A_silF1 and NbRab8A_silR1 , Rab8a1-4RNAi_F1 , Rab8a1-4RNAi_F2 , Rab8a1-4RNAi_F3 , Rab8a1-4RNAi_R1 , Rab8a1-4RNAi_R2 and Rab8a1-4RNAi_R3 , and cloned into the pRNAiGG vector as described elsewhere ( Dagdas et al . , 2018 ) . Silencing of Rab8a was verified using RT-PCR . CFP-GLOX ( Glycolate oxidase ) was generated by PCR amplification from wheat cDNA using primers GLOX-F and GLOX-R followed by Gateway cloning into the entry vector pENTR/D/TOPO ( Invitrogen ) then into the pGWB445 ( CFP ) vector . The sequence for the silencing resistant synthetic Rab8a construct ( Rab8a-1syn ) was obtained by codon shuffling the sequence of Rab8a-1 through a combination of the Integrated DNA Technologies ( IDT ) Codon optimization tool and manual codon shuffling . The construct was then gene synthesized and inserted into the vector by Gibson assembly . AIMpsyn and mAIMpsynAA were synthesized with solid phase peptide synthesis and purified with HPLC . 5 ( 6 ) -Carboxyfluorescein ( Merckmillipore , Chem851082 ) and 5 ( 6 ) -CarboxyTAMRA ( Merckmillipore , Chem851030 ) fluorophores were incorporated with double coupling using DIC/K-Oxyma and HCTU/DIEA . Proteins were transiently expressed by agroinfiltration in N . benthamiana leaves and harvested 2 days post agroinfiltration . Protein extraction , purification and western blot analysis steps were performed as described previously ( Bozkurt et al . , 2011; Dagdas et al . , 2016 ) . Polyclonal anti-GFP ( Chromotek ) , anti-tBFP ( tRFP ) ( Evrogen ) antibodies produced in rabbit , monoclonal anti-RFP ( Chromotek ) and anti-FLAG ( Sigma-Aldrich ) antibodies produced in mouse , monoclonal anti-GFP ( Chromotek ) and anti-HA ( Chromotek ) produced in rat were used as primary antibodies . For secondary antibodies anti-mouse antibody ( Sigma-Aldrich ) , anti-rabbit ( Sigma-Aldrich ) and anti-rat ( Sigma-Aldrich ) antibodies were used . For RNA extraction , 100 mg of leaf tissue was excised and frozen in liquid nitrogen . RNA was then extracted using either GeneJET Plant RNA Purification Kit ( Thermo Scientific ) or TRIzol RNA Isolation Reagent ( Invitrogen ) according to producers’ recommendations . RNA concentration was measured using NanoDrop Lite Spectrophotometer ( Thermo Scientific ) . For Figure 2—figure supplement 2 and µg of extracted RNA underwent DNase treatment using RQ1 RNase-Free DNase ( Promega ) . Two μg of RNA was then used for cDNA synthesis using SuperScript IV Reverse Transcriptase ( Invitrogen ) . cDNA was then amplified using Phusion High-Fidelity DNA Polymerase ( New England Biolabs ) with the appropriate primer pairs ( TKey Resources Table ) . GAPDH is used to normalize the transcription levels of the genes . Microscopy analyses were carried out on live leaf tissue 3–4 days post agroinfiltration . To minimize the damage of live tissue , leaf discs of N . benthamiana were cut using a cork borer and mounted onto Carolina observation gel ( Carolina Biological Supply Company ) . For BODIPY-dodecanoic acid ( BODIPY-C12 , Invitrogen ) staining , 10 μM was infiltrated into the leaf tissue 5 hr prior to observation . For PexRD54 AIM peptide experiments in leaf tissue , a solution of 10 μM of peptide in agroinfiltration buffer or buffer alone was infiltrated in leaves 3 hr prior to observation . For imaging in roots , seedlings were collected at 3 weeks old and the roots placed in 2 mL tubes containing 5 μM peptide solution in agroinfiltration buffer or buffer alone for 3 hr prior to observation . Confocal florescence microscopy was performed using Leica SP5 and SP8 resonant inverted confocal microscope ( Leica Microsystems ) using 63x and 40x water immersion objective , respectively . In order to excite fluorescent tagged proteins , Diode laser excitation was set to 405 nm , Argon laser to 488 nm and the Helium-Neon laser to 561 nm and their fluorescent emissions detected at 450–480 , 495–550 , and 570–620 nm to visualize BFP , GFP , and RFP fluorescence , respectively . Sequential scanning between lines was done to avoid spectral mixing from different fluorophores and images acquired using multichannel . Maximum intensity projections of Z-stack images were processed using ImageJ ( 2 . 0 ) to enhance image clarity . Images for quantification of autophagosome numbers were obtained from Z stacks consisting of 1 . 3 μm depth field multi-layered images with similar settings for all samples unless stated otherwise . To detect and quantify punctate structures in one channel ( green channel or red channel or blue channel ) and to validate colocalization an overlay of two or three channel , where applicable , was acquired ( green channel and/or red channel and/or blue channel ) . Z stacks were separated into individual images with the ImageJ ( 2 . 0 ) program and analyzed . Autopahgosome quantification was done using ATG8CL as a marker protein as described before ( Dagdas et al . , 2016 ) . Colocalization of puncta was analyzed using ImageJ ( 2 . 0 ) with a modified version of the colocalization macro described elsewhere ( Pampliega et al . , 2013 ) with thresholding set as manual to avoid background cytoplasmic signals in each image . Boxplots were generated with mean of punctate numbers generated from stacks obtained in three to six independent biological experiments . Statistical differences were analysed by Welch Two Sample t-test in R . Measurements were significant when p<0 . 05 ( * ) and highly significant when p<0 . 001 ( *** ) . The image processing algorithm calculates the gradient of the image to identify the boundaries of the puncta . We then algorithmically identify the enclosed regions formed by the boundaries and counted the number of puncta in each figure . For the case of co-localisation , the co-ordinates of the centres of the puncta/clusters from each channel were calculated and compared to see if they lie within a small tolerance for each puncta and channel . The puncta/clusters satisfying the abovementioned conditions were considered to be co-localized and were counted . | With its long filaments reaching deep inside its prey , the tiny fungi-like organism known as Phytophthora infestans has had a disproportionate impact on human history . Latching onto plants and feeding on their cells , it has caused large-scale starvation events such as the Irish or Highland potato famines . Many specialized proteins allow the parasite to accomplish its feat . For instance , PexRD54 helps P . infestans hijack a cellular process known as autophagy . Healthy cells use this ‘self-eating’ mechanism to break down invaders or to recycle their components , for example when they require specific nutrients . The process is set in motion by various pathways of molecular events that result in specific sac-like ‘vesicles’ filled with cargo being transported to specialized compartments for recycling . PexRD54 can take over this mechanism by activating one of the plant autophagy pathways , directing cells to form autophagic vesicles that Phytophthora could then possibly use to feed on or to destroy antimicrobial components . How or why this is the case remains poorly understood . To examine these questions , Pandey , Leary et al . used a combination of genetic and microscopy techniques and tracked how PexRD54 alters autophagy as P . infestans infects a tobacco-related plant . The results show that PexRD54 works by bridging two proteins: one is present on cellular vesicles filled with cargo , and the other on autophagic structures surrounding the parasite . This allows PexRD54 to direct the vesicles to the feeding sites of P . infestans so the parasite can potentially divert nutrients . Pandey , Leary et al . then went on to develop a molecule called the AIM peptide , which could block autophagy by mimicking part of PexRD54 . These results help to better grasp how a key disease affects crops , potentially leading to new ways to protect plants without the use of pesticides . They also shed light on autophagy: ultimately , a deeper understanding of this fundamental biological process could allow the development of plants which can adapt to changing environments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2021 | An oomycete effector subverts host vesicle trafficking to channel starvation-induced autophagy to the pathogen interface |
RNA-binding proteins play myriad roles in regulating RNAs and RNA-mediated functions . In bacteria , the RNA chaperone Hfq is an important post-transcriptional gene regulator . Using live-cell super-resolution imaging , we can distinguish Hfq binding to different sizes of cellular RNAs . We demonstrate that under normal growth conditions , Hfq exhibits widespread mRNA-binding activity , with the distal face of Hfq contributing mostly to the mRNA binding in vivo . In addition , sRNAs can either co-occupy Hfq with the mRNA as a ternary complex , or displace the mRNA from Hfq in a binding face-dependent manner , suggesting mechanisms through which sRNAs rapidly access Hfq to induce sRNA-mediated gene regulation . Finally , our data suggest that binding of Hfq to certain mRNAs through its distal face can recruit RNase E to promote turnover of these mRNAs in a sRNA-independent manner , and such regulatory function of Hfq can be decoyed by sRNA competitors that bind strongly at the distal face .
In all kingdoms of life , RNA-binding proteins ( RBPs ) orchestrate the post-transcriptional fates of RNAs by modulating their turnover , structure , and localization , and often as a companion of regulatory RNAs . As one of the most abundant RBPs in bacterial cells , Hfq is an important and prevalent post-transcriptional gene regulator ( Brennan and Link , 2007; Vogel and Luisi , 2011; Updegrove et al . , 2016 ) . Acting as a chaperone of sRNA-mediated gene regulation , Hfq protects small RNAs ( sRNAs ) from degradation and promotes sRNA–mRNA duplex formation ( Brennan and Link , 2007; Vogel and Luisi , 2011; Updegrove et al . , 2016 ) . Binding of sRNAs to target mRNAs further leads to changes in the translation activity and the stability of the mRNAs ( Storz et al . , 2011; Wagner and Romby , 2015 ) . Moreover , other functions of Hfq in regulating translation and degradation of mRNAs independent of the sRNA-mediated regulatory pathway have also been reported ( Urban and Vogel , 2008; Vytvytska et al . , 2000; Pei et al . , 2019; Hajnsdorf and Régnier , 2000; Mohanty et al . , 2004 ) . Loss of Hfq compromises the fitness of bacterial cells , especially under harsh conditions , and abolishes the virulence of bacterial pathogens ( Sobrero and Valverde , 2012; Tsui et al . , 1994 ) . Hfq binds broadly to cellular mRNAs , sRNAs , and ribosomal RNAs ( rRNAs ) ( Sittka et al . , 2008; Tree et al . , 2014; Chao et al . , 2012; Melamed et al . , 2016; Melamed et al . , 2020; Andrade et al . , 2018; Dos Santos et al . , 2020 ) . Hfq interacts with RNAs through multiple interfaces of its homohexameric structure . The surface containing the N-terminal α-helices is referred to as the ‘proximal face’ , whereas the opposite surface is referred to as the ‘distal face’ , and the outer ring as the ‘rim’ ( Figure 1a ) . The proximal face preferably binds U-rich sequences , while the distal face prefers A-rich sequences , with the exact composition of the A-rich motif varying from species to species ( Horstmann et al . , 2012; Link et al . , 2009; Robinson et al . , 2014; Salim et al . , 2012; Someya et al . , 2012 ) . The rim can also interact with UA-rich RNAs through the patch of positively charged residues ( Dimastrogiovanni et al . , 2014; Murina et al . , 2013; Peng et al . , 2014 ) . Finally , the unstructured C-terminal end of Hfq can also interact with certain RNAs to promote the exchange of RNAs ( Robinson et al . , 2014; Dimastrogiovanni et al . , 2014; Santiago-Frangos et al . , 2016 ) . The most refined model describing the interactions between Hfq and sRNAs/mRNAs has sorted sRNAs into two classes ( Schu et al . , 2015 ) . The proximal face of Hfq is generally important for the binding of the sRNAs through their poly-U tail of the Rho-independent terminator , independent of sRNA class . Class I sRNAs ( such as RyhB and DsrA ) then use the rim as the second binding site , whereas class II sRNAs ( such as ChiX and MgrR ) use the distal face as the second binding site ( Schu et al . , 2015 ) . In addition , the preferred target mRNAs of the two classes of the sRNA are proposed to have the complementary binding sites on Hfq , i . e . , class I sRNA-targeted mRNAs ( or class I mRNAs ) binding to the distal face , and class II sRNA-targeted mRNAs ( or class II mRNAs ) binding to the rim , to efficiently form sRNA–mRNA complexes ( Schu et al . , 2015 ) . As a pleiotropic regulator , Hfq establishes additional interactions with other essential protein factors . Particularly , RNase E , the key ribonuclease in the bacterial inner membrane for processing and turnover of rRNAs and mRNAs , is known to interact with Hfq through its C-terminal scaffold region ( Bruce et al . , 2018; Ikeda et al . , 2011; Morita et al . , 2005; Worrall et al . , 2008 ) . The Hfq-RNase E interactions can promote the degradation of the sRNA-targeted mRNA ( Morita et al . , 2005; Afonyushkin et al . , 2005; Pfeiffer et al . , 2009; Prévost et al . , 2011 ) . While Hfq is an abundant RBP in bacterial cells ( Ali Azam et al . , 1999; Kajitani et al . , 1994 ) , it is still considered to be limiting , given the abundance of cellular RNAs . Particularly , in vitro studies on specific sRNAs demonstrate that Hfq binds RNAs tightly with a dissociation constant in the range of nanomolar , and the Hfq–RNA complexes are stable with a lifetime of >100 min ( Fender et al . , 2010; Salim and Feig , 2010; Olejniczak , 2011 ) . However , under stress conditions , induced sRNAs can regulate target mRNAs within minutes , raising a long-standing question of how sRNAs can rapidly access Hfq that might be tightly bound by pre-existing cellular RNAs . To address this question , a model of RNA exchange on Hfq , that is , an RNA can actively displace another RNA from Hfq , was proposed to account for the fast sRNA-mediated stress response ( Vogel and Luisi , 2011; Updegrove et al . , 2016; Wagner , 2013 ) . While in vitro biophysical experiments can measure the affinity of RBP binding to many different RNAs under many different controllable conditions , it is difficult to replicate the concentrations , compartmentalization , crowding , competitive binding , and changes in cellular conditions that can affect the behavior and function of RBPs in live cells . Therefore , the mechanism ( s ) that can recycle Hfq from pre-bound RNAs in live cells remains to be elucidated . To address this question in a cellular context , we measured the diffusivity of Hfq in live Escherichia coli cells , using single-molecule localization microscopy ( SMLM ) ( Manley et al . , 2008 ) , with a rationale that the diffusivity is affected by the molecular weight of the molecules ( Mika and Poolman , 2011 ) , and therefore can report the interactions between Hfq with different cellular components . By measuring Hfq diffusivity under a variety of cellular conditions in combination with other biochemical assays , we demonstrate that Hfq dynamically changes its interactions with different RNAs . Specifically , the two classes of sRNA can gain access to mRNA pre-bound Hfq through different mechanisms . Finally , our data suggest that binding of Hfq to certain mRNAs through its distal face can recruit RNase E to promote turnover of these mRNAs in a sRNA-independent manner .
Hfq was tagged by a photo-switchable fluorescent protein ( FP ) , mMaple3 ( Wang et al . , 2014 ) , at the C-terminus , and the fused hfq gene was integrated into the genomic locus to replace the wild-type ( WT ) hfq ( denoted as ‘hfq-mMaple3’ , Materials and methods ) . The strain harboring hfq-mMaple3 showed comparable growth curve as the WT strain , whereas Δhfq strain showed a growth defect ( Figure 1—figure supplement 1 ) . In addition , Hfq-mMaple3 showed activity comparable to WT Hfq protein , as revealed by northern blots of RyhB-mediated sodB mRNA degradation and MicA-mediated ompA mRNA degradation ( Figure 1—figure supplement 2 ) . We performed single-particle tracking using SMLM in two dimensions ( 2D ) . Images were collected at a rate of 174 frames per second with 2 . 4 ms exposure time for each frame . Imaging conditions and parameters for applying tracking algorithm were optimized using fixed samples as the control ( Figure 1—figure supplement 3 ) . We first tracked Hfq-mMaple3 in live cells grown at exponential phase ( referred to as ‘no treatment’ , or ‘NT’ case ) . In the NT condition , Hfq-mMaple3 exhibited a relatively uniform distribution within the cell ( Figure 1b ) , consistent with the distribution revealed by the earlier live-cell imaging with Hfq tagged by a different FP ( Dendra2 ) ( Persson et al . , 2013 ) . Quantification of Hfq-mMaple3 localization with DNA stained by Hoechst revealed a slightly higher cytoplasm enrichment than nucleoid or membrane localization in the NT condition ( Figure 1c ) . We did not observe a helical organization along the longitudinal direction of the bacterial cell ( Taghbalout et al . , 2014 ) , membrane localization ( Diestra et al . , 2009 ) , or cell pole localization ( Kannaiah et al . , 2019 ) , as reported in a few fixed-cell experiments . In addition , we calculated the one-step displacement ( osd ) speed of individual Hfq-mMaple3 protein at each time step and plotted osd speed as a function of the cellular coordinate in a diffusivity speed map ( Figure 1b ) . The speed map and quantification of the average osd speed suggest that Hfq diffuses similarly within the nucleoid and cytoplasmic region , but slightly more slowly in the membrane ( Figure 1d ) , which could be due to the association with RNase E in the inner membrane . We first tested the effect of mRNA on Hfq-mMaple3 diffusivity by treating the cells with rifampicin ( Figure 2a ) . Rifampicin is an antibiotic that inhibits transcription and results in the loss of most cellular RNAs . We estimated the effective diffusion coefficient ( D ) by fitting the linear function to the mean squared displacement ( MSD ) as a function of time lag ( Δt ) ( Figure 2b ) . Transcription inhibition increased the diffusivity of Hfq-mMaple3 ( Figure 2b ) , suggesting that a fraction of Hfq-mMaple3 proteins is associated with cellular RNAs , consistent with a previous report ( Persson et al . , 2013 ) . We also generated several control constructs: mMaple3 protein alone , mMaple3 fused to an engineered synthetic antibody sAB-70 ( Mukherjee et al . , 2018 ) , and mMaple3-fused scFv-GCN4 used in the SunTag imaging approach ( Tanenbaum et al . , 2014 ) , which are not reported to bind to any RNA . None of them showed any changes in diffusivity upon rifampicin treatment ( Figure 2—figure supplement 1 ) , confirming that the change in diffusivity is not due to the change in the cellular milieu . It should be noted that under the current rifampicin treatment ( 200 µg/mL concentration for 15 min ) , mRNAs , which have an average half-live of 1–4 min ( Chen et al . , 2015 ) , are preferentially degraded , compared to tRNAs ( Svenningsen et al . , 2017 ) and rRNAs ( Blundell and Wild , 1971 ) . While many sRNAs show long half-lives when target-coupled degradation is reduced in the absence of mRNAs upon rifampicin treatment ( Massé et al . , 2003; Zhang et al . , 2002 ) , some sRNAs do have short half-lives ( Vogel et al . , 2003 ) . Therefore , rifampicin treatment might also reduce the fraction of Hfq bound by sRNAs . However , our data suggest that binding of sRNA to RNA-free Hfq or to mRNA-associated Hfq does not change the diffusion coefficients of corresponding species ( see sections below ) . Therefore , we interpreted that the change in the diffusion coefficient upon rifampicin treatment primarily reflected the binding of mRNAs to Hfq . We next introduced point mutations on Hfq-mMaple3 that are shown to affect RNA binding: Q8A and F42A at the proximal face , R16A and R19D at the rim , and Y25D and K31A at the distal face ( Schu et al . , 2015; Zhang et al . , 2013 ) , and imaged these mutant Hfq-mMaple3 proteins under NT and rifampicin treated conditions . With rifampicin treatment , all Hfq mutants exhibited similar diffusivity . However , mutations on different faces changed Hfq diffusivity in the NT case to different levels , suggesting that mutations on different faces changed the ability of Hfq to bind cellular mRNAs . Specifically , both proximal face mutants ( Q8A and F42A ) exhibited similar diffusivity as the WT Hfq-mMaple3; both rim mutations ( R16A and R19D ) had a minor increase in diffusivity under NT condition; and both distal face mutations ( Y25D and K31A ) led to the largest increase in the diffusivity under NT condition , with the diffusion coefficients close to the rifampicin treated case ( Figure 2c ) . Comparison of the WT and the mutant Hfq-mMaple3 proteins supports conclusions that Hfq binds mRNAs in the cell and that binding of mRNAs is primarily achieved through the interactions with the distal face of Hfq , whereas the rim also contributes to the mRNA binding in a minor way . To analyze the subpopulations of Hfq-mMaple3 under different conditions , we plotted one-step squared displacement ( osd2 ) in a histogram . Consistent with D values , distribution of osd2 overall shifted to larger values with rifampicin treatment compared to the NT case ( Figure 2—figure supplement 2 ) . We fit the cumulative probability density function ( CDF ) of osd2 with double populations ( Figure 2—figure supplement 2; Bettridge et al . , 2021 ) . WT Hfq and all mutants with rifampicin treatment showed consistently 86–98% fast population with average osd2 of 0 . 048–0 . 053 µm2 , and the remaining 2–14% slow population with average osd2 of 0 . 0022–0 . 0054 µm2 ( Supplementary file 1 ) . In the NT case , WT Hfq showed ( 57 ± 1 ) % slow population and ( 43 ± 1 ) % fast population . Consistent with the previous interpretation ( Persson et al . , 2013 ) , we assigned the slow population as the mRNA-associated fraction , and the fast population as mRNA-free fraction . This result is consistent with the previous hypothesis that Hfq proteins are largely occupied in the cell ( Vogel and Luisi , 2011; Updegrove et al . , 2016; Wagner , 2013 ) . Y25D and K31A mutants had the most compromised mRNA binding ability , with ( 23 ± 10 ) % and ( 26 ± 4 ) % of the population being mRNA-associated under NT condition , respectively ( Figure 2d ) . It is worth noting that as sRNA binding to RNA-free Hfq or to mRNA-associated Hfq does not change the diffusion coefficients of corresponding species ( see sections below ) , it is possible that a subpopulation of mRNA-free or mRNA-associated Hfq might be sRNA-associated Hfq , or a sRNA-mRNA-Hfq ternary complex , respectively . In addition , the remaining 2–14% slow population after rifampicin treatment may reflect Hfq interactions with rRNAs or possibly DNA ( Malabirade et al . , 2018; Orans et al . , 2020 ) . We next examined the effect of sRNAs on the diffusivity of Hfq-mMaple3 . We ectopically induced expression of different sRNAs , including RyhB , a class I sRNA , ChiX , a class II sRNA , and an sRNA that is less clearly defined between these two classes , SgrS ( Schu et al . , 2015 ) from the same vector . Whereas overexpression of RyhB or SgrS did not cause any noticeable changes in the Hfq-mMaple3 diffusivity or mRNA-associated fraction , overexpression of ChiX dramatically increased its diffusivity and lowered the mRNA-associated fraction ( Figure 3a and b ) . As described above , the distal face is the primary binding site for mRNAs in the cell ( Figure 2c and d ) . Since ChiX requires binding at both the proximal and distal faces , we expected the diffusivity of Hfq to increase after shifting from mRNA-associated Hfq to ChiX-associated Hfq . Due to the relatively small molecular weight of sRNAs ( ~50–300 nucleotides in length ) , sRNA-associated Hfq-mMaple3 has similar diffusivity as free Hfq-mMaple3 . We then checked if ChiX could compete with mRNAs for binding to Hfq in vitro using electrophoretic mobility shift assay ( EMSA ) . A radiolabeled fragment of ptsG mRNA was pre-incubated with purified Hfq protein and then chased with unlabeled ChiX . Consistent with the in vivo results , ChiX can effectively displace ptsG from Hfq ( Figure 3a , b , and e ) . Overexpression of RyhB or SgrS , in contrast , did not cause any significant changes in the Hfq-mMaple3 diffusivity or the corresponding mRNA-associated fraction ( Figure 3a and b ) . We reasoned that there might be two possibilities . First , since class I sRNAs bind through the proximal face and the rim of Hfq , it can bind to the mRNA-free Hfq or co-occupy the mRNA-associated Hfq to generate sRNA-associated Hfq or sRNA–mRNA–Hfq ternary complex , respectively . Second , class I sRNAs cannot effectively compete against mRNAs for Hfq binding; therefore , most Hfq proteins remain associated with mRNAs . To distinguish these two possibilities , we examined the abundance of RyhB and SgrS compared to ChiX . Since the stabilities of RyhB and SgrS are highly dependent on Hfq ( Massé et al . , 2003 ) , if the second hypothesis is correct , then we would expect a much lower cellular level of RyhB and SgrS compared to ChiX . We performed droplet digital PCR ( ddPCR ) in the same conditions as the tracking assays , and the result showed that RyhB or SgrS level was comparable to ChiX ( Figure 3d ) . It should be noted that while ChiX level was almost fivefold of the RyhB level when normalized to the reads of 16S rRNA ( Figure 3d , upper panel ) , ChiX level was about 50% of RyhB level when normalized to the reads from empty vector ( representing the induction fold change ) ( Figure 3d , lower panel ) . We , therefore , reasoned that the difference between ChiX and RyhB when normalizing to the 16S rRNA was very likely due to the different efficiency during reverse transcription and PCR steps for these two targets . SgrS level was higher than ChiX with both normalizations . The observation supports that the stability of RyhB or SgrS is not compromised even though they do not displace mRNAs from Hfq , and therefore the first possibility that they either occupy the free Hfq or co-occupy on Hfq with the mRNA is more likely . To further corroborate the observation that SgrS and RyhB can co-occupy Hfq with non-target mRNA , we performed the same EMSA competition assay using RyhB as an example ( Figure 3f ) . When chasing with increasing concentration of RyhB , the band intensity of ptsG-Hfq complex decreased with an increased intensity of free ptsG and the appearance of an additional upper-shifted band that did not appear when chasing with ChiX ( Figure 3f , red arrow ) . This result supports the possibility of the RyhB-ptsG-Hfq ternary complex formation . In the EMSA assay , we also observed direct displacement of ptsG fragment by RyhB , albeit less efficiently than by ChiX , which was not indicated by the in vivo imaging results . The exact cause of the discrepancy is unclear , but we speculate that it could result from the differences between the cellular conditions and in vitro setting . Nevertheless , the EMSA results still support that RyhB can have different mechanisms to gain access to mRNA-occupied Hfq , and that it is structurally possible to have RyhB co-occupy with a non-target mRNA on Hfq . To summarize , our results collectively suggest that representative sRNAs for both class I and class II sRNAs can access mRNA-occupied Hfq in vivo . It is possible that the mechanisms can be generalized to other members of the two sRNA classes , that is , class I sRNAs can co-occupy the Hfq protein with mRNAs through different binding sites , whereas class II sRNAs can directly compete against the mRNAs at the distal face . Interestingly , fluorescence in situ hybridization ( FISH ) showed a much stronger signal for RyhB compared to ChiX ( Figure 3—figure supplement 2 ) , even though their levels were similar as revealed by ddPCR ( Figure 3d ) . The weaker hybridization signal for ChiX is very likely a reflection of the larger protected region by Hfq on both distal and proximal faces , hindering FISH probe binding . We next sought to understand the molecular features that made ChiX a strong competitor for Hfq binding . When overexpressed in the hfq Q8A-mMaple3 background ( proximal face mutation ) , ChiX lost its capability to displace mRNAs from the mutant Hfq ( Figure 3a and b ) , suggesting that additional binding affinity provided by the proximal face of Hfq is critical for displacing other RNAs from the distal face . E . coli Hfq prefers an ( A-A-N ) n sequence for distal face binding , where N can be any nucleotide , and each monomer binds to one A–A–N repeat ( Robinson et al . , 2014 ) . ChiX contains four AAN motifs ( Figure 3c ) . We tested the effect of AAN motifs on conferring the competitive binding to Hfq over mRNAs . We generated and overexpressed ChiX mutants with one or two AAN motif ( s ) deleted ( Figure 3c ) and found that the fraction of remaining mRNA-associated Hfq increased , when the number of AAN motifs decreased ( Figure 3b and c ) . Notably , the levels of WT ChiX in the hfq Q8A-mMaple3 background , and the ChiX mutants in the WT hfq-mMaple3 background remained similar as the WT ChiX in WT hfq-mMaple3 background ( Figure 3d ) , indicating that the observed difference was not due to a change in the cellular ChiX level . The C-terminal region of RNase E serves as a scaffold for the degradosome protein components ( RNA helicase RhlB , enolase , and polynucleotide phosphorylase [PNPase] ) . Hfq has been demonstrated to interact with the C-terminal scaffold region of RNase E , although it is still under debate whether such interaction is direct or mediated by RNA ( Bruce et al . , 2018; Ikeda et al . , 2011; Morita et al . , 2005; Worrall et al . , 2008 ) . To study whether the interaction with RNase E affects the diffusivity of Hfq , we imaged Hfq-mMaple3 in two RNase E mutant strains . The rne131 mutant strain has RNase E truncated by the last 477 amino acid residues ( Lopez et al . , 1999 ) ; therefore , while it maintains its nuclease activity , this mutant cannot interact with Hfq . The rneΔ14 mutant has a smaller fraction of the C-terminal scaffold ( residues 636–845 ) deleted , encompassing the Hfq , RhlB , and enolase binding regions and two RNA-binding domains ( Leroy et al . , 2002 ) . In both RNase E mutant backgrounds , the diffusivity of WT Hfq-mMaple3 became less sensitive to transcription inhibition by rifampicin compared to the WT rne background ( Figure 4a ) . For WT Hfq-mMaple3 , ~75% of Hfq-mMaple3 became mRNA-free upon rifampicin treatment in the RNase E mutant backgrounds compared to ~89% of mRNA-free Hfq in the WT rne background ( Figure 4b ) . Hfq Y25D-mMaple3 ( the distal face mutant ) , which is deficient in mRNA binding , showed minimal sensitivity to rifampicin treatment in the rne mutant backgrounds , the same as in the WT rne background ( Figure 4a and b ) . These observations suggest that without the Hfq-RNase E interaction , more mRNAs remained bound to Hfq , and hint that Hfq-RNase E interaction may help recycle Hfq from the mRNA-associated form through degradation of mRNAs . To investigate whether the increased mRNA-associated form in the absence of Hfq-RNase E interaction is primarily contributed by the regulation of sRNAs , we imaged Hfq Q8A-mMaple3 in the rne mutant backgrounds . Q8A mutation in the proximal face of Hfq broadly disrupts the binding and stabilization of sRNAs ( Schu et al . , 2015 ) . If sRNA-dependent regulation is the sole pathway in changing Hfq from the mRNA-associated form to the mRNA-free form , we would expect a great increase in the mRNA-associated fraction of the Q8A mutant in the rne mutant backgrounds compared to the WT Hfq with rifampicin treatment . However , we observed a minor difference in the mRNA-associated fraction between Q8A mutant ( ~33% considering both rne131 and rneΔ14 backgrounds ) and the WT Hfq ( ~25% ) ( Figure 4b ) , suggesting that the recycling of Hfq from the mRNA-associated form to the mRNA-free form can be affected by Hfq-RNase E interaction in addition to the sRNA-dependent regulatory pathway . As our results above indicate that Hfq-RNase E interaction contributes to the recycling Hfq from the mRNA-associated form , likely through degradation , we hypothesized that Hfq-RNase E interaction might play a role in the turnover of specific Hfq-bound mRNAs . To test this hypothesis , we used northern blots to measure the half-lives of selected mRNAs that are known to interact with Hfq in four backgrounds: ( 1 ) WT hfq-mMaple3 + WT rne , ( 2 ) WT hfq-mMaple3 + rne131 mutant , ( 3 ) hfq Y25D-mMaple3 ( distal face mutant ) + WT rne , and ( 4 ) hfq Y25D-mMaple3 + rne131 mutant ( Figure 5 ) . To test whether such regulation can occur in the absence of the corresponding sRNAs , in addition to the genetic background of hfq-mMaple3 and rne , we also knocked out the corresponding sRNA regulators of the selected mRNAs ( ∆ryhB∆fnrS for sodB , ∆cyaR∆micA for ompX and ∆ryhB∆spfΔrybB for sdhC; herein simplified as ‘∆sRNA’ ) . The choice of knocked-out sRNAs covers all sRNAs identified in a global mapping of sRNA-target interactions for the selected mRNAs in log phase ( Melamed et al . , 2016 ) . In the WT rne background , these three mRNAs showed 47% to 2 . 2-fold increase in the half-lives in the hfq Y25D-mMaple3 background compared to WT hfq-mMaple3 ( Figure 5d ) . In the rne131 mutant background , while all mRNAs showed increased half-lives of 1 . 4 to 3 . 7-fold compared to the WT rne background , consistent with a compromised activity in the rne131 mutant ( Lopez et al . , 1999 ) , the differences in the mRNA half-lives between WT hfq-mMaple3 and hfq Y25D-mMaple3 backgrounds were largely diminished ( Figure 5e ) . This result indicates that in the absence of Hfq-RNase E interaction , association with Hfq or not does not change the mRNA turnover . To further exclude the contributions by potentially unknown sRNAs , we compared the lifetime of sodB , ompX , and sdhC in the hfq Q8A-mMaple3 background in addition to knocking out corresponding sRNAs ( Figure 5 ) . The half-lives of these mRNAs increased by 16–45% in the hfq Q8A background compared to WT hfq background , smaller than the increase observed in the hfq Y25D background ( Figure 5d ) . The increase of mRNA half-life in the hfq Q8A background can either be due to contributions by unknown sRNA regulators or due to other possible regulatory pathways by Hfq through binding at the proximal face . One of such regulatory pathways may be Hfq-mediated polyadenylation , which involves binding of Hfq at the Rho-independent termination site and promotes mRNA degradation ( Hajnsdorf and Régnier , 2000; Mohanty et al . , 2004 ) . Despite these two possibilities , the increase in the mRNA half-life due to Y25D mutation cannot be fully explained by sRNA-mediated regulation . These results collectively support that besides the sRNA-mediated pathway , Hfq can facilitate the turnover of certain mRNAs by binding to the mRNAs through the distal face and bridging them to RNase E for degradation . RNase E has two mechanisms for substrate recognition and subsequent endonuclease cleavage . RNase E can directly access and cleave RNA substrates with certain sequence preferences ( Chao et al . , 2017; Clarke et al . , 2014 ) . Alternatively , RNase E can recognize the 5’ monophosphate on RNA substrates for catalytic activation ( Mackie , 1998; Jiang and Belasco , 2004; Bandyra et al . , 2012 ) . In this 5’-end dependent mechanism , RppH , a pyrophosphohydrolase , is needed to convert the 5’ triphosphate of the RNA substrate to 5’ monophosphate ( Celesnik et al . , 2007; Deana et al . , 2008 ) . To test whether this Hfq-mediated mRNA turnover is dependent on the 5’-end decapping , we used sdhC as an example , and compared its half-life in the backgrounds of ∆sRNA∆rppH and ∆sRNA∆rppH hfq Y25D . In the ∆sRNA∆rppH background , sdhC’s half-life was ~2 . 6-fold of the half-life in the ∆sRNA background , suggesting that the action of RppH contributes to the endogenous turnover rate of sdhC in general ( Figures 5d and 6a , b; Deana et al . , 2008 ) . However , in the ∆sRNA∆rppH background , additional Y25D mutation on Hfq caused ~90% increase in the half-life compared to that in the presence of WT Hfq ( Figure 6a and b ) . The half-life increase caused by Hfq Y25D mutation in the ∆rppH background was comparable with the half-life increase in the WT rppH background ( 120% increase in half-life in the background of ∆sRNA hfq Y25D compared to ∆sRNA [Figure 5d] for sdhC ) . These results suggest that the decapping by RppH is not required for the Hfq-mediated regulation of mRNA turnover , at least for the case of Hfq regulation on sdhC mRNA . As our model suggests that binding of Hfq to the mRNA through the distal face can regulate the mRNA turnover , we reasoned that sRNAs that can effectively compete for Hfq binding against mRNAs may decoy Hfq from this regulatory function . To test this , we again used sdhC as an example and measured its half-life in the presence of ChiX , which is a strong competitor for Hfq binding ( Figure 3 ) . In the presence of vector control , sdhC exhibited comparable half-life compared to the case without any plasmid ( Figures 5d and 6c , d ) . The presence of WT ChiX increased the half-life by ~70% , whereas the mutant ChiX without two AAN motif did not cause a significant increase in the half-life of sdhC , consistent with its reduced binding ability to Hfq ( Figures 3a , b , 6c and d ) . These results further support our model of Hfq-mediated regulation of mRNA turnover and demonstrate that the presence of strong Hfq binding sRNAs can modulate the strength of Hfq’s regulation .
Using single-particle tracking , we resolved different diffusivity states of Hfq proteins in live cells , reporting on the interactions with different cellular RNAs . Specifically , free Hfq and sRNA-associated Hfq proteins ( collectively termed as ‘mRNA-free Hfq’ ) have a high diffusivity , and association of mRNAs to form mRNA-associated Hfq or sRNA-mRNA-Hfq ternary complex ( collectively termed as ‘mRNA-associated Hfq’ ) reduces the diffusivity of Hfq ( Figure 7a ) . Our results are reminiscent of a previously proposed model of Hfq interacting with sRNAs and mRNAs in a face-dependent manner ( Schu et al . , 2015 ) . During exponential growth , Hfq proteins are largely occupied by mRNAs . The distal face of Hfq is the primary binding site for cellular mRNAs , while the rim has a minor binding role . These observations suggest that the majority of the Hfq-bound mRNAs are class I mRNAs , and a minority are class II mRNAs , consistent with the previous findings that most of the sRNAs are class I sRNAs ( Schu et al . , 2015 ) . Under the conditions when specific sRNAs are highly induced , both classes of sRNAs can easily access Hfq upon induction , albeit with different mechanisms ( Figure 7b ) . Our data demonstrate that class II sRNAs , such as ChiX , can effectively displace class I mRNAs from the distal face , consistent with the proposed RNA exchange model . Interestingly , our data indicate that class I sRNAs do not necessarily need to displace mRNA from Hfq . Instead , they can directly co-occupy Hfq through the binding faces that are non-overlapping with the class I mRNA binding face . In both cases , the mRNA-associated Hfq proteins are in standby mode for sRNA binding if needed . The displacement of mRNA by the class II sRNA requires both the interactions at the proximal face of Hfq and higher AAN motif number to outcompete mRNAs for the binding at the distal face . In addition , we propose that the competitive binding by the class II sRNA is likely to occur stepwise , with binding at the proximal face happening first , followed by the displacement of mRNA from the distal face , which is supported by the observation that with the Hfq proximal face mutation , ChiX cannot displace mRNAs , even with a strong AAN motif . We observed from live-cell tracking experiments that recycling of Hfq from the mRNA-associated form to the mRNA-free form upon rifampicin treatment is compromised in the RNase E mutant backgrounds , where regions including the Hfq binding site are deleted . This observation suggests that RNase E can facilitate the recycling of Hfq from the mRNA-associated form through mRNA degradation . Additional half-life measurements on a few selected mRNAs under various genetic backgrounds further demonstrate that Hfq can facilitate the turnover of certain mRNAs through binding with its distal face and recruiting RNase E , and this regulation is most likely to be independent of their corresponding sRNA regulators . Interestingly , we observed that sRNA competitors , such as ChiX , which can outcompete mRNAs for binding at the distal face , can decoy Hfq from regulating mRNA turnover , the same effect as the distal face Y25D mutation . Similar observation was reported previously that ChiX can titrate Hfq from translationally repressing transposase mRNA ( Ellis et al . , 2015 ) . Considering our mutational work on ChiX-Hfq interaction , it should be possible to engineer synthetic sRNAs to tune Hfq-RNA interactions and Hfq regulatory functions in vivo . Mechanisms of sRNA-independent Hfq-mediated regulation on mRNA turnover have been reported . First , binding Hfq , or Hfq in complex with other proteins such as Crc , at the ribosome binding site of the mRNA can repress translation ( Urban and Vogel , 2008; Vytvytska et al . , 2000; Pei et al . , 2019 ) , therefore indirectly increasing the mRNA degradation due to de-protection of the mRNA by translating ribosomes against RNase E . Since this translation-dependent regulation of Hfq does not require Hfq-RNase E interaction , if this mechanism applied to the mRNAs we tested , we would expect the difference in half-life between the cases of WT hfq and hfq Y25D to be similar in the WT rne and rne131 backgrounds . The observation that the difference in half-life due to Hfq binding is eliminated in the rne131 background suggests that the turnover of these mRNAs is not primarily through regulation at the translation level . Second , binding of Hfq may recruit polyA polymerase ( PAP ) and PNPase to stimulate polyadenylation at the 3’ end , and therefore promote degradation ( Hajnsdorf and Régnier , 2000; Mohanty et al . , 2004 ) , an action that may also involve interactions with the C-terminal scaffold region of RNase E . However , this mechanism also cannot fully explain our results , as Hfq-stimulated polyadenylation prefers Hfq binding at the 3’ termini of mRNAs containing Rho-independent transcription terminator ( Mohanty et al . , 2004 ) , whereas in our selected mRNAs , they do not all utilize Rho-independent termination , and the binding sites of Hfq on the tested mRNAs are within the 5’ UTR and CDS region containing A-rich motif , based on the CLIP-seq analysis of Hfq ( Tree et al . , 2014 ) . Therefore , it is more likely that the regulation for the selected mRNAs is through recruitment of RNase E rather than through Hfq-stimulated polyadenylation mechanism . Nevertheless , we expect Hfq can potentially regulate mRNA turnover through a combination of these mechanisms in a gene-specific manner , which explains our observation that depending on the specific mRNA , Hfq can facilitate the turnover rate through distal face binding to different extent . The endonuclease activity of RNase E has been shown to either be dependent on 5’-monophosphate of the RNA substrate , or independent , with the former mechanism requiring RppH to convert 5’-triphosphate cap to 5’-monophosphate . Using sdhC mRNA as an example , our results demonstrate that the 5’-monophosphate dependent pathway contributes insignificantly to this specific Hfq-mediated regulation of mRNA turnover . While our results reveal that Hfq binding contributes to mRNA turnover through recruitment of RNase E , more mechanistic details remain to be further elucidated . First , it is under debate whether Hfq-RNase E interaction is direct or mediated by RNAs ( Bruce et al . , 2018; Ikeda et al . , 2011; Morita et al . , 2005; Worrall et al . , 2008 ) . While our data show that Hfq can promote mRNA degradation without their matching regulatory sRNAs , it remains to be investigated whether other cellular RNAs may participate in bridging the interaction . Second , while our data demonstrate that deletion of the scaffold region of RNase E abolishes the Hfq-mediated regulation on mRNA turnover , the same scaffold region also includes the binding sites of other degradomes components; thus we cannot exclude the possibility that these protein components also participate in the regulation . Further experiments are needed to answer these mechanistic questions .
Transfer of the Linker-mMaple3-Kan sequence at the 3’ end of chromosomal hfq gene was achieved by following the PCR-based method ( Datsenko and Wanner , 2000 ) with a few modifications . First , a PCR ( PCR1 ) was performed using plasmid pZEA93M as template to amplify the mMaple3 sequence ( oligos EM4314-4293 ) . Then , to add sequence homology of hfq gene , a second PCR ( PCR2 ) was performed using PCR1 product as template ( oligos EM4313-4293 ) . The final PCR product ( PCR3 ) , containing a flippase recognition target ( FRT ) ‐flanked kanamycin resistance cassette , was generated from pKD4 plasmid as template with PCR2 product and oligo EM1690 as primers carrying extensions homologous to the hfq gene . PCR3 was then purified and transformed to WT ( EM1055 ) , hfq Y25D ( KK2562 ) , hfq Q8A ( KK2560 ) , hfq K31A ( AZZ41 ) , or hfq R16A ( KK2561 ) strains containing the pKD46 plasmid using electroporation , to obtain strains with hfq-Linker-mMaple3-Kan , hfq Y25D-Linker-mMaple3-Kan , hfq Q8A-Linker-mMaple3-Kan , hfq K31A-Linker-mMaple3-Kan , and hfq R16A-Linker-mMaple3-Kan , respectively . Hfq mutations F42A and R19D were obtained by performing PCRs on fusion strains KP1867 ( Hfq-linker-mMaple3 ) with oligos EM4704-1690 ( Hfq F42A ) or EM4705-1690 ( Hfq R19D ) . Fragments were then transformed to WT ( EM1055 ) containing the pKD46 plasmid , following induction of the λ Red . P1 transduction was used to transfer the linked FP and the antibiotic resistance gene into a WT ( EM1055 ) , rne131 ( EM1377 ) , or rne∆14 ( EM1376 ) strains . fnrS , micA , spf , chiX , and rppH knockouts were obtained through transformation of PCR products into EM1237 after induction of λ red and selecting for kanamycin or chloramphenicol resistance . P1 transduction was used to transfer the knockout mutations and the antibiotic resistance gene into appropriate strains . Selection was achieved with kanamycin , chloramphenicol , or tetracycline . When necessary , FRT-flanked antibiotic resistance cassettes were eliminated after transformation with pCP20 , as described ( Datsenko and Wanner , 2000 ) . P1 transduction was also used to transfer hfq-Linker-mMaple3-Kan , hfq Y25D-Linker-mMaple3-Kan and hfq Q8A-Linker-mMaple3-Kan in the ΔsRNA or rne131-ΔsRNA strains . All constructs were verified by sequencing and are listed in Supplementary file 2 . Oligonucleotides used for generating constructs are listed in Supplementary file 3 . E . coli MG1655 sRNA genes sgrS , ryhB , and chiX were inserted into the pET15b vector ( kind gift from Perozo lab ) to create plasmids pET15b-RyhB , pET15b-SgrS , and pET15b-ChiX using Gibson Assembly ( in house ) using oligos listed in Supplementary file 2 . The chiX ΔAAN mutants were made using site directed mutagenesis . Primers EH159 , EH160 , and EH161 homologous to chiX while excluding the AAN domain were used to amplify the plasmid . The products were phosphorylated ( NEB M0201S ) and ligated ( NEB M0202S ) before transformation . Cloning of pBAD-micA was performed by PCR amplification of micA ( oligos EM2651-2652 ) on WT strain ( EM1055 ) . The PCR product was digested with SphI and cloned into a pNM12 vector digested with MscI and SphI . Two DNA fragments encoding the Fab fragment of the sAB-70 synthetic antibody were PCR amplified from pRH2 . 2-70-4D5-EA plasmid with OSA996/OSA997 and OSA998/OSA999 primer pairs . The third fragment ( plasmid backbone ) was amplified with OSA992/OSA993 primer pair from pZESA93M plasmid , and these three DNA fragments were joined to construct pZEFabM with RAIR assembly ( Watson and García-Nafría , 2019 ) . Construction of pZEGCN4MSD plasmid encoding the scFv-GCN4 binding protein was performed by amplifying the scFv-GCN4 reading frame ( with oligos OSA1010/OSA1011 ) from pHR-scFv-GCN4-sfGFP-GB1-NLS-dWPRE plasmid . The plasmid backbone was PCR amplified with OSA1008/OSA1009 DNA oligos using pZESA93M as a template . The fragments were joined into pZEGCN4MSD with RAIR assembly . All constructs were verified by sequencing and are listed in Supplementary file 2 . Oligos used for generating the constructs are listed in Supplementary file 3 . Overnight cultures of E . coli strains were diluted by 1:100 in MOPS EZ-rich defined medium ( Teknova ) . 0 . 2% glucose was used as the carbon source for imaging Hfq-mMaple3 WT and mutants under NT and rifampicin treated conditions . 0 . 2% fructose was used as the carbon source with 100 µg/mL ampicillin for cases with sRNA overexpression , mMaple3 control , mMaple3 fused sAB-70 , and mMaple3 fused scFv-GCN4 . Cultures were grown at 37°C aerobically . Plasmid-encoded sRNAs were induced by 1 mM IPTG when the OD600 of the cell culture was ~0 . 1 . Induced cells were grown for ~45 min before imaging . Plasmid-encoded mMaple3 protein ( with 100–400 μM IPTG ) , mMaple3 fused sAB-70 ( with 1 mM IPTG ) , and mMaple3 fused scFv-GCN4 ( with 1 mM IPTG ) were expressed and imaged in the same way . For the rifampicin treatment , rifampicin was added to a final concentration of 200 μg/mL when the OD600 of the cell culture was ~0 . 2 , and the cells were incubated for 15 min before imaging . The bacterial strains were grown overnight in LB or MOPS EZ-rich medium containing 0 . 2% glucose . Cultures were diluted to 6 × 106 cells/mL in their respective medium and samples were prepared in triplicate by mixing 50 µL of cells and 50 µL of fresh medium to obtain 3 × 106 cells/mL . Assay was performed in Microtest plate , 96-well , flat base , polystyrene , sterile ( Sarstedt ) and growth was monitored using Epoch 2 Microplate Spectrophotometer reader ( BioTek ) with the following settings: OD = 600 nm , Temperature = 37°C , Reading = every 10 min for 22 hr , Continuous shaking . Total RNA was extracted following the hot-phenol protocol as described ( Aiba et al . , 1981 ) . To test the function of the mMaple3-tagged Hfq and compare that with the WT Hfq , cells were grown in LB to the OD600 of 0 . 5 and either RyhB was induced by adding of 2 . 2′-dipyridyl in a WT hfq or in an hfq-mMaple3 background , or MicA was induced by addition of 0 . 1% arabinose ( ara ) in a ΔmicA WT hfq or in a ΔmicA hfq-mMaple3 background ( pBAD-micA ) . Determination of RNA half-life was performed in MOPS EZ-rich defined medium ( Teknova ) with 0 . 2% glucose by addition of 500 μg/mL rifampicin to the culture at the OD600 of 0 . 5 before total RNA extraction . Northern blots were performed as described previously ( Desnoyers and Massé , 2012 ) with some modifications . Following total RNA extraction , 5–10 μg of total RNA was loaded on polyacrylamide gel ( 5% acrylamide 29:1 , 8 M urea ) and 20 μg was loaded on agarose gel ( 1% , 1× MOPS ) . Radiolabeled DNA and RNA probes used in this study are described in Supplementary file 3 . The radiolabeled RNA probes used for northern blot analysis were transcribed with T7 RNA polymerase from a PCR product to generate the antisense transcript of the gene of interest ( Desnoyers et al . , 2009 ) . Membranes were then exposed to phosphor storage screens and analyzed using a Typhoon Trio ( GE Healthcare ) instrument . Quantification was performed using the Image studio lite software ( LI-COR ) . The decay rate of mRNA degradation was calculated as previously described ( Moffitt et al . , 2016 ) . Briefly , the intensity of northern blot at each time point upon adding rifampicin was normalized to the intensity at time zero and was fit by a piecewise function in the log space:lnI ( t ) ={lnI ( 0 ) , t≤ αlnI ( 0 ) −k ( t−α ) , t>αwhere I ( t ) is the normalized intensity at time t , I ( 0 ) is the normalized intensity at time zero , k is the rate of exponential decay , and α is the duration of the initial delay before the exponential decay begins . The reported half-lives ( τ ) are calculated by τ=log ( 2 ) /k . Droplet digital PCR ( ddPCR ) was performed on total RNA extracted following the hot-phenol protocol ( Aiba et al . , 1981 ) from cells grown in MOPS EZ-rich defined medium containing 0 . 2% fructose ( Teknova ) with 50 µg/mL ampicillin . 1 mM IPTG was added at OD600 = 0 . 1 for 1 hr before total RNA extraction . Samples were treated with 8 U Turbo DNase ( Ambion ) for 1 hr . RNA integrity was assessed with an Agilent 2100 Bioanalyzer ( Agilent Technologies ) . Reverse transcription was performed on 1 . 5 µg total RNA with Transcriptor reverse transcriptase , random hexamers , dNTPs ( Roche Diagnostics ) , and 10 U of RNase OUT ( Invitrogen ) following the manufacturer’s protocol in a total volume of 10 µL . ddPCR reactions were composed of 10 µL of 2× QX200 ddPCR EvaGreen Supermix ( Bio-Rad ) , 10 ng ( 3 µL ) cDNA , 100 nM final ( 2 µL ) primer pair solutions , and 5 µL molecular grade sterile water ( Wisent ) for a 20 µL total reaction . Primers are listed in Supplementary file 3 . Each reaction mix ( 20 µL ) was converted to droplets with the QX200 droplet generator ( Bio-Rad ) . Droplet-partitioned samples were then transferred to a 96-well plate , sealed , and cycled in a C1000 deep well Thermocycler ( Bio-Rad ) under the following cycling protocol: 95°C for 5 min ( DNA polymerase activation ) , followed by 50 cycles of 95°C for 30 s ( denaturation ) , 59°C for 1 min ( annealing ) , and 72°C for 30 s ( extension ) followed by post-cycling steps of 4°C for 5 min and 90°C for 5 min ( Signal stabilization ) and an infinite 12°C hold . The cycled plate was then transferred and read using the QX200 reader ( Bio-Rad ) either the same or the following day post-cycling . The concentration reported is copies/μL of the final 1× ddPCR reaction ( using QuantaSoft software from Bio-Rad ) ( Taylor et al . , 2015 ) . Hfq was purified following the previously described procedure ( Prévost et al . , 2007 ) with modifications . Briefly , strain EM1392 containing pET21b-hfq was grown at 37°C in LB medium supplemented with 50 μg/mL ampicillin and 30 μg/mL chloramphenicol until it reached an OD600 = 0 . 6 . Hfq expression was induced by addition of 5 mM IPTG ( Bioshop ) for 3 hr . Cells were pelleted by centrifugation ( 15 min , 3825 g ) and resuspended in 4 mL Buffer C ( 50 mM Tris-HCl pH 7 . 5 , 1 mM EDTA , 50 mM NH4Cl , 5% glycerol ) ( Zhang et al . , 2002 ) supplemented with 30 U Turbo DNase ( Ambion ) . Cells were lysed by sonication for 4 min ( amplitude 25% , cycles of 5 s sonication , 5 s on ice ) and samples were cleared by centrifugation ( 45 min , 12 , 000 g ) . The supernatant was incubated at 80°C for 10 min , centrifuged again ( 20 min , 12 , 000 g ) , and cleared by filtration . The protein extract was loaded onto a 1 mL HiTRAP Heparin column HP ( GE Healthcare Life Sciences , 17-0406-01 ) equilibrated with Buffer A ( 50 mM Tris-HCl pH 8 . 0 , 50 mM NaCl , 50 mM KCl , 1 mM EDTA , 5% glycerol ) . After washes , the protein was eluted with a linear NaCl gradient ( 0 . 05–1 M NaCl ) in Buffer A . Fraction samples were loaded on SDS-PAGE and stained with Coomassie-Blue . Hfq-containing fractions were dialyzed against a dialysis buffer ( 50 mM Tris-Cl pH 7 . 5 , 1 mM EDTA pH 8 . 0 , 5% Glycerol , 0 . 25 M NH4Cl ) . Glycerol concentration was brought up to 10% and protein content was quantified by BCA assay ( Thermo Scientific ) . DNA templates containing a T7 promoter were synthesized by PCR amplification on genomic DNA using oligonucleotides EM88-EM1978 ( T7-ryhB ) , T7-ChiX ( F ) -T7-ChiX ( R ) ( T7-chiX ) , or T7-ptsG ( F ) -T7-ptsG ( R ) ( T7-ptsG ) . Briefly , templates were incubated for 4 hr at 37°C in RNA Transcription Buffer ( 80 mM HEPES-KOH pH 7 . 5 , 24 mM MgCl2 , 40 mM DTT , 2 mM spermidine ) in the presence of 5 mM NTP , 40 U porcine RNase Inhibitor ( in house ) , 1 μg pyrophosphatase ( Roche ) , and 10 μg purified T7 RNA polymerase ( in house ) . Samples were treated with 2 U Turbo DNase ( Ambion ) and purified on polyacrylamide gel ( 6% acrylamide:bisacrylamide 19:1 , 8 M urea ) . When necessary , transcripts were dephosphorylated using 10 U Calf Intestinal Phosphatase ( NEB ) and were 5’ end-radiolabeled with [γ-32P]-ATP using 10 U T4 polynucleotide kinase ( NEB ) . Radiolabeled transcripts were purified on polyacrylamide gel ( 6% acrylamide:bisacrylamide 19:1 , 8 M urea ) . EMSA were performed as previously described ( Morita et al . , 2012 ) . To determine binding affinity of Hfq to RyhB , ChiX , and ptsG , radiolabeled RNA was heated for 1 min at 90°C and put on ice for 1 min . RNA was diluted to 20 nM in modified Binding Buffer 2 ( 10 nM Tris-HCl pH 8 . 0 , 1 mM DTT , 1 mM MgCl2 , 20 mM KCl , 10 mM Na2HPO4-NaH2PO4 pH 8 . 0 , 12 . 5 μg/mL yeast tRNA ) and mixed with specific concentrations of Hfq ( 0–200 nM ) . Samples were incubated for 15 min at 37°C and reactions were stopped by addition of 1 μL of non-denaturing loading buffer ( 1× TBE , 50% glycerol , 0 . 1% bromophenol blue , 0 . 1% xylene cyanol ) . For competition assays , 20 nM of radiolabeled ptsG was first incubated for 15 min at 37°C with 100 nM Hfq ( as described above ) . Specific concentrations of RyhB or ChiX ( 0–100 nM ) were added to the samples and incubation was carried out for 15 min at 37°C . Reactions were stopped by addition of 1 μL non-denaturing loading buffer . Samples were loaded on native polyacrylamide gels ( 5% acrylamide:bisacrylamide 29:1 ) in cold TBE 1X and migrated at 50 V , at 4°C . Gels were dried and exposed to phosphor storage screens and analyzed using a Typhoon Trio ( GE Healthcare ) instrument . When applicable , quantification was performed using the Image studio lite software ( LI-COR ) and data was fitted using nonlinear regression ( GraphPad Prism ) . Sample preparation for fixed cells was performed mostly according to the protocol previously reported ( Park et al . , 2018a; Fei et al . , 2015 ) . Briefly , ~10 mL of cell culture was collected and fixed with 4% formaldehyde in 1× PBS for 30 min at room temperature ( RT ) . The fixed cells were then permeabilized with 70% ethanol for 1 hr at RT . Permeabilized cells can be stored in 70% ethanol at 4°C until the sample preparation . FISH probes were designed and dye-labeled as in the previous reports ( Park et al . , 2018a; Fei et al . , 2015 ) . Hybridization was performed in 20 μL of hybridization buffer ( 10% dextran sulfate [Sigma D8906] and 10% formamide in 2× SSC ) containing specific sets of FISH probes at 30°C in the dark overnight . The concentration of FISH probes was 50 nM . After hybridization , cells were washed three times with 10% FISH wash buffer ( 10% formamide in 2× SSC ) at 30°C . Imaging was performed on a custom built microscopy setup as previously described ( Park et al . , 2018b ) . Briefly , an inverted optical microscope ( Nikon Ti-E with 100× NA 1 . 49 CFI HP TIRF oil immersion objective ) was fiber-coupled with a 647 nm laser ( Cobolt 06–01 ) , a 561 nm laser ( Coherent Obis LS ) and a 405 nm laser ( Crystalaser ) . A common dichroic mirror ( Chroma zt405/488/561/647/752r-UF3 ) was used for all lasers , but different emission filters were used for different fluorophores ( Chroma ET700/75M for Alexa Fluor 647 and Chroma ET595/50M for mMaple3 ) . For imaging Hoechst dye , a LED lamp ( X-Cite 120LED ) was coupled with a filter cube ( Chroma 49000 ) . The emission signal was captured by an EMCCD camera ( Andor iXon Ultra 888 ) with slits ( Cairn OptoSplit III ) , enabling fast frame rates by cropping the imaging region . During imaging acquisition , the Z-drift was prevented in real time by a built-in focus lock system ( Nikon Perfect Focus ) . For live-cell single particle tracking , 1 mL of cell culture was centrifuged at 1500 g for 5 min and 970 µL of the supernatant was removed . The remaining volume was mixed well and ~1 . 5 µL was covered by a thin piece of 1% agarose gel on an ethanol-cleaned-and-flamed coverslip sealed to a custom 3D printed chamber . The agarose gel contained the same concentration of any drug or inducer used in each condition . Exceptions include rifampicin , which was at 100 μg/mL in the gel , due to high imaging background caused by high concentration of rifampicin , and IPTG for mMaple3 alone control culture , which was eliminated in the gel , due to the high abundance mMaple3 already induced by IPTG in the culture . The power density of the 561 nm laser for single-particle tracking was ~2750 W/cm2 , and the power density of the 405 nm laser was ~7 W/cm2 ( except for mMaple3 alone control where ~4 . 5 W/cm2 was used due to high abundance of mMaple3 ) . 1 . 5× Tube lens was used for the microscope body , and 2 × 2 binning mode was used for the camera . In this way , the effective pixel size became larger ( 173 nm instead of original 130 nm ) , receiving 77% more photons per pixel . Ten frames with 561 nm excitation were taken after each frame of 405 nm photo-conversion . About 13 , 000 frames were collected per movie at a rate of 174 frames per second . For fixed-cell control experiment for tracking parameter optimization , imaging was performed using the exact same imaging parameters as in live-cell measurements for a fair comparison . In cases imaging DNA together , Hoechst dye ( Thermo 62249 ) was added to the ~30 µL of cell culture before imaging at ~20 µM final concentration and imaged by the LED lamp ( 12% ) with 500 ms exposure time . Imaging acquisition was conducted by NIS-Element ( Nikon ) software , at RT . The SMLM images are reconstructed as previously descried ( Fei et al . , 2015 ) , by a custom code written in IDL ( Interactive Data Language ) . Briefly , all the pixels with an intensity value above the threshold were identified in each frame . The threshold was set at three times of the standard deviation of the individual frame pixel intensity . Among those pixels , the ones having larger values than surrounding pixels in each 5 × 5 pixel region were identified as possible peak candidates , and 2D Gaussian function was fit to a 7 × 7 pixel region surrounding these candidates . Candidates with failed fitting were discarded , and precise peak positions were defined for the remaining ones . The horizontal drift , which often occurred during the imaging acquisition , was corrected by fast Fourier transformation analysis . We used a MATLAB coded tracking algorithm to generate diffusion trajectories , which was modified by Sadoon and Yong ( Sadoon and Wang , 2018 ) based on the previously developed code ( Crocker and Grier , 1996 ) . Per each time step of ~5 . 76 ms , 400 nm was empirically chosen to be the maximum one-step displacement to reduce artificial diffusion trajectories connected between different molecules , using a fixed cell sample as a control ( Figure 1—figure supplement 3B ) . Trajectories longer than five time steps were used to calculate effective diffusion coefficient ( D ) . MSD as a function of time lag ( Δt ) was fit with a linear function ( MSD = D×Δt ) . D values are reported in related figures . For analysis using one-step displacement ( osd ) , trajectories longer than three time steps were used . Enrichment at a certain region ( nucleoid , membrane , or cytoplasm ) of a cell is defined as follows: ( #oflocalizationsintheregion ) / ( total#oflocalizationsinthecell ) ( areaoftheregion ) / ( totalareaofthecell ) Here the area of a cell refers to the two-dimensional area of the cell from the differential interference contrast ( DIC ) image . The area of the nucleoid region was defined from the Hoechst image ( nucleoid staining ) and calculated by our custom MATLAB code ( Reyer et al . , 2018 ) . Membrane region was determined as the boundary region from the DIC image , and the cytoplasm region was defined as the total cell region minus the nucleoid and the membrane regions . The analysis of mRNA-associated and mRNA-free population of Hfq was performed by double population fitting of the cumulative probability density function ( CDF ) of one-step squared displacement ( osd2 ) according to a previous report ( Bettridge et al . , 2021 ) :CDF ( osd2 ) =1−∑i=1nPie−osd2/4Di′twhere n is the number of diffusion states , D’i is the diffusion coefficient of ith state , Pi is the fraction of ith state population , and ∑i=1nPi=1 . We found that a two-state model ( n=2 ) fit better than one-state model ( n=1 ) , whereas a three-state model ( n=3 ) did not further improve the fitting ( Figure 2—figure supplement 2a ) . Therefore , we used a two-state model for fitting all Hfq tracking data , and the fast-diffusing state ( D1 , P1 ) was assigned as the mRNA-free fraction and the slow-diffusing state ( D2 , P2 ) assigned as the mRNA-associated . CDFs of rifampicin treatment cases in the WT rne background were fit first , the D1 range of rifampicin treatment cases under WT rne , that is , <D1 , rif>±2×std ( D1 , rif ) , was then used to constrain the D1 values in the fitting for all other cases ( Figure 2—figure supplement 2c and Supplementary file 1 ) . All CDF fittings were conducted in OriginPro with Levenberg-Marquardt iteration algorithm . Osd speed was calculated as osd/Δt0 ( Δt0 is the time interval between two consecutive frames , i . e . , 5 . 76 ms ) . For comparison with D values from the linear fitting of MSD , one-step diffusion coefficient Di≡4D’i values were reported in Supplementary file 1 and Figure 2—figure supplement 2d . | Messenger RNAs or mRNAs are molecules that the cell uses to transfer the information stored in the cell’s DNA so it can be used to make proteins . Bacteria can regulate their levels of mRNA molecules , and they can therefore control how many proteins are being made , by producing a different type of RNA called small regulatory RNAs or sRNAs . Each sRNA can bind to several specific mRNA targets , and lead to their degradation by an enzyme called RNase E . Certain bacterial RNA-binding proteins , such as Hfq , protect sRNAs from being degraded , and help them find their mRNA targets . Hfq is abundant in bacteria . It is critical for bacterial growth under harsh conditions and it is involved in the process through which pathogenic bacteria infect cells . However , it is outnumbered by the many different RNA molecules in the cell , which compete for binding to the protein . It is not clear how Hfq prioritizes the different RNAs , or how binding to Hfq alters RNA regulation . Park , Prévost et al . imaged live bacterial cells to see how Hfq binds to RNA strands of different sizes . The experiments revealed that , when bacteria are growing normally , Hfq is mainly bound to mRNA molecules , and it can recruit RNase E to speed up mRNA degradation without the need for sRNAs . Park , Prévost et al . also showed that sRNAs could bind to Hfq by either replacing the bound mRNA or co-binding alongside it . The sRNA molecules that strongly bind Hfq can compete against mRNA for binding , and thus slow down the degradation of certain mRNAs . Hfq could be a potential drug target for treating bacterial infections . Understanding how it interacts with other molecules in bacteria could provide help in the development of new therapeutics . These findings suggest that a designed RNA that binds strongly to Hfq could disrupt its regulatory roles in bacteria , killing them . This could be a feasible drug design opportunity to counter the emergence of antibiotic-resistant bacteria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2021 | Dynamic interactions between the RNA chaperone Hfq, small regulatory RNAs, and mRNAs in live bacterial cells |
Transmission of respiratory pathogens such as SARS-CoV-2 depends on patterns of contact and mixing across populations . Understanding this is crucial to predict pathogen spread and the effectiveness of control efforts . Most analyses of contact patterns to date have focused on high-income settings . Here , we conduct a systematic review and individual-participant meta-analysis of surveys carried out in low- and middle-income countries and compare patterns of contact in these settings to surveys previously carried out in high-income countries . Using individual-level data from 28 , 503 participants and 413 , 069 contacts across 27 surveys , we explored how contact characteristics ( number , location , duration , and whether physical ) vary across income settings . Contact rates declined with age in high- and upper-middle-income settings , but not in low-income settings , where adults aged 65+ made similar numbers of contacts as younger individuals and mixed with all age groups . Across all settings , increasing household size was a key determinant of contact frequency and characteristics , with low-income settings characterised by the largest , most intergenerational households . A higher proportion of contacts were made at home in low-income settings , and work/school contacts were more frequent in high-income strata . We also observed contrasting effects of gender across income strata on the frequency , duration , and type of contacts individuals made . These differences in contact patterns between settings have material consequences for both spread of respiratory pathogens and the effectiveness of different non-pharmaceutical interventions . This work is primarily being funded by joint Centre funding from the UK Medical Research Council and DFID ( MR/R015600/1 ) .
Previous outbreaks of Ebola ( Mbala-Kingebeni et al . , 2019 ) , influenza ( Khan et al . , 2009 ) , and the ongoing COVID-19 pandemic have highlighted the importance of understanding the transmission dynamics and spread of infectious diseases , which depend fundamentally on the underlying patterns of social contact between individuals . Together , these patterns give rise to complex social networks that influence disease dynamics ( Eubank et al . , 2004; Ferrari et al . , 2006; Firth et al . , 2020; Zhang et al . , 2020 ) , including the capacity for emergent pathogens to become endemic ( Ghani and Aral , 2005; Jacquez et al . , 1988 ) , the overdispersion of the offspring distribution underlying the reproduction number ( Delamater et al . , 2019 ) and the threshold at which herd immunity is reached ( Fontanet and Cauchemez , 2020; Mistry et al . , 2021 ) . They can similarly modulate the effectiveness of non-pharmaceutical interventions ( NPIs ) , such as school closures and workplace restrictions , that are typically deployed to control and contain the spread of infectious diseases ( Prem et al . , 2020 ) . Social contact surveys provide insight into the features of these networks , which is typically achieved through incorporating survey results into mathematical models of infectious disease transmission frequently used to guide decision making in response to outbreaks ( Chang et al . , 2021; Davies et al . , 2020 ) . Such inputs are necessary for models to have sufficient realism to evaluate relevant policy questions . However , despite the known importance of contact patterns as determinants of the infectious disease dynamics , our understanding of how they vary globally remains far from complete . Reviews of contact patterns to date have focused on high-income countries ( HICs ) ( Hoang et al . , 2019 ) . This is despite evidence that social contact patterns differ systematically across settings in ways that have material consequences for the dynamics of infectious disease transmission and the evolution of epidemic trajectories ( Prem et al . , 2017; Walker et al . , 2020 ) . Previous reviews have also primarily explored the total number of contacts made by individuals ( Hoang et al . , 2019 ) and/or how these contacts are distributed across different age/sex groups ( Horton et al . , 2020 ) . Whilst these factors are a vital component underpinning disease spread , recent work has also underscored the importance of the characteristics of contacts ( such as the location , duration , and extent of physical contact ) in determining transmission risk ( Thompson et al . , 2021 ) . Here , we carry out a systematic review of contact surveys ( conducted prior to the emergence of COVID-19 ) in lower-income , lower-middle and upper-middle-income countries ( LICs , LMICs and UMICs , respectively ) . Alongside previously published data from HICs ( Kwok et al . , 2018; Kwok et al . , 2014; Leung et al . , 2017; Mossong et al . , 2008 ) , we collate individual participant data ( IPD ) on social contacts from published work spanning 27 surveys from 22 countries and over 28 , 000 individuals . We use a Bayesian framework to explore drivers and determinants of contact patterns across a wider range of settings and at a more granular scale than has previously been possible . Specifically , we assess the influence of key factors such as age , gender , and household structure on both the total number and characteristics ( such as duration , location , and type ) of contact made by an individual , and explore how the comparative importance of different factors varies across different settings . We additionally evaluate the extent and degree of assortativity in contact patterns between different groups , and how this varies across settings .
The mean , median , and interquartile range of total daily unique contacts were calculated for subgroups including country income status , individual study , survey methodology ( diary-based or questionnaire/interview-based ) , survey day ( weekday/weekend ) , and respondent characteristics such as age , sex , employment/student status , and household size . Detailed description of data assumptions for each study can be found in Supplementary file 4 . A negative binomial regression model was used to explore the association between the total number of daily contacts and the participant’s age , sex , employment/student status , and household size , as well as methodology and survey day . Incidence rate ratios from these regressions are referred to as ‘contact rate ratios’ ( CRRs ) . A sensitivity analysis was carried out that excluded additional contacts ( such as additional work contacts , group contacts , and number missed out , which were recorded separately and in less detail by participants compared to their other contacts [Ajelli and Litvinova , 2017; Kumar et al . , 2018; Leung et al . , 2017; Zhang et al . , 2020] ) . Logistic regressions were used to explore determinants of contact duration ( <1 hr/1 hr+ ) and type ( physical/non-physical ) , using the same explanatory variables as in the total contacts analyses . There were differences in the contact duration categories defined by studies , and the threshold of 1 hr for longer durations was used to maximise sample size , by allowing inclusion of all available data . An additional sensitivity analysis , weighing all studies equally within an income stratum , explored the impact of study size on the estimated CRRs and ORs for all main outcomes ( total contacts , duration , and whether physical ) . The proportion of contacts made at each location ( home , school , work , and other ) was explored descriptively and contacts made with the same individual in separate locations/instances were considered as separate contacts . All analyses were done in a Bayesian framework using the probabilistic programming language Stan , using uninformative priors in all analyses and implemented in R via the package brms ( Bürkner , 2018 , Bürkner , 2017 ) . All analyses were stratified by three income strata ( LICs and LMICs were combined to preserve statistical power ) and included random effects by study , to account for heterogeneity between studies . The only exceptions to this were any models adjusting for methodology which did not vary by study . The effect of each factor was explored in an age- and gender-adjusted model . All models exploring the effect of student status or employment status were restricted to children aged between 5 and 18 years and adults over 18 , respectively . In the remaining models including all ages , age was adjusted as a categorical variable ( < 15 , 15–65 , and over 65 years ) . CRRs , odds ratios ( ORs ) , and their associated 95% credible intervals are presented for all regression models . Here , we report estimates adjusted for age and gender ( referred to as adjCRR or adjOR ) . Studies which collated contact-level data were used to assess assortativity of mixing by age and gender for different country-income strata by calculating the proportions of contacts made by participants that are male or female and those that belong to three broad age groups ( children , adults , and older adults ) .
A total of 3409 titles and abstracts were retrieved from the databases , and 313 full-text articles were screened for eligibility ( Appendix 1—figure 1 ) . This search identified 19 studies with suitable contact data from LIC , LMIC , and UMIC settings – individual-level data were obtained from 16 of these studies , including one study from an LIC , six studies from an LMIC , and nine studies from an UMIC . These were analysed alongside four HIC studies from Hong Kong and Europe . The majority of the studies collected data representative of the general population , through random sampling and included a combination of both rural and urban sites ( see Appendix 1 for further details ) . Although most studies included respondents of all ages , one study restricted their participants to ages over 18 years ( Dodd et al . , 2016 ) , one to ages over 15 years ( Mahikul et al . , 2020 ) , one to ages over 6 months ( Huang et al . , 2020 ) , one study only collected contact data on infants under 6 months ( Oguz et al . , 2018 ) , and another on contacts of children under 6 years and their caregivers ( Neal et al . , 2020 ) . The distribution of participant age groups in each study was also dependent on the sampling method . For instance , two studies focused on school and university students and their contacts , thereby oversampling older children and young adults ( Ajelli and Litvinova , 2017; Stein et al . , 2014 ) . Details of the identified studies and a full description of the systematic review findings can be found in Appendix 1 and Supplementary file 6 . In total , this meta-analysis yielded 28 , 503 participants reporting on 413 , 069 contacts . All studies contained information on main demographic variables such as age and gender . Availability of other variables analysed here for each study are listed in Supplementary file 7 . All studies reported the number of contacts made in the past 24 hr of ( or day preceding ) the survey . The definitions of contacts were broadly similar across studies ( Supplementary file 6 ) . Specifically , contacts were defined as skin-to-skin ( physical ) contact or a two-way conversation in the physical presence of another person . All studies scored above 65% of the items on the AXIS risk of bias tool , suggesting good or fair quality ( Supplementary file 3 ) . Among all participants 47 . 5% were male , 30 . 1% were aged under 15 years and 7 . 2% were aged over 65 years . The majority ( 83 . 4% ) of participants were asked to report the number of contacts they made on a weekday . A large proportion ( 34 . 1% ) of respondents lived in large households of six or more people but this was largely dependent on income setting ( LIC/LMIC = 63 . 2% , UMIC = 35 . 9% , HIC = 4 . 9% ) . Among school-aged children ( 5–18 years ) , 88 . 1% were students , and 59 . 1% of adults aged over 18 were employed . The median number of contacts made per day across all the studies was 9 ( IQR = 5–17 ) , and was similar across income strata ( LIC/LMIC = 10[5–17] , UMIC = 8[5–16] , HIC = 9[5–17]; Table 1 ) . There was a large variation in contact rates across different studies , with the median number of daily contacts ranging from 4 in a Zambian setting ( Dodd et al . , 2016 ) to 24 in an online Thai survey ( Stein et al . , 2014 ) . When stratifying by study methodology , median daily contacts was higher in diary-based surveys compared to interview-/questionnaire-based surveys , which was true across all income strata ( Table 1 , Appendix 2—figure 1 ) . Overall , children aged 5–15 had the highest number of daily contacts ( Figure 1A–C ) , although there was substantial variation between studies and across income strata in how the number of daily contacts varied with age ( Figure 1A–C ) . Across UMICs and HICs , the number of daily contacts made by participants decreased with age , with this decrease most notable in the oldest age groups ( adjCRR for 65+ vs . <15 years [95%CrI]: UMIC = 0 . 67[0 . 63–0 . 71] and HIC = 0 . 57[0 . 54–0 . 60] ) . By contrast , there was no evidence of contact rates declining in the oldest age groups in LICs/LMICs ( adjCRR for 65+ vs . < 15 years [95%CrI] = 0 . 94[0 . 89–1 . 00] ) . We observed contrasting effects of gender on the number of daily contacts , with men making more daily contacts compared to women in LICs/LMICs after accounting for age ( adjCRR = 1 . 17 , 95%CrI:1 . 15–1 . 20; Figure 1D ) , but no effect of gender on total daily contacts for other income strata ( CRR[95%CrI]: UMIC = 1 . 01[0 . 98–1 . 04] , HIC = 0 . 99[0 . 97–1 . 02] ) . There were also differences in the number of daily contacts made according to the methodology used and whether the survey was carried out on a weekday or over the weekend – in both instances , contrasting effects of these factors on the number of daily contacts according to income strata were observed ( Figure 1D , F ) . We also examined the influence of factors that might influence both the total number and location ( home , work , school , and other ) of the contacts individuals make . Across all income strata , students ( defined as those currently in education , attending school , and aged between 5 and 18 years ) made more daily contacts than non-students aged between 5 and 18 ( adjCRR [95%CrI]: LIC/LMIC = 1 . 26[1 . 16–1 . 37] , UMIC = 1 . 18[1 . 03–1 . 35] and HIC = 1 . 54[1 . 42–1 . 66]; Figure 1D–F ) . Similarly , we observed strong and significant effects of employment in all income strata , with adults who were employed having a higher number of total daily contacts compared to those not in employment ( adjCRR [95%CrI]: LIC/LMIC = 1 . 17[1 . 12–1 . 23] , UMIC = 1 . 07[1 . 03–1 . 13] , HIC = 1 . 60[1 . 54–1 . 65]; Figure 1D–F ) . The number of daily contacts made at home was proportional to the participant’s household size ( Appendix 2—figure 2 ) . Total daily contacts increased with household size ( Figure 2A , Appendix 2—figure 1 ) across all income-strata; individuals living in large households ( 6+ members ) had 1 . 47 ( 95%CrI:1 . 32–1 . 64 ) ( LIC/LMICs ) , 2 . 58 ( 95%CrI:2 . 37–2 . 80 ) ( UMICs ) , and 1 . 51 ( 95%CrI:1 . 40–1 . 63 ) ( HICs ) times more daily contacts than those living alone , after accounting for age and gender ( Figure 1E–F ) . Sensitivity analyses excluding additional contacts ( as defined in Materials and methods ) showed little difference in effect sizes for total daily contacts , and were strongly correlated with the effect sizes shown in Figure 1D–F ( Appendix 2—figure 3 ) . Motivated by this suggestion of strong , location-related ( school , work , and household ) effects on total daily contact rates , we further explored the locations in which contacts were made . Contact location was known for 314 , 235 contacts , 42 . 7% of which occurred at home ( 13 . 1% at work , 12 . 5% at school , and 31 . 7% in other locations ) . Across income strata , there was significant variation in the proportion of contacts made at home – being highest in LICs/LMICs ( 68 . 3% ) and lowest in HICs ( 37 . 0% ) ( Figure 2B ) . Age differences were also observed in the number of contacts made at home , particularly for LICs/LMICs ( Figure 2C , D ) . Relatedly , a higher proportion of contacts occurred at work and school ( 14 . 6% and 11 . 3% ) in HICs compared to LICs/LMICs ( 3 . 9% and 5 . 2% , respectively; Appendix 2—figure 4 ) . Strong , gender-specific patterns of contact location were also observed . Across all income strata males made a higher proportion of their contacts at work compared to females , although this difference was largest for LICs/LMICs ( Appendix 2—figure 4 and Appendix 2—figure 5 ) . Further , we found significant variation between income strata in median household size ( seven in LICs/LMICs , five in UMICs , and three in HICs ) . This trend of decreasing household size with increasing country income was consistent with global data ( Figure 2E ) . The larger households observed for LIC/LMIC settings were also more likely to be intergenerational – in LICs/LMICs , 59 . 4% of participants aged over 65 lived in households of at least six members compared to 17 . 5% in UMICs and only 2 . 2% in HICs . Data on the type of contacts ( physical and non-physical ) were recorded for 20 , 910 participants . The mean percentage of physical contacts across participants was 56 . 0% and was the highest for LICs/LMICs ( 64 . 5% ) . At the study level , the highest mean percentage of physical contacts was observed for a survey of young children and their caregivers conducted in Fiji ( Neal et al . , 2020 ) ( 84 . 0% ) and the lowest in a Hong Kong contact survey ( Leung et al . , 2017 ) ( 18 . 9% ) . Physical contact was significantly less common among adults compared to children under 15 years in all settings ( ORs ranged between 0 . 22 and 0 . 48 ) ( Figure 3A–F ) . Despite the proportion of physical contacts generally decreasing with age , there was a higher proportion observed for adults aged 80 or over ( Figure 3A–C ) . Contacts made by male participants were more likely to be physical compared to female participants in UMICs ( adjOR = 1 . 13 , 95%CrI = 1 . 10–1 . 16 ) and HICs ( adjOR = 1 . 09 , 95%CrI = 1 . 07–1 . 12 ) , but in LICs/LMICs men had a lower proportion of physical contacts than women ( adjOR = 0 . 81 , 95%CrI = 0 . 79–0 . 83; Figure 3D–F ) . Most physical contacts made by women in LICs were made at home ( 73 . 5% ) , whilst for HICs this was just 41 . 4% – similar differences across income strata were observed for men , although the proportions were always lower than observed for women ( 62 . 4% for LIC/LMICs and 36 . 4% for HICs ) . Increasing household size was generally associated with a higher proportion of contacts being physical ( for households of 6+ members compared to one member: adjCRR[95%CrI]: LIC/LMIC = 1 . 73[1 . 48–2 . 02] , UMIC = 1 . 30[1 . 12–1 . 52] , HIC = 1 . 57[1 . 48–1 . 67]; Figure 3D–F ) . Employment was associated with having a significantly lower proportion of physical contacts in LICs/LMICs ( adjOR = 0 . 83 , 95%CrI:0 . 79–0 . 87 ) and HICs ( adjOR = 0 . 71 , 95%CrI:0 . 69–0 . 73 ) , but not in UMICs ( adjOR = 1 . 11 , 95%CrI:1 . 03–1 . 19 ) . The proportion of physical contacts among all contacts was the highest for households ( 70 . 4% ) , followed by schools ( 58 . 5% ) , community ( 55 . 7% ) , and work ( 33 . 6% ) ( Appendix 2—figure 6 ) . Data on the duration of contact ( <1 or ≥1 hr ) were available for 22 , 822 participants . The percentage of contacts lasting at least 1 hr was 63 . 2% and was highest for UMICs ( 76 . 0% ) and lowest for LICs/LMICs ( 53 . 1% ) . Across both UMICs and HICs , duration of contacts was lower in individuals aged over 15 years compared to those aged 0–15 , with the extent of this disparity most stark for HICs ( for ages 65+ compared to <15 years: adjCRR [95%CrI]: LIC/LMIC = 0 . 61[0 . 57–0 . 64] , UMIC = 0 . 61[0 . 58–0 . 65] , HIC = 0 . 35[0 . 33–0 . 37]; Figure 4A–F ) . We observed contrasting effects of gender across income strata: males made longer-lasting contacts than females in UMICs ( adjOR = 1 . 11 , 95%CrI = 1 . 08–1 . 14; Figure 4D–F ) , but not in LIC/LMICs ( adjOR = 0 . 92 , 95%CrI = 0 . 90–0 . 95 ) or HICs ( adjOR = 0 . 98 , 95%CrI = 0 . 97–1 . 00 ) . Participants reported shorter contacts on weekends compared to weekdays in LICs/LMICs ( adjOR = 0 . 91 , 95%CrI: 0 . 88–0 . 95 ) , and HICs ( adjOR = 0 . 95 , 95%CrI: 0 . 92–0 . 97 ) , but not in UMICs ( adjOR = 1 . 12 , 95%CrI = 1 . 03–1 . 21 ) . Contacts lasting over an hour as a proportion of all contacts was highest for households ( 72 . 7% ) , followed by schools ( 67 . 9% ) , community ( 47 . 0% ) , and work ( 44 . 0% ) . However , it was only in HICs that there was a significant effect of being a student ( adjOR = 1 . 18 , 95%CrI: 1 . 09–1 . 27; Figure 4D–F ) on the proportion of contacts lasting ≥1 hr . For all income strata , the proportion of contacts >1 hr increased with increasing household size ( Figure 4D–F ) . The sensitivity analysis weighing all studies equally within an income group yielded similar results to those from the main analysis ( range of Pearson’s correlation coefficients between main analysis and sensitivity analysis effect sizes: 0 . 92–1 . 00; Appendix 2—figure 7 and Appendix 2—table 1 ) , and any differences are discussed in Appendix 2 . Twelve studies collected information on the gender of the contact and eight studies contained information on age allowing assignment of contacts to one of the three age groups described in Materials and methods ( Appendix 2 ) . We found evidence to suggest that contacts were assortative by gender for all income strata , as participants were more likely to mix with their own gender ( Appendix 2—table 2 and Appendix 2—table 3 ) . Mixing was also assortative by age , with participants more likely to contact individuals who belonged to the same age group this degree of age assortativity was lowest for LICs/LMICs , where only 29% of contacts made by adults were with individuals of the same age group . By contrast , in HICs we observed a higher degree of assortative mixing , with most contacts ( 51 . 4% ) made by older adults occurring with individuals belonging to the same age group .
Understanding patterns of contact across populations is vital to predicting the dynamics and spread of infectious diseases , as well understanding the control interventions likely to have the greatest impact . Here , using a systematic review and individual-participant data meta-analysis of contact surveys , we summarise research exploring these patterns across a range of populations spanning 28 , 503 individuals and 22 countries . Our findings highlight substantial differences in contact patterns between income settings . These differences are driven by setting-specific sociodemographic factors such as age , gender , household structure , and patterns of employment , which all have material consequences for transmission and spread of respiratory pathogens . Across the collated studies , the total number of contacts was highest for school-aged children . This is consistent with previous results from HICs ( Béraud et al . , 2015; Fu et al . , 2012; Hoang et al . , 2019; Ibuka et al . , 2016; Lapidus et al . , 2013 ) and shown here to be generally true for LICs/LMICs and UMICs also . Interestingly however , we observed differences in patterns of contact in adults across income strata . Whilst contact rates in HICs declined in older adults , this was not observed in LICs/LMICs , where contact rates did not differ in the oldest age group compared to younger ages . This is consistent with variation in household structure and size across settings , with nearly two-thirds of participants aged 65+ in included LIC/LMIC surveys living in large , likely intergenerational , households ( 6+ members ) , compared to only 2% in HICs . HICs were also characterised by more assortative mixing between age groups , with older adults in LICs/LMICs more likely to mix with individuals of younger ages , again consistent with the observed differences between household structures across the two settings . These results have important consequences for the viability and efficacy of protective policies centred around shielding of elderly individuals ( i . e . those most at risk from COVID-19 or influenza ) . In these settings other strategies may be required to effectively shield vulnerable populations , as has been previously suggested ( Dahab et al . , 2020 ) . Our results support the idea of households as a key site for transmission of respiratory pathogens ( Thompson et al . , 2021 ) , with the majority of contacts made at home . Our analysis highlights that the number of contacts made at home is mainly driven by household size . However , the relative importance of households compared to other locations is likely to vary across settings . We observed significant differences across income settings in the distribution of contacts made at home , work , and school . The proportion of contacts made at home was highest for LIC/LMICs , where larger average household sizes were associated with more contacts , more physical contacts , and longer lasting contacts . By contrast , participants in HICs tended to report more contacts occurring at work and school . The lower number of contacts at work in LIC/LMIC may be explained by the types of employment ( e . g . agriculture in rural surveys ) and a selection bias ( women at home/homemakers more likely to be surveyed in questionnaire-based surveys ) . Our analyses similarly highlighted significant variation in the duration and nature of contacts across settings . Contacts made by female participants in LICs/LMICs were more likely to be physical compared to men , whilst the opposite effect was observed for HICs and UMICs , potentially reflecting context-specific gender roles . In all settings , we observed a general decline of physical contacts with age , except in the very old ( Mossong et al . , 2008 ) , potentially reflecting higher levels of dependency and the need for physical care . Altogether , these results suggest differences between settings in the comparative importance of different locations ( such as the household or the workplace ) to transmission of SARS-CoV-2 , a finding which would likely modulate the impact of different NPIs ( such as workplace or school closures , stay at home orders , etc . ) . Moreover , it suggests that previous estimates of NPI effectiveness primarily derived from European data and settings ( Brauner et al . , 2021 ) may be of limited generalisability to non-European settings characterised by different structures and patterns of social contact . However , beyond highlighting heterogeneity in where and how transmission is likely to occur , it remains challenging to disentangle exactly how these differences in contact patterns would shape patterns of transmission . Whilst the collated data provide a cross-sectional snapshot into the networks of social contact underpinning transmission , they remain insufficient to completely resolve this network or its temporal dynamics . Our results therefore do not consider key features relevant to population-level spread and transmission ( such as overall network structure or the extent of repeated contacts , which would be most likely to occur with household members ) which previous work has demonstrated can have a significant impact on infectious disease dynamics , both in general terms ( Bansal et al . , 2010; Keeling and Eames , 2005 ) and with COVID-19 ( Rader et al . , 2020 ) . It is in this context that recent results generating complete social networks ( including both the frequency and identity of an individual’s contacts ) from high-resolution GPS data represent promising developments in understanding social contact networks and how they shape transmission ( Firth et al . , 2020 ) . There are important caveats to these findings . Data constraints limited the numbers of factors we were able to explore – for example , despite evidence ( Kiti et al . , 2014 ) suggesting that contact patterns differ across rural and urban settings , only three studies ( Kiti et al . , 2014; le Polain de Waroux et al . , 2018b; Neal et al . , 2020 ) contained information from both rural and urban sites , allowing classification . Similarly , we were unable to examine the impact of socioeconomic factors such as household wealth , despite experiences with COVID-19 having highlighted strong socioeconomic disparities in both transmission and burden of disease ( De Negri et al . , 2021; Routledge et al . , 2021; Ward et al . , 2021; Winskill et al . , 2020 ) and previous work suggesting that poorer individuals are less likely to be employed in occupations amenable to remote working ( Loayza , 2020 ) . A lack of suitably detailed information in the studies conducted precludes analysis of these factors but highlights the importance of incorporating economic questions into future contact surveys , such as household wealth and house square footage . Other factors also not controlled for here , but that may similarly shape contact patterns include school holidays or seasonal variations in population movement and composition that we are unable to capture given the cross-sectional nature of these studies . Another important limitation to these results is that we are only able to consider a limited set of contact characteristics ( the location and duration of the contact and whether it was physical ) . Previous work has highlighted the importance of these factors in determining the risk of respiratory pathogen transmission ( Chang et al . , 2021; Dunne et al . , 2018; le Polain de Waroux et al . , 2018a; Neal et al . , 2020; Thompson et al . , 2021 ) , but only a limited number of studies reported whether a contact was ‘close’ or ‘casual’ ( Kwok et al . , 2018; Kwok et al . , 2014; le Polain de Waroux et al . , 2018b ) and whether the contact was made indoors or outdoors ( Wood et al . , 2012 ) ; both factors likely to influence transmission risk ( Bulfone et al . , 2021; Chu et al . , 2020 ) . More generally , the relevance and comparative importance of different contacts to transmission likely varies according to the specific pathogen and its predominant transmission modality ( e . g . aerosol , droplet , fomite , etc . ) . It is therefore important to note that these results do not provide a direct indication of explicit transmission risk , but rather an indicator of factors likely to be relevant to transmission . Relatedly , it is also important to note that the studies collated here were conducted over a wide time-period ( 2005–2018 ) . In conjunction with the cross-sectional nature of the included studies , this precludes us from being able to examine for potential time-related trends in contact patterns . Additionally , the collated surveys were all carried out prior to the onset of the SARS-CoV-2 pandemic . Previous work has documented significant alterations to patterns of social contact in response to individual-level behaviour changes or government implemented NPIs aimed at controlling SARS-CoV-2 spread , and that these changes are dynamic and time-varying ( Gimma et al . , 2021; McCreesh et al . , 2021 ) . A detailed understanding of the impact of changing contact patterns on disease spread necessarily requires both an understanding of baseline contact patterns ( as detailed in the studies collated here ) and what changes have occurred as a result of control measures – however this latter data remains sparse and is available for only a limited number of settings ( Jarvis et al . , 2021; Jarvis et al . , 2020; Liu et al . , 2021 ) . Description of contact location was also coarse and precluded more granular analyses of specific settings , such as markets , which have previously been shown to be important locations for transmission in rural areas ( Grijalva et al . , 2015 ) . Heterogeneity between studies was larger for LICs/LMICs and UMICs , which we partly accounted for , through fitting random study effects . These study differences may be attributed to the way individual contact surveys were conducted , making comparisons of contact patterns among surveys more difficult ( e . g . prospective/retrospective diary surveys , online/paper questionnaires , face-to-face/phone interviews , and different contact definitions ) . For instance , there is evidence suggesting that prospective reporting , which is less affected by recall bias , can often lead to a higher number of contacts being reported ( Mikolajczyk and Kretzschmar , 2008 ) and a lower probability of casual or short-lasting contacts being missed . The relatively high contact rates observed in HICs may be explained by the fact that all but two HIC surveys used diary methods . Our study highlights that a unified definition of ‘contact’ and standard practice in data collection could help increase the quality of collected data , leading to more robust and reliable conclusions about contact patterns . Whilst we aggregate results by income strata due to the limited availability of data ( particularly in LICs and middle-income countries ) , it is important to note that the outcomes considered here are likely to be shaped by several different factors other than country-level income . Whilst some of these factors will be correlated with a country’s income status ( e . g . household size Walker et al . , 2020 ) , many others will be unique to a particular setting or geographical area or correlate only weakly with country-level data . Examples include patterns of employment , the role of women , and other contextual factors . These analyses are therefore intended primarily to provide indications of prevailing patterns , rather than a definitive description of contact patterns in a specific context and highlight the significant need for further studies to be carried out in a diversity of different locations . Despite these limitations however , our results highlight significant differences in the structure and nature of contact patterns across settings . These differences suggest that the comparative importance of different locations and age groups to transmission will likely vary across settings and have critical consequences for the efficacy and suitability of strategies aimed at controlling the spread of respiratory pathogens such as SARS-CoV-2 . Most importantly , our study highlights the limited amount of work that has been undertaken to date to better understand and quantify patterns of contact across a range of settings , particularly in LICs and middle-income countries , which is vital in informing control strategies reducing the spread of such pathogens . | Infectious diseases , particularly those caused by airborne pathogens like SARS-CoV-2 , spread by social contact , and understanding how people mix is critical in controlling outbreaks . To explore these patterns , researchers typically carry out large contact surveys . Participants are asked for personal information ( such as gender , age and occupation ) , as well as details of recent social contacts , usually those that happened in the last 24 hours . This information includes , the age and gender of the contact , where the interaction happened , how long it lasted , and whether it involved physical touch . These kinds of surveys help scientists to predict how infectious diseases might spread . But there is a problem: most of the data come from high-income countries , and there is evidence to suggest that social contact patterns differ between places . Therefore , data from these countries might not be useful for predicting how infections spread in lower-income regions . Here , Mousa et al . have collected and combined data from 27 contact surveys carried out before the COVID-19 pandemic to see how baseline social interactions vary between high- and lower-income settings . The comparison revealed that , in higher-income countries , the number of daily contacts people made decreased with age . But , in lower-income countries , younger and older individuals made similar numbers of contacts and mixed with all age groups . In higher-income countries , more contacts happened at work or school , while in low-income settings , more interactions happened at home and people were also more likely to live in larger , intergenerational households . Mousa et al . also found that gender affected how long contacts lasted and whether they involved physical contact , both of which are key risk factors for transmitting airborne pathogens . These findings can help researchers to predict how infectious diseases might spread in different settings . They can also be used to assess how effective non-medical restrictions , like shielding of the elderly and workplace closures , will be at reducing transmissions in different parts of the world . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"epidemiology",
"and",
"global",
"health"
] | 2021 | Social contact patterns and implications for infectious disease transmission – a systematic review and meta-analysis of contact surveys |
Homeostatic regulation of the partial pressure of CO2 ( PCO2 ) is vital for life . Sensing of pH has been proposed as a sufficient proxy for determination of PCO2 and direct CO2-sensing largely discounted . Here we show that connexin 26 ( Cx26 ) hemichannels , causally linked to respiratory chemosensitivity , are directly modulated by CO2 . A ‘carbamylation motif’ , present in CO2-sensitive connexins ( Cx26 , Cx30 , Cx32 ) but absent from a CO2-insensitive connexin ( Cx31 ) , comprises Lys125 and four further amino acids that orient Lys125 towards Arg104 of the adjacent subunit of the connexin hexamer . Introducing the carbamylation motif into Cx31 created a mutant hemichannel ( mCx31 ) that was opened by increases in PCO2 . Mutation of the carbamylation motif in Cx26 and mCx31 destroyed CO2 sensitivity . Course-grained computational modelling of Cx26 demonstrated that the proposed carbamate bridge between Lys125 and Arg104 biases the hemichannel to the open state . Carbamylation of Cx26 introduces a new transduction principle for physiological sensing of CO2 .
CO2 is the unavoidable by-product of cellular metabolism . Humans produce approximately 20 moles of CO2 per day ( Marshall and Bangert , 2008 ) . The dissolved CO2 can readily combine with water , aided by carbonic anhydrase , to form H2CO3 , which dissociates rapidly to H+ and HCO3− . In any physiological solution therefore , the partial pressure of CO2 ( PCO2 ) will be in equilibrium with , and inescapably related to , the pH and the concentration of HCO3− of that solution . Regulation of PCO2 is thus a vital homeostatic function that is linked to acid-base balance . As might be expected , chemosensory reflexes regulate the frequency and depth of breathing to ensure homeostatic control of blood gases . The field of respiratory chemosensitivity has been dominated by ‘reaction theory’ which posits that pH is a sufficient signal for detection of changes in PCO2 ( Loeschcke , 1982 ) . Many investigators therefore equate pH-sensing with CO2-sensing . There are several areas of the medulla oblongata which contain neurons that respond to changes in pH/CO2 , especially near the highly vascularised ventral surface . For example a population of neurons highly sensitive to pH/CO2 have been described in the retrotrapezoid nucleus ( RTN ) ( Mulkey et al . , 2004 , 2006; Guyenet et al . , 2008 ) and the medullary raphé nucleus ( Richerson , 2004; Ray et al . , 2011 ) . Despite the acceptance of pH-sensing as the predominant mechanism by which PCO2 is measured , there is substantial evidence for an additional and independent effect of molecular CO2 ( Eldridge et al . , 1985; Shams , 1985; Huckstepp and Dale , 2011 ) . For example , if pH is carefully controlled at the medullary surface , an increase in PCO2 at constant pH will still enhance breathing by as much as a pH change at constant PCO2 ( Shams , 1985 ) . We have recently shown that connexin 26 ( Cx26 ) hemichannels , open in response to increases in PCO2 at constant extracellular pH and are an important conduit for the CO2-dependent , as opposed to pH-dependent , release of ATP ( Huckstepp et al . , 2010a ) . Cx26 hemichannels contribute to the chemosensory control of breathing ( Huckstepp et al . , 2010b; Wenker et al . , 2012 ) . Hemichannels of two closely related connexins , Cx30 and Cx32 , also exhibited CO2-sensitive opening ( Huckstepp et al . , 2010a ) . Despite this evidence , widespread acceptance of direct sensing of CO2 requires a detailed molecular explanation of any putative transduction system . One possible way that CO2 can interact with proteins is via carbamylation—the formation of a covalent bond between the carbon of CO2 and a primary amine group . For example , CO2 forms carbamate bonds with haemoglobin ( Kilmartin and Rossi-Bernardi , 1971 ) and the plant enzyme RuBisCo ( Lundqvist and Schneider , 1991 ) . Here we document an important new advance—the mechanism by which CO2 binds directly to Cx26 , most probably via carbamylation of a lysine residue , to cause hemichannel opening . Our work establishes a new field of direct CO2 sensing that can be mediated by CO2-dependent carbamylation of certain β connexins . As these are widely distributed in the brain and other tissues , it is likely that direct detection of CO2 via this mechanism is important in many different physiological processes .
To document the extent to which this sensitivity to CO2 is limited within the β connexin family ( Figure 1F ) , and to form the basis of a bioinformatic comparison to identify possible CO2 binding motifs , we investigated whether another β connexin , Cx31 , might also be sensitive to CO2 . We expressed , in HeLa cells , either rat Cx31 or rat Cx26 tagged at the C-terminal with mCherry and used a previously described dye loading assay ( Huckstepp et al . , 2010a ) to test whether the cells could load with carboxyfluorescein ( CBF ) in a CO2-dependent manner . As expected from our previous work , HeLa cells expressing the Cx26 readily loaded with CBF when exposed to this dye in the presence of elevated PCO2 ( 55 mmHg , at pH 7 . 5 , Figure 1A , B ) . However , HeLa cells expressing Cx31 failed to dye load in a CO2-dependent manner ( Figure 1A , B ) . As the connexins were tagged with mCherry , we could verify the presence of fluorescent puncta in both the Cx26 and Cx31 expressing HeLa cells ( Figure 1—figure supplement 2 ) . To check for the existence of functional hemichannels in the Cx31-expressing HeLa cells , we removed extracellular Ca2+ as a positive control . This manipulation will open all types of connexin hemichannel . Parental HeLa cells do not load with dye when Ca2+ is removed from the medium ( Figure 3—figure supplement 1 ) ; they therefore possess virtually no endogenous hemichannels . The removal of extracellular Ca2+ readily caused loading of CBF into the Cx31-expressing HeLa cells ( Figure 1A , inset ) , demonstrating the presence of functional Cx31 hemichannels . 10 . 7554/eLife . 01213 . 003Figure 1 . Identification of the motif in Cx26 that imparts CO2 sensitivity . ( A ) Dye loading assay demonstrates CO2-dependent loading of carboxyfluorescein into HeLa cells expressing Cx26 , but not into those expressing Cx31 . The inset in Cx31 shows that these hemichannels are expressed and functional in the membrane by utilizing a zero Ca2+ stimulus to open them and allow dye loading . Scale bars 40 µm . ( B ) Cumulative probability plots of pixel intensity in the control and following exposure to PCO2 of 55 mmHg . Each curve is comprises the measurements of mean pixel intensity for at least 40 cells . ( C ) Sequences ( from mouse ) for Cx26 , 30 , 32 and 31 to show K125 and four following amino acids that are present in Cx26 , Cx30 and Cx32 , but absent from Cx31 . R104 in Cx26 and 30 , and K104 in Cx32 and Cx31 are also highlighted . Accession numbers: Cx26 , NP_032151; Cx30 , AAH13811; Cx32 , AAH26833; and Cx31 , NP_001153484 . ( D ) The structure of Cx26 drawn from the 2zw3 PDB file , cytoplasmic face of the channel upwards . On each subunit K125 and R104 are drawn . ( E ) Detail from the structure of Cx26 ( dashed square ) showing the orientation of K125 ( red ) towards R104 ( dark grey ) . The short distance between the two amino acid side chains suggests that this gap could be bridged by carbamylation by CO2 of K125 and a subsequent salt bridge with R104 . ( F ) Phylogenetic tree showing relationship between Cx26 and other β connexins . The three CO2 sensitive connexins are very closely related to each other while Cx31 is more distant . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 00310 . 7554/eLife . 01213 . 004Figure 1—figure supplement 1 . Expression of Cx26 in HeLa cells imparts sensitivity to CO2 . Top , diagram of dye loading protocol , showing the methods for assessing background dye loading under control conditions ( no additional CO2 ) , dye loading during a hypercapnic challenge ( PCO2 55 mmHg ) and the use of the hemichannel blockers to probe the identity of the hemichannel required for CO2 sensitive dye loading . Bottom , cumulative probability plots demonstrating that exposure of Cx26-expressing HeLa cells to elevated PCO2 caused dye loading compared to the background control . This CO2-dependent loading was blocked by 100 µM carbenoxolone ( Carb ) , but unaffected by probenecid or ruthenium red ( RuRed ) , demonstrating that the heterologously expressed Cx26 rather than any potential endogenous pannexin-1 or calhm1 was responsible for the CO2 sensitivity . Minimum of 40 cells measured from three independent repetitions of each treatment . The slight reduction in dye loading caused by carbenoxolone compared to the background control loading is to be expected , as Cx26 will partially gate in response to control aCSF which has a PCO2 of 35 mmHg . Parental HeLa cells do not exhibit CO2-dependent dye loading ( Figure 3—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 00410 . 7554/eLife . 01213 . 005Figure 1—figure supplement 2 . Expression of connexin variants in HeLa cells . The connexin variants were tagged with mCherry , allowing verification of successful of expression . Examples of wild type Cx26 and Cx31 are shown along with two mutant variants . Scale bar 40 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 005 The CO2-sensitivity in the β connexins therefore appears to be limited to the three closely related connexins , Cx26 , Cx30 and Cx32 , and Cx31 has no sensitivity to increases in PCO2 ( Figure 1F ) . We hypothesized that CO2 carbamylated a lysine residue in Cx26 to induce conformational change and hence opening of the hemichannel . We therefore compared the sequences of the three CO2-sensitive connexins to Cx31 to identify a lysine present in all three CO2 sensitive connexins but absent from Cx31 ( Figure 1C ) . This analysis revealed K125 and four further amino acids as forming a motif that was absent from Cx31 . The existing crystal structure for Cx26 ( Maeda et al . , 2009 ) , shows that K125 is at the end of an alpha helix and that the sequence KVREI ( ‘carbamylation motif’ ) orients K125 towards R104 on the neighbouring subunit ( Figure 1D ) . The distance from K125 to R104 is only 6 . 5 Å ( Maeda et al . , 2009 ) , strongly suggesting that if K125 were to be carbamylated it could form a salt bridge between these two residues in adjacent subunits ( Figure 1E ) . Interestingly , R104 is present in Cx30 , but conservatively substituted by a lysine residue in Cx32 ( Figure 1C ) , which has a lower sensitivity to CO2 than Cx26 ( Huckstepp et al . , 2010a ) . Our analysis predicts that if we were to introduce the putative carbamylation-motif of Cx26 into Cx31 , the resulting mutant Cx31 ( mCx31 ) should be sensitive to CO2 as the lysine introduced into the sequence should be able to form a salt bridge with the native residue K104 in mCx31 once carbamylated ( Figure 1C–E and 2A ) . We therefore made mCx31 in which the motif TQKVREI was introduced in place of K123H124 of the native connexin ( Figure 2A ) . This insertion/substitution maintained the correct orientation of the K125 with respect to K104 of Cx31 . HeLa cells expressing mCx31 displayed clear CO2-dependent dye loading ( Figure 2B , C ) . We confirmed the CO2 sensitivity of mCx31 expressing HeLa cells by performing whole cell patch camp recordings . mCx31-expressing cells exhibited a conductance change of 3 . 3 ± 0 . 84 nS ( mean ± SEM , n = 8 , Supplementary file 1 ) when exposed to elevated PCO2 ( Figure 2D ) . Cells , expressing wild type Cx31 showed no CO2-dependent changes in their whole cell conductance ( mean conductance change −0 . 002 ± 0 . 023 nS , n = 6 , Supplementary file 1 , Figure 2D ) . 10 . 7554/eLife . 01213 . 006Figure 2 . Insertion of the identified motif into Cx31 creates a CO2-sensitive hemichannel . ( A ) Comparison of the WT and mutated Cx31 amino acid sequence to show the insertion of the K125 and surrounding residues . ( B ) The dye loading assay demonstrates gain of CO2-sensitivity in mCx31 . Scale bar 40 µm . ( C ) Cumulative probability of mean pixel density of 40 cells in five independent replications . ( D ) Whole cell patch clamp recordings from HeLa cells expressing mCx31 and Cx31 . Recordings were performed under voltage clamp at a holding potential of −50 mV with a constant 10 mV step to assess whole cell conductance . The cells expressing mCx31 show a clear conductance change on exposure to a change in PCO2 , whereas cells expressing wild type Cx31 showed no such change ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 006 To demonstrate that K125 is the key residue for interaction with CO2 , we made mCx31K125R , by inserting TQRVREI into Cx31 in place of K123H124 . Unlike lysine , the arginine side chain cannot be carbamylated by CO2 as its pKa ( 12 . 5 ) is much higher than that of lysine ( 10 . 5 ) , therefore this variant should have no sensitivity to CO2 . mCx31K125R did indeed lack sensitivity to CO2 ( Figure 3A–C , Figure 3—source data 1 ) . This was not because the mutant channel failed to assemble correctly , as the positive control of zero Ca2+ dye loading demonstrated the presence of functional hemichannels ( Figure 3A , Figure 3—figure supplement 1 ) . Next we investigated the relevant residues in Cx26 itself . The carbamate bridge that we propose must involve K125 ( being carbamylated ) and R104 ( forming the salt bridge with the carbamylated lysine ) . We therefore made mutations that individually disrupted both of these functions: K125R to prevent carbamylation , and R104A to disrupt formation of the salt bridge . Neither Cx26K125R nor Cx26R104A exhibited sensitivity to CO2 sensitivity . Nevertheless the positive controls demonstrated the presence of functional mutant hemichannels in the expressing HeLa cells ( Figure 3A , Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 01213 . 007Figure 3 . K125 and R104 are essential residues for CO2 sensitivity . ( A ) Insertion of the motif ( Figure 2A ) from Cx26 but with the mutation K125R into Cx31 ( mCx31K125R ) does not give a gain of CO2 sensitivity indicating that this is an essential residue . Introducing the mutations K125R or R104A into Cx26 destroys the CO2-sensitivity of Cx26 . Insets show the zero Ca2+ positive controls to demonstrate the presence of functional hemichannels in the cells . Scale bars , 40 µm . ( B ) Cumulative probability distributions demonstrate that none of these mutant channels are sensitive to CO2 . ( C ) Summary data demonstrating: gain of function in the mCx31 hemichannel and subsequent loss in mCx31K125R; and loss of function in the Cx26K125R and Cx26R104A mutants . The graphs shown the median of the median change in pixel intensity from five independent replications for each type of hemichannel . KW: Kruskal-Wallis ANOVA , pairwise comparisons by the Mann-Whitney U-test . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 00710 . 7554/eLife . 01213 . 008Figure 3—source data 1 . Median differences in pixel intensity between CO2 and control dye loading experiments for the various connexin hemichannel variants in the histograms of Figure 3C and statistical analysis: Kruskal-Wallis anova , pairwise Mann-Whitney tests and false discovery rate procedure . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 00810 . 7554/eLife . 01213 . 009Figure 3—figure supplement 1 . All connexin variants form functional hemichannels capable of opening in response to zero Ca2+ . Dye loading of wild type HeLa cells and HeLa cells expressing Cx26 , Cx31 and the mutated versions used in this study . In all cases zero Ca2+ ( red trace ) increases the amount of dye loading in connexin expressing cells over the control ( light grey trace ) . For reference the dye loading caused by CO2 is also shown ( dark grey trace ) . Neither CO2 nor zero Ca2+ altered dye loading in parental HeLa cells devoid of heterologous connexin expression , demonstrating that these cells have virtually no endogenous hemichannels . For Cx26 , the amount of dye loading caused by CO2 and zero Ca2+ was similar . In all other connexin variants zero Ca2+ caused dye loading but CO2 had no effect . The cumulative probability distributions comprise all measurements from five independent replications of every set of hemichannel variants that is the entire dataset arising from the image analysis of these connexin variants . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 009 To test further our prediction that the carbamylated K125 forms a salt bridge with R104 , we made the mutation K125E in Cx26 . The insertion of glutamate in place of the lysine has the potential to act as an analogue of the carbamylated K125 . If our hypothesis is correct , this mutated channel should be constitutively open , as the carboxyl group of the E125 should be able to form a salt bridge with R104 . We found that HeLa cells expressing Cx26K125E readily loaded with dye under control conditions and exhibited no sensitivity to CO2 ( Figure 4 ) . The Cx26K125E-expressing HeLa cells showed much greater loading under control conditions than parental HeLa cells ( Figure 4 , Figure 4—source data 1 ) . To further confirm that the constitutive dye loading occurred via the misexpressed connexin , we demonstrated that carbenoxolone ( 100 µM ) completely blocked CO2-dependent dye loading in HeLa cells expressing Cx26K125E ( Figure 4 , Figure 4—source data 1 ) . 10 . 7554/eLife . 01213 . 010Figure 4 . Engineering an analogue of the carbamylated lysine residue , Cx26K125E , creates a constitutively open hemichannel that no longer responds to CO2 . ( A ) HeLa cells expressing Cx26K125E readily load with dye under control conditions . Increasing the PCO2 does not give a further increase in dye loading . This dye loading was blocked by 100 µM carbenoxolone ( Carb ) , indicating that it occurred through the heterologously expressed connexin . Scale bar 40 µm . ( B ) Cumulative probability plots comparing the median pixel intensities of at least 40 cells per experiment and five independent repetitions for the control , hypercapnic and carbenoxolone treatments with that of parental HeLa cells ( four independent repetitions ) . ( C ) Summary data showing the median of the median pixel intensity for the three conditions for Cx26K125E and the background loading for parental HeLa cells . Pairwise comparisons by the Mann-Whitney U-test; KW Kruskall-Wallis Anova . Neither the difference between control and CO2 nor the difference between Cx26K125E treated with carbenoxolone and parental HeLa cells is significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 01010 . 7554/eLife . 01213 . 011Figure 4—source data 1 . Median pixel intensity values for histogram in Figure 4C and statistical analysis: Kruskal-Wallis anova and pairwise Mann-Whitney tests . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 011 Reasoning that if bridge formation between subunits was key to opening the hemichannel , we also made the further mutation R104E . In the mutated channel E104 has the potential to form a salt bridge in the reverse direction with K125 and we predicted that if this were to happen such a mutant hemichannel should also be constitutively open . We found that HeLa cells expressing Cx26R104E did indeed load with dye under control conditions and that this enhanced dye loading was blocked with carbenoxolone ( Figure 5 , Figure 5—source data 1 ) . 10 . 7554/eLife . 01213 . 012Figure 5 . Bridging in the reverse direction: the mutation R104E forms a salt bridge with K125 in Cx26R104E to create a constitutively open hemichannel that no longer responds to CO2 . ( A ) HeLa cells expressing Cx26R104E readily load with dye under control conditions . Increasing the PCO2 does not give a further increase in dye loading . This dye loading was blocked by 100 µM carbenoxolone ( Carb ) , indicating that it occurred through the heterologously expressed connexin . Scale bar 40 µm . ( B ) Cumulative probability plots comparing the median pixel intensities of at least 40 cells per experiment and five independent repetition for the control , hypercapnic and carbenoxolone treatments . ( C ) Summary data showing the median of the median pixel intensity for the three conditions for Cx26R104E . Pairwise comparisons by the Mann-Whitney U-test; KW Kruskall-Wallis Anova . The difference between control and CO2 is not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 01210 . 7554/eLife . 01213 . 013Figure 5—source data 1 . Median pixel intensity values for histogram in Figure 5C and statistical analysis: Kruskal-Wallis anova and pairwise Mann-Whitney tests . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 013 Although our experimental data point to the importance of carbamylation of K125 and the formation of a salt bridge to R104 in the adjacent subunit , it is not clear how this would lead to opening of the Cx26 hemichannel . Course-grained modelling reduces protein atomistic complexity for more efficient computational studies of harmonic protein dynamics and is particularly suited to examining hemichannel opening over millisecond time scales ( Sherwood et al . , 2008 ) . We therefore built coarse-grained elastic network models ( ENM ) to gain insight into the mechanism by which CO2 maintains Cx26 in the open state . In an ENM the Cα-atom co-ordinates of an atomic resolution structure are used to represent a protein structure . The total global protein motion within the ENM consists of a defined number of modes , each of a characteristic frequency and representing a particular harmonic motion within the protein . ENMs are known to reproduce the global low frequency modes of protein motion well in comparison to experimental data ( Delarue and Sanejouand , 2002; Valadie et al . , 2003 ) . We used the co-ordinates from a high-resolution crystal structure to construct an ENM ( Tirion , 1996 ) for the Cx26 hexamer in the unbound and CO2-bound states . CO2 was represented in the ENM by the inclusion of six additional Hookean springs between residues K125 and R104 of neighbouring monomers ( Figure 6A ) . 10 . 7554/eLife . 01213 . 014Figure 6 . Elastic network model ( ENM ) of Cx26 demonstrating that CO2 binding restricts the motion of the hemichannel and biases it to the open state . ( A ) Left , diagram of Cx26 from the 2zw3 structure , indicating the ENM ( black lines ) superimposed on the tertiary structure of Cx26 and showing the position of the hookean springs ( white lines ) introduced to simulate binding of CO2 to K125 and bridging to R104 . Right , ENM of Cx26 seen end on from the cytoplasmic side of the channel showing the six springs ( white lines ) that represent CO2 binding . ( B ) Frequency modes of channel motion plotted for CO2 bound against those without CO2 bound . The grey scale on the right indicates the similarity of the modes between the CO2 bound and unbound states . Note that there is a high degree of similarity between most modes in the bound and unbound state , indicating that binding of CO2 reorders the modes of motion . In the graph , the modes that fall on the dotted line ( x = y ) have not changed between the two states . Mode 1 without CO2 bound ( closing of hemichannel ) moves to Mode 9 with CO2 bound ( dashed upward arrow ) indicating that it contributes much less to the total channel motions when CO2 is bound . Most of the other modes fall below the dotted line , indicating that they become more frequent when CO2 is bound . ( C ) Vectors indicating the Mode 1 motions of the α helices without CO2 bound ( left ) and with CO2 bound ( right ) . The binding of CO2 and creation of the carbamylation bridge between subunits greatly restricts hemichannel motion . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 014 Analysis of the model revealed that in the absence of CO2 the lowest frequency mode ( mode 1 ) represented an opening/closing motion that was able to fully occlude the hemichannel central pore ( Video 1 ) . Addition of springs representing CO2-binding to the ENM restricted the closing motions of the hemichannel and thus connexin 26 was maintained in the open state ( Video 2 ) . We examined the overlap in the ordering of the modes in the unbound and CO2 bound states to gain insight into how this occurs . A significant reordering of the lowest frequency modes to higher frequencies was observed in the presence of CO2 rather than the removal of any modes from the total protein motion ( Figure 6B ) . Mode 1 , the lowest frequency mode that represents the opening/closing mode , represented about 40% of the total protein motion in the absence of CO2 . In the presence of CO2 this closing mode is reordered through a change in its frequency as mode 9 , which accounts for only about 2% of the total motion ( Figure 6B , C ) . CO2 therefore opens Cx26 through a reordering of the normal modes of global protein motion such that the normal closing motion of Cx26 no longer significantly contributes to the total motion of the hemichannel . We infer from this that the carbamate bridge formed between Cx26 monomers represents a constraining force that hinders hemichannel closure . 10 . 7554/eLife . 01213 . 015Video 1 . Hemichannel mode 1 motions in absence of CO2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 01510 . 7554/eLife . 01213 . 016Video 2 . Hemichannel mode 1 motions in presence of CO2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 016
Evidence from six different experimental tests supports our hypothesis that CO2 forms a carbamate bridge between K125 and R104 on the adjacent subunit to open the Cx26 hemichannel . Firstly , we demonstrated the sufficiency of the carbamylation motif to confer CO2 sensitivity by inserting it into a CO2-insensitive connexin , Cx31 . Secondly , we showed that K125 of the carbamylation motif was essential for this motif to confer CO2 sensitivity on Cx31 . Thirdly and fourthly , we demonstrated that the mutations K125R and R104A in Cx26 ( to prevent carbamate bridging at either end of the bridge ) destroyed the CO2 sensitivity of this connexin . Fifthly , by exploiting glutamate as an analogue of the carbamylated K125 ( in Cx26K125E ) , we demonstrated a gain of function—Cx26K125E was constitutively open , yet had lost sensitivity to CO2 . Finally , we further tested the bridging concept by demonstrating that the bridge is in effect bidirectional: the mutated hemichannel Cx26R104E , in which E104 can bridge to K125 in the reverse direction , was also constitutively open , but had no sensitivity to CO2 . Although we have not directly demonstrated CO2 binding to Cx26 , our extensive testing of this hypothesis through selective mutations leads us to conclude that CO2 interacts with Cx26 directly and that no other protein is required for CO2 sensitivity . This interaction is most probably via carbamylation of K125 . Interestingly , the mutations Cx26K125E and Cx26K125R can be considered respectively as open-state and closed-state analogues of the wild type channel . Collectively , our data strongly suggests that CO2 binds to the intracellular surface of Cx26 and must therefore cross the membrane to reach this site . This could occur either direct diffusion through the membrane bilayer , potentially via Cx26 itself , or via other CO2 permeable channels ( Boron et al . , 2011 ) . Amongst its many other functions , Cx26 can therefore be regarded as a receptor for CO2 . Interestingly , this mechanism of modulation applies to both Cx30 and Cx32 , which can both potentially form a carbamate bridge at equivalent residues to Cx26 . In the case of Cx32 this would involve bridging to K104 rather than R104 ( in Cx26 and 30 ) . Cx26 can co-assemble with both Cx30 and Cx32 to form heteromeric hemichannels ( Forge et al . , 2003; Yum et al . , 2007 ) . Our structural studies predict that , as Cx30 and Cx32 have K125 and either R104 or K104 , carbamate bridges could form in such heteromeric hemichannels and that they should also therefore be CO2-sensitive . Carbamylation involves formation of a labile covalent bond between the carbon of CO2 and a primary amine . For this to occur the amine must be in a restricted hydration space so that it is not protonated . Some examples of physiologically significant carbamylation are known . The carbamylation of the N-terminal amines of haemoglobin contributes to the Bohr effect ( Kilmartin and Rossi-Bernardi , 1971 ) , whereby the affinity of haemoglobin for O2 is reduced in the presence of elevated CO2 . However in mammalian systems no other examples of carbamylation by CO2 have been described . In C3 photosynthetic plants , the enzyme RuBisCo , that participates in the Calvin cycle and carbon fixation is activated by carbamylation of a specific lysine residue ( Lundqvist and Schneider , 1991 ) . Several microbial enzymes are also carbamylated ( Maveyraud et al . , 2000; Golemi et al . , 2001; Young et al . , 2008 ) . Despite this precedent , the functional significance of CO2-carbamylation and its potential as a transduction principle for the measurement of CO2 has been almost completely overlooked in vertebrate physiology . The mechanism of formation of a salt bridge between a carbamylated lysine and an appropriately oriented arginine on the neighbouring subunit is a unique mechanism for modulation of an ion channel and establishes carbamylation as a mechanistic basis for the direct signalling of PCO2 in mammalian physiology . This carbamylation of a lysine to transduce the concentration of CO2 into a biological signal is somewhat equivalent to the nitrosylation of a cysteine residue by NO/nitrite . It establishes a CO2-dependent signalling paradigm in which the concentration of CO2 is signalled by ATP release via Cx26 from the chemosensory cell and consequent activation of neighbouring cells , or potentially by a Ca2+ influx through the Cx26 hemichannel ( Fiori et al . , 2012 ) to initiate a Ca2+ wave within the chemosensory cell itself and further Ca2+-dependent signalling processes .
All connexin genes except Cx26R104A , Cx26K125E and Cx26R104E were synthesised by Genscript USA and subcloned into the pCAG-GS mCherry vector . The sequence for wild type Cx26 and Cx31 genes were respectively take from accession numbers NM_001004099 . 1 and NM_019240 . 1 . To produce Cx26R104A , Cx26K125E and Cx26R104E site directed mutagenesis was performed using Quikchange II site directed mutagenesis kit . All wild type and mutant genes were sequenced to verify that the correct sequence was present . Hela cells were maintained in Dulbecco’s modified Eagle’s medium ( DMEM ) ( Sigma-Aldrich Company Ltd , Gillingham , UK ) , 10% FCS ( Biosera Europe , Labtech International Ltd , Uckfield , UK ) , 1:1000 pen/strep and supplemented with 3 mM CaCl2 . Cells were grown in a humidified 5% CO2 incubator at 37°C . The connexin proteins were expressed via transient transfection . Cells were plated in six-well plates at 1 × 105 cells per well for Cx26 and its mutants and 5 × 104 cells per well for Cx31 and its mutants . Following the GeneJuice transfection reagent ( Merck-Millipore UK , Merck KGaA , Darmstadt , Germany ) user protocol , cells were transfected with 1 µg of the appropriate DNA . Experiments were performed when the connexin proteins had reached the cell membrane . This was found to be approximately 2 days for Cx26 and its mutants and approximately 3 days for Cx31 and its mutants . Connexin expressing HeLa cells were plated on cover slips . A coverslip was placed in a small flow chamber and the cells were exposed to either: control aCSF with 200 µM carboxyfluorescein for 10 min; isohydric hypercapnic aCSF with 200 µM carboxyfluorescein for 10 min; or zero Ca2+ , 1 mM EGTA-containing aCSF plus 200 µM carboxyfluorescein for 10 min . This was followed by control aCSF plus 200 µM carboxyfluorescein for 5 min and then thorough washing for 30 min with control aCSF . These protocols are summarized in Figure 7 . 10 . 7554/eLife . 01213 . 017Figure 7 . Dye loading protocols . The control background loading tests for any potential CO2-insensitive pathways of dye loading that are constitutively active in the HeLa cells . Hypercapnic dye loading uses the 50 mM HCO3− aCSF to test CO2-sensitive loading under conditions of isohydric hypercapnia ( PCO2 55 mmHg ) . The zero Ca2+ positive control tests for the presence of functional hemichannels in those cases where the misexpressed hemichannels exhibit no sensitivity to CO2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01213 . 017 The cells were then imaged by epifluorescence ( Scientifica Slice Scope ( Scientifica Ltd , Uckfield , UK ) , Cairn Research OptoLED illumination ( Cairn Research Limited , Faversham , UK ) , 60x water Olympus immersion objective , NA 1 . 0 ( Scientifica ) , Hamamatsu ImageEM EMCCD camera ( Hamamatsu Photonics K . K . , Japan ) , Metafluor software ( Cairn Research ) ) . Using ImageJ , the extent of dye loading was measured by drawing a region of interest ( ROI ) around individual cells and calculating the mean pixel intensity for the ROI . The mean pixel intensity of the background fluorescence was also measured in a representative ROI , and this value was subtracted from the measures obtained from the cells . All of the images displayed in the figures reflect this procedure in that the mean intensity of the pixels in a representative background ROI has been subtracted from every pixel of the image . At least 40 cells were measured in each condition , and the mean pixel intensities plotted as cumulative probability distributions . For the dye loading experiments , the median pixel intensities of the control and CO2 dye loading conditions ( minimum of five independent repetitions ) were compared by a Kruskal Wallace ANOVA and pairwise comparions by the Mann-Whitney test . The false discovery rate procedure ( Curran-Everett , 2000 ) was used to determine whether the multiple pairwise comparisons remained significant . Cover slips containing non-confluent cells were placed into a perfusion chamber at 28°C in sterile filtered standard aCSF . Standard patch clamp techniques were used to make whole-cell recordings . The intracellular fluid in the patch pipette contained: K-gluconate 120 mM , CsCl 10 mM , TEACl 10 mM , EGTA 10 mM , ATP 3 mM , MgCl2 1 mM , CaCl2 1 mM , sterile filtered , pH adjusted to 7 . 2 with KOH . All whole-cell recordings were performed at a holding potential of −40 mV with regular steps of 5 s to −50 mV to assess whole-cell conductance . Elastic network model ( ENM ) simulations were performed based on its regular implementation using pdb file 2ZW3 , where all the Cα atoms in the protein within a given cut-off radius are joined with simple Hookean spring ( Tirion , 1996; Rodgers et al . , 2013a ) . The spring constants were set to a constant value of 1 kcal mol−1 Å−2 with a cut-off radius of 8 Å . The presence of CO2 molecules were represented in the ENM by the inclusion of an additional Hookean spring between residues K125 and R104 of each set of neighbouring monomers ( Rodgers et al . , 2013b ) . The first six modes , that is the lowest frequency modes , represent the solid body translational and rotational motions of the protein and are thus ignored from the analysis . | A number of gaseous molecules , including nitric oxide and carbon monoxide , play important roles in many cellular processes by acting as signalling molecules . Surprisingly , however , it has long been assumed that carbon dioxide – a gaseous molecule that is produced during cellular metabolism – is not a signalling molecule . Controlling the concentration of carbon dioxide ( CO2 ) in a biological system is essential to sustain life , and it was thought that the body used pH – which is the concentration of hydrogen ions – as a proxy for the level of CO2 . The concentration of CO2 is related to pH because CO2 reacts with water to form carbonic acid , which quickly breaks down to form hydrogen ions and bicarbonate ions . This close relationship has led many researchers to equate pH-sensing with CO2-sensing , and to suggest that a physiological receptor for CO2 does not exist . Recent research into structures called connexin hemichannels has challenged this view . Researchers found that when pH levels were held constant , increasing the level of CO2 caused the structures to open up , suggesting that CO2 could be directly detected by the hemichannels . Each hemichannel contains six connexin subunits , but the details of how the CO2 molecules interact with the individual connexin subunits to open up the hemichannels remained mysterious . Now Meigh et al . show that CO2 molecules bind to a specific amino acid ( lysine ) at a particular place ( residue 125 ) in one of the connexin subunits to form a carbamate group . This group then interacts with the amino acid ( arginine ) at residue 104 in a neighbouring connexin subunit to form a carbamate bridge between the two subunits . This leads to structural changes that cause the gap junction hemichannels to open and release signals that can activate other cells . Since connexin hemichannels are found throughout the human body , these results suggest that CO2 might act as a signalling molecule in processes as diverse as the control of blood flow , breathing , hearing and reproduction . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2013 | CO2 directly modulates connexin 26 by formation of carbamate bridges between subunits |
The chromatophore of purple bacteria is an intracellular spherical vesicle that exists in numerous copies in the cell and that efficiently converts sunlight into ATP synthesis , operating typically under low light conditions . Building on an atomic-level structural model of a low-light-adapted chromatophore vesicle from Rhodobacter sphaeroides , we investigate the cooperation between more than a hundred protein complexes in the vesicle . The steady-state ATP production rate as a function of incident light intensity is determined after identifying quinol turnover at the cytochrome bc1 complex ( cytbc1 ) as rate limiting and assuming that the quinone/quinol pool of about 900 molecules acts in a quasi-stationary state . For an illumination condition equivalent to 1% of full sunlight , the vesicle exhibits an ATP production rate of 82 ATP molecules/s . The energy conversion efficiency of ATP synthesis at illuminations corresponding to 1%–5% of full sunlight is calculated to be 0 . 12–0 . 04 , respectively . The vesicle stoichiometry , evolutionarily adapted to the low light intensities in the habitat of purple bacteria , is suboptimal for steady-state ATP turnover for the benefit of protection against over-illumination .
Energy for most life on Earth is provided by sunlight harvested by photosynthetic organisms , which have evolved a wide variety of mechanisms for utilizing light energy to drive cellular processes ( Blankenship , 2014 ) . These organisms absorb sunlight and subsequently utilize the Förster mechanism ( Sener et al . , 2011 ) and quantum coherence ( Strümpfer et al . , 2012; Panitchayangkoon et al . , 2010; Scholes , 2010 ) for efficient excitation energy transfer , followed by conversion of light energy into chemical energy ( Feniouk and Junge , 2009 ) . The light harvesting system of purple bacteria ( Hu et al . , 2002; Cartron et al . , 2014 ) is claimed to be the earliest of the current photosynthetic lineages ( Xiong et al . , 2000 ) and exhibits , at the supra-molecular level as well as at the level of individual proteins , less complexity than the thylakoid membranes of the more ubiquitous cyanobacteria and plants ( Kirchhoff et al . , 2002 ) . In the purple bacterium Rhodobacter ( Rba . ) sphaeroides the basic photosynthetic unit is the chromatophore ( Cogdell et al . , 2006; Strümpfer et al . , 2011; Cartron et al . , 2014 ) , a 50–70 nm diameter vesicle , as shown in Figure 1 , formed through invagination of the intracytoplasmic membrane ( Tucker et al . , 2010; Gubellini et al . , 2007 ) and comprising over a hundred protein complexes ( Jackson et al . , 2012; Woronowicz and Niederman , 2010; Woronowicz et al . , 2013 ) . The proteins that constitute the chromatophore are primarily the light harvesting ( LH ) complexes , photosynthetic reaction centers ( RCs ) , cytbc1 complexes , and ATP synthases , which cooperate to harvest light energy for photophosphorylation . The architecture of the chromatophore , reported in ( Şener et al . , 2007 , 2010; Cartron et al . , 2014 ) , has been determined by combining atomic force microscopy ( AFM ) ( Bahatyrova et al . , 2004; Olsen et al . , 2008 ) , cryo-electron microscopy ( cryo-EM ) ( Qian et al . , 2005; Cartron et al . , 2014 ) , crystallography ( Koepke et al . , 1996; McDermott et al . , 1995; Papiz et al . , 2003; Jamieson et al . , 2002 ) , optical spectroscopy ( Hunter et al . , 1985; Sener et al . , 2010 ) , mass spectroscopy ( Cartron et al . , 2014 ) , and proteomics ( Jackson et al . , 2012; Woronowicz and Niederman , 2010; Woronowicz et al . , 2013 ) data . The composition of the chromatophore depends on growth conditions such as light intensity ( Adams and Hunter , 2012; Woronowicz et al . , 2011b , 2011a ) and can also be influenced by mutations ( Siebert et al . , 2004; Hsin et al . , 2010b ) . 10 . 7554/eLife . 09541 . 003Figure 1 . Atomic structural model of a low-light-adapted chromatophore vesicle from Rba . sphaeroides . The model is based on AFM , EM , crystallography , mass spectroscopy , proteomics , and optical spectroscopy data ( Cartron et al . , 2014 ) . The inner diameter of the vesicle is 50 nm . The model considered in this study is a variant of the one reported in ( Cartron et al . , 2014 ) , which features 63 LH2 complexes ( green ) , 11 dimeric and 2 monomeric RC-LH1-PufX complexes ( LH1:red; RC:blue; PufX:lime ) , 4 cytbc1 ( magenta ) , and 2 ATP synthases ( orange ) , as well as 2469 BChls and 1542 carotenoids . Proteins are shown in surface representation . ( A ) Proteins and BChls of the chromatophore . Some of the light harvesting proteins are rendered transparent to reveal their BChl pigments . BChls are represented by their porphyrin rings only . See Video 1 presenting the vesicle . ( B ) Close-up of chromatophore showing its lipid membrane ( transparent ) along with its proteins colored as in ( A ) . The membrane of 16 , 000 lipids contains the quinone/quinol pool of about 900 molecules . Energy conversion in the chromatophore proceeds in three stages: ( I ) light harvesting and electron transfer reducing the quinone pool; ( II ) quinone/quinol diffusion and exchange of quinols for quinones at cytbc1 ( thereby generating a proton gradient across the vesicle membrane ) as well as diffusive motion of cytochrome c2 inside the chromatophore shuttling single electrons from cytbc1 to RC; ( III ) utilization of the proton gradient for ATP synthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 09541 . 00310 . 7554/eLife . 09541 . 004Video 1 . Chromatophore structural model . A movie that shows a detailed structural model for the low-light adapted chromatophore vesicle as displayed in Figure 1 ( Cartron et al . , 2014 ) . Presented is a rotating view of the vesicle comprising LH2 complexes ( green ) , dimeric RC-LH1-PufX complexes ( red-blue-lime green ) , dimeric cytbc1 complexes ( magenta ) , and ATP synthases ( orange ) . For half of the model , proteins are shown in solid surface representation , and for the other half , proteins are shown as transparent surfaces with bacteriochlorophylls ( BChls ) , represented by their porphyrin rings , shown as solid surfaces . DOI: http://dx . doi . org/10 . 7554/eLife . 09541 . 004 The chromatophore displays organizational principles for the integration of multiple processes ( Sener et al . , 2010 ) . Evolutionary competition at the organism level has driven photosynthetic subsystems toward optimal and robust function ( Noy et al . , 2006; Noy , 2008; Scholes et al . , 2011 ) that can guide the design of artificial light harvesting devices such as biohybrid antennas ( Harris et al . , 2013 ) and nanopatterned light harvesting ( LH ) complex arrays ( Reynolds et al . , 2007; Vasilev et al . , 2014; Patole et al . , 2015 ) . The development and improvement of such artificial , biological , or biohybrid light harvesting systems may alleviate mankind’s future energy demand ( Blankenship et al . , 2011 ) . The functional principles displayed by the chromatophore and prevalent also in other photosynthetic systems include efficient excitonic coupling between components ( Hu et al . , 1997 , 1998; van Grondelle and Novoderezhkin , 2006b; Olaya-Castro et al . , 2008; Sener et al . , 2011 ) , the utilization of quantum coherence ( Ishizaki and Fleming , 2009a; Strümpfer et al . , 2012 ) , photoprotection by carotenoids ( Damjanović et al . , 1999 ) , accommodation of thermal fluctuations , studied through experimental ( Visscher et al . , 1989; van Grondelle et al . , 1994; Pullerits et al . , 1994; Gobets et al . , 2001; Janusonis et al . , 2008; Freiberg et al . , 2009 ) as well as theoretical ( Damjanović et al . , 2002; Şener and Schulten , 2002; Ishizaki and Fleming , 2009b; van Grondelle and Novoderezhkin , 2006b; Strümpfer and Schulten , 2011 , 2012a ) methods . The chromatophore exhibits also the features of modularity , repair , and assembly of components ( Hsin et al . , 2010a ) , high quantum yield of organelle-scale pigment networks ( Şener et al . , 2007 , 2010; Cartron et al . , 2014 ) , isolation of the electron transfer chains ( Şener and Schulten , 2008 ) , co-accommodation of competing functions such as efficient energy transfer and diffusion in the quinone/quinol pool ( Lavergne et al . , 2009; Sener et al . , 2010 ) , as well as adaptation to changing external conditions ( Adams and Hunter , 2012; Woronowicz et al . , 2011a; Niederman , 2013; Woronowicz et al . , 2013 ) . Energy conversion in the chromatophore proceeds in three stages as discussed below: ( i ) light harvesting and charge separation , converting quinone into quinol at a RC; ( ii ) diffusion of quinone/quinol in the chromatophore membrane and cytochrome c2 diffusion inside the chromatophore vesicle , resulting , at cytbc1 , in the generation of a proton gradient as well as a transmembrane electrochemical gradient across the chromatophore membrane ( henceforth referred to collectively as proton gradient ) ; ( iii ) utilization of the proton gradient , culminating in ADP binding and ATP release at ATP synthase . The quinone/quinol as well as the generated proton-motive force function as energy buffers between light harvesting and ATP synthesis stages . The proton gradient along with the redox states of the quinone/quinol pool are influenced by the enzymes succinate dehydrogenase , NADH dehydrogenase , cytochrome c oxidase , and ubiquinol oxidase ( Bowyer et al . , 1985; Klamt et al . , 2008 ) . A summary for the energy conversion processes in the chromatophore can be found in ( Klamt et al . , 2008; Sener et al . , 2014 ) . In addition to ATP synthesis , the chromatophore utilizes the generated proton motive force also for NADH production via NADH dehydrogenase ( Klamt et al . , 2008 ) and , thereby , for control of the quinone/quinol pool redox state . Other channels for proton gradient depletion are flagellar motility ( Kojadinovic et al . , 2013 ) and proton leak across the vesicle membrane . In the present study , we focus on the overall energy conversion characteristics of the molecular components identified in the current structural model ( Cartron et al . , 2014 ) , namely LH2 , RC-LH1 , cytbc1 , and ATP synthase , where NADH dehydrogenase plays an indirect role . Efficient energy conversion requires some degree of robustness with respect to supramolecular organization , since no two chromatophores are likely to be identical . Though chromatophore vesicles share structural motifs ( Bahatyrova et al . , 2004; Cartron et al . , 2014 ) that vary gradually with growth conditions , inevitable irregularities in the distribution of their constituent proteins and their quinone/quinol pools render chromatophores heterogeneous , requiring energy conversion processes to be insensitive to structural inhomogeneity . Robustness in photosynthetic systems had been demonstrated computationally for the excitation transfer step of light harvesting at the single protein level with respect to loss or rearrangement of pigments ( Şener et al . , 2002 ) as well as against fluctuations of pigment site-energies ( Damjanovic et al . , 2002 ) and at the vesicle level against deformations of the pigment network ( Sener et al . , 2010 ) . Efficiency of energy conversion in a photosynthetic system is not straightforward to define , since it involves multiple interrelated subprocesses spanning both quantum mechanical and classical domains over timescales ranging from picoseconds to milliseconds ( Blankenship , 2014; van Amerongen et al . , 2000 ) . A simple measure of conversion efficiency at the light harvesting stage is provided by the quantum yield , q , defined as the probability , upon the absorption of a photon by any pigment of the chromatophore , of charge separation at any RC ready for excitation-induced electron transfer to quinone . The quantum yield is solely a function of pigment network geometry , is independent of incident light intensity , and is found to be close to unity ( Strümpfer et al . , 2012; Sener et al . , 2011 ) for initial chlorophyll light absorption; in case of carotenoid light absorption the quantum yield can be lower due to so-called covalent electronic excitation as argued in ( Ritz et al . , 2000b ) . Since excitation transfer does not constitute a rate-limiting step of photosynthetic energy conversion , the quantum yield is not a major limiting factor for the overall efficiency of the chromatophore . A comprehensive measure of chromatophore efficiency that also permits a limited comparison with photovoltaic systems is the conversion efficiency of captured solar energy to the chemical energy of the final photoproduct , namely ATP . Earlier studies of photosynthetic membrane systems include percolation theory-based models of quinone diffusion in Rhodospirillum ( Rsp . ) photometricum membranes ( Scheuring et al . , 2006 ) , plastoquinone diffusion in thylakoid membranes ( Kirchhoff et al . , 2002 ) , models of dissipative photoprotective behavior in Rsp . photometricum membranes ( Caycedo-Soler et al . , 2010 ) , and stoichiometry-based rate kinetics ( Geyer et al . , 2007 , 2010 ) . In fact , long before any structural information of the light harvesting apparatus of purple bacteria was available , Vredenberg and Duysens ( Vredenberg and Duysens , 1963 ) postulated that the total fluorescence yield can be expressed in terms of the ratio of closed and open RCs , after which random-walk models of excitation transfer were developed using a master equation formalism ( Den Hollander et al . , 1983 ) . Prior to the availability of AFM imaging data ( Bahatyrova et al . , 2004 ) , the supramolecular organization of chromatophores was suggested to feature RCs partially surrounded by LH-complexes facilitating efficient shuttling of quinones ( Joliot et al . , 1990 , 1996; Jungas et al . , 1999 ) . The aim of the present study is to determine , based on a supramolecular structural model ( Cartron et al . , 2014 ) , for the chromatophore of Rba . sphaeroides the ATP production rate as a function of illumination and vesicle stoichiometry along with the corresponding energy conversion efficiency . A low-light adapted chromatophore vesicle model is considered ( Cartron et al . , 2014 ) , since low-light illumination , namely ≲10% of full sunlight , is typical for the habitat of purple bacteria ( Woronowicz and Niederman , 2010; Blankenship , 2014 ) . The quantum yield of excitation transfer for the pigment network geometry shown in Figure 1 is determined in terms of an effective Hamiltonian formulation . The processes subsequent to charge separation and the corresponding rate kinetics of ATP production are described in terms of chromatophore vesicle stoichiometry , instead of at atomic detail , by identifying rate-limiting steps . The organizational optimization of the chromatophore is considered in terms of the dependence of energy conversion on vesicle composition and illumination conditions .
Previous studies showed that the quantum yield of excitation transfer , q , computed through Equation 8 below and discussed in greater detail in Supplementary Materials , is very high , namely , 85–94% , varying gradually with LH2:RC stoichiometry ( Şener et al . , 2007 , 2010 , 2011 ) . For the vesicle presented in Figure 1 the quantum yield , q , has a value of 0 . 91 , consistent with earlier studies ( Şener et al . , 2007 , 2010; Cartron et al . , 2014 ) . Such high value for the quantum yield , close to the ideal limit of 1 , is achieved because loss due to internal conversion and fluorescence arises much more slowly ( rates about ( 1 ns ) −1 ) than excitation transfer or charge separation at RC ( rates about ( 10 ps ) −1 ) . Clearly , the quantum yield does not constitute a limiting factor for the overall energy conversion efficiency in the chromatophore . At very low light intensity , nearly all electronic excitation delivered to RCs contribute to the generation of a proton gradient across the membrane and to eventual ATP synthesis . With increasing light intensity , the cycling time of quinones at the RC , τRC ( I ) as given by Equation 19 , increases; fewer RCs are found in a state available to receive photoexcitation , described by the probability , pRC ( I ) , given by Equation 13 , and resulting in a corresponding loss of electronic excitation . The time-scale with which the quinone/quinol pool redox state adapts to a change in light conditions is reported to be about 0 . 5 s ( Woronowicz et al . , 2011a ) . The ATP turnover rate , kATP , calculated according to Equation ( 20 ) , and the energy conversion efficiency , ηATP , calculated according to Equation ( 21 ) , for a low-light adapted chromatophore vesicle ( Figure 1 ) under steady-state illumination are presented in Figure 2 . At light intensities equivalent to 1% and 3% of full sunlight , the vesicle is found to produce ATP molecules at a rate of 82 s−1 and 118 s−1 , respectively . At the high-light limit , the ATP synthesis rate approaches 158 molecules s−1 . These rates are consistent with experimental observations for continuous light-induced photophosphorylation , reported to be in the range of 0 . 017 molecules per BChl per second ( Saphon et al . , 1975 ) and 0 . 05 ATP molecules per BChl per second ( Clark et al . , 1983 ) , corresponding to ~43 ATP molecules s−1 and ~130 ATP molecules s−1 , respectively , for the vesicle shown in Figure 1 . We note that the Clark estimate was reported for Rhodopseudomonas capsulata . The corresponding energy conversion efficiency , ηATP , at the stated low-light intensities of 1% and 3% of full sunlight , calculated in the present study , are 12% and 7% , respectively . Notably , an upper-limit of 30% was estimated in ( Hellingwerf et al . , 1993 ) for the conversion efficiency of photosynthesis in Rba . sphaeroides . In comparison , the efficiency value , ηATP , computed for the recently established model ( Cartron et al . , 2014 ) of the chromatophore ( Figure 2 ) ranges between 0%–17% . 10 . 7554/eLife . 09541 . 005Figure 2 . ATP production rate and energy conversion efficiency . ( A ) Steady-state ATP production rate , kATP , calculated according to Equation ( 20 ) , and ( B ) energy conversion efficiency , ηATP , calculated according to Equation ( 21 ) , of a chromatophore vesicle as a function of incident light intensity Φ . Solid curves correspond to the vesicle shown in Figure 1; dashed curves represent a similar vesicle with only a single cytbc1 . The vertical lines denote the light intensities corresponding to ( I ) 3% of full sunlight ( 30 W/m2 ) , a typical growth condition for purple bacteria , and ( II ) full sunlight ( 1 kW/m2 ) , respectively . Thus , for light intensities typical for the habitat of purple bacteria ( 1–5% of full sunlight; shaded area ) the energy conversion efficiency ηATP of a chromatophore vesicle is between 0 . 12–0 . 04 . DOI: http://dx . doi . org/10 . 7554/eLife . 09541 . 005 The lower efficiency values , ηATP , for the chromatophore at higher light intensities ( Figure 2B ) does not indicate a failing , since the chromatophore does actually produce slightly more ATP in high-light than in low-light illumination ( Figure 2A ) , but rather reflects the optimization of purple bacteria for a low-light intensity habitat . The saturation of ATP synthesis with increasing light intensity seen in Figure 2 arises because quinol turnover capacity at the cytbc1 complex becomes rate limiting at higher light intensities . The rate limiting property of cytbc1 complexes was suggested by earlier studies ( Lavergne et al . , 2009; Geyer et al . , 2010 ) and is discussed below in connection with Equation ( 14 ) in Materials and methods . The maximal electron processing capacity of all cytbc1 complexes is estimated ( see Equation ( 14 ) in Materials and methods ) to be 2×nBτB-1=320s-1 , where nB=4 is the number of cytbc1 dimers and τB = 25 ms is the quinol turnover time at cytbc1 ( Crofts , 2004 ) and the prefactor 2 accounts for every quinol transferring two electrons . The electron processing capacity at cytbc1 becomes equal to the total RC electron turnover rate , Iq , at a light intensity of 9 W/m2 , i . e . , at approximately 1% of full sunlight . As illumination exceeds this low-light value , RC electron turnover is limited by the electron processing capacity of cytbc1 , leading to a gradual saturation of proton gradient formation and ATP turnover , as seen in Figure 2A , thereby reducing the efficiency of ATP synthesis ( Figure 2B ) . Rate limitation of ATP synthesis by the cytbc1 turnover capacity can be related also to the availability of quinones at the RC . In the absence of bound quinone , excitations delivered to a RC are wasted , except if the excitation escapes from the RC and reaches another RC ready for quinone reduction , especially within the same RC-LH1 dimer . However , with increasing illumination the likelihood of nearby RCs having quinones available also diminishes , excitation energy is lost , and energy conversion efficiency is reduced . The probability of a RC being ready for quinone reduction , pRC ( I ) , is given by Equation ( 13 ) in the Materials and methods section . With increasing illumination , pRC ( I ) decreases , thereby reducing the overall conversion efficiency , ηATP . At 1% of full sunlight , pRC ( I ) assumess the value of 0 . 73 , which , according to Equation ( 13 ) , drops to 0 . 23 at 5% of full sunlight . Remarkably , the role of closed and open RCs in determining the overall efficiency of the photosynthesic apparatus had been already pointed out long before any structural details were known ( Vredenberg and Duysens , 1963 ) . The rate limiting effect of cytbc1 can be further illustrated considering the efficiency of chromatophores with fewer cytbc1 complexes compared to the ones shown in Figure 1 . As indicated by the dotted lines in Figure 2A , B , describing ATP synthesis in a chromatophore with a single cytbc1 dimer , a lower number of cytbc1 dimers results in a reduction of the ATP production rate , kATP , and , accordingly , in a lower conversion efficiency , ηATP . A comparison with plant light harvesting efficiencies is not straightforward: efficiency for biomass production is significantly lower than the aforementioned thermodynamic efficiency; in fact , only as little as 1% of total incident solar energy is stored by crop plants as biomass ( Blankenship et al . , 2011 ) . One might wonder how the chromatophore compares to engineered photovoltaic devices . At peak solar intensity photovoltaic-driven electrolysis is reported to have an energy conversion efficiency of 5–15% ( Blankenship et al . , 2011 ) . However , comparison of efficiency alone overlooks issues such as stability and reclaimability of the energy stored in the final products . More refined measures of efficiency need to include total integrated cost of components , life expectancy , repair and maintenance . Evolutionary pressure toward greater fitness at the organism level results in the composition and architecture of photosynthetic systems to display adaptation toward optimal function ( Xiong et al . , 2000; Şener and Schulten , 2008; Blankenship , 2014 ) . Such adaptation has been reported for the individual protein level; it is not as well understood at the system integration level . For instance , pigment networks of individual light harvesting proteins were reported to display optimality and robustness in their quantum yield with respect to the spatial organization of pigments and the site energy distribution ( Şener et al . , 2002; Noy et al . , 2006; Damjanovic et al . , 2002 ) ; a similar robustness was reported with respect to size-scaling deformations of an entire vesicle ( Sener et al . , 2010 ) . Prior studies did not take into account optimization of the complete energy conversion process , including ATP synthesis , the effects of vesicle composition influenced by growth conditions such as light intensity ( Niederman , 2013; Woronowicz et al . , 2013 ) , the regulation of the redox state of the quinone/quinol pool ( Klamt et al . , 2008 ) , or the effects of cell-scale concentration and connectivity of chromatophores also influenced by light intensity at growth ( Tucker et al . , 2010 ) . In the following , the effect of vesicle composition on the ATP turnover rate is examined in order to determine the degree of optimality of the vesicle composition for ATP production . The vesicle shown in Figure 1 is used as a reference point for comparison with chromatophores of alternate composition , As composition variables , the number of dimeric cytbc1 complexes , nB , and the number of dimeric RC-LH1-PufX complexes , nL , are considered for a two-parameter ( nB , nL ) family of vesicles with the same surface area as the reference vesicle ( Figure 1 ) . The dependence of the steady-state ATP turnover rate , kATP ( nB , nL;I ) , on nB and nL is determined according to Equations 17 , 19 , 20 , where nL=2×nRC . In order to avoid massive computation , vesicles are not constructed explicitly . Instead , the corresponding quantum yield q is estimated by a linear interpolation on the LH2:RC stoichiometry based on earlier reported values ( Şener et al . , 2007 , 2010 , 2011 ) as described in Materials and methods ( Equation 10 ) . Since q varies very little with vesicle composition , the dependence of kATP on composition is dominated primarily by the explicit nB and nRC dependence in Equations 19 , 20 . The rate kATP ( nB , nL;I ) is shown in Figure 3 for light intensities equal to 1% and 3% of full sunlight . The respective ATP synthesis rates for the reference vesicle in Figure 1 under these two illumination conditions are 82 and 118 ATP molecules/s , respectively , ( marked by circles in Figure 3 ) which corresponds to 79% and 50% of the maximum possible rate ( marked by crosses in Figure 3 ) among all possible ( nB , nL ) values at that illumination . Clearly , steady state ATP synthesis is not optimized by the vesicle composition shown in Figure 1 . The turnover rate , kATP , would be improved by an nB:nL ratio that is greater than the native value of 1:3 ( Crofts , 2004; Cartron et al . , 2014 ) , as suggested also by a comparison of the turnover times at cytbc1 and RC ( τB/τL≃8 ) . 10 . 7554/eLife . 09541 . 006Figure 3 . Effect of vesicle composition on steady-state ATP production at different light intensities . Vesicle composition is given in terms of the number of cytbc1 dimers ( nB ) and of RC-LH1-PufX dimers ( nL ) for vesicles featuring identical surface area; LH2 composition of the vesicle is determined by considering the vesicle shown in Figure 1 as a reference point and adjusting the number of LH2 complexes to compensate for the changes in the number of cytbc1 and RC-LH1-PufX dimers to cover the vesicle surface . ATP production rate , kATP , is shown for ( A ) 1% of full sunlight ( 10 W/m2 ) and ( B ) 3% of full sunlight ( 30 W/m2 ) , determined according to Equation ( 20 ) . The two RC-LH1-PufX monomers of the vesicle in Figure 1 were counted as a single dimer for the purposes of this plot . The reference vesicle ( Figure 1 ) is represented by a circle , corresponding to an ATP production rate of 82 s−1 ( 118 s−1 ) , i . e . 79% ( 51% ) of the maximum possible rate among all stoichiometries , for 1% ( 3% ) of full sun light . The optimal vesicle composition for each illumination is represented by a cross; the corresponding LH2 count for optimal composition is 93 ( 74 ) at 1% ( 3% ) of full sunlight as compared with 63 for the reference vesicle ( circle ) . The ATP production rate is marginally greater for vesicles that contain more cytbc1 and LH2 complexes at the expense of fewer RC-LH1-PufX complexes as compared with the reference vesicle . This increase in ATP production rate results from cytbc1 being the rate-limiting component in the energy conversion process . DOI: http://dx . doi . org/10 . 7554/eLife . 09541 . 006 A reason for the aforementioned suboptimal ( nB , nL ) values in native low-light adapted vesicles might be protection against light-induced damage that can arise at high illumination via destruction of the vesicle membrane through overacidification . Though typical illumination levels in habitats of purple bacteria are low , occasional surges in light intensity are inevitable . During sustained ( >1 s ) high illumination intervals , a proton turnover unhindered by a low ( nB=4 ) cytbc1 stoichiometry can exceed the turnover capacity of the ATP synthases , resulting in overacidification of the vesicle interior , harming the integrity of the chromatophore membrane and its proteins . The observed nB value of 4 , apparently suboptimal for most light intensities ( Figure 3 ) , ensures that during sustained over-illumination proton turnover is limited by cytbc1 to a rate below the synthesis capacity of ATP synthase ( Lavergne et al . , 2009; Geyer et al . , 2010 ) , thus preventing overacidification . The ( nB , nL ) value also has an effect on the size of the quinone/quinol pool relevant for intermittent energy storage under fluctuating light conditions , since the number of quinones in the system correlates with the number of RCs ( Comayras et al . , 2005; Woronowicz et al . , 2011a; Cartron et al . , 2014 ) . Energy conversion through the quinone/quinol pool also involves electron exchange processes from outside the chromatophore as furnished , for example , through the enzymes NADH dehydrogenase and succinate dehydrogenase ( Klamt et al . , 2008 ) . The turnover capacities of cytbc1 and ATP synthase are compared in Materials and methods in relation to the rate limitation of energy conversion by the cytbc1 . A single ATP synthase is sufficient to take advantage of proton turnover of an entire chromatophore ( Etzold et al . , 1997 ) . Additional ATP synthases reported in chromatophore vesicles ( Cartron et al . , 2014 ) appear to provide necessary redundancy , since an isolated chromatophore without ATP synthase is non-functional . In this regard , it is of interest that vesicles have been found to occasionally fuse through formation of membrane tubes ( Tucker et al . , 2010 ) permitting passage of protons between neighboring chromatophore vesicles , thereby sharing their proton gradients with the ATP synthases of multiple vesicles , reducing the need for back-up ATP synthases and even permitting less than one ATP synthase per vesicle . Robustness requirements for protecting the vesicle against damage under environmental strain apparently supersede optimality constraints for steady state conditions . A photosynthetic vesicle adapted for steady-state illumination at higher light intensities than considered in this study would require a larger number of cytbc1 to maximize ATP production , along with more than the 1–2 ATP synthases observed per vesicle ( Cartron et al . , 2014 ) .
The combined structural and functional model of a low-light adapted chromatophore ( Cartron et al . , 2014 ) permits a quantitative description of ATP synthesis at different light intensities . The energy conversion efficiency , ηATP , is determined to be ~12%–4% at the low-light conditions typical for purple bacterial habitats ( 1%–5% of full sunlight ) , dropping rapidly to ≲0 . 1% beyond full sunlight conditions . Moderate levels of illumination saturate the bacterial light harvesting apparatus lowering its efficiency , whereas plants and photovoltaic devices function efficiently at high light intensities . The efficiency curve determined in the present study for the purple bacterial chromatophore ( Figure 2B ) indicates specialization for low-light intensities . The primary rate-limiting component among the energy conversion steps in the chromatophore appears to be quinol turnover at cytbc1 , as discussed in Section 4 . 2 . The rate limitation at cytbc1 , as compared with the ATP synthase turnover capacity , prevents the generation of an overly strong proton gradient at sustained high-light conditions , thereby protecting the chromatophore against overacidification of its interior and assuring vesicle integrity . As the light intensity I increases , photoexcitations are more likely to be dissipated as the probability for a RC to have a quinone or semiquinone ready to accept an electron , pRC ( I ) , decreases . The chromatophore composition appears to be suboptimal for ATP production under steady-state illumination . The chromatophore is apparently a highly specialized device that performs its energy conversion function robustly for low average light intensity , while featuring protective measures that dissipate energy at higher light intensity . Robustness against damage , such as overacidification of the membrane due to sustained overillumination , appears to supersede optimality under idealized conditions , such as steady state illumination . The present study focuses on steady state energy harvesting in the chromatophore without explicitly modeling the spatial dynamics of the charge carriers ( quinone/quinol and cytochrome c2 ) , the redox states of the proteins ( RC and cytbc1 ) , proton leakage through the membrane , or the coupling of NADH dehydrogenase to the proton-motive force . A more complete description of chromatophore function requires placement of NADH dehydrogenase , along with possibly succinate dehydrogenase , cytochrome c oxidase , and ubiquinol oxidase in the chromatophore membrane , the presence of which would also affect the energy conversion efficiency determined in this study . The added enzymes need to be described along with their reactions with redox partners located in the cell’s cytoplasm . In particular , a non-steady state formulation is necessary to account for spatial heterogeneity and light intensity dependence of the redox states of the proteins and the charge carriers in the chromatophore . The present study differs from earlier studies in functional modeling of the chromatophore ( Geyer et al . , 2007 , 2010 ) in several respects: first , it is based on an explicit atomic-detail structural model; second , instead of employing many ( over 30 ) adjustable parameters , few experimentally determined rate constants are employed to describe the rate determining steps; third , a steady-state description is chosen such that energy conversion steps that are not rate-limiting can be left out of the kinetic model . Nonetheless , earlier and present studies give similar results for the overall ATP synthesis rate at saturation , since this rate is determined largely by the total turnover capacity of cytbc1 complexes as a rate-limiting component . A key role in chromatophore energy conversion involves proton translocation , generating and using proton motive force . The present treatment does not resolve individual translocation steps , but rather assumes that the individual steps taking place at the overall RC , cytbc1 , and ATP synthase proton reactions can be treated as a single reaction event . Primary conclusions reached presently would not be affected by a more detailed description , i . e . , the model is robust with respect to the neglect of explicit modeling of individual proton translocation steps and proton motive force conversion . Integrative models of organelle function such as the one presented here provide a bridge between experimental methods that do not resolve temporal and spatial detail needed for establishing physical mechanisms and microscopic simulations that span the multiple length and time scales relevant for the function of living cells .
As already stated , the structural model of the chromatophore considered in the present study is a variation of the model reported in ( Cartron et al . , 2014 ) . The primary components of chromatophore vesicles in purple bacteria , as depicted in Figure 1 , are , in order of energy utililization ( Cogdell et al . , 2006 , Cartron et al . , 2014 ) : ( i ) light harvesting complex 2 ( LH2 ) ( Koepke et al . , 1996; Papiz et al . , 2003 ) ; ( ii ) light harvesting complex 1 ( LH1 [Qian et al . , 2008; Sener et al . , 2009] ) ; ( iii ) RC ( Jamieson et al . , 2002; Strümpfer and Schulten , 2012a ) ; ( iv ) cytbc1 ( Crofts , 2004; Crofts et al . , 2006 ) ; and ( v ) ATP synthase ( Feniouk and Junge , 2009; Hakobyan et al . , 2012 ) . RC-LH1 complexes typically form dimeric RC-LH1-PufX complexes facilitated by the polypeptide PufX ( Qian et al . , 2013; Sener et al . , 2009 ) , although monomeric complexes are also found in membranes from photosynthetically grown cells at a ratio of approximately 10% ( Olsen et al . , 2008 ) . The chromatophore in Figure 1 exhibits for the LH2:RC complexes a stoichiometry of 2 . 6:1 and corresponds to a low-light-adapted vesicle as described in ( Sener et al . , 2010; Cartron et al . , 2014 ) . In a typical vesicle , about a hundred protein complexes , LH2 and RC-LH1-PufX , form an efficient light harvesting network ( Şener et al . , 2007 , 2010 ) supplying electronic excitation energy for the conversion of quinones to quinols . The quinols produced at the RC are converted back to quinones by cytbc1 to generate a proton gradient across the chromatophore vesicle membrane , which , in turn , is consumed by the ATP synthase for the synthesis of ATP from ADP and phosphate . The electrons from quinol-to-quinone conversion are shuttled back to the RC by cytochrome c2 acting inside the vesicle . These energy conversion processes are illustrated in Figure 4 . We note that the experimental data ( Saphon et al . , 1975; Clark et al . , 1983 ) used to test the present energy conversion model based on ( Cartron et al . , 2014 ) were not obtained with chromatophores in vivo , but for a suspension of chromatophores in a pH-buffer; the energy conversion processes as coupled to the entire bacterium are inevitably more complex than portrayed here . 10 . 7554/eLife . 09541 . 007Figure 4 . Processes involved in energy conversion in the photosynthetic chromatophore . ( A ) Energy conversion processes starting after initial light absorption are divided into three stages: ( 1 ) quinol production at RC as a result of excitation transfer; ( 2 ) diffusion between RC and cytbc1 of quinone/quinol and cytochrome c2 , together with quinol-to-quinone conversion resulting in a proton gradient across the vesicle membrane; ( 3 ) utilization of proton gradient for ATP synthesis . ( B ) Chromatophore components , in which stages ( 1–3 ) take place , include LH2 ( green ) , LH1 ( red ) -RC ( blue ) , cytbc1 ( purple ) , and ATP synthase ( brown ) complexes as well as the lipid phase ( olive; see also Figure 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09541 . 007 Atomic level structural models of chromatophores have been presented earlier ( Şener et al . , 2007 , 2010; Hsin et al . , 2010b; Sener et al . , 2011; Chandler et al . , 2014 ) for Rba . sphaeroides and Rsp . photometricum and their mutants . The supramolecular organization of the vesicles in Rba . sphaeroides was determined primarily by AFM and EM images of intact membrane domains ( Bahatyrova et al . , 2004; Frese et al . , 2004; Scheuring et al . , 2007; Olsen et al . , 2008; Qian et al . , 2008; Scheuring and Sturgis , 2009 ) , whereas the stoichiometry of light harvesting proteins was determined by optical spectroscopy ( Sener et al . , 2010 ) and mass spectrometry ( Cartron et al . , 2014 ) . Vesicle models were subsequently constructed by mapping planar membrane patches viewed through AFM imaging back onto the parent spherical domains ( Şener et al . , 2007 ) , adjusting for the observed packing density ( Olsen et al . , 2008 ) and the spatial arrangement patterns ( Hsin et al . , 2009; Qian et al . , 2008 ) of the constituting proteins . The chromatophore model shown in Figure 1 comprises , in addition to the aforementioned constituent proteins , 16 , 000 lipids and 900 quinones , corresponding to a system containing 100 million atoms , including solvent . This system has been equilibrated through a 100 ns MD simulation to test the viability of the model employed . However , molecular dynamics simulations of the chromatophore describing energy conversion processes are not considered in the current study , because the large system size combined with timescales of energy conversion reaching milliseconds render a straightforward simulation prohibitive . Instead , the current study aims to describe key rate limiting components of energy conversion processes , such as quinone diffusion and turnover at cytbc1 as discussed below , to guide future simulation efforts . The atomic detail model is used below for the computation of the quantum yield , but rate kinetics subsequent to charge separation is described in terms vesicle stoichiometry only , with key rate constants taken from experimental studies . Early chromatophore models prior to ( Cartron et al . , 2014 ) account only for LH proteins , whereas in proteomics studies , hundreds of different types of non-LH peptides are actually identified , including ATP synthase , cytbc1 , membrane assembly factors , as well as proteins of unknown function ( Jackson et al . , 2012; Woronowicz and Niederman , 2010 ) . Most of these components are notably unresolved in AFM images . Assignment of cytbc1 was recently achieved through EM and AFM studies using gold nanoparticle labeling , revealing separated regions containing one or more cytbc1 , suggested to be located within lipid- and quinone-enriched membrane domains ( Cartron et al . , 2014 ) . It is plausible that cytbc1 induces different curvature profiles in membrane domains compared to the LH-rich constant-curvature regions predominant in AFM images . Such a curvature-induced separation of protein domains is also supported by experimental ( Frese et al . , 2004; Sturgis and Niederman , 1996 ) and computational ( Frese et al . , 2008; Chandler et al . , 2009; Hsin et al . , 2009 , 2010a ) studies that established the role of LH2 and RC-LH1-PufX domains in determining membrane shape . Induced curvature profiles are known to exert a segregating force between different types of proteins in the membrane ( Frese et al . , 2008 ) . Mass spectrometry showed that the RC:cytbc1 stoichiometry is 3:1 ( Cartron et al . , 2014 ) , consistent with earlier observations ( Crofts , 2004; Crofts et al . , 2006 ) , corresponding to approximately 4 cytbc1 dimeric complexes per vesicle . Chromatophore vesicles typically contain 1–2 ATP synthases ( Feniouk et al . , 2002; Cartron et al . , 2014 ) . Proteomics studies suggest preferential co-location of ATP synthase with LH2 subunits ( Woronowicz and Niederman , 2010 ) . Consequently , ATP synthase locations were assigned to LH2-rich regions of the membrane ( Cartron et al . , 2014 ) . The low-light adapted vesicle studied here contains 63 LH2 complexes , 11 dimeric and 2 monomeric RC-LH1-PufX complexes , 4 dimeric cytbc1 complexes , and 2 ATP synthases , in a spherical vesicle of 50 nm inner diameter based on a variation of the model reported in ( Cartron et al . , 2014 ) and shown in Figure 1 ( see also Video 1 ) . Transmembrane proteins beyond those of the light harvesting-cytbc1-ATP synthase model shown in Figure 4 , namely NADH dehydrogenase , succinate dehydrogenase , cytochrome c oxidase , and ubiquinol oxidase , are associated with controlling the redox state of the quinone/quinol pool in the chromatophore ( Klamt et al . , 2008 ) . These proteins , presented schematically in Figure 5 , indirectly couple the chromatophore proton gradient to metabolic reactions in the cytoplasmic part of the bacterial cell . Indeed , the chromatophore structure shown in Figure 1 may accommodate , by removal of LH2 complexes near cytbc1 complexes ( the latter referred to as complex III in the respirasome of mitochondria [Dudkina et al . , 2011] ) , the placement of adjacent NADH dehydrogenase complexes ( referred to as complex I ) in an arrangement similar to that in respirasomes as reported in ( Dudkina et al . , 2011 ) . First simulations , employing the complex I structure reported in ( Baradaran et al . , 2013 ) , have demonstrated that the chromatophore can adapt to the necessary local shape change . 10 . 7554/eLife . 09541 . 008Figure 5 . Regulation of the quinone/quinol-pool redox state in the chromatophore involves transmembrane proteins beyond those included in the light harvesting-RC-cytbc1-ATP synthase system described in ( Cartron et al . , 2014 ) and shown in Figure 4 . Succinate dehydrogenase and NADH dehydrogenase regulate the quinone/quinol redox state in the chromatophore membrane . This regulation influences the light intensity dependence of the ATP production rate in the chromatophore by changing the likelihood of finding available quinones at the RC . Two further proteins involved in redox kinetics are cytochrome c oxidase and ubiquinol oxidase , which are not shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09541 . 008 The overall aim of the present study is to determine the ATP synthesis rate as a function of chromatophore vesicle illumination and composition establishing , thereby , the energy conversion efficiency . The three stages of energy conversion in the chromatophore introduced above are summarized in Figure 4 . These stages span time scales ranging from femto- and picoseconds ( transfer of excitations ) to milliseconds ( diffusion of quinols , quinones , and cytochrome c2; ATP synthesis ) , involving both classical and quantum dynamics . The absorbed light power I is given in units of photons absorbed per second for the entire vesicle , i . e . , it holds approximately , I=Fσtotal where F is the flux of useable photons and σtotal is the total absorption cross-section of the chromatophore determined via the functional absorption cross-section reported in ( Woronowicz et al . , 2011a ) . Key quantities describing energy conversion are the quinone-to-quinol formation rate kQ→QH2 ( I ) , the quinol-to-quinone use rate kQH2→Q ( I ) and the ATP synthesis rate , kATP ( I ) , all functions of the absorbed light power I . For stationary illumination , assumed here , the chromatophore kinetics becomes stationary and , as a result , the rates kQ→QH2 ( I ) , kQH2→Q ( I ) , and kATP ( I ) must be identical , ( 1 ) kQ→QH2 ( I ) =kQH2→Q ( I ) =kATP ( I ) . We note here that every net quinol→quinone conversion event at cytbc1 , due to the so-called Q-cycle ( Crofts , 2004 ) , results in the release of four protons into the vesicle interior ( two at cytbc1 and two at RC , for each quinol passage ) , which coincidentally happens to be , in the present system , the same number of protons as the ones that have to move back over the membrane to produce one ATP molecule at ATP synthase ( based on the assuption of a 12-subunit c-ring of the ATP synthase ) . Under steady state conditions , the rate , kATP ( I ) , is then equal to kQH2→Q ( I ) and kQ→QH2 ( I ) . The quinone/quinol pool in the lipid phase of the vesicle and the proton gradient across the vesicle membrane act as temporary energy buffers between light harvesting and ATP synthesis ( Feniouk and Junge , 2009; Clark et al . , 1983 ) . Under the steady-state conditions assumed here , quinone-quinol pool and redox states of RC/cytbc1 are assumed to feature spatially homogeneous distributions . As a result , individual diffusive processes of quinone/quinol , cytochrome c2 , and protons do not need to be modeled , and the aforementioned energy buffers are determined solely by incident light intensity and quinol→quinone turnover capacity of cytbc1 , the latter constituting the rate limiting conversion process as discussed below . Typical proton diffusion timescales are on the order of microseconds ( Agmon , 1995 ) , i . e . , not rate limiting compared to turnover at cytbc1 and , therefore , do not affect significantly overall conversion rates . In going beyond steady-state conditions , a simulation of quinone mobility in a lamellar chromatophore membrane has recently been achieved for a 20 million atom , 150 ns simulation ( Chandler et al . , 2014 ) ; however , the time scale covered is not long enough to observe long-range positional relaxation of the quinone/quinol pool . This non-stationary behavior needs to be addressed by a coarser description like the ones employed for cell-scale modeling ( Roberts et al . , 2013 ) . In purple bacteria there are proton gradient consumption channels other than ATP synthesis . These channels include: flagellar motility ( Kojadinovic et al . , 2013 ) ; NADH/NADPH synthesis ( Klamt et al . , 2008; Blankenship , 2014 ) involving respiratory protein complexes in the chromatophore vesicle; leakage across the membrane . These channels are not included in the chromatophore kinetics described below , though the influence of NADH dehydrogenase is implicitly accounted for as explained below . The first stage of energy conversion in the chromatophore begins with light absorption by carotenoid and BChl pigments in the light harvesting complexes LH1 and LH2 leading to electronic excitation of individual pigments . Carotenoids transfer excitation within less than a picosecond to a nearby BChl ( Damjanović et al . , 1999; Berera et al . , 2009 ) and also play a role in quenching triplet states of BChls through reverse excitation transfer ( Ritz et al . , 2000a ) . The electronic excitations of BChls embedded in LH1 and LH2 are reviewed in ( Hu et al . , 1998 , 2002; Cogdell et al . , 2006; van Grondelle and Novoderezhkin , 2006b; Kosztin and Schulten , 2014 ) . These excitations form so-called exciton states , excitations shared among LH1 or LH2 BChls ( Ma et al . , 1997; Bradforth et al . , 1995 ) coherently ( Strümpfer et al . , 2012; Ishizaki and Fleming , 2009b; Rebentrost et al . , 2009 ) . Electronic excitation is transferred efficiently in the form of excitons between light harvesting complexes ( Hu et al . , 1997; Ritz et al . , 1998; Janusonis et al . , 2008; Ishizaki and Fleming , 2009b ) . Exciton-based excitation transfer in the chromatophore proceeds within 10–100 ps ( Sener et al . , 2011 ) , first among the LH2s , then from LH2 to LH1 , and finally from LH1 to the four BChls of the RC ( Visscher et al . , 1989; Beekman et al . , 1994; Strümpfer and Schulten , 2012a; Sener et al . , 2009 ) . In the RC , the excitation quickly settles onto the so-called special pair BChls ( Small , 1995; Damjanović et al . , 2000 ) , where it induces the transfer of an electron ( Pawlowicz et al . , 2008; Jordanides et al . , 2004 ) . This transfer proceeds stepwise to reach a quinone molecule , Q , attracted into the RC from the quinone/quinol pool of about 900 molecules ( Cartron et al . , 2014 ) . The quinol and quinone molecules of the pool are inter-converted at RC and cytbc1 ( Crofts , 2004 ) ( see Figure 4 ) . The electron transferred in the RC is joined on the quinone by a proton , turning Q into semi-quinone , QH . Repeating the reaction turns QH into quinol , QH2 . In converting Q to QH2 two electron charges move from near the inside of the chromatophore ( where the special pair BChls are located and the electron potential is low ) to near the cytoplasmic exterior of the chromatophore ( where the quinone is bound and the electron potential is high ) , i . e . , to the cytoplasmic side; the protons are attracted from the exterior of the chromatophore vesicle . Freshly formed quinol is released by the RC into the lipid phase of the chromatophore rejoining the quinone/quinol pool . The efficiency of the 10–100 ps light harvesting step is measured by the so-called quantum yield , q , namely the probability that light absorption leads to electron transfer in a RC with a Q or QH bound to receive the electron . The quantum yield can be calculated as reported in ( Sener et al . , 2011 , 2010 ) . Electronic excitation energy absorbed directly or indirectly ( through carotenoids ) by a BChl is rapidly shared between BChls within individual LH1 and LH2 light harvesting complexes ( Cory et al . , 1998 ) , forming , within about a ps , thermally equilibrated exciton states as established experimentally ( Visser et al . , 1996; Jimenez et al . , 1997; Valkunas et al . , 2007 ) as well as computationally ( Strümpfer and Schulten , 2009; Strümpfer et al . , 2012 ) . The exciton states of the BChls of each complex are determined as eigenstates of the effective Hamiltonian HI , accounting for the Qy excitations and their coupling inside LH1 , LH2 and RC as described in ( Strümpfer et al . , 2012 ) , ( 2 ) HI=∑i=1NIEiI∣i⟩⟨i∣+∑i>j>0N1VijI ( ∣i⟩⟨j∣+∣j⟩⟨i∣ ) . Here , the index I is employed to label one of the pigment-protein complexes , namely one of 63 LH2s , 24 LH1s , and 24 RC complexes for the vesicle shown in Figure 1 , with NI BChls; |i⟩ corresponds to the Qy excitation of BChl i with excited state energy EiI;VijI accounts for the respective Qy-Qy coupling among BChls i and j . Tables S2 and S3 in Supplementary Materials list the BChl coordinates as well as the constants employed in this study and discuss the computation of the quantum yield in greater detail . The coupling VijI in Equation ( 2 ) can be computed for well separated pigments ( rij>1 nm ) using the point-dipole approximation ( Ritz et al . , 2001; Sener et al . , 2011 ) , employing , ( 3 ) Vij=C ( 𝐝^i⋅𝐝^jrij3-3 ( 𝐝^i⋅𝐫ij ) ( 𝐝^j⋅𝐫ij ) rij5 ) , where 𝐝^i is the transition dipole moment unit vector of pigment i , 𝐫ij is the vector joining pigments i and j; the coupling constant C has the value C=348 , 000Å3 cm-1 ( using wavenumbers as unit of energy ) ( Şener et al . , 2007 , 2010 ) . Couplings between closely spaced pigments ( rij<1 nm ) require quantum chemical calculations as described in ( Damjanović et al . , 1999; Tretiak et al . , 2000 ) . The exciton states |α ) =∑iciα|i⟩ and the associated energies ϵα correspond to the eigenstates defined through Hℐ|α ) =ϵα|α ) . As electronic excitations settle within about 1 ps into the Boltzmann-populated excitons ( Strümpfer et al . , 2012; Strümpfer and Schulten , 2009 ) , excitation transfer among LH2 and LH1 involves the excitons , not individual chlorophyll or carotenoid excitations . The rate of excitation transfer between a donor complex I and an acceptor complex J is given by ( Ritz et al . , 2001; Şener et al . , 2007 , 2011 ) ( 4 ) kIJ=2πℏ∑μ∈I∑ν∈JpμI| ( μ|HIJ|ν ) |2Jμν , where HIJ is the matrix of interactions between the excited states of pigments in complexes I and J , and ( 5 ) Jμν=∫dESμD ( E ) SνA ( E ) , is the spectral overlap between donor exciton state |μ ) and acceptor exciton state |ν ) in units of ( 1/energy ) ( Sener et al . , 2011 ) ; SμI ( E ) and SνJ ( E ) are the normalized ( ∫dES ( E ) = 1 ) spectra for emission of the donor ( D ) and absorption of the acceptor ( A ) , respectively; pμI in Equation ( S6 ) are the populations of donor exciton states , which , as stated , become very rapidly ( ~1 ps ) ( Strümpfer and Schulten , 2009 ) Boltzmann-distributed such that pμI are given by ( 6 ) pμI=e-βϵμ∑γ∈Ie-βϵγ . The above description is known as the generalized Förster theory ( Förster , 1948; Novoderezhkin and Razjivin , 1996; Hu et al . , 1997; Sumi , 1999; Scholes et al . , 2001 ) . For reviews see ( van Grondelle and Novoderezhkin , 2006a; Sener et al . , 2011; Strümpfer et al . , 2012 ) . Excitation transfer kinetics in the chromatophore was reported experimentally in ( Woodbury and Parson , 1984; Visscher et al . , 1989; Crielaard et al . , 1994; Hess et al . , 1994 , 1995 ) . Exciton migration across the network of light harvesting complexes in the chromatophore can be described by a rate matrix 𝒦 which is constructed from inter-complex exciton transfer rates kIJ , the latter given by Equation ( S6 ) , as follows ( Sener et al . , 2010 , 2007 ) ( 7 ) ( 𝒦 ) IJ=kJI-δIJ ( ∑MkIM+kdiss+kCSδI , RC ) , where I , J are defined as in Equation ( S6 ) ; kdiss=1/ns is the rate of excitation loss due to internal conversion; kCS=1/ ( 3ps ) is the rate of charge separation at the RC ( Ritz et al . , 2001 ) ; δI , RC assumes the value 1 if complex I is a RC and the value 0 otherwise . The quantum yield , q , is the probability for an absorbed photon to initiate charge transfer at a RC ready for electron transfer; q is given for an initial state vector 𝐏 ( 0 ) by ( Sener et al . , 2011 , 2007; Ritz et al . , 2001 ) ( 8 ) q=-kCS ( 𝟏𝐑𝐂 ) T⋅𝒦-1⋅𝐏 ( 0 ) where the components of the vector ( 𝟏𝐑𝐂 ) are ( 𝟏𝐑𝐂 ) I=δI , RC; the initial state , 𝐏 ( 0 ) , corresponds to every BChl in the system being equally likely to be excited by photon absorption and accordingly is given by ( 9 ) ( 𝐏 ( 0 ) ) I=NI/ ( ∑JNJ ) , where NI is the number of BChls in complex I as indicated above . The effect of the initial state , 𝐏 ( 0 ) , on the quantum yield , q , arising , for example , due to wavelength-dependent absorption , is considered in ( Şener et al . , 2007 ) , with the result that q is altered by less than 3% . Therefore , wavelength dependence of q , through corresponding changes in 𝐏 ( 0 ) , is not considered further in the present study . The quantum yield given by Equation ( 8 ) for the vesicle shown in Figure 1 is 0 . 91 . For alternate vesicle compositions considered in Figure 3 , the quantum yield q is not computed by an explicit construction of vesicles to avoid massive computation; instead , q is approximated as a linear interpolation between the values reported earlier for high LH2:RC and low LH2:RC chromatophore vesicles ( Şener et al . , 2007 , 2010 , 2011 ) , namely between q=0 . 85 and q=0 . 95 . For a vesicle containing nB cytbc1 dimers and nL LH1-RC dimers , the corresponding number of LH2 complexes nLH2 ( nB , nL ) is estimated by the excluded surface resulting from changes in nB and nL with respect to the reference vesicle in Figure 1 . The corresponding quantum yield is estimated according to ( 10 ) q=0 . 91+0 . 0152 ( s0-s ) chosen to reproduce the correct value of q for the reference vesicle ( Figure 1 ) as well as for the low LH2 limit ( Sener et al . , 2011 ) ; here s=nLH2/ ( 2nL ) is the LH2:RC stoichiometry , which for the reference vesicle equals s0=2 . 625; a lower limit of q=0 . 85 is imposed to account for high LH2:RC vesicles ( Şener et al . , 2007 ) , where the linear interpolation breaks down . Validity of the generalized Förster formulation , thus outlined , has been demonstrated by excitation transfer calculations employing the so-called hierarchy equation of motion formalism of stochastic quantum mechanics ( Ishizaki and Fleming , 2009b; Strümpfer and Schulten , 2012b ) ; the calculations show that photoexcitation of chromatophore BChls relaxes into a Boltzmann occupancy of exciton states within approximately 1 ps , i . e . , faster than inter-complex transfer that takes 3–5 ps ( Hess et al . , 1995; Strümpfer and Schulten , 2012b ) . Accordingly , the assumption underlying generalized Förster theory , namely that transfer occurs from a thermally relaxed distribution of exciton states , holds in good approximation . The final step in stage I of energy conversion is the formation of quinol from quinone at the RC . The respective formation rate can be expressed ( 11 ) kQ→QH2 ( I ) =12IqpRC ( I ) , where the prefactor 12 accounts for every quinol requiring two electron transfer events at the RC . Here q is the quantum yield given by Equation ( 8 ) and pRC ( I ) is the probability for the RC to hold a quinone Q or a semiquinone QH , in either case the RC being ready to accept and convert an electronic excitation . The probability pRC ( I ) decreases with increasing I , since the quinone/quinol pool becomes quinol rich/quinone poor at increasing light intensities , due in part to coupling ( Figure 5 ) to chromatophore redox factors , succinate dehydrogenase , NADPH dehydrogenase , and cytochrome c oxidase ( Klamt et al . , 2008 ) . As the quinone/quinol ratio decreases , it becomes less likely for RC to have a quinone/semiquinone available for electron transfer . The stated change in the quinone/quinol pool is crucial for energy conversion control of the chromatophore and comes about through the proton motive force , generated through light harvesting , inducing in the redox factors redox generation of products along with quinone/quinol conversion . The light-condition dependency of the quinone/quinol pool is described in the present model heuristically as explained below in Equation ( 15–19 ) . Under the assumed steady state conditions , the rate kQ→QH2 ( I ) , i . e . , the rate at which RCs release QH2 as given by Equation ( 11 ) , is equal to the rate at which RCs bind fresh quinones . Accordingly holds ( 12 ) 12IqpRC ( I ) =[nRC ( 1−pRC ( I ) ) ]/τRC ( I ) , where nRC=2nL is the number of RCs in the chromatophore ( 24 for the vesicle shown in Figure 1 ) , 1-pRC ( I ) is the fraction of RCs ready to bind a fresh Q , τRC ( I ) is the mean time needed for a RC to become available for binding a new Q after it had just accepted a Q ( Remy and Gerwert , 2003 ) . Below , we refer to τRC ( I ) as the cycling time . The probability pRC ( I ) is assumed , for convenience , to be uniform across all RCs rather than to vary between RCs due to inhomogeneities in the redox state of the quinone/quinol pool . This assumption is strictly valid only when the mixing time of quinols and quinones in the vesicle lipid phase is shorter than the time scales associated with the rates in Equation ( 1 ) . The spatial inhomogeneity of pRC ( I ) can be determined only through the simulation of the diffusive processes in the chromatophore , which is currently prohibitive . It had been suggested in ( Geyer and Helms , 2006 ) that the primary rate-limiting step in the chromatophore is quinol turnover at cytbc1 rather than cytochrome c2 diffusion; in ( Geyer and Helms , 2006 ) it had been estimated that each cytochrome c2 is capable of approximately 80 electron transfers per second and that three cytochrome c2’s per vesicle are sufficient to saturate the turnover capacity of an ATP synthase . A chromatophore vesicle is expected to feature 10–20 cytochrome c2 molecules ( Geyer and Helms , 2006; Cartron et al . , 2014 ) , safely exceeding the necessary number needed for saturation . Therefore , cytochrome c2 kinetics should not be rate limiting for energy conversion in the chromatophore . Using Equation ( 12 ) , pRC ( I ) can be expressed in terms of τRC ( I ) , namely , ( 13 ) pRC ( I ) = ( 1+12IqτRC ( I ) 1nRC ) -1 . According to Equations ( 11 ) and ( 13 ) , τRC ( I ) needs to be determined in order to compute the rate kQ→QH2 ( I ) or , equivalently , kATP ( I ) . The cycling time , τRC ( I ) , arising in Equations ( 12 , 13 ) , depends on light intensity . The cycling time is related to the quinone/quinol stoichiometry , i . e . , the redox state of the quinone/quinol pool: the fewer quinones are present , the longer is the cycling time . The redox state is affected by not only RC and cytbc1 reactions , but also by transmembrane enzymes succinate dehydrogenase and NADH dehydrogenase ( Figure 5 ) . The low-light and high-light limits for the cycling time , τRC , employed below are based on experimental observation ( Woronowicz et al . , 2011b , 2011a; Crofts , 2004 ) instead of direct computation; the reported values of τRC implicitly combine the redox reactions of all enzymes interacting with the quinone/quinol pool , including NADH dehydrogenase . In a stationary state , the rate kATP ( I ) of ATP synthesis is equal to the rate kQ→QH2 ( I ) as stated in Equation ( 1 ) , which according to Equations ( 11 ) and ( 13 ) can be expressed through the cycling time , τRC ( I ) . The condition of equilibrium assumed here might not be valid for rapidly fluctuating light intensities , where spatial inhomogeneities of the vesicle and the quinone/quinol pool are expected to play a nontrivial role on the cycling time . The low-light limit , τL , of the cycling time , τRC ( I ) , is observed to range from 0 . 7 ms for the membrane of an LH2-minus mutant to about 3 ms for the LH2-rich chromatophores adapted to low-light growth conditions ( Woronowicz et al . , 2011b , 2011a ) . In the following , we assume τL=3ms for the low-light growth vesicle shown in Figure 1 . At the high-light limit , the immediate vicinity of a RC contains mostly quinols and the replacement of the converted quinones at the RC becomes rate limited by the turnover at cytbc1 ( Woronowicz et al . , 2011b , 2011a ) . The high-light limit , τH , of the cycling time , τRC ( I ) , can be estimated by considering the total turn-over rate at all RCs , namely nRCτH-1 . In the stationary high I regime , this rate must be equal to the quinol turnover rate at all cytbc1 , namely nBτB-1 , i . e . , it holds ( 14 ) nBτB-1=nRCτH-1 , where nB is the number of cytbc1 dimers ( 4 for the vesicle shown in Figure 1 ) and τB=25 ms is the quinol turnover time at a cytbc1 ( Crofts , 2004 ) . As the last step of energy conversion , the proton gradient , generated at cytbc1 through quinol → quinone conversion , is utilized by ATP synthase for the production of ATP . The ATP turnover rate of the vesicle , kATP ( I ) , under stationary conditions is equal to kQ→QH2 ( I ) given by Equation ( 11 ) . This equality is based on the assumption that ATP synthase of Rba . sphaeroides has an Fo of 12 c-subunits such that four H+ conducted through the Fo-ring of ATP synthase lead to a 120° rotation of the stalk in the F1 part and , thereby , to synthesis of one ATP . Currently , the structure of the ATP synthase of Rba . sphaeroides and the corresponding number of c-subunits is not known experimentally . If the Fo oligomer were to feature , e . g . , 11 or 10 subunits instead , this structural detail would proportionally affect the number of protons required for the rotation of the stalk and subsequent synthesis of each ATP and , therefore , directly influence the estimated energy conversion efficiency of the chromatophore , by 11% or 20% , respectively . Combining Equation ( 1 ) , ( 11 ) , ( 13 ) , the ATP turnover rate can be expressed ( 20 ) kATP ( I ) =12Iq ( 1+12IqτRC ( I ) 1nRC ) -1 , where the cycling time at the RC , τRC ( I ) , is given by Equation ( 19 ) . The overall energy conversion efficiency of the chromatophore , ηATP ( I ) , can be defined as the ratio of formation rate of energy in the form of ADP→ATP synthesis to the total absorption rate of photon energy ( 21 ) ηATP ( I ) =EATPkATP ( I ) EγI , where EATP=4197cm-1 is the ATP hydrolysis energy in the cell ( Berg et al . , 2011 ) and Eγ is the average energy of an absorbed photon , taken to be the photon energy at 850 nm ( 11765 cm-1 ) . Not all the energy of an absorbed photon , Eγ , is available for energy harvesting . The fraction of Eγ available for conversion into chemical energy is the so-called Carnot yield ( Lavergne , 2009 ) described by comparing photochemical energy conversion to the function of a heat engine . This limitation in photochemical energy conversion establishes a theoretical upper limit for photosynthetic energy conversion at broad daylight of approximately 0 . 7 ( Lavergne , 2009 ) . The determination of the energy conversion efficiency , ηATP ( I ) , computed through Equation ( 21 ) , has the shortcoming that the ATP hydrolysis energy , EATP , depends , in principle , on the ADP , ATP , and H+ concentrations in the cytoplasm , which are not modeled explicitly . Nevertheless , Equation ( 21 ) permits a comparison with similar measures of efficiency reported for other photosynthetic or photovoltaic systems ( Blankenship et al . , 2011 ) . | Photosynthesis , or the conversion of light energy into chemical energy , is a process that powers almost all life on Earth . Plants and certain bacteria share similar processes to perform photosynthesis , though the purple bacterium Rhodobacter sphaeroides uses a photosynthetic system that is much less complex than that in plants . Light harvesting inside the bacterium takes place in up to hundreds of compartments called chromatophores . Each chromatophore in turn contains hundreds of cooperating proteins that together absorb the energy of sunlight and convert and store it in molecules of ATP , the universal energy currency of all cells . The chromatophore of primitive purple bacteria provides a model for more complex photosynthetic systems in plants . Though researchers had characterized its individual components over the years , less was known about the overall architecture of the chromatophore and how its many components work together to harvest light energy efficiently and robustly . This knowledge would provide insight into the evolutionary pressures that shaped the chromatophore and its ability to work efficiently at different light intensities . Sener et al . now present a highly detailed structural model of the chromatophore of purple bacteria based on the findings of earlier studies . The model features the position of every atom of the constituent proteins and is used to examine how energy is transferred and converted . Sener et al . describe the sequence of energy conversion steps and calculate the overall energy conversion efficiency , namely how much of the light energy arriving at the microorganism is stored as ATP . These calculations show that the chromatophore is optimized to produce chemical energy at low light levels typical of purple bacterial habitats , and dissipate excess energy to avoid being damaged under brighter light . The chromatophore’s architecture also displays robustness against perturbations of its components . In the future , the approach used by Sener et al . to describe light harvesting in this bacterial compartment can be applied to more complex systems , such as those in plants . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems",
"computational",
"and",
"systems",
"biology"
] | 2016 | Overall energy conversion efficiency of a photosynthetic vesicle |
Many organisms lose the capacity to regenerate damaged tissues as they mature . Damaged Drosophila imaginal discs regenerate efficiently early in the third larval instar ( L3 ) but progressively lose this ability . This correlates with reduced damage-responsive expression of multiple genes , including the WNT genes wingless ( wg ) and Wnt6 . We demonstrate that damage-responsive expression of both genes requires a bipartite enhancer whose activity declines during L3 . Within this enhancer , a damage-responsive module stays active throughout L3 , while an adjacent silencing element nucleates increasing levels of epigenetic silencing restricted to this enhancer . Cas9-mediated deletion of the silencing element alleviates WNT repression , but is , in itself , insufficient to promote regeneration . However , directing Myc expression to the blastema overcomes repression of multiple genes , including wg , and restores cellular responses necessary for regeneration . Localized epigenetic silencing of damage-responsive enhancers can therefore restrict regenerative capacity in maturing organisms without compromising gene functions regulated by developmental signals .
The ability of tissues to regenerate following damage varies greatly between different species ( Li et al . , 2015 ) . Even some vertebrate species are capable of fully regenerating organs that have essentially no regenerative capacity in humans such as heart tissue in adult zebrafish ( Poss et al . , 2002 ) and limbs in urodele amphibians such as newts and salamanders ( Tanaka and Reddien , 2011 ) . Understanding the cellular differences between homologous organs that differ in their regenerative capacity among species could suggest genetic or pharmacological manipulations that improve the regenerative capacity of damaged organs in humans . The ability to regenerate specific organs differs not only between species , but can also change within a single species during its development . In general , for organisms that have determinate growth , i . e . growth that ceases once a genetically pre-determined size is reached ( Sebens , 1987 ) , the capacity for regeneration usually decreases with increasing maturity . For example , embryonic and neonatal mice are able to regenerate myocardial tissue following substantial damage ( Drenckhahn et al . , 2008; Porrello et al . , 2011 ) . However , within the first week of life , this capacity to regenerate is lost almost completely , and the damaged tissue is instead replaced by fibrotic scarring ( Porrello et al . , 2011 ) . Other examples include appendage regeneration in developing Xenopus ( Dent , 1962; Slack et al . , 2004 ) and the progressive loss of digit regeneration in mice and humans ( King , 1979; Borgens , 1982; Reginelli et al . , 1995 ) . The basis for this phenomenon is not well understood . In some instances where tissue re-growth is dependent on stem cell based replacement of tissue ( for review see Tanaka and Reddien , 2011 ) , the loss of a stem cell population could account for the inability to regenerate . However , this explanation cannot account for situations where regeneration is not driven by a stem cell population but rather by the proliferation and de-differentiation of adjacent tissue . In these cases the loss of regenerative capacity in a still-developing organism necessitates a mechanism that selectively inactivates regenerative processes while still allowing cell proliferation and differentiation associated with normal development . Since most genes that function during regeneration also have a variety of functions either in normal development or in maintaining homeostasis in the same tissue , it is presently unclear how their expression could be regulated so as to selectively block regeneration . Genetic studies using Drosophila have provided important insights into the genetic regulation of tissue growth . Many of these studies have examined growth of the imaginal discs , larval epithelial tissues that are precursors of adult structures , such as the wing and the eye ( Cohen , 1993 ) . Imaginal discs are capable of regenerating missing portions following damage ( Worley et al . , 2012 ) . Studies of imaginal disc regeneration were pioneered by the group of Ernst Hadorn and were mostly conducted by transplanting damaged discs into the abdomens of adult female flies ( Ursprung and Hadorn , 1962 ) . More recently , imaginal disc regeneration has been studied in intact larvae by damaging portions of the disc via brief expression of a pro-apoptotic gene in a spatially-restricted manner ( Smith-Bolton et al . , 2009; Bergantiños et al . , 2010 ) . Using either a genetic ablation system ( Smith-Bolton et al . , 2009 ) , or following X-ray irradiation ( Halme et al . , 2010 ) , it was observed that the capacity of the wing-imaginal disc to regenerate progressively diminished during the later stages of the third larval instar as the larva approached the beginning of metamorphosis . Expression of wingless ( wg , Drosophila WNT1 ortholog ) is robustly upregulated in regenerating discs following genetic ablation . However , in more mature discs , which do not regenerate , there is a strong reduction in this wg upregulation ( Smith-Bolton et al . , 2009 ) . We chose to address the basis for reduced regeneration in mature discs by examining the mechanisms underlying the reduction in wg upregulation following damage . The upregulation of WNT proteins is an early and often essential response to tissue damage , observed in diverse species including Hydra and planaria ( Gurley et al . , 2008; Petersen and Reddien , 2008; Lengfeld et al . , 2009 ) , as well as vertebrate species such as zebrafish , axolotls and Xenopus ( Kawakami et al . , 2006; Stoick-Cooper et al . , 2007; Yokoyama et al . , 2007; Lin and Slack , 2008 ) . Multiple studies have demonstrated that WNT signaling is essential for regeneration and is able to augment the process ( Kawakami et al . , 2006; Stoick-Cooper et al . , 2007; Yokoyama et al . , 2007; Lin and Slack , 2008 ) , even in tissues that do not normally undergo regeneration ( Kawakami et al . , 2006 ) . Thus , the mechanisms that link tissue damage in imaginal discs to wg upregulation , and the basis for diminished wg upregulation with increasing maturity , are likely to provide insights into similar processes in diverse organisms . Here we characterize the properties of an enhancer in the major WNT locus of Drosophila ( Schubiger et al . , 2010 ) that mediates the upregulation of both wg and Wnt6 following damage to imaginal discs , and show that deletion of this enhancer impairs regeneration . We show that damage-responsive activation and age-dependent attenuation can each be ascribed to separate modules within the enhancer , and while the damage-responsive module is equally effective in promoting damage-responsive gene expression in mature discs , its activity is overridden by an adjacent element that nucleates epigenetic silencing with increasing effectiveness as the larva matures . Importantly , this epigenetic silencing is restricted to the immediate vicinity of this enhancer , thereby not interfering with the activation of wg and Wnt6 by distinct developmentally regulated enhancers . We also describe a way of overcoming this silencing and improving regeneration in more mature larvae .
A number of enhancers have been identified that regulate the normal pattern of wg expression in the wing disc during L3 ( Neumann and Cohen , 1996; Pereira et al . , 2006; Koshikawa et al . , 2015 ) . However , wg expression following damage appears to be regulated by a distinct enhancer . Examination of genomic DNA fragments spanning the entire WNT gene cluster , including the wg locus , by Schubiger and colleagues ( Schubiger et al . , 2010 ) identified a ~3 kb region , named BRV118 , that was activated in imaginal discs following mechanical injury . In these studies , BRV118 was assumed to primarily regulate wg . However , since BRV118 is located in the middle of a cluster of WNT genes , between wg and Wnt6 ( Figure 3A ) , it is possible that it might regulate multiple WNT genes . 10 . 7554/eLife . 11588 . 011Figure 3 . The BRV118 enhancer is necessary for wg and Wnt6 expression following tissue damage . ( A ) Schematic of the major WNT locus showing the BRV118 enhancer ( blue rectangle ) . Approximate intergenic distances are labeled . ( B ) RNA in situ hybridization to detect wg , Wnt4 , Wnt6 and Wnt10 RNA in day 7 unablated ( top row ) and rnts>egr ablated ( bottom row ) wing discs . Only wg and Wnt6 demonstrate significant upregulation of RNA in response to damage . ( C ) RNA in situ hybridization to detect wg , and Wnt6 RNA in discs ablated with rnts>egr on day 7 and day 9 . Transcription of both wg and Wnt6 in response to damage is absent in a day 9 disc . ( D ) Schematic of the lesion in the wg1 allele that deletes most of the BRV118 enhancer . ( E ) RNA in situ hybridization to detect wg and Wnt6 RNA and Wg protein , in wild type ( top row ) and homozygous wg1 ( bottom row ) day 7 rnts>egr ablated discs . The damage-specific expression of wg and Wnt6 RNA , and Wg protein levels , are reduced in wg1 discs compared to wild type . ( F ) Adult wing sizes following ablation with rnts>egr on day 7 in wild type and wg1 homozygotes , indicating the percentage of animals that eclose with fully regenerated wings . Regeneration of wg1 homozygous discs is significantly reduced compared to wild type . Error bars indicate SD of at least three independent experiments , scoring a total of >200 animals in each experiment . Only untransformed wg1 wings were scored ( for explanation of untransformed see Figure 3—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 01110 . 7554/eLife . 11588 . 012Figure 3—figure supplement 1 . The wg1 allele causes transformation of wing to notum in a proportion of the population . ( A-A’ ) Wing discs dissected from day 7 wg1 homozygous larvae and stained for Wg ( red ) and DAPI ( blue ) , showing a transformed disc ( A ) , in which presumptive wing tissue is replaced by notum , indicated by the duplicated wg notum stripe ( arrowheads ) , and an untransformed wing disc ( A’ ) that is morphologically normal and has wild type wg expression . Transformation is thought to occur at L2 when the wing pouch is specified by wg expression ( Sharma and Chopra , 1976; Whitworth and Russell , 2003 ) , and thus ablation of the pouch by rnts>rpr or rnts>egr does not occur in these discs , ( B-B’ ) A wg1 homozygous adult with one untransformed left wing and a transformed right wing . Wing tissue is replaced by additional notum and associated bristles following transformation ( arrowheads ) . ( C-C’ ) For comparison , a wild type adult with two non-regenerated wings resulting from rnts>rpr ablation . The presence of hinge tissue and normal bristle formation allows clear distinction between ablated versus transformed wings; only untransformed wings were scored in our assays . ( D ) Quantification of the wing to notum transformation frequency in the wg1 homozygous stock used in our experiments . Error bars are SD , n>300 flies . On average 58% of animals eclose with two untransformed wings . The frequency of the variably penetrant transformation phenotype can be shifted by selective breeding of two wing , one wing or no wing flies , but not eliminated ( Sharma and Chopra , 1976 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 012 To investigate this possibility further , we used RNA in situ hybridization to examine wing imaginal discs following genetic ablation using rnts>egr for changes in expression of each of the four WNT genes located near BRV118 . In the absence of tissue ablation , expression patterns of all four genes on day 7 matched previously published descriptions ( Figure 3B ) , including low ubiquitous expression of Wnt10 ( Janson et al . , 2001 ) , and a wg-like pattern for Wnt4 and Wnt6 ( Gieseler et al . , 2001; Doumpas et al . , 2013 ) , with Wnt4 being weaker and more diffuse , as previously reported ( Gieseler et al . , 2001 ) . In day 7 rnts>egr discs , wg RNA was detected at high levels around the damaged tissue , as was Wnt6 RNA ( Figure 3B ) . Neither Wnt4 nor Wnt10 expression was appreciably elevated . Moreover , the upregulation following damage of both wg and Wnt6 RNA was diminished in day 9 discs ( Figure 3C ) . Thus , our findings indicate that expression of both wg and Wnt6 , the two genes that flank BRV118 , is upregulated in response to damage , and that this response declines with larval maturation . To study whether BRV118 regulates the expression of both wg and Wnt6 , we examined their expression in ablated wg1 discs . The BRV118 element is located approximately 8 kb downstream of the wg gene and largely overlaps the region that is deleted in the wg1 allele ( Figure 3D ) . Homozygous wg1 adults are viable , and have the ability to develop normally sized and patterned wings , suggesting this region is not essential for normal development . However , a proportion of flies exhibit a variably-penetrant wing-to-notum transformation , reflecting a semi-redundant requirement of this region early in larval development for wing-pouch specification by wg ( Sharma and Chopra , 1976 ) . This loss of wing tissue is phenotypically distinct and easily distinguishable from ablated wings ( Figure 3—figure supplement 1 ) . In ablated wg1 discs , expression of both wg and Wnt6 was markedly reduced ( Figure 3E ) suggesting that sequences disrupted in the wg1mutant are necessary for the upregulation of both genes . We therefore examined regeneration in wg1 flies following rnts>egr ablation . The size of adult wings after regeneration was considerably reduced in wg1 flies ( Figure 3F ) . Thus , sequence alterations in the wg1 mutant , most likely the deletion of the BRV118 enhancer , compromise regeneration . To further characterize the properties of the BRV118 enhancer , we constructed a reporter transgene consisting of a 2 . 9 kb enhancer fragment and a basal promoter driving eGFP expression ( Figure 4A ) . In the absence of tissue damage , GFP expression was not detected in imaginal discs at any point during L3 , or in any earlier larval stages that were examined , including L2 or L1 ( Figure 4B , and data not shown ) . However , upon ablation with rnts>egr , GFP is strongly expressed in dying cells and in the surrounding tissues in a pattern that closely resembles damage-induced Wg expression ( Figure 4C ) . Importantly , the reporter is also expressed at much higher levels in rnts>egr day 7 discs than in day 9 discs ( Figure 4C ) . Thus , both the activation of damage-responsive expression of wg , as well as the reduced responsiveness with increasing maturity , are likely mediated in significant part at the transcriptional level . Moreover , the 2 . 9 kb BRV118 fragment must include sequences that can mediate both of these aspects of wg transcription . 10 . 7554/eLife . 11588 . 013Figure 4 . BRV118 is a damage-responsive WNT enhancer . ( A ) Schematic of the 2 . 9 kb BRV118-GFP reporter transgene , and the subdivisions used to generate the ~1 kb BRV-A , BRV-B and BRV-C reporters . All transgenes were inserted into the same genomic landing site to enable direct comparison . Numbers indicate nucleotide positions in BRV118 . ( B ) Unablated day 7 disc bearing the BRV118-GFP reporter , stained for Wg ( red ) and DAPI ( blue ) . No GFP is detected during normal development throughout L3 ( green ) or L2 ( insets ) . Discs were staged using the Wg expression patterns: diffuse Wg in the primordial pouch tissue to indicate L2 and the defined D-V stripe and "hinge circles" to indicate L3 ( red ) . ( C , D-D’’ ) Discs bearing the BRV118-GFP ( C ) , BRV-A ( D ) , BRV-B ( D’ ) or BRV-C ( D’’ ) reporters ablated with rnts>egr on day 7 ( top row ) or day 9 ( bottom row ) . The BRV118-GFP reporter ( green ) is expressed in a pattern closely resembling that of damage-induced Wg ( red ) , on both day 7 and day 9 . The BRV-A reporter has weaker activation than the full enhancer on day 7 and on day 9 . The BRV-B reporter has much stronger activation than the full BRV118 and is not attenuated with age . BRV-C is not expressed following ablation . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 01310 . 7554/eLife . 11588 . 014Figure 4—figure supplement 1 . The BRV118 enhancer reporter demonstrates damage-responsive but not developmentally-regulated wg expression . ( A-B ) Discs bearing the BRV118-GFP reporter ablated with rnts>rpr ( A ) and rnts>egr ( B ) on day 7 , stained for GFP ( green ) and Wg ( red ) , showing the enhancer responds to rpr induced cell death at a lower level than egr , similar to wg expression . ( C ) Discs bearing the BRV118-GFP enhancer following ablation with rnts>egr on day 7 and imaged at 0 hr ( immediately at temperature downshift ) , 24 hr , 48 hr and 72 hr of regeneration . Discs are stained for GFP ( green ) and Wg ( red ) , showing the initial overlap of expression patterns between Wg and the enhancer reporter ( at 0 hr and 24 hr ) , followed by divergence of the expression domains , where Wg regains its developmental expression pattern in the hinge and D-V boundary , while GFP expression is maintained in the blastema of the regenerating pouch ( 48 hr and 72 hr ) . This GFP is distinct from cell debris , indicated by DCP-1 staining ( gray , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 014 The BRV118 reporter is also activated in rnts>rpr discs , though not as strongly ( Figure 4—figure supplement 1A–B ) . Similar differences between rnts>egr and rnts>rpr discs were also observed with respect to Wg protein expression . During regeneration , Wg and GFP mostly overlap initially but their patterns of expression differ as regeneration proceeds ( Figure 4—figure supplement 1C ) . Wg resolves into its characteristic developmentally-regulated pattern while GFP expression persists in a broader separate domain within the regenerated tissue , distinct from dead cells ( Figure 4—figure supplement 1D ) . Thus the 2 . 9 kb fragment contains sequences sufficient to drive reporter expression in a pattern characteristic of damaged and regenerating discs , yet lacks sequences necessary for directing the developmentally-regulated patterns of wg expression . To identify damage responsive elements within the BRV118 enhancer , we generated further reporter constructs that each contained one of three non-overlapping fragments of the enhancer , each approximately 1 kb in length ( Figure 4A ) . Of these , the BRV-B fragment directed strong damage-responsive reporter gene expression in both day 7 and day 9 discs ( Figure 4D’ ) . The BRV-A fragment also elicited some damage-responsive expression ( Figure 4D ) , albeit far less than BRV-B , and the BRV-C fragment did not drive any detectable expression ( Figure 4D’’ ) . The lack of age-dependent attenuation of expression driven by the BRV-B fragment indicates that damage-responsive transcription and its age-dependent attenuation can be dissociated , thereby implying that the signaling pathways that activate the damage-responsive element are fully active in imaginal discs of late L3 larvae . To test whether the BRV-B fragment functions as a damage-responsive enhancer in other contexts , we induced tissue damage in a variety of ways . Reporter activation was observed in and surrounding clones of cells that express the pro-apoptotic gene egr in wing , haltere , leg and eye discs ( Figure 5A–A’ , Figure 5—figure supplement 1A–B ) indicating that the reporter can be activated by egr-induced cell death in most , if not all , imaginal discs . Similarly , inducing cell death by irradiation ( Figure 5B ) or physical fragmentation of discs in culture ( Figure 5C–D ) induced reporter gene expression . In all of these experiments , reporter activation was less evident following damage to the notum of the wing disc ( Figure 5A , open arrowheads ) or the larval brain ( Figure 5—figure supplement 1C , open arrowheads ) suggesting that additional mechanisms might regulate the tissue specificity of enhancer activation . Also , consistent with the differences between the full-length enhancer and the damage-responsive BRV-B module , irradiation or a physical cut does not induce Wg protein expression in mature discs , even though the BRV-B reporter is expressed ( Figure 5B , D ) . 10 . 7554/eLife . 11588 . 015Figure 5 . BRV-B is a damage-responsive enhancer in multiple contexts . ( A-A’ ) egr-expressing clones activate BRV-B in some but not all regions of wing and eye discs . Heat-shock induced FLP-out clones expressing Gal4 ( marked by RFP , red ) were generated on day 4 and 5 of development in larvae bearing the BRV-B reporter . Clones were allowed to grow , and subsequently Gal4 was activated by inactivation of Gal80ts to express egr for 24 hr on day 8 . GFP is expressed in and surrounding large dying clones ( closed arrowheads ) in wing discs ( A ) and eye/antennal discs ( A’ ) , but is absent from egr-expressing clones in various parts of each disc ( open arrowheads ) , including the notum of the wing disc . ( B ) Disc from larva bearing the BRV-B reporter irradiated with 45 Gy on day 7 , dissected after 16 hr and stained for cell death marker DCP-1 ( gray ) , Wg ( red ) and GFP ( green ) . The BRV-B reporter is strongly activated throughout the wing pouch and hinge ( arrowhead ) , but not in the presumptive notum ( open arrowhead ) . ( C-D ) Day 7 discs bearing the BRV-B reporter , physically cut ( D , arrowhead indicates cut site ) or left intact ( C ) , and cultured in Schneider’s medium for 16 hr , followed by staining to detect GFP and Wg . Reporter activation is detected specifically along the cut site . ( E-E’ ) Confocal sections through midguts of adult flies bearing the BRV-B reporter following 2 days of feeding 5% dextran solution as control animals , ( E ) or gut-damaging 5% DSS solution ( E’ ) . Guts were stained with anti-Delta ( red ) to show the intestinal stem cell ( ISC ) population , and GFP to detect reporter activity . Damaged gut tissue in DSS fed animals increases ISC number . Strong induction of the BRV-B reporter ( green ) is observed but not in the population that expresses Delta . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 01510 . 7554/eLife . 11588 . 016Figure 5—figure supplement 1 . The BRV-B enhancer reporter is activated by genetic ablation in different tissues . ( A-C ) Temporally controlled expression of egr in clones generated within haltere ( A ) and leg ( B ) discs of larvae bearing the BRV-B reporter , as in Figure 5A–A’ . Expression of egr in clones of day 8 discs results in activation of the reporter in and around dying clones ( closed arrowheads ) . Some areas of the discs are refractory to reporter activation despite egr expression ( open arrowheads ) . Similar expression of egr in clones in the non-regenerating tissue of the larval brain ( C ) does not result in BRV-B activation ( open arrowheads ) . Low level expression of the reporter can be seen in unidentified neurons within of the optic lobe , which is unrelated to ablation ( white arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 016 Wg has also been shown to be necessary for regeneration of the intestine in adult Drosophila following damage induced by the ingestion of cytotoxic agents ( Cordero et al . , 2012 ) . In adults fed 5% dextran sulfate sodium ( DSS ) , a sulfated polysaccharide that injures the intestinal epithelium ( Amcheslavsky et al . , 2009 ) , the BRV-B reporter was activated in cells that have the morphology of enteroblasts and do not express Delta , which is a marker for intestinal stem cells ( Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2007 ) ( Figure 5E’ ) . Control animals fed the non-sulfated polymer dextran showed only scattered expression of GFP at low levels ( Figure 5E ) . Thus the BRV-B enhancer responds to tissue damage induced in a variety of ways , including in situations that are known to activate wg expression in the context of regeneration . In discs damaged in vivo by rpr genetic ablation or in culture by fragmentation , the spatial distribution of BRV-B activation correlates with the expression of the AP-1 reporter ( Figure 6A , F ) . Additionally , as with the full length enhancer , BRV-B is activated more strongly by rnts>egr than by rnts>rpr ( Figure 6B–B’ ) . egr is thought to promote cell death in large part through activation of JNK signaling ( Igaki et al . , 2002 ) , whereas blocking JNK signaling does not prevent rpr-mediated cell death ( Sun and Irvine , 2011 ) . These observations suggest that tissue injury could activate BRV-B in significant part via the JNK/AP-1 pathway . To test whether activation of JNK was sufficient to activate BRV-B driven expression , even in the absence of egr- or rpr-directed ablation , we expressed a constitutively active hemipterous transgene ( UAS-hepCA ) , which encodes an active form of JNK kinase . This strongly induced the reporter expression ( Figure 6C ) . Conversely , in hep hemizygotes , where activity of JNK is greatly reduced , the upregulation of BRV-B following ablation with rnts>rpr is completely abolished ( Figure 6D–D’ ) . To test whether the requirement for JNK signaling was disc autonomous and to rule out systemic influences , discs were dissected , physically fragmented and cultured in the presence of a small molecule JNK inhibitor . These discs failed to upregulate the BRV-B reporter ( Figure 6—figure supplement 1 ) , indicating that JNK signaling in the wounded tissue itself is responsible for activation of the enhancer . Thus , at least under the conditions of these experiments , JNK signaling is necessary and potentially sufficient for BRV-B mediated expression . 10 . 7554/eLife . 11588 . 017Figure 6 . BRV-B reporter activation requires the JNK/AP-1 pathway . ( A ) Basal section of a day 9 disc bearing the AP-1 RFP reporter shows JNK pathway activity ( red ) and the BRV-B GFP reporter ( green ) following ablation by rnts>rpr , imaged at the time of downshift to 18°C . Reporter expression overlaps in wound edge cells , distinct from dead cells ( DCP-1 , gray ) , ( B-B’ ) Day 7 discs bearing the BRV-B reporter , ablated using rnts>rpr ( B ) or rnts>egr ( B’ ) and imaged at the time of downshift to 18°C , demonstrating stronger expression of both Wg ( red ) and BRV-B GFP ( green ) following ablation with egr compared to rpr . ( C ) A day 7 disc bearing the BRV-B reporter expressing rnts>hepCA for 20 hr to activate JNK signaling in the wing pouch . Both Wg ( red ) and the BRV-B GFP ( green ) are strongly expressed . ( D-D’ ) Day 7 discs bearing the BRV-B reporter ablated with rnts>rpr in a wild type ( D ) or hep- hemizygote ( D’ ) , imaged at the time of downshift to 18oC . GFP expression is abolished in the hep- mutant following ablation , while cell death is unaffected ( DCP-1 staining , gray ) . ( E ) Schematic showing three predicted AP-1 binding sites ( red ) in the BRV-B enhancer fragment . The black histogram below shows a measure of the evolutionary conservation of the BRV-B DNA sequence in twelve Drosophila species , mosquito , honeybee and red flour beetle , based on a phylogenetic hidden Markov model ( https://genome . ucsc . edu/ ) . Numbers indicate nucleotide position in the BRV-B element . ( F-F’ ) Day 7 discs bearing an AP-1-RFP reporter and BRV-B ( F ) or BRV-B with AP-1 sites deleted ( BRV-BΔAP-1 , F’ ) physically cut and cultured for 16 hr . Activation of BRV-B ( green ) occurs along the edges of the cut site ( arrowheads ) , coincident with AP-1-RFP expression ( red ) ( F ) . Conversely the BRV-BΔAP-1 reporter is not activated by physical wounding ( F’ ) . ( G ) Basal section of a day 7 disc bearing BRV-BΔAP-1 , following rnts>egr ablation , imaged at the time of downshift to 18oC . Loss of the predicted AP-1 sites abolishes activation of the reporter ( green ) . ( H ) Maximum projection Z-stack of a dlg hemizygous mutant disc from early L3 bearing the BRV-B enhancer . BRV-B reporter expression is maximal in the dorsal hinge region coinciding with the site most prone to neoplastic overgrowth ( Khan et al . , 2013 ) . ( I ) Maximum projection Z-stack of a day 7 disc bearing the BRV-B enhancer following expression of rnts>yki for 20 hr . Hyperplastic growth resulting from yki expression fails to activate BRV-B GFP expression . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 01710 . 7554/eLife . 11588 . 018Figure 6—figure supplement 1 . The BRV-B reporter is activated by physical damage , and is dependent on JNK signaling . Day 7 discs bearing the BRV118-GFP reporter ( green ) , physically fragmented with tungsten wire ( yellow arrowheads ) , and cultured in Robb’s media for 16 hr , in the presence of 1% DMSO vehicle ( left ) or 10 μM of the small molecule JNK pathway inhibitor SP600125 in 1% DMSO ( right ) . Addition of the inhibitor prevents activation of the reporter following damage ( open arrowheads ) , and also constrains the morphological changes ( folding and curling of the disc epithelium ) associated with culturing conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 01810 . 7554/eLife . 11588 . 019Figure 6—figure supplement 2 . Activation of the BRV-B reporter coincides with damage-induced , but not developmentally regulated JNK pathway activity . ( A ) Dorsal-most portion of a late larval wing disc bearing the AP-1 RFP reporter ( red ) and BRV-B GFP reporter ( green ) , showing activation of the JNK pathway in the peripodial cells of the apical notum that will participate in JNK-dependent thorax closure after puparium formation ( open arrowhead ) , but absence of BRV-B expression . ( B ) Stage 13 embryo ( left ) and stage 15 embryo ( right ) bearing the AP-1 RFP reporter ( red ) and the BRV-B GFP reporter ( green ) undergoing dorsal closure , showing activation of the JNK pathway in leading edge cells ( open arrowheads ) , but absence of BRV-B expression . GFP autofluorescence of the gut is visible in the stage 15 embryo ( white arrow ) . DE-cadherin , blue . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 019 The 931bp BRV-B fragment contains three strongly conserved high-consensus sites that match either the vertebrate or Drosophila AP-1 ( Kay/Jra ) motif ( Lee et al . , 1987a; 1987b; Perkins et al . , 1988 ) ( Figure 6E ) . We deleted these sites in the BRV-B reporter to generate BRV-B∆AP1 . Following physical fragmentation , discs bearing the BRV-B reporter strongly expressed GFP along the wound edges , coincident with AP-1-RFP ( Figure 6F ) . In contrast , deletion of the AP-1 sites in BRV-B∆AP1 completely abolished expression of GFP in fragmented discs despite AP-1 activation ( Figure 6F’ ) . Similarly , the BRV-B∆AP1 reporter failed to be activated following rnts>egr ablation ( Figure 6G ) , or after irradiation ( data not shown ) . Together , these data indicate that upon tissue damage , WNT signaling is activated , in significant part , by JNK signaling via the BRV-B enhancer . Consensus sites for several other candidate activators are also present in the BRV-B sequence , including Grainyhead , a transcription factor that has been shown to function with AP-1 in larval epidermal wound repair ( Mace et al . , 2005 ) , and the Dpp signaling transducers Mad and Medea . Several potential binding sites for the WNT pathway transcription factor TCF/Pangolin also exist , consistent with a previous observation that ubiquitously expressed wg can activate this enhancer ( Schubiger et al . , 2010 ) . In this way , wg could activate or maintain its own expression as part of a feed forward mechanism during regeneration . Thus , although a high level of JNK signaling may be able to activate the enhancer on its own , under conditions of damage in vivo JNK signaling may act in concert with other damage-induced signals to activate wg expression via the BRV-B enhancer . We also examined several other situations where JNK signaling is activated in the absence of tissue injury . The BRV-B reporter was expressed in discs mutant for discs large ( dlg ) ( Figure 6H ) or following knockdown of scribble ( data not shown ) . In both situations there is a loss of epithelial apicobasal polarity that results in neoplastic growth . In contrast , expression of yorkie ( yki ) , the growth-promoting co-activator downstream of the Hippo pathway , for 20 hr using the rnts system elicits overgrowth while preserving apico-basal polarity , but does not activate reporter expression ( Figure 6I ) . The BRV-B reporter was also not activated in two situations where JNK is activated under physiological conditions: in the notum of late stage larval discs , which is required for dorsal thorax closure during the pupal stage ( Figure 6—figure supplement 2A ) and dorsal closure of the embryo ( Figure 6—figure supplement 2B ) . These differences may reflect the extent of JNK activation since BRV-B is mostly activated under conditions where JNK is activated at high levels . Alternatively other factors specific to stress or damage may function in combination with JNK to modulate expression . Unlike the 2 . 9 kb BRV118 full enhancer , the expression directed by BRV-B covers a broader region and is also unchanged between day 7 and day 9 ( Figures 4D’ , 7B ) . Thus a negatively acting element must be present in either the BRV-A or BRV-C fragments , to reduce its area of expression and to mediate a further decrease in expression on day 9 . Inclusion of the BRV-A fragment to the reporter ( BRV-AB ) elicited a pattern of expression that was similar to BRV-B ( Figure 7—figure supplement 1A ) . In stark contrast , the addition of the BRV-C region to BRV-B ( BRV-BC , Figure 7A ) resulted in decreased expression in day 7 discs , both in the level and in the region of reporter expression ( Figure 7B ) . Moreover , expression was greatly reduced in day 9 discs in a manner similar to the full-length 2 . 9 kb enhancer fragment ( Figure 7B ) . Thus , a negatively acting element in BRV-C is responsible for limiting the response of BRV118 during regeneration and its ability to do so increases as the disc matures . 10 . 7554/eLife . 11588 . 020Figure 7 . Polycomb-mediated epigenetic silencing of the BRV118 enhancer limits damage-responsive wg expression in mature discs . ( A ) Schematic of the BRV-BC reporter transgene in relation to BRV118 and BRV-B . Numbers indicate nucleotide positions . ( B ) Basal sections of day 7 discs bearing the BRV-B ( left panels ) or BRV-BC ( right panels ) reporters , ablated by rnts>egr on day 7 ( top row ) or day 9 ( bottom row ) . Addition of the BRV-C fragment reduces expression on day 7 and almost completely abolishes expression on day 9 . ( C ) Schematic of the deletion series used to test silencing of BRV-B by BRV-C fragments ( top ) . Predicted binding sites for Brinker ( Brk ) , Ecdysone Receptor ( EcR ) and Pleiohomeotic ( Pho ) are shown as colored bars . The black histogram shows a measure of the evolutionary conservation in BRV-C , as in Figure 6E . Confocal sections of day 7 discs ablated by rnts>egr bearing each BRV-BC derivative ( bottom ) shows that multiple regions across BRV-C are required for silencing of BRV-B . ( D-D’ ) Day 9 discs bearing the BRV-BC reporter following rnts>egr ablation in wild type ( D ) and Pc15/+ heterozygotes ( D’ ) . Both damage-induced Wg ( red ) and reporter GFP expression ( green ) is elevated in Pc15/+ discs compared to wild type . ( E-E’ ) Heat shock induced Pc15 mutant clones ( marked by the absence of RFP ) in a wandering stage larval disc bearing the BRV-BC reporter ( green ) . Following clone induction , discs were irradiated with 45 Gy and dissected after 16 hr . BRV-BC reporter expression is elevated in Pc15 clones ( E’ , -/- homozygous Pc15 ) but significantly lower in the wild-type twin spot ( E’ , red ) . ( F ) Chromatin immunoprecipitation by anti-H3K27me3 followed by qPCR to detect regions surrounding wg and Wnt6 ( numbered primer sets 1–13 ) and the BRV118 region ( lettered primers sets ) . ChIP-qPCR was performed in wild type and homozygous BRV-ΔC day 7 and day 9 discs , and shown as the fold change between the two days . Schematic above shows numbered amplicon positions in the genome , WNT gene loci and previously identified developmental WNT enhancers ( green , DE1: Wing disc hinge and embryonic enhancers , ( Neumann and Cohen , 1996; Von Ohlen and Hooper , 1997 ) , DE2: Notum and leg/antennal disc enhancer ( Pereira et al . , 2006 ) , DE3: Eye/antennal disc and leg enhancer ( Koshikawa et al . , 2015 ) ) . In wild type discs ( black bars ) epigenetic silencing marks increase at the BRV118 locus in day 9 discs versus day 7 , while levels at the wg and Wnt6 coding sequences , and surrounding developmental wg enhancers , remain largely unchanged . Deletion of the BRV-C region of the enhancer from the genome abolishes the increase of H3K27me3 at the enhancer ( yellow bars ) . Error bars are SD of repeats from 3 independent ChIP experiments . ( G-G’ ) Day 9 wild type discs ( G ) or homozygous for BRV-ΔC ( G’ ) following ablation by rnts>rpr . Absence of the BRV-C enhancer fragment produces significantly more Wg in response to damage ( G’ , arrowhead ) . ( H ) Model of damage response regulation in a day 7 ( top ) and a day 9 ( bottom ) disc . JNK signaling activates wg and Wnt6 expression in response to damage via the BRV-B region , but epigenetic silencing of the enhancer , nucleated by BRV-C , in a day 9 disc overrides this activation . ( I ) Assay of adult wing sizes that develop from homozygous BRV-ΔC discs ablated with rnts>rpr on day 9 . Absence of the BRV-C enhancer fragment does not alter regeneration of ablated disc . Error bars are SD of 3 biological repeats , n>200 flies per repeat . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 02010 . 7554/eLife . 11588 . 021Figure 7—figure supplement 1 . Additional characterization of the silencing activity in BRV118 . ( A ) Wing discs bearing the BRV-B ( left ) and BRV-AB ( right ) reporters following ablation with rnts>egr on day 7 ( top ) and day 9 ( bottom ) , stained for GFP ( green ) . Addition of the BRV-A fragment does not reduce expression of the BRV-B reporter following ablation on either day . ( B ) Diagram of predicted Brk binding sites that are deleted in the BRV-BC reporter , yielding the BRV-BCΔBrk reporter transgene . ( C ) Wing discs bearing the BRV-BC and BRV-BCΔBrk reporter transgenes ( green ) following ablation with rnts>egr on day 9 , stained for GFP ( green ) . Loss of the Brk sites does not alleviate silencing by the BRV-C fragment . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 02110 . 7554/eLife . 11588 . 022Figure 7—figure supplement 2 . Additional data on PcG-mediated repression by BRV-C . ( A ) Eye colors resulting from expression of the w+ marker gene associated with the BRV-B ( top row ) or BRV-BC ( bottom row ) reporter transgenes , in wild type ( left panels ) , a pho null heterozygous mutant ( middle panels ) or a phol heterozygote ( right panels ) . Presence of the BRV-C fragment reduces expression of the w+ , resulting in lighter eyes , while this difference is lost in the pho/+ and phol/+ mutants . ( B ) Sequences that match consensus binding sites for PcG proteins within the BRV-C genomic region . Motifs for the following are highlighted: Pleiohomeotic ( Pho , purple , [Mihaly et al . , 1998; Fritsch et al . , 1999; Ringrose et al . , 2003] ) Zeste ( yellow , ( Mihaly et al . , 1998 ) , Grainyhead ( Grh , brown , [Mace et al . , 2005] ) , SP1 ( red , [Brown et al . , 2005] ) , Dorsal Switch Protein 1 ( DSP1 , green , [Déjardin and Cavalli , 2005] ) and GAGA factor ( GAGA , blue , [Strutt et al . , 1997; Ringrose et al . , 2003; Zhu et al . , 2011] ) . ( C ) Control regions for the H3K27me3 ChIP of wing discs . Graph shows level of H3K27me3 binding at the act5c transcribed region ( open chromatin ) and the heterochromatin H23 loci ( closed chromatin ) ( Zhang et al . , 2008 ) in wild type and BRV-ΔC discs , shown as the ratio of binding in day 9 versus day 7 discs . Level of H3K27me3 binding does not change at either locus from day 7 to day 9 . Error bars are SD of repeats from 3 independent ChIP experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 02210 . 7554/eLife . 11588 . 023Figure 7—figure supplement 3 . Cas9-mediated deletion of the BRV-C genomic region . ( A ) Schematic of the CRISPR mediated BRV-C genomic deletion used to generate the BRV-ΔC allele . Numbers indicate nucleotide positions relative to BRV-BC . ( B ) Crossing scheme used to generate the BRV-ΔC allele . Over 100 lines were screened via genomic PCR , yielding a single deletion line . ( C ) PCR showing amplicon of BRV-C genomic region from the parental line ( left ) and BRV-ΔC line ( right ) . ( D ) Sequences of the parental chromosome used to generate the BRV-ΔC allele , and the deletion in relation to the 986 bp BRV-C sequence used in the enhancer reporter transgenes ( shown in blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 02310 . 7554/eLife . 11588 . 024Figure 7—figure supplement 4 . A fragment of BRV-B driving wg rescues damage-induced wg expression in late stage discs , but does not improve regeneration . ( A ) Schematic of the BRV-B3-wg transgene . The B3 fragment is a subdivision of the full BRV-B enhancer . Numbers indicate nucleotide positions . ( B ) Day 7 discs bearing BRV-B and BRV-B3 GFP reporters following ablation by rnts>egr , imaged at the time of downshift . BRV-B3 yields slightly reduced levels of expression compared to BRV-B . ( C ) Apical and basal sections of disc proper showing Wg ( red ) following rnts>rpr ablation on day 9 in wild type ( top ) and BRV-B3-wg bearing discs ( bottom ) . The BRV-B3-wg transgene causes expression of Wg in the basal wound periphery cells ( arrowhead ) , which is absent in the wild type disc . ( D ) Assay of adult wing sizes that develop following ablation with rnts>rpr on day 9 in wild type and BRV-B3-wg discs . Mean differences between wild type and BRV-B3-wg data sets is not statistically significant ( N . S . ) , calculated by two way ANOVA . Error bars are SEM of 3 biological repeats , n>100 flies for each genotype/condition . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 024 To map this element more precisely , we generated deletions from the end of BRV-C furthest away from BRV-B and examined their ability to silence GFP expression on day 9 ( Figure 7C ) . We found that as more of the BRV-C fragment was deleted , there was a progressive loss of silencing , indicating that multiple regions within BRV-C contribute to this activity . However , there was a strong difference in silencing activity between deletions that retained 441 and 183 bp of BRV-C , suggesting the presence of significant silencing activity within this 258 bp region . Sequences at both ends of this 258 bp region have almost complete conservation of nucleotide identity across 12 Drosophila species ( Figure 7C ) , including two predicted binding sites for the transcriptional co-repressor Brinker ( Brk ) . We generated a BRV-BC enhancer with both Brk sites mutated but found no loss of silencing activity ( Figure 7—figure supplement 1B–C ) , indicating either that Brk does not function in mediating the repression , or that its activity is redundant with other mechanisms that are still functional . Thus , unrecognized motifs , likely within the conserved blocks clustered near the predicted Brk binding sites , must mediate the silencing activity associated with this region . In many cases , developmentally-regulated gene silencing is mediated by the recruitment of Polycomb Group ( PcG ) proteins ( Schuettengruber et al . , 2007; Schwartz and Pirrotta , 2007 ) , which act by promoting the trimethylation of histone H3 on lysine 27 ( H3K27 ) ( Cao et al . , 2002 ) . For many transgenes that contain elements capable of recruiting PcG proteins , silencing of the adjacent mini-white gene in the transgene is also observed ( Kassis and Brown , 2013 ) . In these cases , the silencing of the mini-white gene is sensitive to the levels of PcG proteins . Indeed , we observed that transgenic flies bearing the BRV-BC reporter had lighter eyes than those bearing the BRV-B reporter ( Figure 7—figure supplement 2A ) even though both reporter genes were inserted at the identical site in the genome , suggesting that the BRV-C fragment was capable of mediating silencing of surrounding DNA when integrated at a genomic locus outside of the WNT cluster . However , in flies that were heterozygous for pleiohomeotic ( pho ) or pleiohomeotic-like ( phol ) , the eye color was similar to those bearing the BRV-B fragment ( Figure 7—figure supplement 2A ) , indicating that the silencing activity induced by the BRV-C fragment was indeed sensitive to PcG group gene dosage . We examined the sequence of BRV-C and found multiple potential binding sites for proteins shown to be important for PcG silencing at PREs , including 2 conserved pho/phol binding sites ( Figure 7—figure supplement 2B ) . To test whether the silencing activity of BRV-C during disc regeneration is regulated by PcG , we examined the expression of the full length BRV118-GFP reporter in rnts>egr imaginal discs heterozygous for the null allele of Polycomb , Pc15 . In these flies , robust expression of the reporter was observed in day 9 discs , as was expression of the Wg protein expressed from the endogenous wg gene ( Figure 7D–D’ ) . However , due to changes in developmental timing , increased lethality , and the significant transdetermination levels associated with the Pc allele , documented previously ( Lee et al . , 2005 ) , we were unable to test whether Pc gene dose affected levels of regeneration in our genetic ablation system . In order to avoid potential organism-wide pleiotropic effects of the Pc allele , we generated Pc15 clones in developing discs that also had the BRV-BC reporter . Day 9 mosaic discs were irradiated to induce damage and examined for enhancer activity . Despite morphological disruption within the disc epithelium , likely due to the presence of Pc mutant tissue , and widespread cell death associated with irradiation , GFP was clearly expressed in the Pc mutant clones , and to a lesser extent in the surrounding heterozygous tissue , but was absent from the wild type sister clones ( Figure 7E–E’ ) . Thus , Pc is required to prevent damage-induced activation of the BRV-BC reporter in mature discs . To directly examine whether there is a change in the levels of PcG-mediated repression of the wg and Wnt6 genes between day 7 and day 9 , we used chromatin immunoprecipitation ( ChIP ) to examine the levels of H3K27 methylation across the WNT gene cluster from undamaged wing imaginal discs . In comparison to day 7 discs , there was a significant increase in the levels of H3K27 methylation in day 9 discs , which was mostly restricted to the BRV118 enhancer ( Figure 7F ) . Most of the wg and Wnt6 transcription units showed no change in H3K27 methylation , nor was there a significant increase in methylation observed in regions known to regulate aspects of developmental wg expression in third-instar discs ( Neumann and Cohen , 1996; Pereira et al . , 2006; Koshikawa et al . , 2015 ) ( Figure 7F ) . This is consistent with the observation that both wg and Wnt6 are expressed in undamaged discs at this stage of development . Two other genomic regions examined , act5C and the heterochromatin H23 locus , also showed no significant change between day 7 and day 9 ( Figure 7—figure supplement 2C ) . Since the increase in the histone modification most characteristic of PcG-mediated silencing during L3 is mostly restricted to the BRV118 region , this suggests that the damage-responsive enhancer is silenced in day 9 discs , while developmentally-regulated WNT enhancers are not . To examine the importance of the silencing induced by BRV-C in vivo , we used CRISPR-mediated genome editing to generate a deletion that removed 1132bp , including the entire BRV-C fragment ( BRV-∆C , Figure 7—figure supplement 3 ) . Flies homozygous for the deletion are viable and have no observable developmental abnormalities , but often hold their wings in an outstretched position reminiscent of the wgP allele , in which the region encompassing the entire BRV118 enhancer is removed from the wg coding sequence by a genomic inversion ( Buratovich and Armer , 2002 ) . ChIP experiments performed on day 9 BRV-∆C discs demonstrated a pattern of H3K27 methylation that resembles that normally found in wild-type day 7 discs in the region of the BRV118 enhancer ( Figure 7F ) . Thus the sequences within BRV-C are necessary for the increased deposition of H3K27 histone marks by day 9 . To examine the functional consequences of this changed pattern of H3K27 methylation , the level of wg expression was examined in regenerating rnts>rpr day 9 BRV-∆C discs . In the absence of BRV-C , damage-induced wg expression is greatly increased in the disc proper epithelium surrounding the ablated pouch ( Figure 7G–G’ ) , indicating that the BRV-C region is required in vivo to suppress wg expression following damage in day 9 discs ( Figure 7H ) . However , when assayed for regeneration , there was no observable difference between BRV-∆C and flies bearing an intact enhancer region ( Figure 7I ) . We also attempted to restore expression of wg in another way . The BRV-B enhancer directs expression to a wide region surrounding the damaged tissue and only in response to tissue damage . Moreover , the BRV-B enhancer is equally active on day 7 and day 9 . We therefore attempted to generate a transgene that expressed wg directly under the control of BRV-B , independently of Gal4/UAS . However , we did not obtain viable transformants . We were able to obtain transformants where wg was driven by a sub-fragment of BRV-B , BRV-B3 ( Figure 7—figure supplement 4A ) possibly because it drives expression at lower levels ( Figure 7—figure supplement 4B ) . While these transformants have high levels of wg following damage ( Figure 7—figure supplement 4C ) , we did not observe an improvement in regeneration ( Figure 7—figure supplement 4D ) indicating that increased wg expression alone is insufficient to improve regeneration and that the damage-responsive expression of other genes necessary for regeneration might also be silenced in mature discs . Myc is upregulated during regeneration ( Figure 2B , Smith-Bolton et al . , 2009 ) , and is therefore likely to be an important driver of regenerative growth . Moreover , we had previously shown that in immature discs , Myc promoted regeneration , while other growth promoting pathways tested , including CyclinD/CDK4 , JAK/STAT and Rheb did not ( Smith-Bolton et al . , 2009 ) . However , in those experiments , Myc was expressed under the control of Gal4/UAS , and it was therefore not possible to express Myc in cells that were not also expressing egr , and expression of Myc was confined to the time of ablation . To more rigorously test whether Myc could promote regeneration in older discs , we chose to target Myc expression to the blastema by directly driving expression under the control of BRV-B ( Figure 8A ) . In contrast to our experience manipulating wg expression , we were able to obtain viable transformants . 10 . 7554/eLife . 11588 . 025Figure 8 . Circumventing the age-related repression of Myc improves regeneration . ( A ) Schematic of the BRV-B-Myc transgene . Numbers indicate nucleotide positions . ( B-D ) Basal sections of discs following rnts>rpr ablation on day 7 ( left panel ) and day 9 ( center panel ) , and on day 9 in discs bearing BRV-B-Myc transgene ( right panel ) , imaged at the time of downshift . ( B ) Myc expression following ablation . Myc is expressed in basal wound periphery cells of a day 7 disc , but is absent in a day 9 ablated disc . The BRV-B-Myc transgene causes expression of Myc protein in an older disc . ( C- D ) Discs bearing the AP-1 GFP reporter ( C ) or stained with anti-Mmp1 ( D ) , and imaged as in ( B ) . AP-1 reporter expression is strongly expressed in basal cells of the pouch in day 7 ablated discs , coincident with Mmp1 expression . AP-1 reporter expression is found in the ring of wound periphery cells in day 9 ablated discs ( arrowheads ) , while Mmp1 staining is reduced . Day 9 ablated discs bearing BRV-B-Myc have AP-1 GFP expressing cells covering the basal wound surface , resembling the expression pattern of day 7 discs , while Mmp1 expression is stronger than a wild type day 9 disc . ( E-F ) Discs stained for Wg ( E ) or bearing the BRV118-GFP reporter ( F ) following ablation with rnts>egr , and imaged as in ( B-D ) . Declining expression of the reporter on day 9 is rescued in discs bearing the BRV-B-Myc transgene . Similarly , wg expression is stronger on a day 9 bearing the BRV-B-Myc transgene , comparable to expression level in an ablated day 7 disc . ( G-G’ ) A day 9 disc bearing the BRV-BC enhancer with heat shock induced flip-out clones ( marked by RFP , red ) expressing Myc , irradiated with 45 Gy and dissected 16 hr later . Clones expressing Myc respond to damage by activating the BRV-BC enhancer ( G’ , green ) , while the enhancer remains inactive in neighboring damaged tissue not expressing Myc . ( H-H’ ) Day 9 discs as in ( G-G’ ) , in the absence of irradiation . Expression of Myc in clones ( red ) causes low level of BRV-BC enhancer activity ( H’ , green ) , even in the absence of damage . ( I ) Assay of adult wing sizes that develop following ablation with rnts>rpr on day 7 , 8 , 9 and 10 in wild type and BRV-B-Myc discs , quantifying animals that eclose with fully regenerated wings . The BRV-B-Myc expressing animals consistently eclose with more of the full regenerated wings compared to wild type . Mean differences between wild type and BRV-B-Myc data sets is statistically significant , ( p-value calculated by two way ANOVA ) . Error bars are SEM of at least 3 biological repeats , scoring a total of >200 animals in each condition . ( J ) Developmental timing of wild type and BRV-B-Myc expressing larvae following rnts>rpr ablation on day 7 , 8 , 9 and 10 . Delay is measured as hours at which 50% of larvae have pupariated compared to unablated wild type . The BRV-B-Myc transgene ( red bars ) does not alter developmental timing following ablation compared to wild type ( gray bars ) on any day ablated . Mean differences between wild type and BRV-B-Myc data sets is not statistically significant ( N . S . ) , calculated by two way ANOVA . Error bars are SEM of 3 biological repeats , n>100 flies for each genotype/condition . DOI: http://dx . doi . org/10 . 7554/eLife . 11588 . 025 In the absence of tissue damage , flies carrying the BRV-B-Myc transgene displayed no obvious growth abnormalities . However , following tissue ablation using rnts>rpr or rnts>egr , strong expression of Myc was detected even in day 9 discs ( Figure 8B ) . In these discs , cells expressing the AP-1 reporter resembled the cells normally found in day 7 discs in that they were more flattened and Mmp1-expressing cells appeared to be migrating into the ablated portion of the disc ( Figure 8C–D ) . Moreover , expression of Wg protein was increased in day 9 discs to levels comparable to those found in day 7 discs following ablation ( Figure 8E ) . This likely results from an increase in wg transcription , since the full length BRV118 reporter , which includes the entire negatively acting BRV-C element , was also expressed in ablated day 9 discs ( Figure 8F ) . This reactivation of the enhancer by Myc is cell autonomous , as clones expressing Myc in mature discs activate the enhancer following irradiation , while neighboring irradiated cells do not ( Figure 8G–G’ ) . Interestingly , clonal expression of Myc causes limited activation of the enhancer even in the absence of damage , but at a significantly reduced level than with irradiation ( Figure 8H–H’ ) . Thus , at least on day 9 , expression of Myc is able to activate transcription mediated by an enhancer that is normally epigenetically silenced . Finally , as BRV-B driven Myc increases the expression of several genes that are usually absent from day 9 ablated discs , we examined whether regenerative ability was also affected . When compared to wild type , flies bearing BRV-B-Myc had significantly improved regeneration following ablation on day 7 , 8 , 9 and 10 ( Figure 8I ) . This improvement in regeneration occurred without extending the delay in pupariation induced by ablation ( Figure 8J ) . The improvement in regeneration elicited by Myc expression is modest by day 10 , indicating that the loss of regenerative capacity becomes progressively more refractory to reversal even by increased levels of Myc expression . Overall , these data show that using an endogenous damage-responsive enhancer to target Myc expression specifically to cells surrounding the ablated tissue that give rise to the blastema can augment the regenerative process , likely by circumventing or reversing the repression of multiple genes that function in regeneration . Many organisms lose the ability to regenerate damaged tissues as they mature ( Dent , 1962; Reginelli et al . , 1995; Beck et al . , 2003; Slack et al . , 2004; Smith-Bolton et al . , 2009; Porrello et al . , 2011; Cox et al . , 2014 ) . This change often occurs concurrently with a slowing of the growth of the organism , or a major transformation in its developmental state , e . g . metamorphosis in Drosophila ( Smith-Bolton et al . , 2009 ) and Xenopus ( Dent , 1962 ) . The loss of regenerative capacity is likely an important mechanism to balance the successful progression to reproductive adulthood at the cost of forming functionally complete tissue . Very few ‘true’ regeneration-specific genes have been identified ( i . e . genes that are not required at any other time throughout the organism’s life ) , but rather developmentally required pathways are often re-used during regeneration ( Sun and Irvine , 2014 ) . Thus , how regenerative growth can be selectively inhibited without compromising cell proliferation or differentiation remains unknown . Here we have shown that in the Drosophila wing disc this loss of regenerative capacity is achieved in part by the localized epigenetic inactivation of a damage-responsive enhancer that regulates the expression of wg and potentially Wnt6 . This mechanism allows an organism to continue with its normal developmental program while shutting down its regenerative response to tissue damage . Previous studies have demonstrated that the JNK pathway is robustly activated following tissue damage and has an important role in regenerative growth ( Bosch et al . , 2005; Mattila et al . , 2005; Bergantiños et al . , 2010 ) . Our data confirm that JNK is strongly activated following damage , but furthermore , it appears similarly activated in both day 7 and day 9 discs , as assessed by the expression of an AP-1 reporter . Thus , the loss of regeneration that occurs between day 7 and day 9 cannot be attributed to a failure to activate JNK . Despite the similar levels of AP-1 activity , the cellular responses and changes in gene expression elicited by tissue damage differ considerably as the disc matures . Importantly , genes that are known to be downstream targets of the JNK/AP-1 pathway such as Mmp1 ( Uhlirova and Bohmann , 2006 ) and DILP8 ( Colombani et al . , 2012; Katsuyama et al . , 2015 ) have reduced expression on day 9 when compared to day 7 . These changes in gene expression are likely to account for many of the differences in the cellular responses to tissue damage that we observe . In addition to the aforementioned genes , the WNT genes wg and Wnt6 also exhibit a significant decline in damage-induced expression with disc maturity . Our data shows this is due to the highly localized epigenetic silencing of a damage-responsive WNT enhancer , BRV118 , that prevents their expression specifically in response to injury in mature discs , but still allows expression from nearby developmentally regulated enhancers . This mechanism ensures that the contribution of both genes to a regeneration program can be shut off in mature tissues independently of their essential roles in growth and development of the disc . Our inability to detect expression of the BRV118-GFP reporter in unablated discs suggests that the BRV118 enhancer does not have a role in normal development . However , the wg1 allele , which results in an incompletely-penetrant phenotype characterized by a failure to specify the wing pouch , is a deletion whose breakpoints are very close to the boundaries of the BRV118 fragment that we have studied ( Schubiger et al . , 2010 ) . This suggests that a separate element , possibly very close to , but not fully contained within the boundaries of BRV118 , may also be disrupted by the wg1 deletion . The expression profile of regenerating discs suggests the regulation of multiple genes is required during regeneration ( Klebes et al . , 2005; Blanco et al . , 2010; Katsuyama et al . , 2015 ) , and that a significant number of these genes are also involved in developmental processes . Thus , equivalent regeneration specific enhancers , like BRV118 , might also exist for these genes , such as DILP8 and Mmp1 . Both genes are known to be activated by JNK , although damage-responsive enhancers have not yet been characterized . Notably though , the Mmp1-lacZ reporter we used to investigate Mmp1 activation ( Figure 2—figure supplement 2 ) , which accurately reflects Mmp1 protein expression following injury ( Uhlirova and Bohmann , 2006 ) , consists of a ~5 kb intronic region upstream of a lacZ reporter , which , based on its pattern of expression on days 7 and 9 , must possess regulatory regions that allow both damage-induced activation and maturity-dependent silencing . Sequence comparison with BRV118 reveals several AP-1 binding sites that are identical to those found in BRV118 , and multiple consensus sites for PcG repression . This combination of regulatory motifs could therefore reflect a molecular signature of genes that function in regeneration , and thus could potentially be used to identify genes that comprise a regeneration program through genome-wide analyses in the future . Our studies of the regulation of wg expression have shown that , despite similar levels of JNK activation , increased levels of PcG-mediated epigenetic silencing can override the effect of JNK activation and suppress gene expression in late L3 . PcG-mediated silencing is best characterized for its role in the epigenetic silencing of Hox genes during embryonic development in Drosophila ( Beuchle et al . , 2001; Beck et al . , 2010 ) , but also has important functions in imaginal disc development ( Pirrotta et al . , 1995; Maurange and Paro , 2002; Perez et al . , 2011; Mason-Suares et al . , 2013 ) and during regeneration ( Klebes et al . , 2005; Lee et al . , 2005; McClure et al . , 2008; Katsuyama et al . , 2015 ) . Indeed inappropriate cell fate switching following damage in imaginal discs ( transdetermination ) is associated with changes in PcG gene expression ( Klebes et al . , 2005; Lee et al . , 2005; McClure et al . , 2008 ) , and in one instance JNK signaling reduced the extent of PcG mediated repression ( Lee et al . , 2005 ) . A key property of epigenetic regulation by PcG is the ability to simultaneously silence multiple regions across the genome via the activity of a single master regulator complex , and , moreover , this silencing is heritable and thus its activation can maintain the locus in a repressed state through subsequent cell generations ( Déjardin and Cavalli , 2005 ) . Such a mechanism is ideally suited to the sustained and progressive silencing of a regeneration program during the ongoing growth and development of imaginal discs . However , unlike Hox genes , silencing of wg and Wnt6 does not involve the entire transcription unit , but rather , is restricted to a damage-responsive enhancer . A similar local mode of epigenetic regulation has been described for the Drosophila rpr locus , in which epigenetic blocking of an irradiation-responsive enhancer region through enrichment of H3K27me3 prevents rpr expression following irradiation in late embryogenesis ( Zhang et al . , 2008 ) . Importantly , the remainder of the rpr locus itself remains accessible , and is thus responsive to developmental signals required for programmed cell death to occur in the nervous system in late embryogenesis ( Maurange et al . , 2008; Rogulja-Ortmann et al . , 2008 ) . Localized epigenetic silencing of individual regulatory elements is therefore likely an important and potentially pervasive mechanism by which gene expression can be selectively activated or repressed by distinct inputs . But how is this epigenetic silencing limited to just the enhancer ? Elements that are responsible for expression of the "inner circle" of wg expression at the edge of the pouch and for expression in the leg disc are immediately adjacent to the BRV118 enhancer ( Pereira et al . , 2006 ) . Thus , while the BRV-C fragment nucleates PcG-mediated repression that then spreads over the remainder of the BRV118 enhancer , mechanisms must exist that limit spread beyond the borders of the enhancer and thus preserve the activity of the adjacent developmentally-regulated enhancers . Chromatin boundary elements that are able to block the spread of heterochromatin formation have previously been described ( Gaszner and Felsenfeld , 2006 ) and are found in a variety of organisms including Drosophila ( Roseman et al . , 1993; Kahn et al . , 2006; Lin et al . , 2011 ) . Unlike other boundary elements such as insulators that inhibit enhancer-promoter interactions ( Gurudatta and Corces , 2009 ) , these ‘chromatin barrier’ elements can prevent the propagation of repressive histone marks separately from a role in enhancer blocking ( Recillas-Targa et al . , 2002; Lin et al . , 2011 ) . Thus , a similar barrier element might be present within or near BRV118 to limit chromatin modifications to the damage responsive region , yet allow nearby developmental enhancers to remain active . If multiple genes that function in regeneration have a similar bipartite mode of regulation , it is unlikely that expressing just one of these genes at a later stage of development can restore the ability to regenerate . Indeed , we found that restoring wg expression in day 9 discs did not promote regeneration . In contrast , expression of Myc , which is able to increase the levels of expression of both wg and Mmp1 , and possibly the expression of other genes that are similarly regulated , was able to enhance regeneration . However , it is likely that Myc does not promote the expression of all genes that have been silenced in late L3 . Indeed , unlike wg and Mmp1 , we found that the JAK/STAT reporter is not reactivated in mature discs by the presence of Myc ( data not shown ) . In addition , the delay in pupariation is not restored , which possibly results from a failure to restore the damage-responsive DILP8 expression level to that of a younger disc . While we have shown that Myc functions cell autonomously to reactivate BRV118-mediated expression of WNT genes , it is unclear whether Myc reverses the PcG-mediated repression of BRV118 or bypasses it completely . However , since the BRV-B-Myc transgene is only expressed in a small region of the disc , it is not easy to detect a change in the overall level of H3K27 methylation at the WNT locus in these cells with confidence . Additionally , even increasing Myc levels has little effect by day 10 , suggesting that the silencing mechanism has become even more effective . It might be necessary to combine Myc overexpression with other manipulations to restore regeneration at even later stages . Previous studies have implicated Myc as a regulator of chromatin organization ( Amente et al . , 2011 ) and also as a regulator of cellular reprogramming ( Smith and Dalton , 2010; Chappell and Dalton , 2013 ) , and therefore studying the role of Myc in reactivating BRV118-mediated expression might provide a tractable way of understanding the role of Myc in these processes . Overall , our investigation has revealed a mechanism by which genes required for both regeneration and development can be regulated to allow the age-dependent restriction of a regenerative response without affecting normal organismal growth and patterning of tissues . As PcG proteins are highly conserved from flies to vertebrates , as indeed are the targets they regulate ( Ringrose , 2007 ) , it would be of considerable interest to determine whether the loss of regenerative capacity in vertebrates also results from the selective epigenetic silencing of damage-responsive enhancers that regulate orthologs of Drosophila genes that we have studied , such as matrix metalloproteases and WNT genes .
Stocks and crosses were maintained on yeast food at 25°C , except those for ablation experiments , which were maintained at 18°C . Stocks used in this study: rnts>egr ( w1118;; rn-Gal4 , tub-Gal80ts , UAS-egr ) and rnts>rpr ( w1118;; rn-Gal4 , tub-Gal80ts , UAS-rpr ) ( Smith-Bolton et al . , 2009 ) , AP-1-GFP and AP-1-RFP reporters ( Chatterjee and Bohmann , 2012 ) , Stat92E-GFP ( Bach et al . , 2007 ) , bantam-GFP ( Matakatsu and Blair , 2012 ) , MMP1-LacZ ( Uhlirova and Bohmann , 2006 ) , PCNA-GFP ( Thacker et al . , 2003 ) , CycE-GFP ( Deb et al . , 2008 ) , puc-LacZ ( Ring and Martinez Arias , 1993 ) , nos-Cas9 ( Kondo and Ueda , 2013 ) , vkg-GFP ( Buszczak et al . , 2007 ) , dlg1A40 . 2 ( a gift from D Bilder ) . Stocks obtained from the Bloomington stock center: wg1 ( BL2978 ) , UAS-hepCA ( BL6406 ) , hepr75 ( BL6761 ) , UAS-yki ( BL28819 ) , Pc15 FRT2A ( BL24468 ) , phob ( BL1140 ) , phol81A ( BL24164 ) , DILP8-GFP ( BL33079 ) , UAS-Myc ( BL9674 ) , Act5C>FRT . CD2>GAL4 , UAS-RFP ( BL30558 ) , hs-Flp ( BL8862 ) . Genetic ablation experiments and scoring of adult wings was performed essentially as described in ( Smith-Bolton et al . , 2009 ) with each experimental condition compared to a suitable control that was ablated and scored in parallel . Unless otherwise indicated , discs were dissected and fixed for immunofluorescence immediately after the ablation period . The BRV118 enhancer was originally identified by S Carroll , and the BRV118-lacZ transgenic line that initiated this study was obtained from G Schubiger ( Schubiger et al . , 2010 ) . The BRV118-GFP enhancer reporter was generated by cloning 2903 bp of the BRV118 genomic region upstream of the minimal hsp70 promoter and eGFP coding sequence in pEGFPattB ( K Basler ) . Reporter derivatives were generated by replacing the BRV118 enhancer DNA with the genomic regions listed in Supplementary file 1A . All GFP reporters were inserted into the AttP40 landing site via PhiC31 recombination . The BRV-BΔAP1 and BRV-BCΔBrk transgenes were generated using InFusion PCR mutagenesis ( Clontech , Mountain View , CA ) to sequentially delete the consensus sequences ( primers listed in Supplementary file 1A ) . The BRV-B-Myc transgene was generated by replacing the eGFP coding sequence in the BRV-B GFP reporter construct with the Myc coding sequence ( GenBank: AY058627 ) and inserted into the AttP40 landing site . The BRV-B3-Wg transgene was generated by replacing the eGFP coding sequence in BRV-B3 with the wg coding sequence ( GenBank: BT133499 ) and 1079 bp of the 3’UTR to ensure proper transcript subcellular localization ( Simmonds et al . , 2001 ) , and inserting into the VK00013 ( chr:3L ) landing site . Cloning primers for wg and Myc constructs are listed in Supplementary file 1 . Transgenic services were provided by BestGene ( Chino Hills , CA ) . Discs were fixed and stained essentially as in ( Smith-Bolton et al . , 2009 ) , and mounted in ProLong Gold Antifade Reagent ( Cell Signaling , Beverly , MA ) . The following primary antibodies were used in this study: from the DSHB , Iowa City , IA; mouse anti-Wg ( 1:100 , 4D4 ) , mouse anti-Mmp1 ( 1:100 , a combination of 14A3D2 , 3A6B4 and 5H7B11 ) , rat anti-DE-cadherin ( 1:100 , DCAD2 ) , mouse anti-Delta ( 1:100 , C594 . 9B ) . Other antibodies; guinea pig anti-Myc ( 1:100 , a gift from G . Morata ) , mouse anti-PHH3 ( 1:500 , Cell signaling ) , rabbit anti-DCP-1 ( 1:250 , Cell signaling ) , rabbit anti-GFP ( 1:500 , Torrey Pines Laboratories , Secaucus , NJ ) , mouse anti-GFP ( 1:500 AB290 , Abcam , Cambridge , MA ) , rabbit anti-β-galactosidase ( 1:1000 , #559762; MP Biomedicals , Santa Ana , CA ) . Secondary antibodies used were from Cell Signaling , all at 1:500; donkey anti-mouse 555 , donkey anti-rabbit 555 , donkey anti-rat 647 , donkey anti-rabbit 488 and donkey anti-mouse 488 . Nuclear staining was by DAPI ( 1:1000 , Cell Signaling ) . Samples were imaged on a Leica TCS SP2 Scanning confocal or Zeiss Light Sheet Z1 . To generate clonal patches of cells expressing egr , flies of genotype hsflp; BRV-B; Gal80ts , UAS-egr were crossed to act>>Gal4 , UAS-RFP . The progeny were heat shocked at 37°C for 15 min at 84 hr and 96 hr after egg deposition ( AED ) and maintained at 18°C to allow growth of Gal4 expressing clones . Larvae were transferred to 30°C at day 7 of development to inactivate Gal80ts and allow egr expression in clones . Discs and brain tissue were dissected after 24 hr egr expression , fixed and stained as described . To generate Pc15 mutant clones , flies of genotype hsflp; BRV-B; his::RFP FRT2A were crossed to Pc15 FRT2A/TM6Csb . The progeny were maintained at 18°C , heat shocked at 37C for 2 hr at 48 hr and 72 hr AED , and irradiated with 45 Gy on day 9 . After 16 hr recovery , discs were dissected , fixed and stained as described . Control larvae were processed identically , omitting the irradiation step . To generate Myc expressing clones , flies of genotype hsflp; BRV-BC; Act5C>FRT . CD2>Gal4 , UAS-RFP were crossed to UAS-Myc and heat shocked 48 hr AED for 10 min . Larvae were irradiated with 45 Gy on day 9 and dissected as above . For each immunoprecipitation ( IP ) chromatin was prepared from approximately 300 wing discs from day 7 larvae or 200 wing discs from day 9 larvae , and the IP was performed essentially as follows: larvae were dissected in small batches in ice cold Schneider’s medium , fixed in 1% paraformaldehyde in PBS for 10 min at room temperature , the reaction was quenched with 250 mM glycine , washed in PBS three times for 10 min at room temperature , discs were removed from carcasses in lysis buffer ( 10 mM Tris HCl pH8 , 1 mM EDTA , 0 . 1% SDS , 1 mM PMSF , 1% Triton-X100 , Complete protease inhibitor [Roche Diagnostics , Indianapolis , IN] ) , pelleted and snap frozen in liquid nitrogen . Pellets were pooled and resuspended in 500 ul lysis buffer , sonicated with a Biorupter ( Diagenode , Denville , NJ ) on high for 12 . 5 min , 30 s on 30 s off in ice to yield chromatin in the 500–1000 bp range . The sonicated samples were placed into RIPA buffer ( lysis buffer plus 10% DOC and 140 mM NaCl ) , centrifuged on max for 5 min at 4o°C , and transferred to a new tube . 60 μl was removed for ethanol precipitation and run on a 1 . 5% gel to test sonication efficiency . From the remaining chromatin sample 5% was removed as input , and the rest was used for IP with 10 μg anti-H3K27Me3 ( EMD Millipore , Billerica , MA ) in RIPA buffer with overnight incubation at 4°C , or no antibody control . IP samples were purified on Protein-A sepharose 4B beads ( Sigma-Aldrich , St . Louis , MO ) in RIPA buffer for 30 min at room temperature . Beads were washed four times in RIPA buffer , once in RIPA buffer with 500 mM NaCl , once in LiCl2 ChIP buffer ( 10 mM Tris HCl pH8 , 1 mM EDTA pH8 , 1% DOC , 1% NP40 , 250 mM LiCl2 ) and three times in TE buffer ( 10 mM Tris HCl pH8 , 1 mM EDTA pH8 ) for 15 min each wash . Beads were resuspended in 50 ug/ml RNAseA ( New England Biolabs , Ipswich , MA ) for 30 min incubation at 37°C , washed in TE , and resuspended in elution buffer ( 0 . 1M NaHCO3 pH10 , 1% SDS ) . Beads were eluted twice , 15 min each and the eluate combined in a new tube with 40 mM Tris HCl pH8 , 250 mM NaCl , 10 mM EDTA and 0 . 5 mg/ml Proteinase K ( New England Biolabs ) . Samples were incubated at 45°oC for 3 hr for proteinase digestion , then at 65°C overnight to reverse crosslinking , followed by ethanol precipitation . Input samples were also RNAse A and Proteinase K treated and purified as for IP samples . Chromatin was purified , diluted 1:10 to 1:100 for qPCR , which was performed on a StepOnePlus qPCR System ( Life Technologies , Carlsbad , CA ) using primers listed in Supplementary file 1B . Each IP was performed 3 times from independent dissections , while qPCR amplifications were repeated a minimum of 3 times on each chromatin sample . The FlyCas9 system ( Kondo and Ueda , 2013 ) was used to generate the BRV-ΔC genomic deletion . The following annealed primer pairs were cloned into pU6 . 2B vector: 5’-CTTCGAATCGCCCGCTCAGACAGT-3’ and 5’-AAACACTGTCTGAGCGGGCGATTC-3’ , and 5’-CTTCGGCTTTCTGCTATTGTTGCT-3’ and 5’-AAACAGCAACAATAGCAGAAAGCC-3’ . The plasmid was inserted into the AttP2 landing site by PhiC31 recombination , yielding a stable transgenic line expressing 2 guide RNAs that target the 5’ and 3’ ends of BRV-C . This line was crossed to a line expressing germline Cas9 , as described in the mating scheme ( Figure 7—figure supplement 3A ) , and potential deletions identified by genomic PCR screening . Discs from day 7 or day 9 larvae bearing the BRV-B reporter were dissected and wounded by cutting a fragment from the posterior ventral region with tungsten wires . After wounding , discs were cultured in either Schneider’s or Robb’s medium ( Robb , 1969 ) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin for 12 hr on a Nunc Lab-Tek II chamber slide , ( Thermo Fisher Scientific , Waltham , MA ) , before being fixed and stained . To examine BRV-B activation in the absence of JNK activity the small molecule JNK inhibitor SP600125 ( Bennett et al . , 2001 ) was dissolved in 1% DMSO to prevent precipitation in the culture medium , and added at a final concentration of 10 μM . Control discs were cultured in 1% DMSO in culture medium . Discs were visualized directly on a Zeiss Axio Imager M1 without fixation or antibody staining . Density controlled wandering stage larvae were placed on shallow yeast food and irradiated with 45 Gy in an X-ray cabinet ( Faxitron , Tucson , AZ ) , followed by recovery for 16 hr before dissection . Discs were dissected and fixed as for immunofluorescence , and RNA in situs were performed according to established methods for alkaline phosphatase based dig-labelled probe detection . Digoxigenin labelled antisense probes were generated targeting the WNT gene coding sequences using the following primer pairs: WNT4: forward 5’-AGTCGAGTGCCGAACGAGCTG-3’ and reverse 5’-TTGTAAAGGCCTTGCTGCATATCCATGT-3’ , WNT6: forward 5’-ATTCCCGAGAGACGGGTTTCGTG-3’ and reverse 5’-GCGACTATTTACAAGGCTTACATGAGC-3’ , WNT10: forward 5’-GCTACCGAGAGAGTGCGTTCGC-3’ and reverse 5’-CCTGTATATCAGCTCCCAGATCGCG-3’ , Wg: forward 5’-CAACAGTCTCCGGGCCACCAAC-3’ and reverse 5’-CGATTGCATTCGCATTTTTCTGCTCGC-3’ . Sense probes were generated using the same DNA sequences , and in situs were performed to ensure specificity . Control and experimental discs were stained simultaneously for the same duration , mounted in Permount ( Fisher Scientific , Pittsburg , PA ) and imaged on a Zeiss Axio Imager M1 . Newly eclosed adults were cultured in a vial with filter paper soaked with 5% sucrose and 5% dextran or 5% dextran sodium sulphate ( DSS , MP Biomedicals ) in water as a food source . Adults were incubated at 30°C for two days , and filter paper was replaced every 12 hr . Males were dissected and gut tissue was fixed and stained , as in ( Amcheslavsky et al . , 2009 ) . Larvae were density controlled , with 50 larvae to a yeast food vial supplemented with yeast paste , and aged at 18C . Larvae were scored for puparium formation every 12 hr from the beginning of the temperature upshift that activates ablation , and throughout the regeneration period in which the larvae were maintained at 18°C . At least 3 independent ablation experiments ( separate upshifts ) using multiple vials ( >250 larvae per repeat ) were used to generate pupariation curves , which were plotted to estimate the average delay of puparation following damage . | The ability of many animals to regenerate damaged tissues decreases as they age , for example , newborn mice can regenerate damaged heart tissue while older mice cannot . Researchers are trying to discover why older animals lose the ability to regenerate , which may help us to develop therapies that can regenerate damaged tissues in humans . Fruit flies are relatively simple animals that are often used as models in biology experiments . In young fruit fly larvae , there are tissues called imaginal discs that regenerate well after damage; however these discs lose this ability as the larva matures . A gene called wingless is very active in young larvae if the imaginal discs become damaged and helps them to regenerate . Previous studies show that this gene is not as strongly switched on in older larvae after tissue damage . However , since wingless also performs other roles in fruit flies , it is not clear how cells can stop wingless from being highly activated after tissue damage without affecting other important processes . Regions of DNA called enhancers can regulate the activity of genes . Harris et al . studied an enhancer that had previously been shown to drive the activation of wingless following tissue damage . The experiments show that there are two separate sections within the enhancer that control wingless activity . One section activates wingless in response to tissue damage and can perform this role even as the tissue matures . However , in older larvae , the other section alters the properties of the first section’s DNA to reduce its effectiveness . By switching off this section of the enhancer – but not the wingless gene itself – the activity of wingless no longer responds to tissue damage , but can still be regulated by signals that influence other processes . Harris et al . also found that several other genes that are not active in mature tissues can be re-activated by a protein called Myc . Therefore , increasing the production of Myc in cells can promote the regeneration of more mature tissues . The next step is to find out if other genes involved in fly tissue regeneration are regulated in a similar way . A future challenge will be to find out if the same mechanism also limits tissue regeneration in humans and other more complex animals as they age . | [
"Abstract",
"Introduction",
"Results",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"cell",
"biology"
] | 2016 | Localized epigenetic silencing of a damage-activated WNT enhancer limits regeneration in mature Drosophila imaginal discs |
Rabex-5 and Rabaptin-5 function together to activate Rab5 and further promote early endosomal fusion in endocytosis . The Rabex-5 GEF activity is autoinhibited by the Rabex-5 CC domain ( Rabex-5CC ) and activated by the Rabaptin-5 C2-1 domain ( Rabaptin-5C21 ) with yet unknown mechanism . We report here the crystal structures of Rabex-5 in complex with the dimeric Rabaptin-5C21 ( Rabaptin-5C212 ) and in complex with Rabaptin-5C212 and Rab5 , along with biophysical and biochemical analyses . We show that Rabex-5CC assumes an amphipathic α-helix which binds weakly to the substrate-binding site of the GEF domain , leading to weak autoinhibition of the GEF activity . Binding of Rabaptin-5C21 to Rabex-5 displaces Rabex-5CC to yield a largely exposed substrate-binding site , leading to release of the GEF activity . In the ternary complex the substrate-binding site of Rabex-5 is completely exposed to bind and activate Rab5 . Our results reveal the molecular mechanism for the regulation of the Rabex-5 GEF activity .
Endocytosis is a major process which eukaryotic cells use to absorb extracellular materials ( Doherty and McMahon , 2009; Grant and Donaldson , 2009; Huotari and Helenius , 2011 ) . In this process , small GTPase Rab5 functions as a master regulator of the early endosomal biogenesis ( Stenmark , 2009; Mizuno-Yamasaki et al . , 2012; Zeigerer et al . , 2012 ) . Rab5 is localized to early endosomal membrane via its isoprenylated C-terminus and regulates early endosomal fusion through interactions with an array of effectors including Rabaptin-5 ( Stenmark et al . , 1995 ) , Rabenosyn-5 ( Nielsen et al . , 2000 ) , EEA1 ( Mills et al . , 1998; Simonsen et al . , 1998 ) , PI3Ks ( Li et al . , 1995; Christoforidis et al . , 1999 ) , and APPLs ( Miaczynska et al . , 2004 ) . Like other small GTPases , Rab5 exists mainly in two states , the GTP-bound active state and the GDP-bound inactive state , and requires guanosine nucleotide exchange factor ( GEF ) for activation and GTPase-activating protein ( GAP ) for inactivation . Rabex-5 is a specific GEF for Rab5 , Rab17 , and Rab21 ( Horiuchi et al . , 1997; Delprato et al . , 2004; Delprato and Lambright , 2007; Mori et al . , 2013 ) . The GEF domain is located in the middle and consists of a helical bundle ( HB ) domain and a Vps9 domain ( Figure 1—figure supplement 1 ) . Besides , the N-terminal region comprises two distinct ubiquitin-binding domains , a zinc finger domain and a motif interacting with ubiquitin domain , which can interact with ubiquitinated cargoes or adaptors to recruit Rabex-5 to early endosomal membrane ( Lee et al . , 2006; Mattera et al . , 2006; Penengo et al . , 2006; Mattera and Bonifacino , 2008 ) and function as an E3 ligase for Ras ubiquitination to promote Ras endosomal localization ( Xu et al . , 2010; Yan et al . , 2010 ) . The following membrane binding motif domain and the HB domain together compose an early endosomal targeting domain that can direct Rabex-5 to early endosomal membrane ( Zhu et al . , 2007 ) . The C-terminal region consists of a coiled-coil ( CC ) domain and a proline rich region; the CC domain is involved in autoinhibition of the GEF activity and binding of Rabaptin-5 ( Lippe et al . , 2001; Mattera et al . , 2006; Delprato and Lambright , 2007 ) . Rabaptin-5 is a key effector of Rab5 and plays an important role in both homotypic and heterotypic fusions of early endosomes ( Stenmark et al . , 1995; Stenmark , 2009 ) . It is a scaffold protein consisting of primarily four coiled-coil domains , namely C1-1 , C1-2 , C2-1 , and C2-2 domains ( Figure 1—figure supplement 1 ) . The C2-1 domain is responsible for interaction with and recruitment of Rabex-5 to early endosomal membrane to activate Rab5 ( Lippe et al . , 2001; Mattera et al . , 2006; Delprato and Lambright , 2007 ) . Besides , the N-terminal region can mediate interactions with Rab4 and Rab8 ( Vitale et al . , 1998; Omori et al . , 2008 ) ; the middle region can interact with the GAE and GAT domains of GGAs that function as effectors of the Arf family small GTPases in the tethering and fusion of trans Golgi network ( TGN ) ( Mattera et al . , 2003; Miller et al . , 2003; Zhu et al . , 2004a ) ; and the C-terminal region can interact with the GTP-bound Rab5 that recruits Rabaptin-5 to early endosomal membrane ( Vitale et al . , 1998; Zhu et al . , 2004b ) . In addition , Rabex-5 and Rabaptin-5 have been shown to function as neoplastic tumor suppressors and are implicated in human cancers ( Magnusson et al . , 2001; Wang et al . , 2009; Christoforides et al . , 2012; Thomas and Strutt , 2014 ) , and Rabex-5 has also been shown to determine the neurite localization of its substrate Rab proteins and thus plays an important role in the development of hippocampal neurons ( Mori and Fukuda , 2013; Mori et al . , 2013 ) . Previous structural , biochemical , and biological data have demonstrated that Rabex-5 and Rabaptin-5 function together to activate Rab5 in endocytosis; the GEF activity of Rabex-5 could be autoinhibited by its CC domain and activated by binding of the Rabaptin-5 C2-1 domain ( Rabaptin-5C21 ) to the CC domain ( Lippe et al . , 2001; Delprato and Lambright , 2007; Zhu et al . , 2007 , 2010 ) . However , the underlying molecular mechanism is unclear . We report here the crystal structures of a Rabex-5 variant in complex with the dimeric Rabaptin-5C21 ( Rabaptin-5C212 ) and in complex with Rabaptin-5C212 and Rab5 . The structural data together with the in vitro functional data reveal the molecular mechanism for the regulation of the Rabex-5 GEF activity .
To investigate the molecular mechanism of the regulation of the Rabex-5 GEF activity , we were intent to determine the crystal structures of Rabex-5 containing the GEF and CC domains ( residues 132–455 , Rabex-5 ) alone and in complex with the Rabaptin-5 C2-1 domain ( residues 552–642 , Rabaptin-5C21 ) . We were able to obtain Rabex-5 and the Rabex-5-Rabaptin-5C21 ( R2 ) complex with high purity , stability , and homogeneity , but unfortunately failed to grow any crystals for either Rabex-5 or the R2 complex . Partial digestion of the R2 complex with trypsin shows that Rabex-5 could be proteolyzed in the linker between the GEF and CC domains ( Figure 1—figure supplement 2 ) , indicating that the linker is surface exposed with high flexibility which may prevent proper crystal packing . Thus , we constructed a series of Rabex-5 variants containing different forms of linker deletion ( Rabex-5Δ ) . Although none could be crystallized alone , one Rabex-5Δ variant ( residues 132–455Δ393–407 ) led to successful crystallization of the Rabex-5Δ-Rabaptin-5C21 ( R2Δ ) complex . The crystal structure of the Rabex-5Δ-Rabaptin-5C21 complex was determined at 3 . 10 Å resolution ( Table 1 ) , containing one Rabex-5Δ and two Rabaptin-5C21 or one Rabex-5Δ-Rabaptin-5C212 complex per asymmetric unit ( Figure 1A and Figure 1—figure supplement 3A ) . The N- and C-terminal regions of each Rabaptin-5C21 form two α-helices linked together by a short loop to assume a ‘V’ shaped conformation with an inclination angle of about 40°; and the N- and C-terminal α-helices of the two Rabaptin-5C21 dimerize with each other to form two two-helix bundles . In addition , two symmetry-related complexes further dimerize through the N-terminal α-helices of Rabaptin-5C212 ( Figure 1B ) . 10 . 7554/eLife . 02687 . 003Table 1 . Summary of diffraction data and structure refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 003Rabex-5CCRabex-5CC-Rabaptin-5C212Rabex-5Δ-Rabaptin-5C212Rab5-Rabex-5Δ-Rabaptin-5C212Diffraction data Wavelength ( Å ) 0 . 92001 . 00000 . 97930 . 9785 Space groupP21C2P3121P41212 Cell parameters a ( Å ) 46 . 890 . 087 . 2174 . 8 b ( Å ) 40 . 328 . 987 . 2174 . 8 c ( Å ) 51 . 6108 . 0168 . 9149 . 0 α ( ° ) 90 . 090 . 090 . 090 . 0 β ( ° ) 95 . 1102 . 290 . 090 . 0 γ ( ° ) 90 . 090 . 0120 . 090 . 0 Resolution ( Å ) 50 . 0–2 . 0050 . 0–2 . 2050 . 0–3 . 1050 . 0–4 . 60 ( 2 . 07–2 . 00 ) * ( 2 . 28–2 . 20 ) ( 3 . 21–3 . 10 ) ( 4 . 76–4 . 60 ) Observed reflections38 , 44547 , 48279 , 255124 , 340 Unique reflections ( I/σ ( I ) > 0 ) 12 , 74813 , 81613 , 73012 , 699 Average redundancy3 . 0 ( 3 . 0 ) 3 . 4 ( 3 . 0 ) 5 . 8 ( 6 . 0 ) 9 . 8 ( 9 . 0 ) Average I/σ ( I ) 23 . 6 ( 14 . 0 ) 21 . 2 ( 3 . 4 ) 20 . 1 ( 2 . 4 ) 18 . 2 ( 2 . 8 ) Completeness ( % ) 96 . 4 ( 97 . 7 ) 97 . 7 ( 85 . 8 ) 98 . 1 ( 100 . 0 ) 97 . 6 ( 95 . 8 ) Rmerge ( % ) †5 . 3 ( 9 . 3 ) 6 . 0 ( 28 . 0 ) 8 . 2 ( 64 . 3 ) 11 . 7 ( 94 . 3 ) Refinement and structure model Reflections ( Fo ≥ 0σ ( Fo ) ) Working set11 , 43712 , 43310 , 80611 , 982 Test set622691601631 Rwork/Rfree ( % ) ‡19 . 1/23 . 419 . 3/23 . 526 . 4/31 . 525 . 1/34 . 3 No . of atoms1726181430749679 Protein1621162730749679 Water105187–– Average B factor ( Å2 ) All atoms22 . 858 . 572 . 0187 . 3 Main-chain atoms17 . 951 . 872 . 6186 . 9 Side-chain atoms25 . 964 . 070 . 7187 . 7 Water34 . 963 . 5-- RMS deviations Bond lengths ( Å ) 0 . 0180 . 0140 . 0050 . 015 Bond angles ( ° ) 1 . 611 . 371 . 271 . 87 Ramachandran plot ( % ) Most favored99 . 599 . 592 . 193 . 8 Allowed0 . 50 . 57 . 65 . 8 Generously allowed0 . 00 . 00 . 30 . 5*Numbers in parentheses represent the highest resolution shell . †Rmerge = ∑hkl∑i|Ii ( hkl ) i−<I ( hkl ) >|/∑hkl∑iIi ( hkl ) . ‡R = ∑hkl‖Fo|−|Fc‖/∑hkl|Fo| . 10 . 7554/eLife . 02687 . 004Figure 1 . Crystal structure of the Rabex-5Δ-Rabaptin-5C212 complex . ( A ) A ribbon representation of the overall structure of the Rabex-5Δ-Rabaptin-5C212 complex . The HB and Vps9 domains of the Rabex-5 GEF domain are colored in light blue and dark blue , respectively , and the Rabex-5 CC domain ( Rabex-5CC ) in green . The two Rabaptin-5C21 are designated with superscripts A and B , and colored in orange and dark yellow , respectively . The disordered loops of the HB domain ( residues 149 , 161–162 , 174 , 190–204 , and 220–230 ) and the linker between the GEF and CC domains ( residues 369–412Δ393-407 ) are indicated with dotted lines . The autoinhibitory residues of Rabex-5CC are marked in red . ( B ) The dimeric Rabex-5Δ-Rabaptin-5C212 complex . ( C ) An electrostatic surface representation of the amphipathic α-helix of Rabex-5CC . The autoinhibitory residues are located on the N-terminal of the nonpolar surface as indicated with a yellow circle . ( D ) Interactions between Rabex-5CC and Rabaptin-5C212 . Left panel: Rabex-5CC is shown in ribbon representation and Rabaptin-5C212 in electrostatic surface representation . Right panel: Rabex-5CC is shown in electrostatic surface representation and Rabaptin-5C212 in ribbon representation . The interacting residues are shown with side chains . The detailed interactions is available in the Figure 1—source data 1 . ( E ) An electrostatic surface representation of the interactions between the GEF and CC domains of Rabex-5Δ . ( F ) A close-up view of the interactions of the GEF domain with Rabex-5CC and Rabaptin-5C21 . The interacting residues are shown with side chains and the hydrogen bonds are indicated with dotted lines . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 00410 . 7554/eLife . 02687 . 005Figure 1—source data 1 . Interactions between Rabex-5CC and Rabaptin-5C21 in the Rabex-5Δ-Rabaptin-5C212 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 00510 . 7554/eLife . 02687 . 006Figure 1—source data 2 . Interactions between Rabex-5CC and Rabaptin-5C21 in the Rabex-5CC-Rabaptin-5C212 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 00610 . 7554/eLife . 02687 . 007Figure 1—figure supplement 1 . Schematic diagrams showing the domain organizations of Rabex-5 and Rabaptin-5 . Rabex-5 consists of a zinc finger ( ZnF ) domain , a motif interacting with ubiquitin ( MIU ) domain , a membrane binding motif ( MBM ) domain , a helical bundle ( HB ) domain , a Vps9 domain , a coiled-coil ( CC ) domain , and a proline-rich region ( PR ) ( Delprato et al . , 2004; Mattera et al . , 2006; Delprato and Lambright , 2007; Zhu et al . , 2007 ) . The MBM and HB domains compose the early endosomal targeting ( EET ) domain . The HB and Vps9 domains compose the GEF domain . The CC domain contains the binding site for Rabaptin-5 as well as an autoinhibitory element for its GEF activity . Rabaptin-5 consists of four coiled-coil domains designated as C1-1 , C1-2 , C2-1 , and C2-2 ( Vitale et al . , 1998; Mattera et al . , 2003; Miller et al . , 2003; Zhu et al . , 2004a; Zhu et al . , 2004b; Mattera et al . , 2006; Omori et al . , 2008 ) . The binding regions for Rab4 , Rab5 , Rab8 , and the GAE and GAT domains of GGAs are indicated . The C-terminal part of the C2-1 domain contains the binding site for Rabex-5 . The segments of Rabex-5 and Rabaptin-5 whose structures are determined in this work are indicated with black bars . The color scheme of the domains is adopted in all the Figure in this work unless otherwise specified . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 00710 . 7554/eLife . 02687 . 008Figure 1—figure supplement 2 . Trypsin digestion of the Rabex-5-Rabaptin-5C21 complex . ( A ) Trypsin digestion of the Rabex-5-Rabaptin-5C21 complex . At high concentration of trypsin , Rabex-5 is proteolyzed to a stable fragment with a molecular mass similar to the GEF domain . ( B ) Affinity chromatography analysis and ( C ) Western blot analysis of the Rabex-5-Rabaptin-5C21 complex treated with and without trypsin . The N-terminus of Rabex-5 is attached with a His6 tag . After trypsin digestion , the remaining Rabex-5 fragment can still bind to the Ni-NTA beads and be detected by anti-His antibody , suggesting that the N-terminal part is intact and the proteolytic site lies in the linker between the GEF and CC domains . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 00810 . 7554/eLife . 02687 . 009Figure 1—figure supplement 3 . Comparison of the Rabex-5 GEF domain in different structures . ( A ) Representative simulated annealing composite omit map of the Rabex-5Δ-Rabaptin-5C212 complex . The map is contoured at 1 . 0σ with the final structure shown in stick model . ( B ) Superposition of the Rabex-5 GEF domain alone ( pink , PDB code 1TXU ) ( Delprato et al . , 2004 ) , in the Rabex-5 GEF-Rab21 complex ( green , PDB code 2OT3 ) ( Delprato and Lambright , 2007 ) , in the Rabex-5Δ-Rabaptin-5C212 complex ( blue ) , and in the Rab5-Rabex-5Δ-Rabaptin-5C212 complex ( cyan ) . The overall structure of the GEF domain in these structures is very similar with RMSD of ∼0 . 90 Å for 228 Cα atoms . Only slight conformational differences are observed in the αV1-αV2 , αV3-αV4 , and αV5-αV6 loops which are involved in the substrate binding . The regions with similar conformations are colored in gray and the regions with slightly varied conformations are marked in different colors . The invariant “aspartic acid finger” Asp313 in the αV3- αV4 loop is shown with side chain . ( C ) Superposition of the Rabex-5Δ-Rabaptin-5C212 complex and the Rabex-5 GEF-Rab21 complex . For the Rabex-5Δ-Rabaptin-5C212 complex , the GEF and CC domains of Rabex-5 are colored in blue and violet , respectively , and Rabaptin-5C212 in pink . For the Rabex-5 GEF-Rab21 complex , the Rabex-5 GEF domain is colored in light blue; switch I , interswitch , and switch II of Rab21 are colored in red , green and dark yellow , respectively , and the rest region in gray . The substrate-binding site of Rabex-5 in the Rabex-5Δ-Rabaptin-5C212 complex is partially occupied by the three-helix bundle formed by Rabex-5CC and Rabaptin-5C212 . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 00910 . 7554/eLife . 02687 . 010Figure 1—figure supplement 4 . Crystal structure of Rabex-5CC . ( A ) A ribbon representation of the overall structure of Rabex-5CC . There are four Rabex-5CC in the asymmetric unit with monomers A , B , C , and D shown in green , cyan , pink , and gray , respectively; two of them ( monomers A and B or C and D ) dimerize in anti-parallel to form a two-helix bundle and the two dimers further dimerize to form a tight four-helix bundle . Each monomer interacts via the nonpolar surface with the other three in a similar way . For example , monomer A interacts extensively with monomers B and C in anti-parallel which buries a total of solvent accessible surface area of 1880 Å2 and 1692 Å2 , respectively , and interacts with monomer D through their C-terminal regions in parallel which buries a total of solvent accessible surface area of 490 Å2 . Most of the residues responsible for the autoinhibition of the Rabex-5 GEF activity ( Delprato and Lambright , 2007 ) ( shown with side chains in red ) are buried in the interaction interfaces . ( B ) Representative simulated annealing composite omit map of Rabex-5CC . The map is contoured at 1 . 0σ with the final structure shown in stick models . ( C ) A schematic diagram showing the interactions of monomer A with monomers B , C , and D . The interacting residues are colored the same as in ( A ) . The hydrophilic interactions are indicated with solid lines and the hydrophobic interactions with dotted lines . The autoinhibitory residues are marked with red asterisks . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01010 . 7554/eLife . 02687 . 011Figure 1—figure supplement 5 . Crystal structure of the Rabex-5CC-Rabaptin-5C212 complex . ( A ) A ribbon representation of the overall structure of the Rabex-5CC-Rabaptin-5C212 complex . Rabex-5CC is colored in green and the two Rabaptin-5C21 in orange and dark yellow , respectively . The autoinhibitory residues are colored in red . In this complex , Rabex-5CC forms a long α-helix with a length of 52 Å which is slightly shorter than that in the structures of Rabex-5CC ( 65 Å ) and the Rabex-5Δ-Rabaptin-5C212 complex ( 60 Å ) due to disordering of several N-terminal residues . Similar to the Rabex-5Δ-Rabaptin-5C212 complex , Rabex-5CC packs in parallel with the C-terminal regions of Rabaptin-5C212 to form a tight three-helix bundle with its nonpolar surface buried in a hydrophobic surface groove of Rabaptin-5C212 and a total of buried solvent accessible surface area of 2612 Å2 . ( B ) Representative simulated annealing composite omit map of the Rabex-5CC-Rabaptin-5C212 complex . The map is contoured at 1 . 0σ with the final structure shown in stick model . ( C ) The dimeric Rabex-5CC-Rabaptin-5C212 complex . Two Rabex-5CC-Rabaptin-5C212 complexes related by a two-fold crystallographic symmetry form a dimer through the N-terminal regions ( residues 552–592 ) of Rabaptin-5C212 . ( D ) Interactions between Rabex-5CC and Rabaptin-5C212 . Left panel: Rabex-5CC is shown in ribbon representation and Rabaptin-5C212 in electrostatic surface representation . Right panel: Rabex-5CC is shown in electrostatic surface representation and Rabaptin-5C212 in ribbon representation . The interacting residues are shown with side chains . The autoinhibitory residues of Rabex-5CC are involved in the interactions with Rabaptin-5C212 . The interactions between Rabex-5CC and Rabaptin-5C212 are essentially the same as those in the Rabex-5Δ-Rabaptin-5C212 complex . ( E ) A schematic diagram showing the interactions of Rabex-5CC with Rabaptin-5C212 . Rabex-5CC is shown with a ball model with the residues participating in the interactions labeled . Residues of Rabaptin-5C21A and Rabaptin-5C21B are shown on the left side and the right side , respectively , with those involved in the hydrophobic interactions colored in black and the hydrophilic interactions in blue . The detailed interactions are available in Figure 1—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 011 Rabex-5CC ( residues 413–452 ) forms a long amphipathic α-helix ( about 60 Å ) ( Figure 1A–D ) that is in agreement with the prediction by Delprato and Lambright ( 2007 ) . It packs in parallel with the C-terminal α-helices of Rabaptin-5C212 to form a tight three-helix bundle with its nonpolar surface buried in a hydrophobic surface groove of Rabaptin-5C212 . The interactions are dominantly hydrophobic that bury a total of solvent accessible surface area of 2664 Å2 . The residues that were suggested to be involved in the autoinhibition of the GEF activity , including Asn413 , Leu414 , Leu417 , Leu420 , Arg423 , and Ile427 ( Delprato and Lambright , 2007 ) , are located in the N-terminal half of the nonpolar surface , and several of them ( Leu420 , Arg423 , and Ile427 ) are involved in the interactions with Rabaptin-5C212 and buried in the interaction interface ( Figure 1C , D ) . The HB domain ( residues 132–229 ) and Vps9 domain ( residues 230–368 ) of the Rabex-5 GEF domain are well defined except a few surface exposed loops ( Figure 1A ) . The C-terminal helix αC and the following linker ( residues 369–412 with the deleted residues 393–407 ) are also disordered , consistent with our trypsin digestion results ( Figure 1—figure supplement 2 ) . The distance between the visible C-terminal end of the GEF domain and N-terminal end of the CC domain is about 10 Å , which is large enough to accommodate the disordered 29 residues with a loop conformation , suggesting that the positions and conformations of the GEF and CC domains are unlikely constrained by the shortened linker . The overall structure of the GEF domain is very similar to that in the free form ( Delprato et al . , 2004 ) and in complex with Rab21 ( Delprato and Lambright , 2007 ) ( RMSD of ∼0 . 90 Å for 228 Cα atoms ) ( Figure 1—figure supplement 3B ) . At the substrate-binding site , there is a surface groove composed of largely nonpolar residues , which exhibits good chemical and geometrical complementarities with the nonpolar surface of the amphipathic helix of Rabex-5CC ( Figure 1E ) . The GEF domain packs along the three-helix bundle formed by Rabex-5CC and Rabaptin-5C212 ( Figure 1A ) . The interactions involve a small portion of the substrate-binding site , a small portion of the N-terminal region of Rabex-5CC adjacent to the nonpolar surface , and a small portion of the N-terminal region of one Rabaptin-5C21 C-terminal α-helix ( Figure 1F ) . The interaction interface buries a total of solvent accessible surface area of 1040 Å2 , which is much smaller than that between Rabex-5 and Rab21 ( 2400 Å2 ) ( Delprato and Lambright , 2007 ) . Rab21 uses switch I , switch II , and the interswitch region to interact with the substrate-binding site of Rabex-5 ( Delprato and Lambright , 2007 ) . Although the binding sites for switch II and a small portion of the interswitch region of Rab21 are occupied by the three-helix bundle , the binding sites for switch I and a large portion of the interswitch region of Rab21 are exposed to the solvent ( Figure 1—figure supplement 3C ) . Hence , we consider that the substrate-binding site of Rabex-5 is largely exposed to the solvent and partially accessible by the substrate . To explore the conservations of the conformations of Rabex-5CC and Rabaptin-5C21 and the interactions between Rabex-5CC and Rabaptin-5C21 , we also determined the crystal structures of Rabex-5CC alone at 2 . 00 Å resolution and in complex with Rabaptin-5C212 at 2 . 20 Å resolution ( Table 1 ) . In the Rabex-5CC structure , Rabex-5CC also forms a long α-helix ( about 65 Å ) and the four Rabex-5CC in the asymmetric unit form a tight four-helix bundle via the nonpolar surface ( Figure 1—figure supplement 4 ) . In the structure of the Rabex-5CC-Rabaptin-5C212 complex , the asymmetric unit contains one complex ( Figure 1—figure supplement 5 ) . Notably , each Rabaptin-5C21 forms a long α-helix ( about 125 Å ) and two Rabaptin-5C21 dimerize in parallel to form a twisted linear two-helix bundle , which is different from the V-shaped conformation in the R2Δ complex . Otherwise , two symmetry-related complexes also dimerize through the N-terminal regions of Rabaptin-5C212 . Rabex-5CC also forms a long α-helix ( 52 Å ) and packs in parallel with the C-terminal regions of Rabaptin-5C212 to form a tight three-helix bundle . The interactions between Rabex-5CC and Rabaptin-5C212 are essentially the same as those in the R2Δ complex . These results confirm that Rabex-5CC assumes a stable amphipathic α-helix and tends to bury its nonpolar surface via interactions with other proteins; Rabaptin-5C21 may adopt two different conformations and forms a stable homodimer; and the interactions between Rabex-5CC and Rabaptin-5C212 are conserved in different complexes . In the structure of the R2Δ complex , the substrate-binding site of Rabex-5 is partially occupied by Rabex-5CC and Rabaptin-5C212 , implying that further conformational change ( s ) of the R2Δ complex will be required to bind Rab5 . To investigate the molecular mechanism for Rab5 activation by the R2 complex , we solved the crystal structure of the Rab5-Rabex-5Δ-Rabaptin-5C21 ( R3Δ ) complex to 4 . 60 Å resolution ( Table 1; Figure 2A , and Figure 2—figure supplement 1 ) . The asymmetric unit contains two Rab5-Rabex-5Δ-Rabaptin-5C212 complexes related by a twofold non-crystallographic symmetry . The four Rabaptin-5C21 and two Rabex-5 are well defined in the electron density map; however , only one Rab5 is fairly defined while the other is poorly defined , indicating that the bound Rab5 has high flexibility which may explain the poor diffraction quality of the crystal . No nucleotide and/or metal ion are found at the active site of Rab5 and thus the bound Rab5 is in nucleotide-free form . 10 . 7554/eLife . 02687 . 012Figure 2 . Crystal structure of the Rab5-Rabex-5Δ-Rabaptin-5C212 complex . ( A ) A ribbon representation of the overall structure of the Rab5-Rabex-5Δ-Rabaptin-5C212 complex . Only one Rab5-Rabex5Δ-Rabaptin5C212 complex in the asymmetric unit is shown with Rab5 in purple and Rabex-5 and Rabaptin-5C21 in the same colors as in Figure 1A . ( B ) Comparison of the Rab5-Rabex-5Δ-Rabaptin-5C212 complex and the Rabex-5Δ-Rabaptin-5C212 complex based on superposition of the three-helix bundle formed by Rabex-5CC and the C-terminal regions of Rabaptin-5C212 . ( C ) Interactions between the Rabex-5 GEF domain and Rabaptin-5C212 . The GEF domain is shown in ribbon representation in blue and the interacting residues are shown with side chains . Rabaptin-5C212 is shown in surface representation with the interacting residues colored in pink . ( D ) Interactions between Rab5 and the Rabex-5 GEF domain . Rab5 is shown in coil representation with the P-loop , switch I , switch II , and interswitch region colored in purple , orange , blue , and dark green , respectively . Several key residues are shown with side chains . The GEF domain is shown in surface representation with Asp313 colored in yellow . For comparison , Rab21 in its complex with the Rabex-5 GEF domain ( Delprato and Lambright , 2007 ) is shown in coil representation in light blue . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01210 . 7554/eLife . 02687 . 013Figure 2—figure supplement 1 . Crystal structure of the Rab5-Rabex-5Δ-Rabaptin-5C212 complex . ( A ) A ribbon representation of the overall structure of the dimeric Rab5-Rabex-5Δ-Rabaptin-5C212 complex in an asymmetric unit . There are two Rab5-Rabex5-Rabaptin5C212 complexes related by a two-fold non-crystallographic symmetry in the asymmetric unit . Rab5 , Rabex-5Δ , and Rabaptin-5C21 are colored the same as in Figure 2A . Rabaptin-5C21 and Rabex-5 are well defined in the electron density map; however , only one Rab5 is fairly defined while the other is poorly defined . ( B ) Representative simulated annealing composite omit map of the Rab5-Rabex-5Δ-Rabaptin-5C212 complex . The map is contoured at 1 . 0σ with the final structure shown in ribbon model . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01310 . 7554/eLife . 02687 . 014Figure 2—figure supplement 2 . Superposition of Rabaptin-5C212 in different complexes . Rabex-5CC-Rabaptin-5C212: yellow , Rabex-5Δ-Rabaptin-5C212: blue , Rab5-Rabex-5Δ-Rabaptin-5C212: cyan , and GAT-Rabaptin-5C212 ( PDB code 1X79 ) ( Zhu et al . , 2004a ) : light orange . The N-terminal regions of Rabaptin-5C212 in these complexes can be superimposed well ( RMSD of ∼1 . 0 Å for 31 Cα atoms ) ( left panel ) ; however , the C-terminal regions cannot ( RMSD of ∼3 . 6 Å for 42 Cα atoms ) ( right panel ) . The angle between the two regions also differs substantially which is about 40° in the Rabex-5Δ-Rabaptin-5C212 complex , and about 180° in the Rabex-5CC-Rabaptin-5C212 complex , the Rab5-Rabex-5Δ-Rabaptin-5C212 complex , and the GAT-Rabaptin-5C212 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 014 In the R3Δ complex , each Rabaptin-5C21 forms a long α-helix; two of them form a twisted linear two-helix bundle; and the two Rabaptin-5C212 dimerize through the middle regions ( residues 590–600 ) ( Figure 2A and Figure 2—figure supplement 1A ) . The conformation of Rabaptin-5C212 is similar to that in the Rabex-5CC-Rabaptin-5C212 complex and the previously reported GAT-Rabaptin-5C212 complex ( Zhu et al . , 2004a ) but significantly different from that in the R2Δ complex ( Figure 2—figure supplement 2 ) . Similar to that in the R2Δ complex , Rabex5CC also forms a long α-helix and interacts with the C-terminal regions of Rabaptin-5C212 to form a three-helix bundle ( Figure 2A ) . However , the orientations and positions of the Rabex-5 GEF domain and the N-terminal regions of Rabaptin-5C212 in relation to the three-helix bundle are dramatically different ( Figure 2B ) . When the two complexes are superimposed based on the three-helix bundle , the N-terminal regions of Rabaptin-5C212 rotates downwards by about 120° to transform from the V-shaped conformation to the linear conformation . Meanwhile , the GEF domain rotates by about 270° along the vertical axis and about 100° along the horizontal axis , and is dislodged from the three-helix bundle without any interaction . As a result , the substrate-binding site is completely exposed to the solvent for Rab5 binding . The GEF domain of Rabex-5 interacts via a small portion of the opposite side of the substrate-binding site with a small portion of the N-terminal regions of Rabaptin-5C212 ( Figure 2A ) . The Rabex-5-Rabaptin-5C21 interaction involves only Glu232 , Leu236 , Gln239 , Arg243 , and Arg246 of αV1 and Val269 and Ser280 of αV2 of the GEF domain , and Asn568 , Glu572 , Gln579 , and Glu582 of one Rabaptin-5C21 and Ile608 and Asp623 of the other , and the interaction interface buries a total of solvent accessible surface area of 1700 Å2 ( Figure 2C ) . The overall structure of the nucleotide-free Rab5 differs from the nucleotide-bound Rab5 in the conformations of the P-loop and the switch regions ( Merithew et al . , 2001; Zhu et al . , 2004b ) . Nevertheless , the interactions between Rabex-5 and Rab5 are similar to those between Rabex-5 and Rab21 , suggesting that Rabex-5 may activate Rab5 via a similar mechanism as for Rab21 ( Figure 2D ) ( Delprato and Lambright , 2007; Langemeyer et al . , 2014 ) . In the R2Δ complex , Rabaptin-5C212 assumes a V-shaped conformation and the substrate-binding site of the Rabex-5 GEF domain is partially occupied by Rabex-5CC and Rabaptin-5C212 ( Figure 1 , Figure 1—figure supplement 3C ) . However , in the R3Δ complex , Rabaptin-5C212 adopts a linear conformation and the Rabex-5 GEF domain is displaced with a completely exposed substrate-binding site ( Figure 2A , B ) . A modeling study shows that if Rabaptin-5C212 assumes the linear conformation in the R2Δ complex , the middle regions of Rabaptin-5C212 would have steric conflict with part of the Rabex-5 GEF domain ( αV1 , αV6 , and αC helices ) , suggesting that the conformational change of Rabaptin-5C212 is essential for the complete exposure of the substrate-binding site of the GEF domain to bind Rab5 . To investigate which conformation the R2 complex may assume in solution , we performed small angle X-ray scattering ( SAXS ) analyses of the R2 , R2Δ , R3 , and R3Δ complexes . Our SAXS data show that the experimental P ( r ) distributions for R2Δ and R3Δ are similar to these of R2 and R3 , respectively ( Figure 3—figure supplement 1A , B ) , suggesting that deletion of the linker does not significantly affect the structures of R2Δ and R3Δ . The maximum paired-distance of the particle ( Dmax ) , the radius of gyration ( Rg ) , and the Porod volume derived from the SAXS data together show that all of these complexes exist as dimers in solution , consistent with our biochemical and structural data ( Figure 3—source data 1 ) . The theoretical scattering curve calculated from the R2Δ model with the V-shaped conformation fits better with the experimental data for both R2 and R2Δ complexes ( goodness of fit χ = 0 . 71 and 0 . 60 , respectively ) than that calculated from the R2Δ model with the linear conformation ( goodness of fit χ = 0 . 74 and 0 . 65 , respectively ) ( Figure 3A , B ) . In addition , although the theoretical and experimental P ( r ) distributions exhibit some differences , the theoretical P ( r ) distribution of the R2Δ model with the V-shaped conformation agrees better with the experimental P ( r ) distributions of both R2 and R2Δ complexes than that with the linear conformation ( Figure 3—figure supplement 1A , C ) . As such , the solution structures of the R2 and R2Δ complexes can be best described as assuming mainly the V-shaped conformation . Nevertheless , it is plausible that the two segments of the V-shaped conformation and/or the component proteins of the complexes may bear some flexibility and adopt alternative conformation ( s ) in solution . Based on these results , we conclude that the R2 and R2Δ complexes assume mainly the V-shaped conformation as observed in the R2Δ structure with some flexibility in solution . 10 . 7554/eLife . 02687 . 015Figure 3 . SAXS analyses of the R2 and R3 complexes . ( A and B ) Comparison of the experimental data with the theoretical scattering curves calculated from the structure models of R2Δ with the V-shaped conformation observed in the R2Δ structure and the linear conformation observed in the R3Δ structure for the R2 complex ( A ) and the R2Δ complex ( B ) . ( C and D ) Comparison of the experimental data with the theoretical scattering curve calculated from the structure model of R3Δ observed in the R3Δ structure for the R3 complex ( C ) and the R3Δ complex ( D ) . The observed and calculated values of Rg , Dmax , and Porod volume are summarized in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01510 . 7554/eLife . 02687 . 016Figure 3—source data 1 . SAXS analysis parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01610 . 7554/eLife . 02687 . 017Figure 3—figure supplement 1 . SAXS analyses of the R2 and R3 complexes . ( A-B ) SAXS experimental data derived P ( r ) ( pair-distance ) distributions of the R2 and R2Δ complexes ( A ) and the R3 and R3Δ complexes ( B ) . ( C-D ) Theoretical P ( r ) distributions calculated from the structure models of the R2Δ complex with the V-shaped conformation and the linear conformation ( C ) and the R3Δ complex ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 017 Similarly , the theoretical scattering curve calculated from the R3Δ model fits well with the experimental SAXS data for both R3 and R3Δ complexes ( goodness of fit χ = 0 . 68 and 0 . 96 , respectively ) ( Figure 3C , D ) , and the theoretical P ( r ) distribution of the R3Δ model is also in good agreement with the experimental P ( r ) distributions of both R3 and R3Δ complexes ( Figure 3—figure supplement 1B , D ) . These results indicate that the R3 and R3Δ complexes assume mainly the linear conformation as observed in the R3Δ structure in solution . To investigate the biological relevance of the R2Δ and R3Δ structures , we performed in vitro functional assays . We first measured the in vitro GEF activity of different Rabex-5 variants and mutants . Our kinetic data show that Rabex-5 containing the GEF and CC domains possesses a basal GEF activity ( 0 . 93 ± 0 . 03 × 104 M−1·s−1 ) ; the Rabex-5 GEF domain alone exhibits a 3 . 2-fold higher activity ( 2 . 93 ± 0 . 06 × 104 M−1·s−1 ) ; and the R2 complex exhibits a 3 . 3-fold higher activity ( 3 . 07 ± 0 . 08 × 104 M−1·s−1 ) ( Figure 4A , B ) . Moreover , the Rabex-5 mutants containing mutations on the nonpolar surface of Rabex-5CC exhibit relatively higher GEF activity ( 1 . 4–1 . 9 folds ) compared with the wild-type Rabex-5 ( Figure 4—figure supplement 1A ) . These results indicate that the GEF domain itself is constitutively active; the CC domain slightly autoinhibits the GEF activity; and the binding of Rabaptin-5C21 to Rabex-5CC relieves the autoinhibition , which are largely in agreement with the previous biochemical data ( Lippe et al . , 2001; Delprato et al . , 2004; Zhu et al . , 2007; Langemeyer et al . , 2014 ) . However , the magnitude of the autoinhibition by Rabex-5CC is smaller than that reported by Delprato and Lambright ( 2007 ) , which may be caused by differences of the assay systems , for examples , different Rab5 and Rabex-5 constructs , different concentration of the proteins , and different sensitivity of the instruments . 10 . 7554/eLife . 02687 . 018Figure 4 . In vitro functional analyses of the Rabex-5-Rabaptin-5 complex . ( A ) GEF activity of Rabex-5 in different forms . Catalytic efficiency ( kcat/Km ) was obtained from the slope of a linear least-squares-fit of the kobs values against the concentrations of Rabex-5 from two independent measurements . ( B ) Histogram of the catalytic efficiencies of Rabex-5 variants alone and in complexes with different Rabaptin-5C21 mutants or variant . Values are means ± SEM of two independent measurements . R2M1: the R2 complex in which Rabaptin-5C21 contains a quadruple mutation N568A/E572A/Q579A/E582A; R2M2: the R2 complex in which Rabaptin-5C21 contains a double mutation I608A/D623A; R2ΔN: the R2 complex in which the N-terminal half of Rabaptin-5C21 ( residues 552–592 ) is deleted . The complexes in ( A ) and ( B ) were co-expressed and co-purified . ( C ) GST pull-down assays for the interactions between the wild-type and mutant GST-Rabex-5 and the wild-type His6-Rabaptin-5C21 . The gel was stained by Coomassie blue . ( D ) GST pull-down assays for the interactions between the wild-type GST-Rabex-5 and the wild-type and mutant His6-Rabaptin-5C21 . ( E ) Histogram of the catalytic efficiencies of the wild-type and mutant Rabex-5 alone and in complexes with the wild-type Rabaptin-5C21 . ( F ) Histogram of the catalytic efficiencies of the wild-type Rabex-5 alone and in complexes with the wild-type and mutant Rabaptin-5C21 . For the assays in ( E ) and ( F ) , Rabex-5 and Rabaptin-5C21 were expressed and purified separately and then mixed together in a 1:2 molar ratio overnight prior to the assay . Tables of the GEF activities are available in the Figure 4—source data 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01810 . 7554/eLife . 02687 . 019Figure 4—source data 1 . GEF activity of different Rabex-5 variants alone and in complexes with different Rabaptin-5C21 mutants or truncates . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 01910 . 7554/eLife . 02687 . 020Figure 4—source data 2 . GEF activity of different Rabex-5 mutants alone and in complexes with wild-type Rabaptin-5C21 . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 02010 . 7554/eLife . 02687 . 021Figure 4—source data 3 . GEF activity of wild-type Rabex-5 in complexes with different Rabaptin-5C21 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 02110 . 7554/eLife . 02687 . 022Figure 4—figure supplement 1 . Functional roles of the Rabex-5 CC domain ( Rabex-5CC ) in the autoinhibition of the Rabex-5 GEF activity . ( A ) Histogram of the catalytic efficiencies of the wild-type and mutant Rabex-5 . The residues contributed to the autoinhibition on the nonpolar surface of Rabex-5CC ( Delprato and Lambright , 2007 ) were individually mutated to alanine and the GEF activities of these mutants were detected to confirm their functional roles in the autoinhibition . Values are means ± SEM of two independent measurements . ( B ) GST pull-down assay of Rabex-5 with Rabaptin-5C21 . Rabex-5Δ: deletion of the linker region , Rabex-5ΔN and Rabex-5ΔC: deletion of the N- and C-terminal part of the CC domain , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 022 In the R2Δ complex , the linker between the Rabex-5 GEF and CC domains was removed to facilitate the crystallization . We then investigated whether the linker deletion has any effects on the functions of Rabex-5 and the R2 complex . Our kinetic data show that Rabex-5Δ possesses a slightly higher activity ( 1 . 9-fold ) than Rabex-5 and the R2Δ complex exhibits a similar activity ( 0 . 9-fold ) as the R2 complex ( Figure 4A , B ) . Meanwhile , our in vitro GST pull-down assay results show that Rabex-5Δ can bind tightly to Rabaptin-5C21; however , deletion of either the N- or C-terminal half of Rabex-5CC disrupts the interaction ( Figure 4—figure supplement 1B ) . These results indicate that the linker plays a minor role in the autoinhibition of the GEF activity but is not involved in the Rabex-5-Rabaptin-5C21 interaction , and the linker deletion does not affect the function of the R2 complex in the Rab5 activation . In the R2Δ and Rabex-5CC-Rabaptin-5C212 complexes , the interactions between Rabex-5CC and Rabaptin-5C21 are well conserved . To validate the biological relevance of these interactions , we mutated several key residues of both Rabex-5CC and Rabaptin-5C21 at the interaction interface and analyzed their effects on the Rabex-5-Rabaptin-5 interaction . Our in vitro GST pull-down assay results show that mutations L434D , L438D , and W441A of Rabex-5CC , and mutations L599D , L610D , L613D , and L617D of Rabaptin-5C21 abolish the interaction , and mutations L420D and I427D of Rabex-5CC and mutation V624D of Rabaptin-5C21 substantially impair the interaction . In contrast , mutations R423E of Rabex-5CC and E607K of Rabaptin-5C21 have no significant effect on the interaction as these two residues form a salt bridge on the solvent-exposed surface and thus their mutations do not affect the hydrophobic core of the interaction . As negative controls , mutations I439D of Rabex-5CC and I608D of Rabaptin-5C21 have no effect on the interaction as these two residues are located on the polar surface of the three-helix bundle and are not involved in the interaction ( Figure 4C , D ) . In agreement with the GST pull-down results , our kinetic data show that the GEF activity of the L434D , L438D , and W441A Rabex-5 mutants cannot be activated by Rabaptin-5C21 , whereas that of the I439D Rabex-5 mutant can be potentiated by Rabaptin-5C21 ( Figure 4E ) . Similarly , the L610D , L613D , and L617D Rabaptin-5C21 mutants cannot relieve the autoinhibition of Rabex-5; whereas the I608D Rabaptin-5C21 mutant can activate the GEF activity of Rabex-5 ( Figure 4F ) . It is noteworthy that due to partial aggregation of Rabaptin-5C21 , the GEF activity of the mixed R2 complex is only 1 . 6-fold higher compared with Rabex-5 which is weaker than the co-expressed and co-purified R2 complex ( 3 . 3-fold ) . In the R3Δ complex , the Rabex-5 GEF domain interacts via a small surface opposite the substrate-binding site with a small portion of the N-terminal regions of Rabaptin-5C212 ( Figure 2C ) . To explore the functional role of this interaction , we constructed two Rabaptin-5C21 mutants containing a quadruple mutation ( N568A/E572A/Q579A/E582A ) and a double mutation ( I608A/D623A ) of the residues on the interaction interface and detected their effects on the GEF activity of the R2 complex . Our kinetic data show that the GEF activity of these mutant complexes are unaffected ( Figure 4B ) , indicating that this interaction is not essential for the activation of the Rabex-5 GEF activity . Intriguingly , when the N-terminal half of Rabaptin-5C21 ( residues 552–592 ) was removed , the GEF activity of Rabex-5 could not be activated ( Figure 4B ) , indicating that the full-length Rabaptin-5C21 is required for the function of the R2 complex in the Rab5 activation . Taken together , these data indicate that the structures of the R2Δ and R3Δ complexes are functionally relevant , and the interaction between Rabex-5 and Rabaptin-5 is important for the activation of the Rabex-5 GEF activity in Rab5 activation , which is in accord with the previous biochemical data ( Lippe et al . , 2001 ) .
The previous biochemical and biological data showed that Rabex-5 functions together with Rabaptin-5 to activate Rab5 and then to promote the fusion of early endosomes in endocytosis ( Lippe et al . , 2001; Delprato and Lambright , 2007; Zhu et al . , 2007 ) . The GEF activity of Rabex-5 is autoinhibited by its CC domain and is activated by the binding of Rabaptin-5 via its C2-1 domain . However , the molecular mechanism is unknown . In this work , we determined the crystal structures of Rabex-5Δ in complex with Rabaptin-5C21 and in complex with Rabaptin-5C21 and Rab5 , which are validated by biophysical and biochemical analyses . Our structural data show that at the substrate-binding site of the Rabex-5 GEF domain , there is a surface groove composed of largely nonpolar residues . Rabex-5CC forms a stable amphipathic α-helix that tends to bury its nonpolar surface via oligomerization or interaction with the C-terminal regions of Rabaptin-5C212 . The nonpolar surface of Rabex-5CC has good chemical and geometrical complementarities with the nonpolar surface groove of the GEF domain and thus might be able to bind there to block the substrate binding and hence autoinhibit the GEF activity as proposed by Delprato and Lambright ( 2007 ) . Nonetheless , our structural and biochemical data show that although Rabex-5CC alone exists as a stable helix in both solution and crystal structure , it cannot form a stable complex with the GEF domain as shown by both GST pull-down assay and ITC analysis ( data not shown ) . In addition , Rabex-5 and Rabex-5Δ cannot be crystallized alone and Rabex-5 can be easily proteolyzed in the linker region ( Figure 1—figure supplement 2 ) . These results suggest that the CC domain and the linker have high flexibility . Moreover , Rabex-5 itself has a basal GEF activity which is slightly weaker ( about 1/3 ) than that of the constitutively active GEF domain ( Figure 4A , B ) but is not so weak compared with some other GEFs including DSS4 ( a GEF for Ypt1p ) ( Esters et al . , 2001 ) and MSS4 ( a GEF for Rab8 ) ( Zhu et al . , 2001; Itzen et al . , 2006 ) . Mutations of the residues on the nonpolar surface of Rabex-5CC can enhance the GEF activity by 1 . 4–1 . 9 folds ( Figure 4—figure supplement 1A ) . These results together indicate that the binding of Rabex5CC to the GEF domain is not tight , and Rabex-5 alone is not completely autoinhibited . On the other hand , in the structure of the R2Δ complex , Rabaptin-5C212 forms 2 two-helix bundles with a V-shaped conformation , and Rabex-5CC interacts via the nonpolar surface with the C-terminal regions of Rabaptin-5C212 to form a tight three-helix bundle . Meanwhile , the GEF domain folds along the three-helix bundle with a partially occupied substrate-binding site , suggesting that further conformational change is required to completely expose the substrate-binding site for Rab5 binding and activation . Indeed , in the structure of the R3Δ complex , although Rabex-5CC still forms a tight three-helix bundle with the C-terminal regions of Rabaptin-5C212 , Rabaptin-5C212 forms a linear two-helix bundle which is different from the V-shaped conformation but similar to that in the Rabex-5CC-Rabaptin-5C212 complex and the GAT-Rabaptin-5C212 complex ( Zhu et al . , 2004a ) . The GEF domain is dislodged from the three-helix bundle and interacts with the N-terminal regions of Rabaptin-5C212 , and the substrate-binding site is completely exposed to bind Rab5 . Meanwhile , our SAXS analysis results indicate that the R2 and R2Δ complexes mainly assume the V-shaped conformation as observed in the R2Δ structure but have some flexibility in solution , and our biochemical data show that the R2 and R2Δ complexes have full GEF activities as the constitutively active GEF domain and the N-terminal regions of Rabaptin-5C212 are required for the function of the R2 complex in the activation of Rab5 . Based on the structural and biological data in this work and those reported previously , we can propose the molecular mechanism for the regulation of the Rabex-5 GEF activity despite the lack of an intact Rabex-5 structure ( Figure 5 ) . In the free-form Rabex-5 , Rabex-5CC may bind weakly via the nonpolar surface to the substrate-binding site of the GEF domain , leading to the blockage of the substrate-binding site and thus a weak autoinhibition of the GEF activity ( Figure 5 , State I ) . As the binding of Rabex-5CC to the GEF domain is not tight , it might assume alternative conformations . The linker between the GEF and CC domains might help to modulate the conformational flexibility and/or the relative conformations of the two domains and thus plays a minor role in the autoinhibition . One plausible alternative conformation of Rabex-5CC might be similar to that observed in the R2Δ structure with a largely exposed substrate-binding site of the GEF domain . As the interaction of Rabex-5CC with the GEF domain via the nonpolar surface is much tighter than that via the adjacent surface , Rabex-5CC might exist mainly in the autoinhibitory conformational state and partially in the alternative conformational state . This may explain why the free-form Rabex5 exhibits some basal GEF activity . Rabaptin-5C21 might bind to the C-terminal region of the nonpolar surface of Rabex-5CC in the autoinhibitory conformational state and induce conformational change of the α-helix to transform into the conformational state as observed in the R2Δ structure , or directly to the exposed nonpolar surface of Rabex-5CC in the alternative conformational state , leading to the relief of the autoinhibition and the release of the GEF activity ( Figure 5 , State II ) . 10 . 7554/eLife . 02687 . 023Figure 5 . Molecular mechanism of the regulation of the Rabex-5 GEF activity . In the free-form Rabex-5 , the CC domain binds weakly via the nonpolar surface to the substrate-binding site of the GEF domain , leading to occlusion of the substrate-binding site and thus a weak autoinhibition of the GEF activity ( State I ) . The free-form Rabex-5 can directly target to the early endosomes to activate Rab5 at the basal level . In the cells , most Rabex-5 forms a binary complex with Rabaptin-5 which can be recruited to early endosomes via the binding of the C-terminal region of Rabaptin-5 to GTP-bound Rab5 . The binding of Rabaptin-5C21 to Rabex-5 pulls Rabex-5CC away from the GEF domain to form a binary complex with the V-shaped conformation and a largely exposed substrate-binding site , leading to the relief of the autoinhibition ( State II ) . The binding of Rab5 induces further conformational changes of the Rabex-5-Rabaptin-5 complex such that Rabaptin-5C212 transforms from the V-shaped to the linear conformation , and the substrate-binding site of the GEF domain is completely exposed to the solvent to bind and activate Rab5 ( State III ) . The positive feedback loop among Rab5 , its effector Rabaptin-5 , and its GEF Rabex-5 can lead to a robust activation of Rab5 , which then promotes the fusion of early endosomes efficiently . DOI: http://dx . doi . org/10 . 7554/eLife . 02687 . 023 When the Rabex-5-Rabaptin-5 complex is recruited to the early endosomal membrane via the interaction of the C-terminal region of Rabaptin-5 with the GTP-bound Rab5 , Rabex-5 can activate Rab5 locally in a very efficient way . The binding of the GDP-bound Rab5 to Rabex-5 induces further conformational changes of the Rabex-5-Rabaptin-5 complex such that Rabaptin-5C212 transforms from the V-shaped to the linear conformation , the GEF domain is dislodged from Rabex-5CC and the C-terminal regions of Rabaptin-5C212 , and the substrate-binding site is completely exposed to the solvent to bind Rab5 as observed in the R3Δ structure ( Figure 5 , State III ) . In this conformational state , the Rabex-5-Rabaptin-5 complex can facilitate the exchange of GDP- to GTP-bound Rab5 . The not-so-tight interaction of the Rabex-5 GEF domain with Rabex-5CC and Rabaptin-5C212 as observed in the R2Δ structure allows the conformational changes easily during the substrate binding . The conformational changes of the Rabex-5-Rabaptin-5 complex induced by the binding of Rab5 leading to the full activation of the Rabex-5 GEF activity may provide another leverage to ensure the high substrate specificity . In the context of the early endosomal membrane , Rabex-5 can directly target to the early endosomes either through the interaction of the N-terminal ubiquitin binding domain ( UBD ) with the ubiquitinated cargoes or through the early endosomal targeting domain ( EET ) ( Zhu et al . , 2007; Mattera and Bonifacino , 2008 ) . In this case , as the GEF activity of Rabex-5 is autoinhibited by Rabex-5CC , Rabex-5 can only activate Rab5 at the basal level . In the cells , most Rabex-5 forms a stable complex with Rabaptin-5 which can be recruited to early endosomes via the binding of Rabaptin-5 through the C-terminal region to GTP-bound Rab5 ( Lippe et al . , 2001; Zhu et al . , 2010 ) . In this case , Rabaptin-5 not only assists the recruitment of Rabex-5 to the early endosomal membrane , but also activates the GEF activity of Rabex-5 . The positive feedback loop among Rab5 , its effector Rabaptin-5 , and its GEF Rabex-5 can lead to a robust activation of Rab5 , which then promotes the fusion of early endosomes efficiently . As a scaffold protein , Rabaptin-5 comprises several coiled-coil regions which can mediate interactions with different proteins to exert different functions . In addition to acting as the effector of Rab5 to function in the fusion of early endosomes , Rabaptin-5 can interact with Rab4 via the N-terminal region and thus may serve as the effector of Rab4 to function in the endocytic recyling process ( Vitale et al . , 1998 ) . As the C2-1 domain of Rabaptin-5 can interact with the CC domain of Rabex5 and the GAT domain of GGA1 , and Rabaptin-5 is the effector of Rab5 and GGA1 is the effector of Arf1 , this dual interaction might mediate the crosstalk between Rab and Arf GTPases to promote the tethering and fusion of early endosomes and trans-Golgi network-derived vesicles ( Mattera et al . , 2003; Zhu et al . , 2004a; Kawasaki et al . , 2005 ) . In the GAT-Rabaptin-5C212 structure , Rabaptin-5C212 assumes a linear conformation , the N-terminal regions of Rabaptin-5C212 bind one GAT , and two symmetry-related Rabaptin-5C212 dimerize through the C-terminal regions ( Zhu et al . , 2004a ) , which are the binding site for Rabex-5CC . Interestingly , Rabaptin-5C212 assumes a linear conformation in the Rabex-5CC-Rabaptin-5C212 structure but a V-shaped conformation in the R2Δ structure , and in both structures , the C-terminal regions of Rabaptin-5C212 bind one Rabex-5CC and two symmetry-related Rabaptin-5C212 dimerize through the N-terminal regions which are the binding site for GAT . Moreover , in the R3Δ structure , Rabaptin-5C212 also assumes a linear conformation but two Rabaptin-5C212 dimerize through the middle regions ( residues 590–600 ) between the GAT-binding site and the Rabex-5CC-binding site . As the GAT-binding site is unoccupied , Rabaptin-5 may bind Rabex-5 and GGA1 simultaneously . These results demonstrate that Rabaptin-5C21 always forms a dimer which can assume either a V-shaped or a linear conformation and can bind other proteins via different regions . In addition , the dimeric Rabaptin-5C212 prefers to further dimerize to form a dimer of dimers via the regions that are not involved in the interactions with other proteins . In the context of full-length Rabaptin-5 , it is possible that Rabaptin-5C212 might exist in either the V-shaped or the linear conformation depending on the functional state . The dimerization of Rabaptin-5C212 might not only avoid exposure of the hydrophobic surface and minimize the overall energy of the protein in aqueous environment , but also play some functional roles in the tethering and fusion of early endosomes and/or with other vesicles .
The cDNAs corresponding to the Rabex-5 CC domain ( residues 409–455 , Rabex-5CC ) , the Rabex-5 GEF domain ( residues 132–392 ) , the Rabex-5 GEF and CC domains ( residues 132–455 , Rabex-5 ) , the Rabaptin-5 C2-1 domain ( residues 552–642 , Rabaptin-5C21 ) , and Rab5 ( residues 15–184 ) were all amplified by PCR from the cDNA library of human brain cells . The Rabex-5CC and Rabex-5 GEF ORFs were cloned into the pET-28a plasmid ( Novagen , Germany ) with a His6 tag inserted at the N-terminus . The Rab5 ORF was cloned into a modified pET-28a plasmid ( Novagen ) with a His6-sumo tag inserted at the N-terminus . The Rabex-5CC and Rabaptin-5C21 ORFs , and the Rabex-5 and Rabaptin-5C21 ORFs were cloned into the pET-Duet1 plasmid ( Novagen ) with a His6 tag inserted at the N-terminus of Rabex-5CC and Rabex-5 , respectively . The Rabex-5Δ variants containing different deletion forms of the linker between the GEF and CC domains were generated using the Takara MutanBEST Mutagenesis kit ( TakaRa Biotechnology , Japan ) . The Rabex-5 and Rabaptin-5C21 mutants containing point mutations were generated using the QuikChange Site-Directed Mutagenesis kit ( Agilent Technologies , Santa Clara , CA ) . All recombinant proteins were expressed in E . coli BL21 ( DE3 ) Codon-Plus strain ( Novagen ) . The transformed cells were grown at 37°C in LB medium containing 0 . 05 mg/ml ampicillin or kanamycin until OD600 reached 0 . 8 , and then induced with 0 . 25 mM IPTG at 16°C for 24 hr . All the proteins were purified by affinity chromatography using a Ni-NTA column ( Qiagen , Germany ) and gel filtration chromatography using a Superdex 200 16/60 column ( GE Healthcare , Sweden ) in a buffer containing 20 mM Tris–HCl , pH 8 . 0 , 150 mM NaCl , and 1 mM PMSF . The resultant samples were of >95% purity as evaluated by SDS-PAGE . A trypsin stock solution ( 2 . 5 mg/ml ) was diluted to 10−1 to 10−6 times . The Rabex-5-Rabaptin-5C21 complex ( 1 mg/ml ) was mixed with the trypsin solution of different concentrations . The digestion reaction proceeded for 30 min at 4°C and 16°C , respectively , and was then stopped by addition of 10 μg/ml aprotinin to inhibit the activity of trypsin . The reaction mixture was loaded onto Ni-NTA beads , and both the beads and the flow-through were analyzed by SDS-PAGE with Coomassie blue staining and Western blot with anti-His antibody ( 1:3000 , TIANGEN , China ) . For in vitro GST pull-down assay , the Rabaptin-5C21 ORF was cloned into the pET-3E-His plasmid ( Novagen ) with an N-terminal His6 tag , and the Rabex-5 ORF into the pGEX 6P-1 plasmid ( GE Healthcare ) with an N-terminal GST tag . His6-Rabaptin-5C21 was purified by Ni-NTA affinity chromatography and GST-Rabex-5 by glutathione sepharose beads ( GE Healthcare ) . 20 μg GST-Rabex-5 immobilized onto the glutathione sepharose beads were incubated with 100 μg His6-Rabaptin-5C21 at 4°C for 2 hr . The beads were analyzed by SDS-PAGE with Coomassie blue staining . The Rabex-5 GEF activity for Rab5 was determined using the method described previously ( Delprato et al . , 2004 ) . Briefly , Rab5 was mixed with 20-fold excess fluorescent 2’ ( 3’ ) -bis-O- ( N-methylanthraniloyl ) -GDP ( mantGDP , Invitrogen , Carlsbad , CA ) . The mixture was incubated for 2 hr and the free mantGDP was removed by gel filtration using a HiTrap De-salting column ( GE Healthcare ) . The mantGDP-bound Rab5 was diluted to 500 nM in a buffer containing 20 mM Tris–HCl ( pH 8 . 0 ) , 150 mM NaCl , and 2 mM MgCl2 . Nucleotide exchange reaction was initiated by addition of GTP to a final concentration of 1 mM and varied concentrations ( 50–500 nM ) of Rabex-5 or Rabex-5-Rabaptin-5C21 . Dissociation of mantGDP was monitored by measuring the decrease of fluorescence . Samples were excited at 360 nm and the emission was monitored at 440 nm . Fluorescence data were recorded using a Varian Cary Eclipse spectrofluorimeter ( Agilent Technologies ) . Observed pseudo first-order exchange rate constant ( kobs ) was obtained by a nonlinear least-squares-fit of the data at each concentration of Rabex-5 to the exponential equationI ( t ) = ( I0−I∞ ) exp ( −kobst ) +I∞where I ( t ) is the emission intensity at time t , I0 the initial emission intensity , and I∞ the final emission intensity . Catalytic efficiency ( kcat/Km ) was obtained from the slope of a linear least-squares-fit of the kobs values to the linear equationkobs= ( kcat/Km ) [Rabex−5]+kintrwhere kintr is the intrinsic nucleotide exchange rate in the absence of Rabex-5 . The intrinsic exchange rate ( kintr ) of Rab5 is measured to be 0 . 00064 ± 0 . 00002 s−1 . Crystallization was performed using the hanging drop vapor diffusion method at 16°C by mixing equal volumes ( 1 . 0 μl ) of protein solution ( 20 mg/ml ) and reservoir solution . Crystals of the R2Δ complex were grown from drops consisting of the reservoir solution of 2 . 0 M NaH2PO4/K2HPO4 ( pH 7 . 0 ) and 0 . 05% n-octyl-β-D-galactopyranoside . Crystals of Rabex-5CC were grown from drops consisting of the reservoir solution of 0 . 10 M NaAc ( pH 5 . 4 ) , 17 . 5% MPD , and 2% PEG4000 . Crystals of the Rabex-5CC-Rabaptin-5C21 complex were grown from drops consisting of the reservoir solution of 0 . 15 M MgAc2 and 20% PEG3350 . Crystals of the R3Δ complex were grown from drops consisting of the reservoir solution of 1 . 0 M NaH2PO4/K2HPO4 ( pH 5 . 0 ) . All of the diffraction data were collected at −175°C at beamline 17U of Shanghai Synchrotron Radiation Facility , and processed with HKL2000 ( Otwinowski and Minor , 1997 ) . The structure of the R2Δ complex was solved using the molecular replacement ( MR ) method as implemented in Phenix ( Adams et al . , 2010 ) with the structure of the Rabex-5 GEF domain ( PDB code 1TXU ) ( Delprato et al . , 2004 ) as the search model . The structure of the Rabex-5CC-Rabaptin-5C21 complex was solved by MR with the structure of the Rabaptin-5 C2-1 domain ( PDB code 1X79 ) ( Zhu et al . , 2004a ) as the search model . The structure of Rabex-5CC was solved by MR with the structure of Rabex-5CC in its complex with Rabaptin-5C21 as the search model . The structure of the R3Δ complex was solved by MR with the structure of the Rabex-5 GEF-Rab21 complex ( PDB code 2OT3 ) ( Delprato and Lambright , 2007 ) as the search model . Structure refinement was carried out using Phenix ( Adams et al . , 2010 ) , Refmac5 ( Murshudov et al . , 1997 ) , and CNS ( Brunger , 2007 ) , and model building using Coot ( Emsley and Cowtan , 2004 ) . Due to the low resolution of the diffraction data , the structure models of Rab5 , Rabex-5 , and Rabaptin-5C21 in the R3Δ complex were refined as rigid bodies with deformable elastic network and group B-factor restraints ( Schroder et al . , 2010 ) and thus the side-chain orientations in this complex are somewhat uncertain . Stereochemistry of the structure models was analyzed using Procheck ( Laskowski et al . , 1993 ) . Structural analyses were carried out using programs in CCP4 ( Winn et al . , 2011 ) and the PISA server ( Krissinel and Henrick , 2007 ) . All structure figures were generated using PyMOL ( http://www . pymol . org ) . The statistics of the structure refinement and final structure models are summarized in Table 1 . Protein samples were concentrated to 5 mg/ml in 20 mM Tris–HCl ( pH 8 . 0 ) and 150 mM NaCl . Solution scattering experiments were performed at 293 K on a SAXSess mc2 platform ( Anton Paar , Austria ) equipped with a sealed tube source and a CMOS diode array detector . The SAXS data were collected with 2 hr exposure time in 1-hr frame to ensure absence of radiation damage during the course of the experiment . The SAXS data for the buffer were recorded for background subtraction . Inverse Fourier transformation was performed with the GIFT program in the PCG software package ( Anton Paar ) . The maximum paired-distance ( Dmax ) value was extrapolated from the P ( r ) distribution . The radius of gyration ( Rg ) and the Porod volume were calculated using PRIMUS ( Konarev et al . , 2003 ) at the low angle region ( q × Rg ≤1 ) . The theoretical P ( r ) distribution and Rg value for each structure model were calculated using PTRAJ from the AMBER12 package ( Case et al . , 2012 ) . The structure model was assessed against the corresponding solution scattering data using CRYSOL from the ATSAS software ( Svergun et al . , 1995 ) with constant subtraction . The his6-tag and disordered residues were built back into the crystal structures using PyMOL ( http://www . pymol . org ) , and the yielded structure models were optimized using Xplor-NIH ( Schwieters et al . , 2006 ) for optimal packing . For each crystal structure , a total of 800 structure models were generated with Monte Carlo simulated annealing while fixing the coordinates of the atoms observed in the crystal structure . The crystal structures of Rabex-5CC , the Rabex-5CC-Rabaptin-5C212 complex , the Rabex-5Δ-Rabaptin-5C212 complex , and the Rab5-Rabex-5Δ-Rabaptin-5C212 complex have been deposited with the Protein Data Bank under accession codes 4N3X , 4N3Y , 4N3Z , and 4Q9U , respectively . | Cells need to allow various molecules to pass through the plasma membrane on their surface . Some molecules have to enter the cell , whereas others have to leave . Cells rely on a process called endocytosis to move large molecules into the cell . This involves part of the membrane engulfing the molecule to form a ‘bubble’ around it . This bubble , which is called an endosome , then moves the molecule to final destination inside the cell . A protein called Rab5 controls how a new endosome is produced . However , before this can happen , various other molecules—including two proteins called Rabex-5 and Rabaptin-5—must activate the Rab5 protein . Exactly how these three proteins interact with each other was unknown . Zhang et al . used X-ray crystallography to examine the structures of the complexes formed when Rabex-5 and Rabaptin-5 bind to each other , both when Rab5 is present , and also when it is absent . Biochemical and biophysical experiments confirmed that the Rabex-5/Rabaptin-5 complex is able to activate Rab5 . Zhang et al . also found that Rabex-5 , on its own , folds so that the site that normally binds to Rab5 instead binds to a different part of Rabex-5 , thus preventing endocytosis . However , when Rabaptin-5 forms a complex with Rabex-5 , the Rab5 binding site is freed up . The Rabex-5/Rabaptin-5 complex can switch between a V shape and a linear structure . Binding to Rab5 stabilizes the linear form of the complex , which then helps activate Rab5 , and subsequently the activated Rab5 can interact with other downstream molecules , triggering endocytosis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Molecular mechanism for Rabex-5 GEF activation by Rabaptin-5 |
We make choices based on the values of expected outcomes , informed by previous experience in similar settings . When the outcomes of our decisions consistently violate expectations , new learning is needed to maximize rewards . Yet not every surprising event indicates a meaningful change in the environment . Even when conditions are stable overall , outcomes of a single experience can still be unpredictable due to small fluctuations ( i . e . , expected uncertainty ) in reward or costs . In the present work , we investigate causal contributions of the basolateral amygdala ( BLA ) and orbitofrontal cortex ( OFC ) in rats to learning under expected outcome uncertainty in a novel delay-based task that incorporates both predictable fluctuations and directional shifts in outcome values . We demonstrate that OFC is required to accurately represent the distribution of wait times to stabilize choice preferences despite trial-by-trial fluctuations in outcomes , whereas BLA is necessary for the facilitation of learning in response to surprising events .
Learning to predict rewards is a remarkable evolutionary adaptation that supports flexible behavior in complex and unstable environments . When circumstances change , previously-acquired knowledge may no longer be informative and the behavior needs to be adapted to benefit from novel opportunities . Frequently , alterations in environmental conditions are not signaled by external cues and can only be inferred from deviations from anticipated outcomes , that is , surprise signals . When making decisions , humans typically attempt to maximize benefits ( i . e . , amount of reward ) received per invested resource ( i . e . , money , time , physical or cognitive effort ) . We , like many other animals , compute economic value that takes into account rewards and costs associated with available behavioral options and choose the alternative that is expected to result in outcomes of the highest value based on previous experiences under similar conditions ( Padoa-Schioppa and Schoenbaum , 2015; Sugrue et al . , 2005 ) . When the outcomes of choices consistently violate expectations , new learning is needed to maximize reward procurement . However , not every unexpected outcome is caused by meaningful changes in the environment . Even when conditions are stable overall , outcomes of a single experience can still be unpredictable due to small fluctuations ( i . e . , expected uncertainty ) in reward and costs . Such fluctuations complicate surprise-driven learning since animals need to distinguish between true changes in the environment from stochastic feedback under otherwise stable conditions , known as the problem of change-point detection ( Courville et al . , 2006; Dayan et al . , 2000; Gallistel et al . , 2001; Pearce and Hall , 1980; Yu and Dayan , 2005 ) . Both the basolateral amygdala ( BLA ) and orbitofrontal cortex ( OFC ) participate in flexible reward-directed behavior . Representations of expected outcomes can be decoded from both brain regions during value-based decision making ( Conen and Padoa-Schioppa , 2015; Haruno et al . , 2014; Padoa-Schioppa , 2007 , 2009; Salzman et al . , 2007; van Duuren et al . , 2009 ) . Amygdala lesions render animals unable to adaptively track changes in reward availability or benefit from profitable periods in the environment ( Murray and Izquierdo , 2007; Salinas et al . , 1996; Salzman et al . , 2007 ) . Furthermore , a recent evaluation of the accumulated literature on BLA in appetitive behavior suggests that this region integrates both current reward value and long-term history information ( Wassum and Izquierdo , 2015 ) , and therefore may be particularly well-suited to guide behavior when conditions change . Importantly , single-unit responses in BLA track surprise signals ( Roesch et al . , 2010 ) that can drive learning . Similarly , a functionally-intact OFC is required for adaptive responses to changes in outcome values ( Elliott et al . , 2000; Izquierdo and Murray , 2010; Murray and Izquierdo , 2007 ) . Impairments produced by OFC lesions have been widely attributed to diminished cognitive flexibility or inhibitory control deficits ( Bari and Robbins , 2013; Dalley et al . , 2004; Elliott and Deakin , 2005; Winstanley , 2007 ) . However , this view has been challenged recently by observations that selective medial OFC lesions cause potentiated switching between different option alternatives , rather than a failure to disengage from previously acquired behavior ( Walton et al . , 2010 , 2011 ) . Indeed , there is increasing evidence that certain sectors of OFC may not exert a canonical inhibitory control over action , but may instead contribute outcome representations predicted by specific cues in the environment and update expectations in response to surprising feedback ( Izquierdo et al . , 2017; Marquardt et al . , 2017; Riceberg and Shapiro , 2012 , 2017; Rudebeck and Murray , 2014; Stalnaker et al . , 2015 ) . Despite important contributions of both the BLA and OFC to several forms of adaptive value learning , some learning tasks progress normally without the recruitment of these brain regions . For example , the OFC is not required for acquisition of simple stimulus-outcome associations , both in Pavlovian and instrumental context , or for unblocking driven by differences in value when outcomes are certain and predictable . However , the OFC is needed for adaptive behavior that requires integration of information from different sources , particularly when current outcomes need to be compared with a history in a different context ( or state ) as in devaluation paradigms ( Izquierdo et al . , 2004; McDannald et al . , 2011 , 2005; Stalnaker et al . , 2015 ) . Similarly , as has been shown in rats , BLA has an important role in early learning or decision making under ambiguous outcomes ( Hart and Izquierdo , 2017; Ostrander et al . , 2011 ) , and seems to play a limited role in choice behavior when these outcomes are known or reinforced through extended training . These observations hint at important roles for BLA and OFC in learning under conditions of uncertainty . Yet little is known about unique contributions of these brain regions to value learning when outcomes are fluctuating even under stable conditions ( i . e . , when there is expected uncertainty in outcome values ) . Furthermore , the functional dissociation between different OFC subregions ( e . g . ventromedial vs . lateral ) is presently debated ( Dalton et al . , 2016; Elliott et al . , 2000; Morris et al . , 2016 ) . Recently-developed computational models based on reinforcement learning ( RL ) ( Diederen and Schultz , 2015; Khamassi et al . , 2011; Preuschoff and Bossaerts , 2007 ) and Bayesian inference principles ( Behrens et al . , 2007; Nassar et al . , 2010 ) are well suited to test for unique contributions of different brain regions to value learning under uncertainty . These models rely on learning in response to surprise , or the deviation between expected and observed outcomes ( i . e . , reward prediction errors , RPEs ) ; the learning rate , in turn , determines the degree to which prediction errors affect value estimates . Importantly , the RL principles do not only account for animal behavior , but are also reflected in underlying neuronal activity ( Lee et al . , 2012; Niv et al . , 2015 ) . In the present work , we first developed a novel delay-based behavioral paradigm to investigate the effects of expected outcome uncertainty on learning in rats . We demonstrated that rats can detect true changes in outcome values even when they occur against a background of stochastic feedback . Such behavioral complexity in rodents allowed us to assess causal contributions of the BLA and OFC to value learning under expected outcome uncertainty . Specifically , we examined the neuroadaptations that occur in these brain regions in response to experience with different levels of environmental uncertainty and employed fine-grained behavioral analyses partnered with computational modeling of trial-by-trial performance of OFC- and BLA-lesioned animals on our task that incorporates both predictable fluctuations and directional shifts in outcome values .
Our delay-based task was designed to assess animals’ ability to detect true changes in outcome values ( i . e . , upshifts and downshifts ) even when they occur against the background of stochastic feedback under baseline conditions ( expected uncertainty ) . To probe the effects of expected outcome uncertainty on learning in rodents , we first presented a group of naïve rats ( n = 8 ) with two choice options identical in average wait time but different in the variance of the outcome distribution . Each response option was associated with the delivery of one sugar pellet after a delay interval . The delays were pooled from distributions that were identical in mean , but different in variability ( low vs high: LV vs HV; ~N ( µ , σ ) : μ = 10 s , σHV=4s σLV=1 s ) . Following the establishment of stable performance ( defined as no statistical difference in any of the behavioral parameters across three consecutive testing sessions , including choice and initiation omissions , average response latencies and option preference ) , rats experienced value upshifts ( delay mean was reduced to 5 s with variance kept constant ) and downshifts ( delay mean was increased to 20 s ) on each option independently , followed by return to baseline conditions ( Figure 1A , B; Video 1 , Video 2 ) . Each shift and baseline phase lasted five 60-trial testing sessions; therefore , the total duration of the main task was 43 testing days for each animal . Maximal changes in the choice of each option in response to shifts were analyzed with omnibus within-subject ANOVA with shift type ( HV , LV; upshift , downshift ) and shift phase ( pre-shift baseline , shift , post-shift baseline ) as within-subject factors . These analyses identified a significant shift type x phase interaction [F ( 6 , 42 ) =16 . 412 , p<0 . 0001] . Post-hoc analyses revealed no differences in preference at baseline conditions across assessments [F ( 3 . 08 , 21 . 57 ) =0 . 98 , p=0 . 422; Greenhouse-Geisser corrected] , suggesting that rats were able to infer mean option values ( wait times ) and maintain stable choice preferences despite variability in outcomes . 10 . 7554/eLife . 27483 . 003Figure 1 . Task design and performance of intact animals . Our task is designed to investigate the effects of expected outcome uncertainty on value learning . ( A ) Each trial began with stimulus presentation in the central compartment of the touchscreen . Rats ( n = 8 ) were given 40 s to initiate a trial . If 40 s passed without a response , the trial was scored as an ‘initiation omission . ’ Following a nosepoke to the central compartment , the central stimulus disappeared and two choice stimuli were presented concurrently in each of the side compartments of the touchscreen allowing an animal a free choice between two reward options . An animal was given 40 s to make a choice; failure to select an option within this time interval resulted in the trial being scored as ‘choice omission’ and beginning of an ITI . Each response option was associated with the delivery of one sugar pellet after a delay interval . ( B ) The delays associated with each option were pooled from distributions that are identical in mean value , but different in variability: LV ( low variability , shown in blue ) vs . HV ( high variability , shown in red ) ; ~N ( µ , σ ) : μ = 10 s , σHV=4s , σLV=1s . Following the establishment of stable performance , rats experienced value upshifts ( µ = 5 s; σ kept constant ) and downshifts ( μ = 20 s ) on each option independently , followed by return to baseline conditions . Each shift and return to baseline phase lasted for five 60-trial sessions . ( C ) Regardless of the shift type , animals significantly changed their preference in response to all shifts ( all p values<0 . 05 ) . However , significant differences between HV and LV in choice adaptations were observed for both upshifts and downshifts: greater variance of outcome distribution at baseline facilitated behavioral adaptation in response to value upshifts ( HV vs LV difference , p=0 . 004 ) , but rendered animals suboptimal during downshifts ( p=0 . 027 ) ; conversely , low expected uncertainty at baseline led to decreased reward procurement during upshifts in reward . The data are shown as group means for option preference during pre-baseline , shift and post-baseline conditions , ± SEM . The asterisks signify statistical differences between HV and LV conditions . ( D ) The number of initiation omissions was significantly increased during downshift ( p=0 . 004 ) and decreased during upshifts ( p=0 . 017 ) in value , regardless of the levels of expected uncertainty , demonstrating effects of overall environmental reward conditions on motivation to engage in the task . The data are shown as group means by condition +SEM . *p<0 . 05 , **p<0 . 01 . Summary statistics and individual animal data are provided in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 00310 . 7554/eLife . 27483 . 004Figure 1—source data 1 . Summary statistics and individual data for naïve animals performing the task . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 00410 . 7554/eLife . 27483 . 005Video 1 . An animal performing the task during upshift on HV option . During an upshift in value on each option , the mean of the delays to reward was reduced to 5 s with variance kept the same as during baseline conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 00510 . 7554/eLife . 27483 . 006Video 2 . An animal performing the task during downshift on HV option . During a downshift in value on each option , the mean of the delays to reward was increased to 20 s with variance kept constant . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 006 All animals significantly changed their preference in response to all shifts ( Figure 1 , all p values<0 . 05 ) . We then assessed the effects of the overall environmental reward conditions on rats’ motivation to engage in the task . The number of initiation omissions ( i . e . , failure to respond to the central cue presented at the beginning of each trial within 40 s ) was analyzed with omnibus ANOVA with reward conditions ( stable , upshift , and downshift collapsed across HV and LV options ) as within-subject factor . The main effect of condition was significant [F ( 1 . 09 , 7 . 61 ) =16 . 772 , p=0 . 03; Greenhouse-Geisser corrected]: the number of omissions was significantly increased during downshifts ( p=0 . 004 ) and decreased during upshifts ( p=0 . 017 ) in value , revealing that task engagement was sensitive to overall environmental reward rate . Therefore , rodents are able to learn about fundamental directional changes in value means despite stochastic fluctuations in outcome values under baseline conditions ( i . e . , expected uncertainty ) . However , significant differences between HV and LV in choice adaptations were observed for both upshifts and downshifts: greater variance of outcome distribution at baseline facilitated behavioral adaptation in response to value upshifts ( HV vs LV difference , p=0 . 004 ) , but rendered animals suboptimal during downshifts ( p=0 . 027 ) ; conversely , low expected uncertainty at baseline led to decreased reward procurement during upshifts in reward . These effects may be explained by a hyperbolic nature of delay-discounting across species ( Freeman et al . , 2009; Green et al . , 2013; Hwang et al . , 2009; Mazur and Biondi , 2009; Mitchell et al . , 2015; Rachlin et al . , 1991 ) . We hypothesized that experience with different levels of outcome uncertainty would induce long-term neuroadaptations , affecting the response to the same magnitude of surprise signals . Specifically , we assessed expression of gephyrin ( a reliable proxy for membrane-inserted GABAA receptors mediating fast inhibitory transmission; [Chhatwal et al . , 2005; Tyagarajan et al . , 2011] ) and GluN1 ( an obligatory subunit of glutamate NMDA receptors; [Soares et al . , 2013] ) in BLA and OFC . Three separate groups of animals were trained to respond to visual stimuli on a touchscreen to procure a reward after variable delays . The values of outcomes were identical to our task described above but no choice was given . One group was trained under LV conditions , the second under HV ( matched in total number of rewards received ) , and the third control group received no rewards ( n = 8 in each group , total n = 24 ) . Given the limited amount of tissue , we focused on NMDA instead of AMPA receptors based on previous evidence demonstrating dissociable effects of ionotropic glutamate receptors in delay-based decision making ( Yates et al . , 2015 ) . Protein expression analyses revealed unique adaptations to outcome variability in BLA , specifically in GABA-ergic sensitivity . Biochemical measures were analyzed with mixed ANOVA with brain region as a within-subject factor and reward experience ( HV , LV or no reward ) as a between-subject factor . There was a significant main effect of group [F ( 2 , 12 ) =6 . 002 , p=0 . 016] and brain region x group interaction [Figure 2A; F ( 2 , 12 ) =41 . 863 , p<0 . 0001] for gephyrin . A significant main effect of group [F ( 2 , 21 ) =4 . 084 , p=0 . 032] and group x brain region [F ( 2 , 21 ) =5 . 291 , p=0 . 014] interaction were also found for GluN1 expression . Subsequent analyses identified uncertainty-dependent upregulation of gephyrin in BLA [between-subject ANOVA: F ( 2 , 21 ) =45 . 448 , p<0 . 0001 ) , that was maximal following HV training ( all post hoc comparison p values<0 . 05 ) . Similarly , GluN1 showed robust upregulation in response to experienced reward in BLA [Figure 2B; F ( 2 , 21 ) =7 . 092 , p=0 . 004; no reward vs LV p=0 . 045; no reward vs HV p=0 . 002] , however post hoc analyses failed to detect a significant difference between HV and LV training ( p=0 . 637 ) . In OFC , gephyrin was instead downregulated in response to experiences with reward in general [F ( 2 , 12 ) =4 . 445 , p=0 . 036; no reward vs LV p=0 . 045; no reward vs HV p=0 . 042] and did not depend on variability in outcome distribution ( post hoc comparison: HV vs LV , p=1 ) ; no changes were observed in GluN1 [F ( 2 , 21 ) =2 . 359 , p=0 . 119] . 10 . 7554/eLife . 27483 . 007Figure 2 . Region-specific alterations in gephyrin and GluN1 expression induced by experience with outcome uncertainty . Three separate groups of animals were trained to respond to visual stimuli on a touchscreen to get a reward after variable delays . The values of outcomes were identical to the main task but no choice was given . One group was trained under LV conditions , the second under HV ( matched in total number of rewards received ) , and the third control group received no rewards ( n = 8 per group ) . We assessed expression of A gephyrin ( a reliable proxy for membrane-inserted GABAA receptors mediating fast inhibitory transmission ) and B GluN1 ( an obligatory subunit of glutamate NMDA receptors ) in BLA and ventral OFC . Biochemical analyses revealed uncertainty-dependent upregulation in gephyrin in BLA , that was maximal following HV training ( p<0 . 0001 ) . Similarly , GluN1 showed robust upregulation in response to experienced reward in BLA ( no reward vs LV p=0 . 045; no reward vs HV p=0 . 002 ) , however post hoc analyses failed to detect a significant difference between HV and LV training ( p=0 . 637 ) . In ventral OFC , gephyrin was downregulated in response to experiences with reward in general ( no reward vs LV p=0 . 045; no reward vs HV p=0 . 042 ) and did not depend on variability in outcome distribution; no changes were observed in GluN1 . The data are shown as group means by condition +SEM . *p<0 . 05 , **p<0 . 01 Summary statistics and individual animal data are provided in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 00710 . 7554/eLife . 27483 . 008Figure 2—source data 1 . Summary statistics and individual data for GluN1 and gephyrin expression in BLA and OFC . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 008 Therefore , both the BLA and OFC undergo unique patterns of neuroadaptations in response to experience with variability , suggesting that these brain regions may play complementary , yet dissociable , roles in value learning under outcome uncertainty . Given the behavioral complexity that rodents exhibit on our task , we were able to directly test the causal contributions of the BLA and ventromedial OFC to value learning under conditions of expected uncertainty in outcome distribution . The results of lesion studies ( lesion sites are shown in Figure 3 ) were in line with predictions suggested by protein data . Because we were primarily interested in the contributions of the BLA and OFC to surprise-driven learning , we first analyzed the maximal changes in option preference in response to up- and downshifts . This analysis allowed us to control for potential effects of brain lesions on choice behavior under baseline conditions in our task . An omnibus ANOVA with shift type as within- and experimental group ( sham , BLA vs OFC lesion; n = 8 per group; total n = 24 ) as between-subject factors detected a significant main effect of group [F ( 2 , 21 ) =11 . 193 , p<0 . 0001] and group x shift type interaction [F ( 6 , 63 ) =9 . 472 , p<0 . 0001 ) . Subsequent analyses showed significant simple main effects of experimental group on all shift types: upshift on HV [F ( 2 , 21 ) =14 . 723 , p<0 . 0001] , upshift on LV [F ( 2 , 21 ) =5 . 663 , p=0 . 011] , downshift on HV [F ( 2 , 21 ) =19 . 081 , p<0 . 0001] , and downshift on LV [F ( 2 , 21 ) =7 . 189 , p=0 . 004] . The OFC-lesioned rats were less optimal on our task: they changed their option preference to a significantly lesser degree compared to control animals during upshifts on HV ( p=0 . 005 ) and LV ( p=0 . 039 ) , as well as the downshift on LV option ( p=0 . 015; Figure 4A ) . Whereas OFC lesions produced a pronounced impairment in performance , it was less clear if alterations produced by BLA lesions lead to suboptimal behavior . BLA-lesioned animals changed their option preference to a lesser degree on HV upshifts ( p<0 . 0001 ) , but compensated by exaggerated adaptations to HV downshifts ( p<0 . 0001; Figure 4A ) . 10 . 7554/eLife . 27483 . 009Figure 3 . Location and extent of intended lesion ( colored regions ) on standard coronal sections through ventral OFC and BLA . The extent of the lesions was assessed after the completion of behavioral testing by staining for a marker of neuronal nuclei , NeuN . ( A ) Top: representative photomicrograph of a NeuN stained coronal section showing ventral OFC lesion . Bottom: depictions of coronal sections adapted from ( Paxinos and Watson , 1997 ) . The numerals on the lower left of each matched section represent the anterior-posterior distance ( mm ) from Bregma . Light and dark blue represent maximum and minimum lesion area across animals , respectively . Though coordinates were aimed at the ventral orbital region , lesion extent includes anterior medial orbital cortex as well . ( B ) Top: representative photomicrograph of a NeuN stained coronal section showing BLA lesion . Bottom: depictions of coronal sections with numerals on the lower left of each matched section representing the anterior-posterior distance ( mm ) from Bregma . Light and dark red represent maximum and minimum lesion area across animals , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 00910 . 7554/eLife . 27483 . 010Figure 4 . Changes in choice preference in response to value shifts and learning strategies in experimental groups . ( A ) The OFC-lesioned rats ( n = 8 ) were less optimal on our task: they changed their option preference to a significantly lesser degree compared to control animals ( n = 8 ) during upshifts on HV ( p=0 . 005 ) and LV ( p=0 . 039 ) , as well as the downshift on LV option ( p=0 . 015 ) . Conversely , animals with BLA lesions ( n = 8 ) changed their option preference to a lesser degree on HV upshifts ( p<0 . 0001 ) , but compensated by exaggerated adaptations to HV downshifts ( p<0 . 0001 ) . Group means for option preference during pre-baseline , shift and post-baseline conditions are shown in Figure 4—figure supplement 1 . We broke the trials into two types: when the delays fell within distributions experienced for each option at baseline ( expected outcomes ) and those in which the degree of surprise exceeded that expected by chance ( unexpected outcomes ) . Win-stay/lose-shift scores were computed based on trial-by-trial data: a score of 1 was assigned when animals repeated the choice following better than average outcomes ( win-stay ) or switched to the other alternative following worse than average outcomes ( lose-shift ) . Sham-lesioned animals demonstrated increased sensitivity to unexpected feedback ( p values < 0 . 001 ) . Similarly , the ability to distinguish between expected and unexpected outcomes was intact in BLA-lesioned animals ( p values < 0 . 001 ) , although their sensitivity to feedback decreased overall . In contrast , OFC-lesioned animals failed to distinguish expected from unexpected fluctuations . ( C , D ) To examine the learning trajectory we analyzed the evolution of option preference . BLA-lesioned animals were indistinguishable from controls during the shifts on LV option . Whereas , this experimental group demonstrated significantly attenuated learning during the upshift on HV ( p values < 0 . 0001 for all sessions ) and potentiated performance during sessions 3 through 5 on HV downshift ( p values < 0 . 05 ) compared to sham group . Conversely , learning in OFC-lesioned animals was affected on the majority of the shift types: these animals demonstrated significantly slower learning during sessions 3 through 5 during upshift on HV ( p values < 0 . 05 ) , all sessions during upshift on LV ( p values < 0 . 05 ) and sessions 3 through 5 during downshift on LV ( p values < 0 . 05 ) . Session 0 refers to baseline/pre-shift option preference . Despite these differences in responses to shifts in value under conditions of uncertainty , we did not observe any deficits in basic reward learning in either the BLA- or OFC-lesioned animals , shown in Figure 4—figure supplement 2 . The data are shown as group means by condition +SEM . *p<0 . 05 , **p<0 . 01 . Summary statistics and individual animal data are provided in Figure 4—source data 1 and Figure 4—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 01010 . 7554/eLife . 27483 . 011Figure 4—source data 1 . Summary statistics and individual data for changes in choice preference and learning strategies . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 01110 . 7554/eLife . 27483 . 012Figure 4—source data 2 . Summary statistics and individual data demonstrating experimental group differences in response to shifts . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 01210 . 7554/eLife . 27483 . 013Figure 4—figure supplement 1 . Changes in choice behavior in response to value shifts . ( A ) Both lesion groups demonstrated reduced adaptations to value upshifts on HV option ( p<0 . 01 ) . ( B ) . BLA-lesioned animals chose LV option more frequently than controls when its value was increased ( p<0 . 01 ) . ( C , D ) Both BLA- and OFC-lesioned animals also showed reduced HV option preference ( p<0 . 01 ) and increased LV option preference ( p<0 . 05 ) during downshifts compared to sham animals . This pattern of results can be explained by changes in choice behavior even under baseline conditions in BLA- and OFC-lesioned animals that interacted with rats’ ability to learn about shifts in value . Indeed , there were significant group differences in pre-shift baseline preferences . The data are shown as group means for option preference during pre-baseline , shift and post-baseline conditions , ± SEM . *p<0 . 05 , **p<0 . 01 . Summary statistics and individual animal data are provided in Figure 4—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 01310 . 7554/eLife . 27483 . 014Figure 4—figure supplement 2 . The lack of group differences in basic reward learning . Our surgeries took place prior to any exposure to the testing apparatus or behavioral training . Both lesioned groups were indistinguishable from controls at early stages of the task . During pre-training , animals first learned to respond to visual stimuli presented in the central compartment of the screen within 40 s time interval in order to receive the sugar reward ( stimulus response ) . Next , rats learned to initiate the trial by nosepoking the bright white square stimulus presented in the central compartment of the touchscreen; this response was followed by disappearance of the central stimulus and presentation of a target image in one side compartment of the touchscreen ( trial initiation ) . Responses to the target image produced an immediate reward . The last stage of training was administered to familiarize animals with delayed outcomes . The protocol was identical to the previous stage , except the nosepoke to the target image and reward delivery were separated by a 5 s stable delay ( certain 5 s delay ) . ( A , B ) . Animals in all groups took similar number of days to learn to nosepoke visual stimuli on the touchscreen to receive sugar rewards ( p=0 . 796 ) and to initiate a trial ( p=0 . 821 ) . ( C , D ) . There were no group differences in responses to the introduction of a 5 s delay interval during pre-training ( p=0 . 518 ) or the number of sessions to reach stable performance during the initial baseline phase of our uncertainty task ( p=0 . 772 ) . The data are shown as group means ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 014 In addition to examining the maximal changes in option preferences , we analyzed the behavioral data with an omnibus ANOVA with shift type and shift phase ( pre-shift baseline , shift performance , and post-shift baseline ) as within-subject and experimental group as between-subject factors . This test similarly detected a significant shift type x phase x group interaction [F ( 6 . 9 , 72 . 5 ) =7 . 41 , p<0 . 0001; Greenhouse-Geisser corrected , Figure 4—figure supplement 1 ) . Consistent with the preceding analyses , post hoc tests revealed reduced adaptations to value uphifts on the HV option in both lesion groups ( p<0 . 01 ) . However , we also observed more frequent choices of the LV option when its value was increased in BLA-lesioned animals ( p<0 . 01 ) as well as reduced HV option preference ( p<0 . 01 ) and increased LV option preference ( p<0 . 05 ) during downshifts in both lesion groups compared to control animals . This pattern of results may be explained by changes in choice behavior even under baseline conditions in BLA- and OFC-lesioned animals that interacted with rats’ ability to learn about shifts in value . Successful performance on our task required animals to distinguish between the variance of outcome distributions under stable conditions from surprising shifts in value , despite the fact that delay distributions at baseline and distributions during the shift partially overlapped . To evaluate whether the animals in lesioned groups adopted a different strategy and demonstrated altered sensitivity to surprising outcomes , we examined the win-stay/lose-shift responses . Win-stay and lose-shift scores were computed based on trial-by-trial data similar to previous reports ( Faraut et al . , 2016; Imhof et al . , 2007; Worthy et al . , 2013 ) : a score of 1 was assigned when animals repeated the choice following better than average outcomes ( win-stay ) or switched to the other alternative following worse than average outcomes ( lose-shift ) . Win-shift and lose-stay trials were counted as 0 s . To specifically address whether rats distinguished expected fluctuations from surprising changes , we divided the trials into two types: when the delays fell within distributions experienced for each option at baseline ( expected outcomes ) and those in which the degree of surprise exceeded that expected by chance . The algorithm used for this analysis kept track of all delays experienced under baseline conditions before the current trial for each animal individually . On each trial , we found the value of the minimal and maximal delay . If the current delay value fell within this interval , the outcome was classified as expected . If the current delay fell outside of this distribution , the outcome on this trial was classified as unexpected ( surprising ) . Win-stay and lose-shift scores were calculated for each trial type separately and their probabilities ( summary score divided by the number of trials ) for both trial types were subjected to ANOVA with strategy as within-subject and experimental group as between-subject factors . Our analyses indicated significant strategy x experimental group interaction [F ( 6 , 63 ) =9 . 912 , p<0 . 0001] . Critically , sham-lesioned animals demonstrated increased sensitivity to unexpected outcomes compared to predictable fluctuations for both wins and losses ( Figure 4B , p values <0 . 0001 ) . Similarly , the ability to distinguish between expected and unexpected outcomes was intact in BLA-lesioned animals ( p values < 0 . 001 ) , although their sensitivity to feedback decreased overall . In contrast , OFC-lesioned animals failed to distinguish predictable from surprising fluctuations . Interestingly , sham and BLA-lesioned animals demonstrated low win-stay and lose-shift scores when trial outcomes were expected; these animals were more likely to shift after better than average outcomes and persist with their choices after worse outcomes . In addition to feedback insensitivity , such behavior may result from increases in exploratory behavior in response to wins and behavioral inflexibility after losses . Additionally , when outcomes are relative stable and predictable , rats may be more sensitive to long-term reward history and rely less on the outcome of any one given trial . To examine the learning trajectory we analyzed the evolution of option preference during shift conditions . Specifically , we subjected the session-by-session data during each swift to an omnibus ANOVA with testing session ( 1 through 5; session 0 in Figure 4C , D corresponds to pre-shift option preference ) and shift type as within- and experimental group as between subject factors . This analysis revealed a three-way session x shift type x group interaction [F ( 8 . 73 , 91 . 71 ) =8 . 418 , p=0 . 002; Greenhouse-Geisser corrected , Figure 4C , D] . Subsequent analyses identified significant two-way session x group interactions for each shift type [upshift on HV: F ( 5 . 24 , 55 . 04 ) =3 . 585 , p=0 . 006; downshift on HV: F ( 4 . 14 , 43 . 452 ) =25 . 646 , p<0 . 0001; upshift on LV: F ( 2 . 59 , 27 . 14 ) = 4 . 378 , p=0 . 016; downshift on LV: F ( 3 . 69 , 38 . 767 ) =6 . 768 , p<0 . 0001; all Greenhouse-Geisser corrected] . BLA-lesioned animals were indistinguishable from controls during the shifts on LV option . However , this experimental group demonstrated significantly attenuated learning during the upshift on HV ( p values < 0 . 0001 for all sessions ) and potentiated performance during sessions 3 through five during the downshifts on HV ( p values < 0 . 05 ) compared to the sham group . Conversely , learning in OFC-lesioned animals was affected on the majority of the shift types: these animals demonstrated significantly slower learning during sessions 3 through five during upshift on HV ( p values < 0 . 05 ) , all sessions during upshift on LV ( p values < 0 . 05 ) and sessions 3 through five during downshift on LV ( p values < 0 . 05 ) . Despite these differences in responses to shifts in value , we did not observe any deficits in basic reward learning in either the BLA- or OFC-lesioned animals . Our surgeries took place prior to any exposure to the testing apparatus or behavioral training , yet both lesioned groups were indistinguishable from controls at early stages of the task . All animals took a similar number of days to learn to nosepoke visual stimuli on the touchscreen to receive sugar rewards [F ( 2 , 21 ) =0 . 231 , p=0 . 796] and to initiate a trial [F ( 2 , 21 ) =0 . 199 , p=0 . 821] . Similarly , there were no group differences in their responses to the introduction of a 5 s delay interval during pre-training [F ( 2 , 21 ) =0 . 679 , p=0 . 518] or the number of sessions to reach stable performance during the initial baseline phase of our uncertainty task [F92 , 21 ) =0 . 262 , p=0 . 772; Figure 4—figure supplement 2] . We fit different versions of RL models to trial-by-trial choices for each animal separately . Specifically , we considered the standard Rescorla-Wagner model ( RW ) and a dynamic learning rate model ( Pearce-Hall , PH ) . The RW model updates option values in response to RPEs ( i . e . , the degree of surprise ) with a constant learning rate , conversely the PH model allows for learning facilitation with surprising feedback ( i . e . , the learning rate is scaled according to absolute prediction errors ) . We also compared models in which expected outcome uncertainty is learned simultaneously with value and scales the impact of prediction errors on value ( RW+expected uncertainty ) and learning rate ( Full model ) updating . The total number of free parameters , BIC and parameter values for each model and experimental group are provided in Table 1 . The behavior of the control group was best captured by the dynamic learning rate model with RPE scaling proportional to expected outcome uncertainty and facilitation of learning in response to surprising feedback ( Full model; Table 1 , lower BIC values indicate better fit ) . Therefore , rats in our experiment increased learning rates in response to surprise to maximize reward acquisition rate , but only if unexpected outcomes were not likely to result from value fluctuations under otherwise stable conditions . Consistent with attenuated learning observed in animals with BLA lesions , trial-by-trial performance in these animals was best fit by RW+expected uncertainty model , demonstrating selective loss of learning potentiation in response to surprise and preserved RPE scaling with expected uncertainty in these animals , leading to slower learning compared to intact animals during the shifts on HV option . Conversely , performance of OFC-lesioned animals was best accounted for by PH model , suggesting that while these animals still increased learning rates in response to surprise , they were insensitive to expected outcome uncertainty . Furthermore , the overall learning rates were reduced in OFC-lesioned animals ( p=0 . 01 compared to the sham group ) . Finally , we observed significantly lower values of β ( inverse temperature parameter in softmax choice rule ) in both BLA- and OFC-lesioned animals [F ( 2 , 21 ) =4 . 88 , p=0 . 018; sham vs BLA: p<0 . 0001; sham vs OFC: p<0 . 0001] , suggesting that their behavior is less stable , more exploratory and less dependent on the difference in learned outcome values compared to control group . 10 . 7554/eLife . 27483 . 015Table 1 . Model comparison . Lower BIC values indicate better model fit ( in bold ) ; number of free parameters and parameter values ± SEM of the best fitting model are provided for each group . Trial-by-trial choices of the intact animals were best captured by the dynamic learning rate model incorporating RPE scaling proportional to expected uncertainty and facilitation of learning in response to surprising outcomes ( Full model ) . BLA lesions selectively eliminated learning rate scaling in response to surprise ( RW+expected uncertainty model provided the best fit ) . Whereas OFC lesioned animals still increased learning rates in response to surprising events ( PH model ) , RPE scaling proportional to expected outcome uncertainty was lost in this group . Furthermore , the overall learning rates were reduced in OFC-lesioned animals ( p=0 . 01 ) . Finally , we observed significantly lower values of β ( inverse temperature parameter in softmax choice rule ) in both BLA- and OFC-lesioned animals ( p<0 . 0001 ) , suggesting that their behavior is less stable , more exploratory and less dependent on the difference in learned outcome values . Asterisks indicate parameter values that were significantly different from the control group ( in bold ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 015ModelRWPHRW+expected uncertaintyFull# parameters3456BICparameter value ± SEMkα , valueβηα , riskωsham26519 . 3926900 . 6626384 . 1825681 . 70 . 29 ± 0 . 030 . 09 ± 0 . 0114 . 1 ± 0 . 990 . 33 ± 0 . 040 . 56 ± 0 . 083 . 04 ± 0 . 11BLA lesion26201 . 8926864 . 7425153 . 8227162 . 820 . 32 ± 0 . 020 . 07 ± 0 . 017 . 4 ± 0 . 6*n/a0 . 58 ± 0 . 063 . 40 ± 0 . 4OFC lesion24292 . 5423171 . 4624630 . 9223994 . 50 . 3 ± 0 . 050 . 05 ± 0 . 01*5 . 5 ± 0 . 68*0/32 ± 0 . 05n/an/a To gain further insights into outcome representations in our experimental groups , we analyzed the microstructure of rats’ choice behavior . Specifically , we addressed whether BLA and ventral OFC lesions altered animals’ ability to form expectations about the timing of reward delivery . On each trial during all baseline conditions , where the overall values of LV and HV options were equivalent , reward port entries were recorded in 1 s bins during the waiting period ( after a rat had indicated its choice and until reward delivery; histograms of true distributions of the delays and animals’ reward-seeking actions normalized to the total number of reward port entries are shown in Figure 5 ) . These data were analyzed with an ANOVA with time bin as within- and lesion group as between-subject factors . There were no significant differences in the mean of expected reward delivery times across groups [F ( 5 , 42 ) =1 . 064 , p=0 . 394] . Similarly , all groups were matched in the total number of reward port entries [F ( 2 , 21 ) =0 . 462 , p=0 . 636; Figure 5—figure supplement 1] . However , a significant difference in variances of reward port entry distributions was detected [χ2 ( 209 ) =4004 . 054 , p<0 . 0001] . Whereas the distributions of reward-seeking times in BLA-lesioned rats were indistinguishable from control animals’ and the true delays , OFC-lesioned animals concentrated their reward port entries in the time interval corresponding to mean delays , suggesting that while these animals can infer the average outcomes , they fail to represent the variance ( i . e . , expected uncertainty ) . 10 . 7554/eLife . 27483 . 016Figure 5 . Animals with ventral OFC lesions fail to represent expected uncertainty in reward delays . We assessed whether BLA and ventral OFC lesions alter animals’ ability to form expectations about the timing of reward delivery . On each trial during all baseline conditions where the overall value of LV and HV options were equivalent , reward port entries were recorded in 1 s bins during the waiting period . There were no significant differences in the means of expected reward delivery times across groups ( p=0 . 394 ) . Similarly , the groups were matched in the total number of reward port entries ( p=0 . 636 ) as shown in Figure 5—figure supplement 1 . Whereas the distributions of reward-seeking times in BLA-lesioned animals were indistinguishable from control animals’ and the true delays ( A–F ) , OFC-lesioned animals concentrated their reward port entries in the time interval corresponding to mean delays ( G , H ) , suggesting that while these animals can infer the average outcome , they fail to represent the variance ( i . e . , expected uncertainty ) . We also considered the changes in waiting times across our task; these data are shown in Figure 5—figure supplement 1 . Each bar in histogram plots represents mean frequency normalized to total number of reward port entries ±SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 01610 . 7554/eLife . 27483 . 017Figure 5—figure supplement 1 . Total number of reward port entries and changes in waiting time variances across task phases . On each trial during all baseline conditions where the overall value of LV and HV options were equivalent , reward port entries were recorded in 1 s bins during the waiting period . ( A ) All groups of animals were matched in the total number of reward port entries ( p=0 . 636 ) . ( B ) We also considered the changes in waiting times across our task . We calculated the variance of reward port entry times during each baseline ( initial phase of the task and four baselines separating the shifts ) separately for each animal . There was a significant main effect of lesion group on waiting time variances for HV option ( p<0 . 0001 ) with OFC-lesioned animals demonstrating consistently lower variability in their waiting behavior despite experience with shifts . The data are shown as group means ± SEM , **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 017 We also considered the changes in waiting times across our task . We calculated the variance of reward port entry times during each baseline ( initial phase of the task and four baseline separating the shifts ) for each animal . We then subjected the estimated variances to ANOVAs with baseline order ( 1st to 5th ) as within- and lesion group as between-subject factors . Similar to our preceding analysis of combined baselines , we did not detect any group differences in waiting times for the LV option ( all p values>0 . 2 ) . However , there was a significant main effect of lesion group on waiting time variances for the HV option [F ( 2 , 21 ) =117 . 074 , p<0 . 0001; Figure 5—figure supplement 1] with OFC-lesioned animals demonstrating consistently lower variability in their waiting behavior despite the experience with shifts . Importantly , since our analyses only included the waiting time prior to reward delivery , these results suggest that OFC-lesioned animals retain the ability to form simple outcome expectations based on long-term experience , yet their ability to represent the more complex outcome distributions is compromised . To assess group differences in uncertainty-seeking or avoidance , we subjected HV option preference data under baseline conditions to an ANOVA with time ( five repeated baseline tests separating the value shifts ) as within- and lesion group as between-subject factors . In addition to their effects on value learning , lesions to both the BLA and ventral OFC induced an uncertainty-avoidant phenotype with animals in both experimental groups demonstrating reduced preference for the HV option under baseline conditions compared to the control group at the beginning of testing [time x group interaction: F ( 4 . 37 , 45 . 87 ) = 8 . 484 , p<0 . 0001; post hoc sham vs BLA: p=0 . 002; sham vs OFC: p=0 . 002 , Figure 6] . BLA-lesioned animals continued to avoid the uncertain option for the entire duration of our experiment ( all p values < 0 . 05 , except for baseline three assessment when this group was not different from control animals ) . However , OFC-lesioned animals increased their choices of the HV option during baseline conditions with repeated testing: they were indistinguishable from controls during baselines 3 and 4 and even demonstrated a trend for higher preference than the control group during the last baseline [post hoc test , OFC vs sham: p=0 . 059] . 10 . 7554/eLife . 27483 . 018Figure 6 . BLA and ventral OFC lesions induce uncertainty-avoidance . We observed significantly reduced preference for the HV option under baseline conditions in both experimental groups compared to control animals at the beginning of testing ( sham vs BLA: p=0 . 002; sham vs OFC: p=0 . 002 ) . BLA-lesioned animals continued to avoid the risky option for most of the experiment ( all p values < 0 . 05 , except for baseline three assessment when this group was not different from control animals ) . OFC-lesioned animals progressively increased their choices of HV option during baseline conditions with repeated testing: they were indistinguishable from controls during baselines 3 and 4 and even demonstrated a trend for higher preference than control group during the last baseline [post hoc test , OFC vs sham: p=0 . 059] . The data are shown as group means by condition ±SEM , *p<0 . 05 , **p<0 . 01 . Summary statistics and individual animal data are provided in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 01810 . 7554/eLife . 27483 . 019Figure 6—source data 1 . Summary statistics and individual data for HV option preference following lesions . DOI: http://dx . doi . org/10 . 7554/eLife . 27483 . 019
One of the most difficult challenges faced by an animal learning in an unstable habitat is correctly distinguishing between true changes in the environment that require new learning from stochastic feedback under mostly stable conditions . Indeed , the problem of change-point detection has long been studied in relation to modulation of learning rates in RL and Bayesian learning theory ( Behrens et al . , 2007; Courville et al . , 2006; Dayan et al . , 2000; Gallistel et al . , 2001; Pearce and Hall , 1980; Pearson and Platt , 2013; Yu and Dayan , 2005 ) . Long-term neuroadaptations in response to experience with outcome uncertainty may benefit learning by altering signal-to-noise processing ( Hoshino , 2014; Liguz-Lecznar et al . , 2015; Rössert et al . , 2011 ) , such that only those surprising events that exceed the levels of expected variability in the environment produce neuronal responses and affect behavior . We directly assessed the changes in expression of gephyrin ( a reliable proxy for membrane-inserted GABAA receptors mediating fast inhibitory transmission; [Chhatwal et al . , 2005; Tyagarajan et al . , 2011] ) and GluN1 ( an obligatory subunit of glutamate NMDA receptors; [Soares et al . , 2013] ) in BLA and ventral OFC in three separate groups of animals following prolonged experience with low and high levels of expected uncertainty in outcome distribution . Both gephyrin and GluN1 showed robust uncertainty-dependent upregulation in the BLA that was maximal after experience with highly uncertain conditions . Conversely , within the ventral OFC , gephyrin was downregulated following reward experience in general and did not depend on the degree of uncertainty in outcomes . However , our experiments did not include a certain control group ( i . e . , animals receiving rewards following a predictable delay on all trials ) . Therefore , we cannot exclude the possibility that changes in protein expression in the OFC in response to reward experience required some , albeit small , levels of outcome uncertainty . Adaptations to expected uncertainty at the protein level are likely to diminish responses to subsequent trial-by-trial surprise signals in BLA . Concurrent increases in the sensitivity to excitation and inhibition benefit signal-to-noise processing , providing further evidence in support of this view ( Hoshino , 2014; Liguz-Lecznar et al . , 2015; Rössert et al . , 2011 ) . To detect environmental changes , animals need to compare current prediction errors to the levels of expected outcome uncertainty . Previous work has shown that GABA-ergic interneurons in BLA gate the information flow and determine the signal intensity that is passed to postsynaptic structures ( Wolff et al . , 2014 ) . The intrinsic excitability of pyramidal neurons ( Motanis et al . , 2014; Paton et al . , 2006 ) and activity of interneurons in the BLA are shaped by reward experiences , possibly via a dopamine-dependent mechanism ( Chu et al . , 2012; Merlo et al . , 2015 ) . Upregulation of functional GABAA receptors as suggested by our data may decrease sensitivity to surprising events when outcome variability is high even under mostly stable conditions , while increases in GluN1 could support learning facilitation when the environment changes . Several psychiatric conditions such as anxiety , schizophrenia , obsessive compulsive and autism spectrum disorders , share pathological uncertainty processing as a core deficit , manifesting as a preference for stable , certain outcomes ( Winstanley and Clark , 2016a , 2016b ) . Interestingly , recent studies have similarly implicated mutations in the gephyrin gene as risk for autism and schizophrenia ( Chen et al . , 2014; Lionel et al . , 2013 ) . Future research may address the role of this synaptic organizer in surprise-driven learning and decision making under uncertainty in animal models of these disorders . Contrary to the pattern of neuroadaptations observed in BLA , gephyrin in the OFC was downregulated in response to reward mean , but not expected uncertainty . These changes in protein expression may leave OFC responsivity to noisy value signals intact or even amplified , suggesting that one of its normal functions is to encode the richness of the outcome distribution or expected uncertainty signal . Indeed , previous reports demonstrated that at least some subpopulations of OFC neurons carry expected uncertainty representations during option evaluation and outcome receipt ( Li et al . , 2016; van Duuren et al . , 2009 ) . Based on these findings we hypothesized that BLA and ventral OFC may play complementary , yet dissociable , roles in decision making and learning under uncertainty . Lesions to the ventral OFC produced a pronounced behavioral impairment on our task . These animals failed to change their choice preference in response to the majority of shifts . Paradoxically , the results of computational modeling revealed that responsivity to surprising outcomes was facilitated in these rats . Specifically , performance of OFC-lesioned animals was best accounted for by the PH model , suggesting that while these animals still increased learning rates in response to surprise ( i . e . , absolute prediction errors ) , they were insensitive to expected outcome uncertainty . Due to the lack of prediction error scaling based on the variability in experienced outcomes , OFC-lesioned animals treated every surprising event as indicative of a fundamental change in the value distribution and updated their expectations , rendering trial-by-trial value representations noisier , preventing consistent changes in preference . Because the delay distributions encountered during baseline and shift conditions in our task partially overlapped , inability to ignore meaningless fluctuations in outcomes would lead to unstable choice behavior and attenuated learning . Complementary analyses of win-stay/ lose-shift strategy provide further support for potentiated sensitivity to surprising feedback in these animals: increased responsivity to both wins and losses emerged following ventral OFC lesions . Note that increased reliance on this strategy is highly suboptimal under stochastic environmental reward conditions ( Faraut et al . , 2016; Imhof et al . , 2007; Worthy et al . , 2013 ) . Furthermore , we observed significantly reduced β ( inverse temperature parameter in softmax decision rule ) values in OFC-lesioned group , indicating a noisier choice process and decreased reliance on learned outcome values in these animals . These results are in agreement with previous findings demonstrating increased switching and inconsistent economic preferences following ventral OFC lesions in monkeys ( Walton et al . , 2010 , 2011 ) . Similarly , lesions to ventromedial prefrontal cortex , encompassing the ventral OFC , in humans render subjects unable to make consistent preference judgments ( Fellows and Farah , 2003 , 2007 ) . Importantly , human subjects with OFC damage cannot distinguish between degrees of uncertainty ( Hsu et al . , 2005 ) . Similarly , previous work has implicated this brain region in prediction of reward timing ( Bakhurin et al . , 2017 ) . We directly addressed whether BLA and ventral OFC lesions alter animals’ ability to form expectations about the expected uncertainty in timing of reward delivery on our task . Whereas the distributions of reward-seeking times in BLA-lesioned animals were indistinguishable from control animals’ and the true delays , OFC-lesioned animals concentrated their reward port entries in the time interval corresponding to mean delays , suggesting that while these animals can infer the average outcomes , they fail to represent the variance ( i . e . , expected uncertainty ) . These findings are consistent with emerging evidence that more ventromedial regions , unlike lateral , OFC may be critical for decision making involving outcome uncertainty , but not response inhibition or impulsive choice behavior as suggested previously ( Stopper et al . , 2014 ) . Although frequently framed as a deficit in inhibitory control ( Bari and Robbins , 2013; Dalley et al . , 2004; Elliott and Deakin , 2005 ) , medial OFC lesions or inactivations induce analogous effects in probabilistic reversal learning tasks where surprising changes in the reward distribution occur against the background of stochastic outcomes during the baseline conditions . For example , a recent study in rodents systematically compared the contributions of five different regions of the frontal cortex to reversal learning ( Dalton et al . , 2016 ) . The results revealed unique contributions of the OFC to successful performance under probabilistic , but not deterministic conditions . Intriguingly , inactivations of the medial OFC impaired both the acquisition and reversal phases , suggesting that this subregion might be critical for many types of reward learning under conditions of expected outcome uncertainty . Since our lesions also intruded on medial OFC , our present observations are in agreement with these findings and suggest that one of the normal functions of more ventromedial sectors of OFC might be to stabilize value representations by adjusting responses to surprising outcomes based on expected outcome uncertainty . Similar to previous work demonstrating that the OFC is not required for acquisition of simple stimulus-outcome associations or for unblocking driven by differences in value when outcomes are certain and predictable ( Izquierdo et al . , 2004; McDannald et al . , 2011 , 2005; Stalnaker et al . , 2015 ) , we observed intact performance in OFC-lesioned animals during training to respond for rewards . It has been previously proposed that the OFC may provide value expectations that can be used to calculate RPEs to drive learning under more complex task conditions ( Schoenbaum et al . , 2011a , Schoenbaum et al . , 2011b ) . Although this initial proposal was based on findings after targeting more lateral OFC subregions , our observations are generally consistent with this view and add a nuanced perspective . Specifically , if the OFC is needed to provide expectations about the value to which the observed outcomes are then compared , lesions of this brain region may result in attenuated learning driven by violation of expectations . The results of computational modeling in our work revealed a reduction in learning rates in OFC-lesioned animals consistent with this account . Yet our data provide further evidence that the representations of expected outcomes in ventral OFC are not limited to a single-point estimate of value , but also include information about expected uncertainty of variability in outcomes . This would allow an animal not only to detect if the outcomes violate expectations , but also to assess whether such surprising events are meaningful and informative to the current state of the world . If such events are important , an animal will shift its behavior , but if they may have occurred by chance , choices should remain unchanged . Finally , more recently it has also been suggested that the OFC represents an animal’s current location within an abstract cognitive map of the task it is facing ( Chan et al . , 2016; Schuck et al . , 2016; Wilson et al . , 2014 ) , particularly when task states are not signaled by external sensory information , but rather need to be inferred from experience . In our task , animals may similarly represent different conditions , stable environment vs . shifted value , as separate states . Attenuated learning may result from state misrepresentations , where an animal incorrectly infers that it is currently in a stable environment and persists with the previous choice policy , despite the shift in value . As has been reported recently , neuronal activity in the lateral OFC organizes the task space according to the sequence of behaviorally significant events , or trial epochs . Conversely , neural ensembles in more medial OFC do not track the sequence of the events , but instead segregate between states depending on the trial value ( Lopatina et al . , 2017 ) . In our study , ventromedial OFC may be especially well-positioned to encode upshifts and downshifts in value on long timescales , and loss of this function could cause an inability to recover appropriate state representations at the time of option choice . Taken together with previous findings , our results implicate the OFC in representing fine-grained value distributions , including the expected uncertainty in outcomes ( that may be task state-dependent ) . Consequently , lacking access to the complex outcome distribution , animals with OFC lesions over-rely on the average cached value . Whereas OFC lesions produced a pronounced impairment in performance on our uncertainty task , whether alterations induced by BLA lesions lead to suboptimal behavior is less clear . These animals changed their option preference to a lesser degree on HV upshifts , but compensated by exaggerated adaptations to HV downshifts . More detailed analyses of session-by-session data revealed specific alteration in responses to surprising value shifts under HV , but not LV , conditions in this group . Consistent with attenuated learning observed in animals with BLA lesions , trial-by-trial performance in this group was best fit by a RW+expected uncertainty model , demonstrating a selective loss of learning rate scaling in response to surprise and preserved RPE scaling with expected outcome uncertainty , leading to slower value learning compared to intact animals during the HV upshift . Note that suboptimal performance during even two or three sessions in our task ( each session lasting 60 trials ) means that BLA-lesioned animals are less efficient at reward procurement for 120–180 experiences . In naturalistic settings , such an early-learning impairment can have detrimental consequences . In agreement with the results of computational modeling , BLA-lesioned animals were less likely to adopt the win-stay/lose-shift strategy compared to the control group , demonstrating decreased sensitivity to surprising outcomes . Whereas the lack of learning facilitation can account for reduced changes in preference in response to HV upshifts in BLA-lesioned animals , it may seem at odds with potentiated responses to downshifts on this option . Our computational modeling results suggest that control animals potentiate their learning in response to highly surprising outcomes , which leads to greater behavioral adaptations in the first few sessions during the shifts . In BLA-lesioned animals , this function is lost , and learning proceeds at the same rate . This results in significantly reduced choice adaptations throughout the HV upshift sessions . Yet BLA-lesioned animals adapt much more to the downshift on HV option . This difference appears to be in the performance asymptote , as learning still progresses linearly in BLA-lesioned group . A couple of factors may drive this effect . Firstly , as discussed earlier hyperbolic discounting leads to a larger impact of short delays on behavior . Immediate or short-delayed rewards encountered during upshift on HV option will potentiate learning in control animals early on during the shift , but fail to do so in BLA lesioned animals . During the downshift on HV option , as delays get longer , differences in waiting times become less meaningful as there is a smaller effect of larger delays on perceived outcome values . Thus , learning will be potentiated , but only briefly in control animals , but will still proceed linearly in BLA-lesioned rats . Additionally , potentiated responses to downshifts on HV option in this group may result from uncertainty avoidance interacting with surprise-driven learning . Indeed , we observed a consistent increase in uncertainty aversion in BLA-lesioned animals . Our computational models did not include an explicit uncertainty avoidance parameter as we were primarily interested in exploring alterations in learning . Previous findings have implicated the BLA in updating reward expectancies when the predictions and outcomes are incongruent and facilitating learning in response to surprising events ( Ramirez and Savage , 2007; Savage et al . , 2007; Wassum and Izquierdo , 2015 ) . Indeed , predictive value learning in the amygdala involves a neuronal signature that accords with an RL algorithm ( Dolan , 2007 ) . Specifically , single-unit responses in the BLA correspond to the unsigned prediction error signals ( Roesch et al . , 2010 ) that are necessary for learning rate scaling in both RL and Bayesian updating models . The BLA utilizes positive and negative prediction errors to boost cue processing , potentially directing attention to relevant stimuli and potentiating learning ( Chang et al . , 2012; Esber and Holland , 2014 ) as demonstrated in downshift procedures with reductions in reward amount . These effects are frequently interpreted as surprise-induced enhancement of cue associability . Notably , a similar computational role for the amygdala has been proposed based on Pavlovian fear conditioning in humans , where cue-shock associations were also probabilistic , highlighting the general role for the amygdala in fine-tuning learning according to the degree of surprise ( Li et al . , 2011 ) . Taken together , the accumulated literature suggests that this contribution of the BLA is apparent for both appetitive and aversive outcomes , for cues in different sensory modalities , and as we demonstrate here , the role is not limited to changes in outcome contingencies , but also supports learning about surprising changes in delay costs . In addition to their effects on value learning , lesions to both the BLA and ventral OFC induced an uncertainty-avoidant phenotype with animals in both experimental groups demonstrating reduced preference for the HV option under baseline conditions compared to control group at the beginning of testing . Similarly , previous findings demonstrated that lesions or inactivations of the BLA shift the behavior away from uncertain options and promote choices of safer outcomes ( Ghods-Sharifi et al . , 2009; Zeeb and Winstanley , 2011 ) . However , inactivations of the medial OFC have been shown to produce consistent shifts toward the uncertain option ( Winstanley and Floresco , 2016b ) . Despite demonstrating pronounced risk-aversion at the beginning of the task , OFC-lesioned animals in our experiments progressively increased their preference for HV option with experience , suggesting that the effects on stable choice preference depend critically on the timing of OFC manipulations . In summary , we show that both BLA and ventral OFC are causally involved in decision making and value learning under conditions of outcome uncertainty . Functionally-intact BLA is required for facilitation of learning in response to surprise , whereas ventral OFC is necessary for an accurate representation of outcome distributions to stabilize value expectations and maintain choice preferences .
Behavioral training was conducted in operant conditioning chambers ( Model 80604 , Lafayette Instrument Co . , Lafayette , IN ) that were housed within sound- and light- attenuating cubicles . Each chamber was equipped with a house light , tone generator , video camera , and LCD touchscreen opposing the pellet dispenser . The pellet dispenser delivered 45 mg dustless precision sucrose pellets . Software ( ABET II TOUCH; Lafayette Instrument Co . , Model 89505 ) controlled the hardware . All testing schedules were programmed by our group and can be requested from the corresponding author . During habituation , rats were required to eat five sugar pellets out of the dispenser inside of the chambers within 15 min before exposure to any stimuli on the touchscreen . They were then trained to respond to visual stimuli presented in the central compartment of the screen within 40 s time interval in order to receive the sugar reward . During the next stage of training , animals learned to initiate the trial by nose-poking the bright white square stimulus presented in the central compartment of the touchscreen within 40 s; this response was followed by disappearance of the central stimulus and presentation of a target image in one of the side compartments of the touchscreen ( immediately to the left or right of the initiation stimulus ) . Rats were given 40 s to respond to the target image , which was followed by an immediate reward . The last stage of training was administered to familiarize animals with delayed outcomes . The protocol was identical to the previous stage , except the nosepoke to the target image and reward delivery were separated by a 5 s stable delay . Across all stages of pre-training , failure to respond to a visual stimulus within the allotted time resulted in the trial being scored as an omission and beginning of a 10 s ITI . All images used in pre-training were pulled from the library of over 100 visual stimuli and were never the same as the images used in behavioral testing described below . This was done to ensure that none of the visual cues acquired incentive value that could affect subsequent performance . Criterion for advancement into the next stage was set to 60 rewards collected within 45 min . Task design and behavior of intact animals are illustrated in Figure 1 , Video 1 and Video 2 . Our task is designed to assess the effects of expected outcome uncertainty on learning . We have elected to focus on reward rate ( outcome value was determined by delay to reward receipt ) rather than reward magnitude to avoid the issue of satiety throughout the testing session . Each trial began with stimulus ( bright white square ) presentation in the central compartment of the touchscreen . Rats were given 40 s to initiate a trial . If 40 s passed without a response , the trial was scored as an ‘initiation omission’ . Following a nosepoke to the central compartment , the central cue disappeared and two choice stimuli were presented concurrently in each of the side compartments of the touchscreen allowing an animal a free choice between two reward options . In our task stimulus-response side assignments were held constant for each animal to facilitate learning . Side-stimulus assignments were counterbalanced across animals , and held constant between sessions . Each response option was associated with the delivery of one sugar pellet after a delay interval . The delays associated with each option were pooled from distributions that are identical in mean value , but different in variability ( LV vs HV; ~N ( µ , σ ) : μ = 10 s , σHV=4s σLV=1s ) . An animal was given 40 s to make a choice; failure to select an option within this time interval resulted in the trial being scored as ‘choice omission’ and beginning of an ITI . Therefore , rats were presented with two options identical in mean ( 10 s ) but different in the variance of the delay distribution ( i . e . , expected outcome uncertainty ) . Following the establishment of stable performance ( defined as no statistical difference in any of the behavioral parameters across three consecutive testing sessions ) , rats experienced reward upshifts ( delay mean was reduced to 5 s with variance kept constant ) and downshifts ( 20 s ) on each option independently , followed by return to baseline conditions . Thus , in upshifts rats were required to wait less on average for a single sugar pellet , whereas in downshifts , rats were required to wait longer , on average . The order of shift experiences was counterbalanced across animals . Animals were given one testing session per day that was terminated when an animal had collected 60 rewards or when 45 min had elapsed . Each shift and return to baseline phase lasted for five sessions . Therefore , rats experienced a total number of 43 sessions with varying delays . We first trained a group of naïve rats ( n = 8 ) on this task to probe the ability to distinguish true changes in the environment from stochastic fluctuations in outcomes under baseline conditions in rodents . The animals in lesion experiments ( n = 24: n sham = 8 , n BLA lesion = 8; n OFC lesion = 8 ) were tested under identical conditions . Each animal participated in a single experiment . For each experiment , rats were randomly assigned into groups . Three separate groups of animals were trained to respond to visual stimuli on a touchscreen to procure a reward after variable delays . The values of outcomes were identical to our task described above but no choice was given . One group was trained under LV conditions , the second under HV ( matched in total number of rewards received ) , and the third control group received no rewards ( n = 8 in each group; total n = 24 ) . The training criterion was set to a 60 sugar pellets for three consecutive days to mimic the baseline testing duration in animal trained on our main task . Rats were euthanized 1d after the last day of reward experience with an overdose of sodium pentobarbital ( 250 mg/kg , i . p . ) and decapitated . The brains were immediately extracted and two mm-thick coronal sections of ventral OFC and BLA were further rapidly dissected , using a brain matrix , over wet ice at 4°C . To prepare the tissues for the assays 0 . 2 mL of PBS ( 0 . 01 mol/L , pH 7 . 2 ) containing a protease and phosphatase inhibitor cocktail ( aprotinin , bestatin , E-64; leupeptin , NaF , sodium orthovanadate , sodium pyrophosphate , β-glycerophosphate; EDTA-free; Thermo Scientific , Rockford , IL; Product # 78441 ) was added to each sample . Each tissue was minced , homogenized , sonicated with an ultrasonic cell disrupter , and centrifuged at 5000 g at 4°C for 10 min . Supernatants were removed and stored at +4°C until ELISA assays were performed ( within 24 hr period ) . Bradford protein assays were also performed to determine total protein concentrations in each sample . The assays were performed according to the manufacturer’s instructions . The sensitivity of the assays is 0 . 1 ng/ml for gephyrin ( Cat# MBS9324933 ) and GluN1 ( Cat# MBS724735 , MyBioSource , Inc , San Diego , CA ) and the detection range is 0 . 625 ng/ml - 20 ng/ml . The concentration of each protein was quantified as ng/mg of total protein accounting for dilution factor and presented as percent of no reward group . Excitotoxic lesions of BLA ( n = 8 ) and ventral OFC ( n = 8 ) were performed using aseptic stereotaxic techniques under isoflurane gas ( 1–5% in O2 ) anesthesia prior to behavioral testing and training . Before surgeries , all animals were administered 5 mg/kg s . c . carprofen ( NADA #141–199 , Pfizer , Inc . , Drug Labeler Code: 000069 ) and 1cc saline . After being placed into a stereotaxic apparatus ( David Kopf; model 306041 ) , the scalp was incised and retracted . The skull was then leveled to ensure that bregma and lambda are in the same horizontal plane . Small burr holes were drilled in the skull to allow cannulae with an injection needle to be lowered into the BLA ( AP: −2 . 5; ML: ± 5 . 0; DV: −7 . 8 ( 0 . 1 μl ) and −8 . 1 ( 0 . 2 μl ) from skull surface ) or OFC ( 0 . 2 μl , AP =+3 . 7; ML = ±2 . 0; DV = −4 . 6 ) . The injection needle was attached to polyethylene tubing connected to a Hamilton syringe mounted on a syringe pump . N-Methyl-D-aspartic acid ( NMDA , Sigma-Aldrich; 20 mg/ml in 0 . 1 m PBS , pH 7 . 4; Product # M3262 ) was bilaterally infused at a rate of 0 . 1 μl/min to destroy intrinsic neurons . After each injection , the needle was left in place for 3–5 min to allow for diffusion of the drug . Sham-lesioned group ( n = 8 ) underwent identical surgical procedures , except no NMDA was infused . All animals were given one-week recovery period prior to food restriction and subsequent behavioral testing . During this week , the rats were administered 5 mg/kg s . c . carprofen ( NADA #141–199 , Pfizer , Inc . , Drug Labeler Code: 000069 ) and their health condition was monitored daily . The extent of the lesions was assessed by staining for NeuN , a marker for neuronal nuclei . After the termination of training , animals were sacrificed by pentobarbital overdose ( Euthasol , 0 . 8 mL , 390 mg/mL pentobarbital , 50 mg/mL phenytoin; Virbic , Fort Worth , TX ) and transcardial perfusion . Brains were post-fixed in 10% buffered formalin acetate for 24 hr followed by 30% sucrose for 5 days . Forty µm coronal sections containing the OFC and BLA were first incubated for 24 hr at 4°C in solution containing primary NeuN antibody ( Anti-NeuN ( rabbit ) , 1:1000 , EMD . Millipore , Cat . # ABN78 ) , 10% normal goat serum ( Abcam , Cambridge , MA , Cat . # ab7481 ) , and 0 . 5% Triton-X ( Sigma , St . Louis , MO , Cat . # T8787 ) in 1X PBS , followed by three 10 min washes in PBS . The tissue was then incubated for 4 hr in solution containing 1X PBS , Triton-X and a secondary antibody ( Goat anti-Rabbit IgG ( H+L ) , Alexa Fluor 488 conjugate , 1:400 , Fisher Scientific , Catalog #A-11034 ) , followed by three 10 min washes in PBS . Slides were subsequently mounted and cover-slipped , visualized using a BZ-X710 microscope ( Keyence , Itasca , IL ) , and analyzed with BZ-X Viewer software . Lesions were determined by comparison with a standard rat brain atlas ( Paxinos and Watson , 1997 ) . We fit different versions of reinforcement learning models to trial-by-trial choices for each animal separately . Specifically , we considered the standard Rescorla-Wagner model ( RW ) and a dynamic learning rate model ( Pearce-Hall , PH ) . Trials from all sessions were treated as contiguous . Option values were updated in response to RPE , δt , weighted by the learning rate , α ( constrained to the interval [0 1] ) . The expected value for each option was updated according to delta rule:Qt+1←Qt+α∗δt . The δt is the difference between current outcome Vt and expected value Qt . Given that the value of each outcome was determined by delay to reward of a constant magnitude , Vt was specified as 1/ ( 1-kD ) , where D is the duration of delay and k [0 , +∞] is a free parameter setting the steepness of the discounting curve . In dynamic learning rate models ( PH and PH+expected uncertainty described below ) , α was updated in response to the degree of surprise ( absolute δt ) according to:αt+1←|δt|∗η+ ( 1−η ) ∗αt . We set initial α for HV and LV options to the same value , but allowed independent updating with experience . We also considered models in which expected outcome uncertainty is learned simultaneously with value and scales the impact of prediction errors on value ( RW+expected uncertainty ) and learning rate ( Full model ) updating . Uncertainty prediction errors are the difference between squared expected and realized RPEs . Expected uncertainty expectations are subsequently updated according to delta rule . Therefore , in the Full model:Qt+1←Qt+αt∗δt/ω∗exp ( σt′ ) ; where ω [1 , +∞] is a free parameter determining individual sensitivity to expected uncertainty . αt+1←η∗|δt|/ω∗exp ( σt′ ) + ( 1−η ) ∗αt . αt+1′←σt′+αrisk∗δrisk , t;δrisk , t=δt2−δt′ The option choice probability on each trial was determined according to a softmax rule with an inverse-temperature parameter β; ∝ exp ( β*Qt ) . The model parameters were estimated to maximize probability of obtaining the observed vector of choices given the model and its parameters ( by minimizing negative log likelihood computed based on the difference between predicted choice probability and the actual choice on each trial using fmincon in MatLab ) . We used Bayesian information criterion ( BIC ) instead of AIC as a more conservative measure to determine the best model . The total number of free parameters , BIC and parameter values for each model and experimental group are provided in Table 1 . Software package SPSS ( SAS Institute , Inc . , Version 24 ) and MatLab ( MathWorks , Natick , Massachusetts; Version R2016b ) were used for statistical analyses . Statistical significance was noted when p-values were less than 0 . 05 . Shapiro Wilk tests of normality , Levene's tests of equality of error variances , Box's tests of equality of covariance matrices , and Mauchly's tests of sphericity were used to characterize the data structure . Protein expression data were analyzed with univariate ANOVA with reward experience group ( HV , LV , or no reward ) as the between-subject factor . Maximal changes in choice of each option in response to shifts were analyzed with omnibus ANOVA with shift type ( HV , LV; upshift , downshift ) and shift phase ( pre-baseline , shift , post-baseline ) as within-subject factors ( total number of animals , n , in this analysis = 8 ) . Similar analyses were performed on data obtained from lesion experiments with an additional between-subject factor of experimental group ( sham , BLA vs OFC lesion; total n = 24 , n = 8 per group ) . Furthermore , we subjected the session-by-session data during each swift to an omnibus ANOVA with testing session ( 1 through 5 ) and shift type as within- and experimental group as between subject factors . | Nobody likes waiting – we opt for online shopping to avoid standing in lines , grow impatient in traffic , and often prefer restaurants that serve food quickly . When making decisions , humans and other animals try to maximize the benefits by weighing up the costs and rewards associated with a situation . Many regions in the brain help us choose the best options based on quality and size of rewards , and required waiting times . Even before we make decisions , the activity in these brain regions predicts what we will choose . Sometimes , however , unexpected changes can lead to longer waiting times and our preferences suddenly become less desirable . The brain can detect such changes by comparing the outcomes we anticipate to those we experience . When the outcomes are surprising , specific areas in the brain such as the amygdala and the orbitofrontal cortex help us learn to make better choices . However , as surprising events can occur purely by chance , we need to be able to ignore irrelevant surprises and only learn from meaningful ones . Until now , it was not clear whether the amygdala and orbitofrontal cortex play specific roles in successfully learning under such conditions . Stolyarova and Izquierdo trained rats to select between two images and rewarded them with sugar pellets after different delays . If rats chose one of these images they received the rewards after a predictable delay that was about 10 seconds , while choosing the other one produced variable delays – sometimes the time intervals were either very short or very long . Then , the waiting times for one of the alternatives changed unexpectedly . Rats with healthy brains quickly learned to choose the option with the shorter waiting time . Stolyarova and Izquierdo repeated the experiments with rats that had damage in a part of the amygdala . These rats learned more slowly , particularly when the variable option changed for the better . Rats with damage to the orbitofrontal cortex failed to learn at all . Stolyarova and Izquierdo then examined the rats’ behavior during delays . Rats with damage to the orbitofrontal cortex could not distinguish between meaningful and irrelevant surprises and always looked for the food pellet ( i . e . anticipated a reward ) at the average delay interval . These findings highlight two brain regions that help us distinguish meaningful surprises from irrelevant ones . A next step will be to examine how the amygdala and orbitofrontal cortex interact during learning and see if changes to the activity of these brain regions may affect responses . Advanced methods to non-invasively manipulate brain activity in humans may help people who find it hard to cope with changes; or individuals suffering from substance use disorders , who often struggle to give up drugs that provide them immediate and predictable rewards . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Complementary contributions of basolateral amygdala and orbitofrontal cortex to value learning under uncertainty |
Changes in expression patterns may occur when organisms are presented with new environmental challenges , for example following migration or genetic changes . To elucidate the mechanisms by which the translational machinery adapts to such changes , we perturbed the tRNA pool of Saccharomyces cerevisiae by tRNA gene deletion . We then evolved the deletion strain and observed that the genetic adaptation was recurrently based on a strategic mutation that changed the anticodon of other tRNA genes to match that of the deleted one . Strikingly , a systematic search in hundreds of genomes revealed that anticodon mutations occur throughout the tree of life . We further show that the evolution of the tRNA pool also depends on the need to properly couple translation to protein folding . Together , our observations shed light on the evolution of the tRNA pool , demonstrating that mutation in the anticodons of tRNA genes is a common adaptive mechanism when meeting new translational demands .
The process of gene translation is fundamental to the function of living cells , and as such its apparatus is highly conserved across the tree of life ( Müller and Wittmann-Liebold , 1997; Itoh et al . , 1999; Wolf et al . , 2001 ) . Yet , the capacity of the translation machinery to adaptively evolve is crucial in order to support life in changing environments . Therefore , a key open question is to identify the mechanisms by which the translation machinery adapts to changing conditions . A thoroughly studied aspect of translation that demonstrates its adaptation capacities is the different proportions by which synonymous codons are used , a phenomenon known as ‘codon usage bias’ . Although differential use of codons can be the result of neutral processes such as mutational biases and the genomic GC content ( Urrutia and Hurst , 2001; Rao et al . , 2011 ) , natural selection also influences codon usage bias . Indeed , it has been demonstrated that codon choice affects expression level , protein folding , translational accuracy , and other translational features ( Akashi , 1994; Parmley and Hurst , 2007; Zhou et al . , 2009; Hudson et al . , 2011 ) . Since both neutral and selective processes govern codon usage bias , the balance between selection , mutational bias and drift is crucial in shaping the codon usage of each species ( Bulmer , 1991 ) . Importantly , although the selective advantage offered by alternative synonymous codons is considered to be moderate , it was recently demonstrated that selection can still shape codon usage patterns in vertebrates even with their small effective population sizes ( Doherty and McInerney , 2013 ) . Notably , the differential usage of codons represents the evolution of the ‘demand’ aspect of translation , namely the codon usage of all expressed genes . Yet , the adaptation mechanisms of the ‘supply’ , namely the expression level of each tRNA type that is loaded with an amino acid , are not fully understood . While ribosomal genes do not exhibit appreciable changes in response to environmental alterations ( Müller and Wittmann-Liebold , 1997; Itoh et al . , 1999; Wolf et al . , 2001 ) , tRNA genes may provide an important source of evolutionary plasticity for fine tuning translation . tRNAs constitute a fundamental component in the process of translation , linking codons to their corresponding amino acids ( Widmann et al . , 2010 ) . tRNA genes are classified into gene families according to their anticodon , with each gene family containing between one and several copies scattered throughout the genome . Importantly , it has been experimentally observed for Saccharomyces cerevisiae ( Tuller et al . , 2010 ) and Escherichia coli ( Dong et al . , 1996 ) that the cellular concentrations of each tRNA family in the cell ( i . e . , the tRNA pool ) correlate with its genomic tRNA copy number ( Percudani et al . , 1997; Kanaya et al . , 1999 ) . Notably , the rate-limiting step of polypeptide elongation is the recruitment of a tRNA that matches the translated codon ( Varenne et al . , 1984 ) . Thus , the translation efficiency is defined as the extent to which the tRNA pool can accommodate the transcriptome ( Sharp and Li , 1987; Dos Reis et al . , 2004; Stoletzki and Eyre-Walker , 2007 ) , thereby affecting protein production and accuracy . In general , highly expressed genes exhibit a marked codon usage bias toward ‘optimal’ codons , whose corresponding tRNA gene copy number is high ( Sharp and Li , 1986a , 1986b ) . The evolutionary force that acts to maintain optimal translation efficiency of such genes was coined ‘translational selection’ ( Dos Reis et al . , 2004 ) . It was previously suggested that translational selection acts to maintain a balance between codon usage and tRNA availability . On the one hand , there is a selective pressure to increase the frequency of preferred codons in highly expressed genes . On the other hand , changes in the tRNA pool may also occur , for example duplication of tRNA genes for which high codon demand exists . Thus , codon frequencies and tRNA copy numbers coevolve toward a supply versus demand balance that facilitates optimal protein production ( Higgs and Ran , 2008; Gingold et al . , 2012 ) . The fitness effects of an unmet translational demand and its potential role in shaping the tRNA pool are not fully characterized . Evolutionary changes to the tRNA pool were appreciated mainly via bioinformatics studies ( Rawlings et al . , 2003; Withers et al . , 2006; Higgs and Ran , 2008; Bermudez-Santana et al . , 2010; Rogers et al . , 2010 ) and only a handful of experimental findings have been reported , which rely on genetic manipulations ( Byström and Fink , 1989; Von Pawel-Rammingen et al . , 1992; Aström et al . , 1993 ) or direct mutagenesis ( Saks et al . , 1998 ) . Sequence analyses of divergent genomes have demonstrated that both the sequence and copy number of tRNA genes may change among various species or strains . However , it is unclear whether the observed variations in the tRNA pool are a consequence of an adaptive process due to unbalanced translational demand or the result of random genomic processes , as tRNA genes are a known source of genomic instability ( McFarlane and Whitehall , 2009 ) . Further , the forces that direct and maintain low copy tRNA families remain unclear . Specifically , it is not clear whether translational selection acts only to favor optimal codons or also acts in particular cases to keep other codons deliberately as ‘non-optimal’ by maintaining their tRNA supply at low level . Encoding genes with optimal codons might not always lead to higher protein expression levels ( Kudla et al . , 2009 ) . Similarly , the use of ‘slow codons’ may not always result in lower levels of protein expression as they could have functional roles in improving expression , for example when enriched at the beginning of a transcript in order to reduce the energy of the RNA structure ( Goodman et al . , 2013 ) or to efficiently allocate ribosomes along the mRNA ( Tuller et al . , 2010 ) . Additionally , it has been proposed that non-optimal codons may play a role in governing the process of cotranslational folding by slowing down translation , which supports proper folding between domain boundaries ( Thanaraj , 1996; Kramer et al . , 2009; Cabrita et al . , 2010; Wilke and Drummond , 2010; Pechmann and Frydman , 2012 ) . Yet , the contribution of non-optimal codons to proper protein folding was observed only for specific genes ( Crombie et al . , 1992; Komar et al . , 1999; Cortazzo et al . , 2002; Tsai et al . , 2008; Zhang et al . , 2009; Zhou et al . , 2013 ) . Thus , the extent and relevance of this phenomenon to the global folding state of the proteome needs to be substantiated . To elucidate the importance of restoring translational equilibrium , we used an experimental evolution approach . To this end , we genetically perturbed the tRNA pool of the budding yeast S . cerevisiae . In this yeast , the genetic code is decoded by 42 different tRNA families that make up a total of 274 tRNA genes ( Chan and Lowe , 2009 ) . Each tRNA gene family ranges from 1–16 copies , with 6 tRNA families consisting of only a single copy . In a recent study ( Bloom-Ackermann et al . , In press ) , we have systematically manipulated the tRNA pool in S . cerevisiae by individually deleting most tRNA genes from its genome . Here , we focus on one particular deletion strain that showed the most extreme fitness reduction among the viable deletion mutants in this tRNA deletion library . This tRNA exists in only one copy in the genome , thus after its deletion the cell is left without a tRNA with the corresponding anticodon . Lab-evolution experiments performed on this strain demonstrated that the translational balance was rapidly restored by mutations in other tRNA genes that compensated for the tRNA deletion . An extensive bioinformatic analysis revealed that a similar evolutionary trend is widespread in nature too , suggesting that the anticodon mutations we observed in the lab recapitulate an existing mechanism that shapes the tRNA pool . To shed light on the constraints that shape the size of tRNA gene families , we artificially overexpressed singleton tRNAs , rather than deleting them . We found that when low copy tRNAs were overexpressed , the protein quality control machinery was challenged due to increased proteotoxic stress . This observation suggests that low tRNA availability for particular codon can serve an essential means to ensure proper cotranslation folding of proteins .
To demonstrate the importance of the balance between codon usage and the cellular tRNA pool we created a yeast strain in which the single copy of the arginine tRNA gene , tR ( CCU ) J , was deleted ( designated ΔtRNAArgCCU ) . Consequently , in this deletion strain , the arginine codon AGG cannot be translated with its fully matched tRNA and it is presumably translated by another arginine tRNA , tRNAArgUCU , owing to a wobble interaction ( Begley et al . , 2007 ) . This shortage in tRNA supply is particularly evident given the demand: AGG is the second most highly used codon for arginine in the yeast genome ( Supplementary file 1A ) . Indeed , the ΔtRNAArgCCU strain showed a severe growth defect compared to the wild-type strain ( Figure 1A , blue and green curves , respectively ) . This growth difference demonstrates the effect of translational imbalance on cellular growth . Although the deletion mutant of this single copy tRNA is viable ( Clare et al . , 1988; Kawakami et al . , 1993 ) , its severe growth defect also reveals the inability of the wobble interactions to fully compensate for the tRNA gene deletion . 10 . 7554/eLife . 01339 . 003Figure 1 . The growth defect associated with deletion of a singleton tRNA gene was rapidly rescued during the lab-evolution experiment . ( A ) Growth curve measurements of wild-type ( WT ) ( green ) , ΔtRNAArgCCU ( blue ) and the evolved deletion ( red ) are shown in optical density ( OD ) values over time during continuous growth on rich medium at 30°C . ( B ) The mutation that was found to recover the deletion phenotype in the evolved strains is shown on the secondary structure of tRNAArgUCU . The UCU anticodon nucleotides are marked with black circles , and the red circle indicates the mutation that occurred in the anticodon , that is T→C transition . ( C ) MutΔtRNAArgCCU in which the same mutation that was found in the evolved strain was deliberately engineered , exhibits similar growth as the WT . Growth curve measurements of WT ( green ) and of MutΔtRNAArgCCU ( magenta ) are shown in OD600 values over time during continuous growth on rich medium at 30°C . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 003 To learn how genomes adapt to translational imbalances , we performed lab-evolution experiments on the ΔtRNAArgCCU strain , employing the procedure of daily growth and dilution to a fresh medium ( Lenski et al . , 1991 ) . The deletion strain was grown under optimal laboratory conditions ( rich medium at 30°C ) and was diluted daily into a fresh medium by a factor of 120 , corresponding to approximately 7 generations per cycle . Every 50 generations , the growth of the evolving population was compared to both the wild-type and the ancestor ΔtRNAArgCCU strains . Strikingly , after 200 generations we observed a full recovery of the growth defect of the ancestor strain ΔtRNAArgCCU , as the growth curve of the evolved population was indistinguishable from that of the wild-type strain ( Figure 1A , red curve ) . Similar dynamics were observed in all four independent evolutionary lines . In search of the potential genetic adaptations underlying this rapid recovery , we first looked for genetic alterations in other arginine tRNA genes . We found a single point mutation in another arginine tRNA gene that codes for tRNAArgUCU . This mutation changed the anticodon triplet of tRNAArgUCU from UCU to CCU ( i . e . , T→C transition ) . Consequently , the evolved tRNAArgUCU perfectly matched the AGG codon ( Figure 1B ) . Unlike the singleton tRNAArgCCU , there are 11 copies of tRNAArgUCU in the yeast genome . Although each of the 4 independent lab-evolution experiments showed the exact same solution ( that is , a mutation in the anticodon of a tRNAArgUCU gene ) , 3 different copies of this gene were changed among the 4 lines ( i . e . , 1 of the 11 copies of tRNAArgUCU was mutated in 2 repetitions; see ‘Materials and methods’ ) . To confirm that a single point mutation in the anticodon of tRNAArgUCU is sufficient to fully compensate for the growth defect of ΔtRNAArgCCU , we artificially inserted the same T→C mutation into the deletion ΔtRNAArgCCU mutant . We inserted the mutation into 1 of the 11 copies of the tRNAArgUCU genes , a copy that resides on chromosome XI , and thus spontaneously mutated in 1 of the evolution lines . Indeed , the artificially mutated strain , termed as MutΔtRNAArgCCU , showed a full recovery of the deletion adverse phenotype ( Figure 1C ) . This indicates that the T→C mutation in the anticodon is sufficient for the full recovery of the tRNAArgCCU deletion phenotype . In general , all copies of each tRNA gene family tend to be highly similar in sequence in S . cerevisiae ( Chan and Lowe , 2009 ) . In particular , the sequences of the 11 copies of tRNAArgUCU are 100% identical to each other . Yet , the 2 arginine tRNA , tRNAArgUCU , and tRNAArgCCU , differ in 21 of their 72 nucleotides ( including the third anticodon position , Figure 2A ) . Thus , the evolutionary solution that occurred in our experiments created a ‘chimeric’ tRNA with a CCU anticodon , whereas the rest of the tRNA sequence ( termed as the ‘tRNA scaffold’ ) remained as tRNAArgUCU . The sequence identity among all members of the tRNAArgUCU family suggests a functional role for the tRNA scaffold in addition to that of the anticodon ( Schultz and Yarus , 1994; Konevega et al . , 2004; Cochella and Green , 2005; Olejniczak et al . , 2005; Saks and Conery , 2007; Schmeing et al . , 2011 ) . Therefore , it is surprising that the chimeric tRNA performed just as well as the deleted tRNAArgCCU in terms of cell growth , despite the major sequence differences between the two tRNA scaffolds . Thus , we raised the hypothesis that more challenging growth conditions may expose possible inadequacies in the chimeric tRNA . To test this notion , we compared the rescued strain , MutΔtRNAArgCCU , which carries the chimeric tRNA , to the wild-type strain under an array of challenging conditions . Surprisingly , under all the examined conditions , we observed no significant growth difference between the two strains ( Figure 2B ) . Hence , the chimeric tRNA provides a direct in vivo indication that the scaffolds of tRNAs , which encode for the same amino acid , may be interchangeable in terms of their effect on cellular growth under the conditions we tested . 10 . 7554/eLife . 01339 . 004Figure 2 . The growth of MutΔtRNAArgCCU carrying the chimeric tRNA compared to wild-type ( WT ) under different conditions . ( A ) The sequence of the chimeric tRNA is drawn showing the scaffold of tRNAArgUCU with the mutated CCU anticodon . The anticodon triplet is marked with black circles . The evolved mutation is marked with a red circle . All 20 nucleotide differences between tRNAArgUCU and tRNAArgCCU are marked with blue background , next to which , in green letters , the original nucleotide of tRNAArgCCU are written . ( B ) Growth curve measurements of WT ( green ) and of MutΔtRNAArgCCU ( magenta ) are shown in OD600 values over time during continuous growth . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 00410 . 7554/eLife . 01339 . 005Figure 2—figure supplement 1 . Quadruple deletion of tRNAserGCU is lethal . We perturbed the tRNA pool in a wild-type strain by deletion of an entire serine tRNA family . Here , the supply of tRNASerGCU was eliminated by deletion of all four identical gene copies of this family located on chromosomes IV , VI , XII and X . A complete deletion of this gene family is lethal , indicating that the tRNASerGCU is essential in Saccharomyces cerevisiae . To validate that the lethality is indeed due to the deletion of the tRNASerGCU genes and not due to an unintentional perturbation of other putative genetic features in the vicinity of the deleted tRNASerGCU copies , we introduced a plasmid with the tRNASerGCU gene . As expected , the quadruple deletion strain was viable when supplemented with a plasmid carrying the tRNASerGCU gene . ( A and C ) BY384 and tRNASerGCU quadruple deletion strains , respectively , with a plasmid harboring a tRNASerGCU gene grown on yeast extract/peptone/dextrose ( YPD ) plates . ( B and D ) BY384 and tRNASerGCU quadruple deletion strains , respectively , grown on liquid medium that contains uracil ( YPD ) to allow the growth of cells that lost the plasmid and then plated on 5-fluoro-orotic acid ( 5-FOA ) plates ( see ‘Materials and methods’ ) . As expected , normal size colonies were only observed in A , B and C . This means that the quadruple deletion strain with the plasmid could not lose the tRNA gene because of its essentiality . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 00510 . 7554/eLife . 01339 . 006Figure 2—figure supplement 2 . A chimeric serine tRNA can rescue the lethality of the quadruple deletion . Here , rather than inserting a plasmid with the original tRNASerGCU gene , we complemented the tRNASerGCU family deletion strain with a plasmid containing a chimeric serine tRNA with a GCU anticodon . The chimeric tRNA has an alternative scaffold derived from a different serine tRNA , tRNASerCGA , that differs in 22 positions from the deleted serine tRNA family . This strain formed normal size colonies on a yeast extract/peptone/dextrose ( YPD ) plate . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 006 To examine the generality of our observation , we once again perturbed the tRNA pool in a wild-type ( WT ) strain by deletion of an entire serine tRNA family , tRNASerGCU that has four copies in the genome . A complete deletion of this gene family is lethal , indicating that the tRNASerGCU is essential in S . cerevisiae . Although we could not evolve that strain , we did find that this quadruple deletion strain was viable when supplemented with a centromeric plasmid carrying the tRNASerGCU gene . Thus , the lethality is conferred directly from the tRNA loss and is not due to other indirect effects ( Figure 2—figure supplement 1 ) . We hypothesized that , as with tRNAArgCCU , other chimeric serine tRNAs that carry a GCU anticodon , yet with the scaffold of another tRNA for serine , would also prevent the observed lethality . Indeed , a strain carrying a chimeric tRNA with a scaffold of tRNASerCGA and the GCU anticodon is viable on the background of the tRNASerGCU-family deletion ( Figure 2—figure supplement 2 ) . Therefore , we concluded that the identity of the anticodon is essential for the function of the tRNASerGCU gene family . Thus , it appears that for the examined tRNAs the anticodon is a dominant feature in terms of cellular fitness , overshadowing other sequence elements . Although the anticodon of a tRNA gene was rapidly mutated under our laboratory conditions , thus regaining proper translational equilibrium , it is unclear to what extent this mechanism naturally occurs in species across the tree of life . To address this question , we performed a systematic bioinformatics screen for tRNA switching events in nature . We defined an anticodon-switching event as a case of a tRNA whose nucleotide sequence is closer to a tRNA gene with a different anticodon than to a tRNA gene with the same anticodon . To this end , we downloaded all the known tRNA sequences from the Genomic tRNA Database ( Chan and Lowe , 2009 ) , a collection that stores the tRNA pools of 524 species . We masked the anticodon triplet as ‘NNN’ in all tRNA genes , aligned all tRNA sequences from each species individually and inferred a maximum likelihood phylogenetic tree for each alignment . For each tRNA sequence , we calculated the shortest phylogenetic distance to another tRNA with the same anticodon ( designated dsame ) and the shortest distance to another tRNA with a different anticodon ( designated ddiff ) . For each species , we defined its set of tRNA switching events as those in which ddiff<dsame ( see ‘Materials and methods’ , Figure 3—source data 1 ) . Our analysis included 416 eubacterial , 68 eukaryotic , and 40 archaeal species . We found that tRNA switching events are present in all domains of life , as we detected at least 1 tRNA switching event per species in 8 bacteria , 58 eukarya , and 1 archaeal species ( Figure 3A ) . A retrospective counting revealed that most switching events occurred due to a mutation in the first position in the anticodon triplet that corresponds to the third codon position ( see details in ‘Materials and methods’ ) . For comparison , we masked as ‘NNN’ additional triplets of nucleotides within the tRNA molecule , and found a higher percentage of discrepancies compared to the anticodon triplet ( Figure 3—figure supplement 1; Supplementary file 1B ) . 10 . 7554/eLife . 01339 . 007Figure 3 . Anticodon switching is a widespread phenomenon in nature . ( A ) Number of species with at least one tRNA switching event in each domain of life . ( B ) The anticodon UUC convergently evolved in Mus musculus . A maximum likelihood phylogeny of tRNA sequences in M . musculus that decode glutamic acid ( Glu ) codons . Branch lengths express average nucleotide substitutions per site . Decimals on internal branches express branch support . ( C ) A comparison of nucleotide sequences for glutamic acid tRNA genes in M . musculus with anticodon UUC ( top , tRNA1547 and tRNA359 ) , ‘switched’ UUC tRNAs ( middle , tRNA286 and tRNA754 ) , and CUC tRNAs ( bottom , tRNA1002 , tRNA745 , tRNA303 , tRNA999 , tRNA996 tRNA709 , tRNA1001 , tRNA1912 and tRNA81 ) . The anticodon triplet is boxed in gray . Red vertical bars indicate differences between sequences . ( D ) The anticodon UAC convergently evolved in Homo sapiens . A maximum likelihood phylogeny of tRNA sequences in H . sapiens encoding for valine ( Val ) is shown . ( E ) A comparison of nucleotide sequences for H . sapiens tRNAs with anticodons UAC ( top , tRNA6 ) , a ‘switched’ UAC tRNA ( middle , tRNA40 ) , and an AAC tRNA ( bottom , tRNA136 ) . The number of genes is according to the tRNA database . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 00710 . 7554/eLife . 01339 . 008Figure 3—source data 1 . Table of anticodon switchings in different species across the tree of life . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 00810 . 7554/eLife . 01339 . 009Figure 3—figure supplement 1 . A comparison of discrepancy proportions at the anticodon triplet vs control triplets . Six control triplets were chosen; their locations are marked on a tRNA structure . ( A–F ) A comparison of the proportion of switched tRNAs for each species when the anticodon was masked versus the proportion of alternative discrepancies when the control triplet was masked . Each point corresponds to one species . A tRNA was identified as ‘switched’ if its nucleotide sequence phylogenetically clustered with other tRNAs with dissimilar mask triplets ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 009 Figure 3 demonstrates two examples of tRNA switching events , the first in Mus musculus and the second in Homo sapiens . In the first example , the phylogeny of tRNA sequences with glutamic acid anticodons is presented ( Figure 3B ) . Notably , six out of the eight tRNAs with a UUC anticodon in M . musculus were clustered together in our analysis , while two other copies of the same anticodon identity were clustered closer to tRNA genes with a CUC anticodon ( Figure 3C ) . The second example demonstrates a switching event for tRNA genes encoding for valine anticodons . In this study , a tRNA with a UAC anticodon was clustered with CAC and AAC tRNA genes and not with the other four UAC tRNAs ( Figure 3D , E ) . Interestingly , the CAC and AAC tRNA genes are intermixed in the tree , suggesting that anticodon switching was prevalent in the evolution of CAC and AAC tRNA genes in H . sapiens ( Figure 3D , E ) . Also of interest , the switching events shown in mouse were not found in human and vice versa . Thus , in each of these two mammals the switching examples shown here probably occurred after they split from their common ancestor . In general , inspecting the relationship across species between the size of the tRNA pool and the number of detected switching events revealed a modest correlation , and in particular species with same size of tRNA repertoire manifested tRNA switching to different extents ( not shown ) . This analysis suggests that future examination of the tRNA switching phenomenon in individual species could be of interest . After demonstrating the prevalence of anticodon switching , we refocused on our lab-evolution results . The switching events that we observed ( from tRNAArgUCU to tRNAArgCCU ) suggest that the effective gene copy number of each tRNA anticodon set can change during evolution , presumably due to demand-to-supply changes . Given that a single point mutation can functionally convert one tRNA into another , an interesting question emerges: why does the genome maintain a single copy of tRNAArgCCU ? T to C mutations must have occurred in evolution but they appear to have been selected against so as to preserve only a single copy of the CCU anticodon tRNA . Consistent with this hypothesis is the observation that other yeast species maintain tRNAArgCCU at a single copy ( Supplementary file 1C ) . We thus reasoned that an artificial increase in the copy number of a rare tRNA , but not of an abundant one , might result in a deleterious effect on the cells . Indeed , transformation of a multi-copy plasmid containing a tRNAArgCCU gene to a wild-type strain ( termed as WTmultiCCU ) resulted in a substantial growth reduction when compared to wild-type cells carrying an empty multi-copy plasmid ( termed WTmultiControl ) . In contrast , when we created a strain with a similar multi-copy plasmid that contains the abundant tRNAArgUCU gene , designated as WTmultiUCU , a growth profile much closer to that of WTmultiControl was observed ( Figure 4A ) . These findings are consistent with the evolutionary tendency for yeast to keep a low copy number of tRNAArgCCU and suggest that a high copy number of such rare tRNAs is deleterious to cells . 10 . 7554/eLife . 01339 . 010Figure 4 . WTmultiCCU experiences a growth defect compared to WTmultiUCU and demonstrates higher levels of misfolded proteins . ( A ) Growth curve measurements of WTmultiControl ( blue ) , WTmultiUCU ( brown ) and WTmultiCCU ( khaki ) are shown in optical density ( OD ) values over time during continuous growth . The WTmultiCCU strain carrying a high copy number plasmid harboring tRNAArgCCU demonstrates slower growth compared to cells with an empty plasmid or with a tRNAArgUCU plasmid that is mainly characterized by a longer growth delay ( lag phase ) . ( B ) A demonstration of a WTmultiCCU cell in which the mCherry-Von Hippel–Lindau ( VHL ) proteins appear with a punctum phenotype when the protein quality control machinery is saturated with misfolded proteins . ( C ) A demonstration WTmultiUCU cell in which the quality control machinery is not occupied with other proteins; mCherry-VHL is localized to the cytosol . ( D ) WTmultiCCU , WTmultiUCU and WTmultiControl were transformed with a VHL-mCherry containing plasmid and visualized under the microscope; 1000 cells per strain were counted for either cytosolic or punctum localization of the VHL protein . The fold change in the number of cells containing puncta was then deduced by normalization to the WTmultiControl population . The 95% confidence interval is indicated . ( E ) The mRNA fold change of six representative heat-shock genes measured by real-time quantitative PCR ( RT-qPCR ) . Presented values are the mean of two biological repetitions ± SEM . The significance of the fold change differences was examined using a t test , with *p<0 . 001 or **p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 01010 . 7554/eLife . 01339 . 011Figure 4—figure supplement 1 . Multiple copies of rare tRNASerCGA gene are deleterious compared to abundant tRNASerAGA . Growth curve measurements of WTmultiControl ( blue ) , WTmultiAGA ( brown ) and WTmultiCGA ( khaki ) are shown in optical density ( OD ) values over time during continuous growth on rich medium at 30°C . The WTmultiCGA strain carrying a high copy number plasmid harboring tRNASerCGA demonstrates slower growth compared to cells with an empty plasmid or with a tRNASerAGA plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 01110 . 7554/eLife . 01339 . 012Figure 4—figure supplement 2 . Multiple copies of the rare tRNAGlnCUG gene are deleterious compared to abundant tRNAGlnUUG . Growth curve measurements of WTmultiControl ( blue ) , WTmultiUUG ( brown ) and WTmultiCUG ( khaki ) are shown in optical density ( OD ) values over time during continuous growth on rich medium at 30°C . The WTmultiCUG strain carrying a high copy number plasmid harboring tRNAGlnCUG demonstrates slower growth compared to cells with an empty plasmid or with a tRNAGlnUUG plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 01210 . 7554/eLife . 01339 . 013Figure 4—figure supplement 3 . Addition of low copy number tRNAArgCCU is deleterious compared to low copy number tRNAArgUCU when grown in heat . We created two strains harboring low copy number ( CEN ) plasmids carrying either tRNAArgUCU or tRNAArgCCU in addition to the endogenous copy . We termed these strains WTextraUCU and WTextraCCU , respectively . Growth curve measurements of WTextraUCU ( green ) and WTextraCCU ( red ) are shown in optical density ( OD ) values over time during continuous growth on rich medium at 39°C . Cells with the centromeric tRNAArgCCU plasmid showed a modest growth defect compared to cells with the centromeric tRNAArgUCU plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 01339 . 013 To demonstrate the generality of our findings , we employed the same assays in two additional cases . First , we examined 2 serine tRNAs , the singleton tRNASerCGA and tRNASerAGA that is found in the genome in 11 copies . In the second case , we focused on two glutamine tRNAs , the singleton tRNAGlnCUG and tRNAGlnUUG that is found in the genome in nine copies . In both the cases , we observed that the wild-type strain supplemented with multiple copies of a singleton tRNA exhibit impaired growth compared to the same strain supplemented with the abundant tRNA for the same amino acid ( Figure 4—figure supplements 1 and 2 ) . Since the changes in tRNA family sizes during evolution likely occur gradually , perhaps one copy at a time , we also examined the effect of adding low copy number plasmids carrying either tRNAArgUCU or tRNAArgCCU . The cells with the tRNAArgCCU plasmid showed a modest growth defect compared to the cells with tRNAArgUCU plasmid , yet only when grown at 39°C ( Figure 4—figure supplement 3 ) . Why is it essential to keep certain tRNAs at a low level ? One interesting possibility is that rare tRNAs are essential for the process of cotranslation folding , presumably because low abundance tRNAs provide a pause in translation that might be needed for proper folding ( Thanaraj , 1996; Drummond and Wilke , 2008; Cabrita et al . , 2010; Pechmann and Frydman , 2012 ) . Other deleterious effects that may stem from a high copy number of tRNAArgCCU could be misincorporation of arginine into non-arginine codons , or the misloading of arginine tRNA molecules with other amino acids . These potential sources of errors are not mutually exclusive and can each contribute to the observed growth defect by exerting a protein folding stress . To examine the possibility that the growth defect associated with multiple copies of tRNAArgCCU is indeed associated with such a proteotoxic stress , we used an established method that examines the load on the protein quality control machinery of the cell ( see ‘Materials and methods’ ) ( Kaganovich et al . , 2008 ) . In this assay , we transformed cells with a plasmid harboring the human gene , von Hippel–Lindau ( VHL ) , fused to a fluorescent tag ( mCherry ) . Fluorescently tagged VHL that is present as aggregated puncta ( Figure 4B ) , and not as a disperse cytosolic localization ( Figure 4C ) , indicates that the protein quality control machinery is saturated due to high levels of misfolding in the cell’s endogenous proteins . We transformed the VHL-mCherry plasmid to each of the multi-copy tRNA strains , WTmultiCCU , WTmultiUCU and WTmultiControl , and monitored the level of proteotoxic stress by quantifying the number of cells with puncta phenotype in each population . The fold change in those cells was then deduced by normalization to the WTmultiControl population . We found that while WTmultiUCU exhibited similar amount of cells with puncta as the WTmultiControl , the WTmultiCCU exhibited a threefold increase ( Figure 4D ) . The proteotoxic stress experienced by the two strains overexpressing tRNAs was further assessed by measuring the induction level of an array of heat-shock proteins ( HSPs ) using real-time quantitative PCR ( RT-qPCR ) . Since the HSPs have been shown to undergo induction under proteotoxic stress , they are an excellent indicator for this stress ( McClellan et al . , 2005; Kaganovich et al . , 2008 ) . Indeed , we found a significant upregulation in mRNA levels for all the examined HSP genes in the WTmultiCCU strain , but not in the WTmultiUCU strain ( Figure 4E ) . These findings further demonstrate that increasing the copy number of a rare tRNA gene , but not of an already abundant one , results in proteotoxic stress in the cell .
Genomic duplications , deletions , and anticodon mutations shape tRNA gene families , yet the evolutionary scenarios that trigger changes in the tRNA pool have not been thoroughly explored . In our evolution experiments , a translational imbalance was imposed by a tRNA gene deletion that compromised growth and drove the tRNA pool to adapt to a novel translational demand . Importantly , organisms may experience equivalent imbalances when their gene expression changes due to altered environmental conditions or upon migrating to a new ecological niche ( Gingold et al . , 2012 ) . This scenario is particularly feasible given that the genes needed in various environments do show differences in codon usage , for example respiration as opposed to fermentation in yeast ( Man and Pilpel , 2007 ) . Indeed , when faced with different environmental challenges , transcriptional changes affect the codon usage of the transcriptome ( Gingold et al . , 2012 ) , and hence the demand for the various tRNAs , and may thus cause translational imbalances . To maintain optimal protein production , the tRNA pool is under pressure to restore the translational balance by accommodating the new translational demands . On a short timescale , the tRNA pool might respond non-genetically by changing expression profiles of the tRNAs ( Tuller et al . , 2010; Saikia et al . , 2012; Pavon-Eternod et al . , 2013a , 2013b ) . Yet , if changes in demand-to-supply persist , a genetic change in the tRNA pool might become beneficial evolutionarily . In this work , we demonstrate how anticodon mutations provide a rapid mechanism to alter the tRNA pool . We propose that during evolution , novel translational requirements can be addressed by anticodon shifting of tRNA copies more readily than by duplications and deletions of tRNA genes . The tRNA pool can evolve to meet new translation demands by adjusting the ratios of tRNA families that code for the same amino acid . Within a single mutational event , anticodon switching holds the potential to rapidly change the ratios of tRNAs within the pool , by increasing the copy number of one tRNA family at the expense of a counterpart . A similar solution could be obtained by a sequence of genomic duplications and deletions of tRNA genes . These alternatives are likely to fixate less frequently than anticodon switching , as they may carry negative effects due to duplications or deletions of adjacent unrelated genetic features . Furthermore , our systematic search for tRNA switching events throughout the tree of life revealed the prevalence of tRNA anticodon mutations in nature . This observation is consistent with the results of our lab-evolution experiment and may be the evolutionary outcome to novel translational demands in the wild . Studies on methionine tRNAs have previously shown that the scaffold sequences determine their function as either initiator or elongator Met tRNA ( Von Pawel-Rammingen et al . , 1992; Aström et al . , 1993; Kolitz and Lorsch , 2010 ) . Yet , the initiator and the elongator represent extreme cases of tRNAs that are used at different stages in translation . The present chimeric tRNAs that emerged in our lab-evolution experiments successfully replaced the deleted tRNA , despite differences in 20 nucleotide positions between the 2 tRNA scaffolds . If tRNA scaffolds are interchangeable in terms of the effect of their function on the fitness , what can explain the high sequence similarity observed among tRNA gene copies of the same family in yeast ? It is possible that the sequence of the tRNA scaffold is indeed important under specific conditions that were not examined in this work , or that our measurements were not sensitive enough to detect small selective disadvantages that can act against chimeric tRNAs in nature . Under these scenarios , the high sequence similarity can be explained by purifying selection that maintains sequence identity within tRNA families . Yet , there is also a possibility that the sequence similarity is not due to purifying selection but the result of ‘concerted evolution’ , an evolutionary process that maintains sequence identity by frequent recombination events among copies of the same gene family ( Munz et al . , 1982; Amstutz et al . , 1985; Teshima and Innan , 2004 ) . This possibility implies that the high conservation observed within tRNA gene families is not due to functionality , but is rather the result of neutral evolution . At present , it is not possible to determine which of the two possibilities explains best the observed sequence identity . If a single point mutation in one of the tRNAArgUCU copies enables it to function like a tRNA from a different family , what were the evolutionary constraints that have left some families with more members while others with fewer ? Is purifying selection acting to deliberately maintain low levels of certain tRNAs ? Such selection would render their corresponding codons ‘non-optimal’ . To examine potential adaptive functions of tRNA family sizes , we tested the consequences of increasing the sizes of several tRNA families . We found that keeping low copy tRNA families is adaptive , as increasing their copy number can result in a proteotoxic stress due to problems in protein folding . Most of the published work on the functionality of codons that correspond to rare tRNAs have so far tested how modified codon usage of specific proteins influences their proper folding ( Crombie et al . , 1992; Komar et al . , 1999; Cortazzo et al . , 2002; Tsai et al . , 2008; Zhang et al . , 2009; Zhou et al . , 2013 ) . In contrast , we took a different approach , in which no protein coding gene sequence is modified , but rather the tRNA supply is manipulated . Thus , the effect we generated could be exerted on all genes , and we could indeed detect it as a global proteotoxic stress in the cell . Our observations are consistent with the theory that programmed pauses in the translation process could promote proper folding during translation ( Thanaraj , 1996; Kramer et al . , 2009; Cabrita et al . , 2010; Wilke and Drummond , 2010 ) . The overexpression of the rare tRNA could have thus impaired with cotranslation folding . Yet , there could be additional reasons for the observed proteotoxic stress , which are not necessarily mutually exclusive . First , overexpression of tRNAArgCCU may result in misincorporation of arginine into non-arginine codons . Second , other aminoacyl tRNA synthetases may aminoacylate an incorrect amino acid to the highly expressed tRNA . Misloading will result in the incorporation of a different amino acid instead of arginine . A potential part of the observed proteotoxic stress due to misincorporation still remains to be studied . Yet , such an effect should be relevant not only for the overexpression of the rare tRNAArgCCU but also for the overexpression of the abundant tRNAArgUCU . Our results show a sever proteotoxic stress only upon expression of the rare tRNA , thus landing more support to the intriguing hypothesis , that the proteotoxic phenotypes observed are due to converting a slow translating codon , scattered in many genes in the genome , into a fast one . This notion is consistent with and complementary to the picture that emerges from single gene-based analyses . When facing the need to adapt , the tRNA pool ( that is the supply ) provides evolutionary plasticity to the translation machinery . The ability of the tRNA pool to change rapidly can be mainly attributed to its unique architecture in the form of multimember gene families . Only on a much longer evolutionary timescale , will the genome-wide codon usage of genes change so as to further fine tune the translational balance . Notably , the plasticity of the tRNA genes is constrained by the need to maintain proper protein folding ( Drummond and Wilke , 2008 ) . Thus , the need to accommodate changes in codon usage demands acts together with protein folding constrains to shape the tRNA pool in the living cells .
The following S . cerevisiae strains were used in this study: ΔtRNAArgCCU ( based on Y5565 , genetic background: ΔtR ( CCU ) J::Hyg , MATα , can1Δ::MFA1pr-HIS3 mfα1Δ::MFα1pr-LEU2 lyp1Δ ura3Δ0 leu2Δ0 ) ( Bloom-Ackermann et al . , In press ) was used for lab-evolution experiments . MutΔtRNAArgCCU is based on ΔtRNAArgCCU and carries a mutation ( T→C transition ) in tR ( UCU ) K gene plus a URA3 selection marker . BY384 ( MATa leu2Δ1 lys2Δ202 trp1Δ63 ura3-52 his3Δ200 ) was used to generate a complete deletion of the tRNASerGCU gene family . BY4741 ( MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ) and BY4742 ( MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ) were used for examining the effect of increasing tRNA gene copy number . Plasmids used in this study to express tRNA genes were pRS316 ( CEN , URA3 ) , pRS425 ( 2µ , LEU2 ) , and pRS426 ( 2µ , URA3 ) . For the rescue assays of the quadruple serine deletion , the pQF50 ( 2µ , URA3 ) and pQF150 ( 2µ , LEU2 ) were used . For the protein quality control assays the pGAL-VHL-mCherry ( 2µ , LEU2 ) plasmid was used ( Kaganovich et al . , 2008 ) . For additional information on plasmids and primers see Supplementary file 2 . Cultures were grown at 30°C in either rich medium ( 1% bacto-yeast extract , 2% bacto-peptone and 2% dextrose [YPD] ) or synthetic medium ( 0 . 67% yeast nitrogen base with ammonium sulfate and without amino acids and 2% dextrose , containing the appropriate supplements for plasmid selection ) . Protein quality control assays were performed on synthetic medium supplemented with 2% galactose as a carbon source . All chemicals used to create the media were manufactured by BD . All sugars , nucleic acids and amino acids were manufactured by Sigma-Aldrich . Lab-evolution experiments were carried out by serial dilution . Cells were grown on 1 . 2 ml of YPD at 30°C until reaching stationary phase and then diluted by a factor of 1:120 into fresh media ( 6 . 9 generations per dilution ) . This procedure was repeated daily until population growth under the applied condition matched the wild type . In all measurements of evolved populations , we used a population sample and not selected clones . The cultures were grown at the relevant condition , and optical density ( OD ) 600 measurements were taken during the growth at 30–45 min intervals until reaching early stationary phase . Qualitative growth comparisons were performed using 96-well plates ( Thermo Scientific ) in which 2 strains were divided on the plate in a checkerboard manner on the plate to cancel out positional geographical effects . For each strain , a growth curve was obtained by averaging over 48 wells . Strains were grown for 2 days in a non-selective liquid medium , which contains uracil ( YPD ) , to allow growth of cells that lost the plasmid containing the URA3 counterselectable marker ( Boeke et al . , 1984 ) . Then , 100 µl were plated on a YPD plate and replicated on the following day on either YPD or standard 5-fluoro-orotic acid ( 5-FOA ) ( US Biological ) plates to identify potential colonies that lost the plasmid . Following 2 days of incubation at 30°C , growth of the colonies was scored . We used a previously published method that allows examination of the protein quality control of the cell ( Kaganovich et al . , 2008 ) . This assay provides an indication for the protein unfolding stress in cells by assessing the load on the protein quality control machinery . In this assay , the cells were introduced with a high copy number plasmid that contains the human gene VHL fused to a fluorescent tag ( mCherry ) . VHL is a naturally unstructured protein and is dependent on two additional proteins ( Elongin B and C ) for proper folding in human cells . Expressing VHL in yeast cells , which lack VHL’s complex partners , leads to misfolding of the translated proteins . Under normal conditions , the misfolded VHL proteins are handled by the cell’s quality control machinery . When the quality control machinery is not saturated , the fluorescently tagged VHL appears in the cytosol . However , under stress , in which the quality control machinery is fully occupied , misfolded proteins in the cytosol are processed into dedicated inclusions ( JUNQ and IPOD ) and form punctum structures . Hence , a punctum phenotype of the VHL-mCherry construct is an indication for cells that experience proteotoxic stress and saturation of the protein quality control machinery . Wild-type yeast cells harboring the pGAL-VHL-mCherry ( CHFP ) fusion plasmid ( Kaganovich et al . , 2008 ) and either an additional empty plasmid or the tRNA overexpression plasmid , were grown overnight on SC+2% raffinose , diluted into SC+2% galactose and grown at 30°C for 6 hr . The cells were visualized using an Olympus IX71 microscope ( Olympus ) controlled by Delta Vision SoftWoRx 3 . 5 . 1 software , with X60 oil lens . Images were captured by a Photometrics Coolsnap HQ camera with excitation at 555/28 nm and emission at 617/73 nm ( mCherry ) . The images were scored using ImageJ image processing and analysis software . The percentage of cells harboring VHL-CHFP foci ( Puncta ) in the overexpression strains were normalized to the level in a control strain carrying an empty plasmid . To characterize the extent of anticodon switching across the tree of life , we first downloaded all tRNA sequences from the Genomic tRNA Database ( Chan and Lowe , 2009 ) . The sequences in this database were discovered with the tRNAscan algorithm ( Lowe and Eddy , 1997 ) , which finds tRNA sequences by scanning genomic DNA . We removed psuedogene tRNAs , which are defined as those tRNAs with a COVE score less than 40 . 0 ( Lowe and Eddy , 1997 ) . For each remaining tRNA sequence , we masked its anticodon triplet as ‘NNN’ . We next grouped all tRNA sequences by their species , and then aligned the sequences for each species using Muscle with default settings ( Edgar , 2004 ) . For each species , we inferred a maximum likelihood phylogeny of its tRNA sequences using RAxML with the GTRCAT model ( Stamatakis , 2006 ) . We calculated statistical support for tree branches using SH-like approximate likelihood ratio test ( Anisimova and Gascuel , 2006 ) . We next interrogated each species’ tRNA phylogeny using DendroPy ( Sukumaran and Holder , 2010 ) . Specifically , we identified those tRNA sequences harboring an anticodon that appears in the genome more than once; for each of these tRNA sequences , we found the shortest distance to another tRNA with the same anticodon ( dsame ) and the shortest distance to another tRNA with a different anticodon ( ddiff ) . We labeled tRNAs as putatively ‘switched’ if ddiff<dsame . Cultures were grown in rich medium at 30°C to a cell concentration of 1 × 107 cells/ml . Then , RNA was extracted using MasterPure kit ( Epicentre-illumina ) ( EPICENTER Biotechnologies ) , and used as a template for quantitative RT-PCR using light cycler 480 SYBR I master kit ( Roche Applied Science ) and the LightCycler 480 system ( Roche Applied Science ) , according to the manufacturer’s instructions . In all , 4 independent lab-evolution experiments that started with ΔtRNAArgCCU as the ancestral strain showed full recovery of the deletion phenotype after 200 generations . In each of the evolved populations a mutation in one of the copies of tRNAArgUCU was found to change the anticodon from UCU to CCU . The genomic copies of tRNAArgUCU that were found to carry the mutation were: tR ( UCU ) K , tR ( UCU ) G1 and tR ( UCU ) D that was changed in two of the independent cultures . Out of 4245 anticodon switching events that we detected , the first position in the anticodon was changed in 2540 cases while the second and third were only involved in 1448 and 1330 cases , respectively . | Genes contain the blueprints for the proteins that are essential for countless biological functions and processes , and the path that leads from a particular gene to the corresponding protein is long and complex . The genetic information stored in the DNA must first be transcribed to produce a messenger RNA molecule , which then has to be translated to produce a string of amino acids that fold to form a protein . The translation step is performed by a molecular machine called the ribosome , with transfer RNA molecules bringing the amino acids that are needed to make the protein . The information in messenger RNA is stored as a series of letters , with groups of three letters called codons representing the different amino acids . Since there are four letters—A , C , G and U—it is possible to form 64 different codons . And since there are only 20 amino acids , two or more different codons can specify the same amino acid ( for example , AGU and AGC both specify serine ) , and two or more different transfer RNA molecules can take this amino acid to the ribosome . Moreover , some codons are found more often than others in the messenger RNA molecules , so the genes that encode the related transfer RNA molecules are more common than the genes for other transfer RNA molecules . Environmental pressures mean that organisms must adapt to survive , with some genes and proteins increasing in importance , and others becoming less important . Clearly the relative numbers of the different transfer RNA molecules will also need to change to reflect these evolutionary changes , but the details of how this happens were not understood . Now Yona et al . have explored this issue by studying yeast cells that lack a gene for one of the less common transfer RNA molecules ( corresponding to the codon AGG , which specifies the amino acid arginine ) . At first this mutation resulted in slower growth of the yeast cells , but after being allowed to evolve over 200 generations , the rate of growth matched that of a normal strain with all transfer RNA genes . Yona et al . found that the gene for a more common transfer RNA molecule , corresponding to the codon AGA , which also specifies arginine , had mutated to AGG . As a result , the mutated yeast was eventually able to produce proteins as quickly as wild type yeast . Moreover , further experiments showed that the levels of some transfer RNAs are kept deliberately low in order to slow down the production of proteins so as to ensure that the proteins assume their correct structure . But does the way these cells evolved in the lab resemble what happened in nature ? To address this question Yona et al . examined a database of transfer RNA sequences from more than 500 species , and found evidence for the same codon-based switching mechanism in many species across the tree of life . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"cell",
"biology"
] | 2013 | tRNA genes rapidly change in evolution to meet novel translational demands |
The prokaryotic CRISPR ( clustered regularly interspaced palindromic repeats ) -associated protein , Cas9 , has been widely adopted as a tool for editing , imaging , and regulating eukaryotic genomes . However , our understanding of how to select single-guide RNAs ( sgRNAs ) that mediate efficient Cas9 activity is incomplete , as we lack insight into how chromatin impacts Cas9 targeting . To address this gap , we analyzed large-scale genetic screens performed in human cell lines using either nuclease-active or nuclease-dead Cas9 ( dCas9 ) . We observed that highly active sgRNAs for Cas9 and dCas9 were found almost exclusively in regions of low nucleosome occupancy . In vitro experiments demonstrated that nucleosomes in fact directly impede Cas9 binding and cleavage , while chromatin remodeling can restore Cas9 access . Our results reveal a critical role of eukaryotic chromatin in dictating the targeting specificity of this transplanted bacterial enzyme , and provide rules for selecting Cas9 target sites distinct from and complementary to those based on sequence properties .
CRISPR ( clustered regularly interspaced palindromic repeats ) prokaryotic adaptive immune systems have yielded transformative tools for manipulating eukaryotic genomes . Most notably , the CRISPR-associated Cas9 protein from Streptococcus pyogenes , together with a single chimeric guide RNA ( sgRNA ) , provides a programmable endonuclease that has revolutionized our ability to edit genomes ( Doudna and Charpentier , 2014 ) . Cas9 has been further modified to a nuclease-dead form ( dCas9 ) to provide a programmable DNA-binding protein that can be fused to effector domains , making it possible to turn on or off targeted genes , mark specific genomic loci with fluorescent proteins , or alter epigenetic marks ( Doudna and Charpentier , 2014; Qi et al . , 2013; Gilbert et al . , 2013; Maeder et al . , 2013; Chen et al . , 2013; Ma et al . , 2015; Hilton et al . , 2015; Kearns et al . , 2015 ) . A central challenge in implementing these tools is identifying effective and specific sgRNAs . While much of the effort to define relevant rules has focused on the sequence of the target site and sgRNA ( Chari et al . , 2015; Doench et al . , 2014; Wang et al . , 2014; Xu et al . , 2015 ) , these only partially predict Cas9 activity and suggest that additional determinants likely exist . Chromatin structure may represent a key parameter governing Cas9 efficacy in eukaryotic cells . CRISPR evolved in archea and bacteria ( Makarova et al . , 2015 ) and is likely not optimized to explore and modify large , chromatin-bound eukaryotic genomes , a hypothesis supported by several studies that point to a correlation between rates of DNA binding and cleavage in regions of open chromatin as measured by DNase I hypersensitivity ( Chari et al . , 2015; Singh et al . , 2015; Wu et al . , 2014 ) . Additionally , recent single-molecule imaging studies have shown that dCas9 explores euchromatin more frequently than it does heterochromatin ( Knight et al . , 2015 ) . We hypothesized that nucleosomes , the basic unit of chromatin structure , are an important impediment to Cas9 recognition . Here , we addressed this hypothesis in vivo by leveraging large datasets collected from over 30 sgRNA tiling and genome-scale genetic screens ( Gilbert et al . , 2014 ) . We found that regions of high nucleosome occupancy in vivo , as determined by MNase-seq ( micrococcal nuclease sequencing ) ( Johnson et al . , 2006; Valouev et al . , 2011 ) , were strongly depleted of highly active sgRNAs for CRISPR interference ( CRISPRi ) ( Gilbert et al . , 2013; 2014 ) and nuclease-active Cas9 . We complemented these results with in vitro experiments demonstrating that formation of a nucleosome provides a direct and profound block to dCas9 binding and Cas9 cleavage . Despite this strong barrier to Cas9 activity , we found that addition of a chromatin remodeling enzyme to chromatinized DNA in vitro can restore Cas9 access to nucleosomal DNA , highlighting one route by which CRISPR may still be able to modify chromatin in vivo . Our results reveal a fundamental aspect of the mechanism by which this transplanted bacterial enzyme interacts with eukaryotic chromatin , and provide a new dimension for selecting highly active sgRNAs .
In order to study the features of chromatin that affect CRISPRi activity , we integrated data from whole-genome screens testing for a wide range of phenotypes performed with a previously described CRISPRi library targeting each gene with ~10 sgRNAs ( Gilbert et al . , 2014 ) . We selected 30 screens performed in the cell line K562 expressing dCas9-KRAB ( ML , BA , BB , CYP , MK , YC , JF , and JN , personal communication ) and set a threshold for high-confidence hit genes , which allowed us to assess the relative strength of phenotypes for the sgRNAs targeting these genes . Specifically , we analyzed 18 , 380 sgRNAs targeting 1539 genes , and generated 'activity scores' by normalizing sgRNA phenotypes to the average of the 3 sgRNAs with the strongest phenotypes for each gene ( Figure 1A and Supplementary file 1 ) . To assess how CRISPRi activity varies with respect to the transcription start site ( TSS ) , we plotted the average sgRNA activity score as a function of distance to the FANTOM-annotated TSS ( Forrest et al . , 2014 ) ( Figure 1B ) . This analysis revealed a robust , periodic pattern of activity , with peaks at ~190 bp intervals relative to the TSS . This periodicity was highly reminiscent of patterns previously described for nucleosomes ( Jiang and Pugh , 2009 ) . Indeed , analysis of K562 MNase-seq data from the ENCODE consortium ( Valouev et al . , 2011; ENCODE Project Consortium , 2012 ) revealed that the average nucleosome signal was strongly anti-correlated with CRISPRi activity ( Figure 1B ) , suggesting that high nucleosome occupancy leads to low CRISPRi activity . 10 . 7554/eLife . 12677 . 003Figure 1 . CRISPRi activity anti-correlates with nucleosome occupancy . ( A ) Workflow for generating CRISPRi activity scores from pooled genetic screens . The resulting values are distributed around 0 for inactive sgRNAs and around 1 for highly active sgRNAs . ( B ) Average CRISPRi activity and MNase-seq signal relative to the TSS . Green line represents average CRISPRi activity score of all sgRNAs within a 50 bp window around each position . Purple line represents the K562 MNase-seq signal at each position averaged across all genes analyzed . ( C ) Average CRISPRi activity and MNase-seq signal for genes grouped by expression value . Genes were grouped into lower expression ( light lines; N=240 ) , higher expression ( heavy lines; N=368 ) , and medium expression ( omitted for clarity; N=930 ) , and analyzed as in ( B ) . Expression values were obtained as fragments per kilobase million ( FPKM ) from ENCODE K562 RNA-seq data . Average activity at positions with fewer than 10 sgRNAs within the 50bp window was not calculated . ( D ) Quantification of the amplitude of periodic CRISPRi activity . Peak and trough coordinates were obtained by calculating the local maxima and minima of the activity traces from analyses in ( B ) and ( C ) . Peak 0 was defined as the local maximum closest to the TSS , peak 1 was defined as the next maximum downstream of peak 0 , and troughs were defined as the minima immediately downstream of the respective peaks . ( E ) CRISPRi activity and target site nucleosome occupancy for individual sgRNAs . Target site nucleosome occupancy was calculated from the average MNase-seq signal at all genomic coordinates across the length of the sgRNA protospacer and the protospacer adjacent motif ( PAM ) . sgRNAs were then binned by the target site nucleosome occupancy , displayed as box-and-whisker plots , and labeled according to the minimum value within the bin except where indicated . P-values were calculated by a two-tailed Mann-Whitney test comparing each bin to the =0 . 0 bin . ( F ) Linear regression for CRISPRi activity . The squared Pearson correlation was calculated for the sgRNA activity scores compared to the indicated individual parameters ( bars 1 , 2 , and 4 ) or linear fits of multiple parameters ( bars 3 and 6 ) . sgRNA activity scores were corrected for sequence and length features ( bar 5 ) by subtracting the linear fit of those two features . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 003 To further explore this inverse relationship between CRISPRi activity and nucleosome organization , we exploited the previous observation that nucleosome phasing is more pronounced in highly expressed genes ( Valouev et al . , 2011 ) . We grouped genes by their expression in K562 ( ENCODE Project Consortium , 2012; Djebali et al . , 2012 ) and analyzed CRISPRi activity and nucleosome occupancy within each group . In support of a connection , we found that peak to trough amplitudes of both features were larger for highly expressed genes ( Figures 1C , D ) . We also analyzed the MNase signal at each sgRNA target site to determine whether nucleosome occupancy could explain variation in CRISPRi activity between individual sgRNAs ( Figure 1E ) . Consistent with the hypothesis that nucleosomes exclude dCas9 , almost all of the highly active sgRNAs targeted sites with low MNase-seq signal . However , not all sgRNAs targeting sites with low nucleosome occupancy were highly active , matching previous findings that sgRNA and target DNA sequence features also influence efficiency of the CRISPR system ( Chari et al . , 2015; Doench et al . , 2014; Wang et al . , 2014; Xu et al . , 2015 ) . To exclude the possibility that differences in sequence features alone between nucleosome-bound and -free regions could explain the periodicity of CRISPRi activity ( rather than the presence of nucleosomes per se ) , we performed linear regression using MNase signal , sgRNA length ( Xu et al . , 2015; Gilbert et al . , 2014 ) , and a validated sgRNA sequence scoring algorithm ( Doench et al . , 2014 ) . We found that each parameter individually correlated with sgRNA activity ( p<10-58 for each; Figure 1F ) . Importantly , correction for sequence and length features had minimal impact on the ability of the MNase signal to predict CRISPRi activity ( p<10-85 after correction ) . Indeed , a linear fit of all three features provided a still stronger correlation ( p<10-221; Figure 1F , far right column ) , suggesting that incorporating nucleosome occupancy in future sgRNA design algorithms could significantly improve predictive value . We have recently developed a comprehensive algorithm for predicting highly active CRISPRi sgRNAs that , by accounting for nucleosome occupancy , higher order sequence features , and non-linear relationships in these parameters , shows even greater correlation with this dataset that was already enriched for active sgRNAs using our original CRISPRi library design principles ( Gilbert et al . , 2014 ) ( cross-validation R2=0 . 32; Horlbeck et al . , manuscript in preparation ) . In order to generate a dataset for evaluating the effect of nucleosome positioning on nuclease-active Cas9 in K562 cells , we took advantage of our previously described library densely tiling sgRNAs in 10kb windows around the TSS of 49 genes known to modulate susceptibility to the toxin ricin ( Gilbert et al . , 2014; Bassik et al . , 2013 ) and tested Cas9-expressing cells for resistance or sensitivity to ricin ( Supplementary file 2 ) . We observed phenotypes consistent with the expected knockdown phenotype primarily in coding sequences ( CDS ) but also in some promoter regions , consistent with recent results showing that modifications introduced by Cas9 can disrupt cis-regulatory regions ( Canver et al . , 2015 ) . Analysis of CDS-targeting sgRNAs revealed that strong phenotypes were found predominantly in regions of low MNase-seq signal ( Figure 2A ) , although this relationship was less pronounced than for CRISPRi . This may be due to decreased phasing of nucleosomes within the gene body ( Jiang and Pugh , 2009 ) . When we analyzed all sgRNAs in the library , thus incorporating information from cis-regulatory regions where nucleosomes are well phased , we found the effect of nucleosome position to be even stronger ( Figure 2—figure supplement 1 ) . Although nucleosome occupancy was predictive of CRISPR activity independent of sgRNA sequence features , a much stronger correlation was obtained when all features were considered ( Figure 2B ) . Therefore , nucleosome organization likely represents an important feature for CRISPR sgRNA design and should be considered a key contributing factor in interpreting future tiling mutagenesis experiments of coding and non-coding regions . 10 . 7554/eLife . 12677 . 004Figure 2 . Cas9 nuclease activity anti-correlates with nucleosome occupancy . ( A ) Cas9 nuclease phenotypes and target site nucleosome occupancy for individual sgRNAs targeting CDS regions . Ricin susceptibility phenotypes for each sgRNA are expressed as a Z score and are positive if the phenotype matches the expected knockdown phenotype . Target site nucleosome occupancy was calculated as in Figure 1E . sgRNAs were then binned by the target site nucleosome occupancy , displayed as box-and-whisker plots , and labeled according to the minimum value within the bin except where indicated . P-values were calculated by a two-tailed Mann-Whitney test comparing each bin to the =0 . 0 bin . ( B ) Linear regression for Cas9 nuclease phenotypes . The squared Pearson correlation was calculated for the sgRNA activity scores compared to the indicated individual parameters ( bars 1 , 2 , and 4 ) or linear fits of multiple parameters ( bars 3 and 6 ) . sgRNA activity scores were corrected for sequence and length features ( bar 5 ) by subtracting the linear fit of those two features . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 00410 . 7554/eLife . 12677 . 005Figure 2—figure supplement 1 . Cas9 nuclease activity anti-correlates with nucleosome occupancy at all target sites . Cas9 nuclease phenotypes and target site nucleosome occupancy for individual sgRNAs . As in Figure 2A , but incorporating all CDS- and non-CDS-targeting sgRNAs in the ricin-susceptibility gene tiling library . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 005 Our in vivo experiments reveal a strong anti-correlation between Cas9-mediated downstream phenotypic outputs and nucleosome positioning , but do not directly report on the ability of Cas9 to access nucleosomal DNA . Factors other than Cas9 access could contribute to the observed correlation in our data . For example , ( d ) Cas9 may bind or cut equally well in nucleosome-bound and un-bound regions , but may exert the observed modulation of gene expression through interference with transcriptional pausing , splicing , regulatory looping , or binding of important regulatory factors , processes which also correlate with nucleosome organization ( Jonkers and Lis , 2015; Denisov et al . , 1997; Naftelberg et al . , 2015; Segal and Widom , 2009; Tilgner and Guigó , 2010; Thurman et al . , 2012; Hsieh et al . , 2015 ) . To determine whether ( d ) Cas9 activity is indeed directly affected by the presence of nucleosomes , we turned to a purified in vitro reconstituted system . Mouse histones were recombinantly expressed , purified , and assembled into a nucleosome using 147bp of the Widom 601 positioning sequence ( Figures 3A , B ) ( Lowary and Widom , 1998 ) . Conveniently , the 601 sequence contains numerous NGG protospacer adjacent motifs ( PAMs ) required for Cas9 recognition , spanning the full length of the DNA at different helical positions , allowing us to test the effects of position and solvent accesibility within the nucleosome ( Figure 3C ) . We first tested the ability of Cas9 to cleave nucleosomal DNA . We pre-loaded purified Cas9 ( histidine tagged and HaloTagged ) with in vitro transcribed sgRNA , then introduced either naked 601 DNA or that same DNA assembled into a nucleosome ( Figures 3A , D and Figure 3—figure supplement 1A ) . Fluorescent labeling of the DNA allowed us to visualize the cleavage products on a denaturing Urea-PAGE gel . In agreement with our in vivo data , the nucleosome protected its DNA from cleavage by Cas9 , and complete protection from cleavage was observed to be independent of target position within the nucleosome ( Figures 3E , F and Figure 3—figure supplement 1B , C ) . While our manuscript was under review , a study by Hinz and colleagues reported a similar in vitro finding that Cas9 nuclease activity is inhibited within the nucleosome but not at adjacent linker sequences ( Hinz et al . , 2015 ) . Additionally , previous single-molecule and biochemical studies have established that nucleosomal DNA undergoes transient unwrapping or breathing at the entry and exit sites , creating a gradient of accessibility along the nucleosome ( Li and Widom , 2004; Li et al . , 2005; Luger and Hansen , 2005; Choy and Lee , 2012; Tomschik et al . , 2009; Polach and Widom , 1995 ) . This property is often credited for the observed position-dependent binding patterns of many transcription factors and DNA-binding proteins . Interestingly , our data suggests that the cleavage activity of Cas9 in vitro is not detectably influenced by this effect , although less stable nucleosomes than the Widom 601 nucleosome or target sites closer to the nucleosome edge may exhibit more breathing and thus more accessibility than those tested here . 10 . 7554/eLife . 12677 . 006Figure 3 . Cas9 nuclease activity is blocked by the presence of a nucleosome in vitro . ( A ) Schematic of the experimental setup for in vitro cleavage assays . Mononucleosomes were assembled by salt gradient dialysis of purified mouse histone octamer with the minimal nucleosome positioning sequence , Widom 601 ( 147 bp ) . Prior to assembly , DNA was 5’-end-labeled with Cy3 and histone H2B was fluorescently labeled using an introduced cysteine ( T115C ) coupled to Alexa Fluor 647 . Purified His-tagged and HaloTagged Cas9 pre-loaded with in vitro transcribed sgRNAs were added to naked DNA or assembled nucleosomes , and DNA cleavage products were visualized using a denaturing gel imaged for Cy3-DNA fluorescence . See also Figure 3—figure supplement 1A . ( B ) Confirmation that fully occupied , well-positioned nucleosomes were assembled . After assembly using salt gradient dialysis , the produced nucleosomes were visualized using a native PAGE gel imaged in the Cy3 and Alexa Fluor 647 channels . Full incorporation of the free DNA into a nucleosome occupying a single position on the DNA is indicated by the presence of a single-shifted band containing all of the detectable Cy3-DNA and Alexa Fluor 647-H2B signal . ( C ) Available PAMs and solvent accessibility of the 601 nucleosome positioning sequence . ( Above ) A schematic of the 601 sequence . The location of the histone octamer when assembled into a nucleosome is indicated by the gray oval . The location of PAMs within the double-stranded sequence are indicated with arrows spanning the 3 bp of the PAM , pointing in the 5’ to 3’ orientation of the NGG motif . The arrows are colored according to solvent accessibility at the center of the PAM as calculated from the crystal structure of the 601 nucleosome ( PDBID 3LZ0 , Vasudevan et al . , 2010 ) . ( Below ) Crystal structure of the 601 nucleosome . For clarity , the surface area of the histones in the crystal structure has been made transparent . The DNA in the crystal structure is colored according to solvent accessibility using the same color scale as the PAMs above . Residues colored teal are less accessible , while residues colored fuchsia are more accessible by solvent . ( D ) Experimental conditions and timeline for cleavage assays . ( E ) Denaturing PAGE gel showing the results of a cleavage assay targeting the indicated PAMs . Cleavage reactions containing naked DNA were loaded on the left half of the gel , while reactions containing nucleosomes were loaded on the right . The DNA was imaged via a Cy3 fluorophore attached to the 5’ end of the sgRNA-complimentary strand . A negative control was conducted with an sgRNA that had no sequence complementarity to the 601 sequence used ( non-sense guide ) . See also Figure 3—figure supplement 1B , C for additional controls . ( F ) Quantification of the gel in ( D ) . For each lane , percent cleavage was determined by calculating the percent of the total band signal corresponding to cleaved DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 00610 . 7554/eLife . 12677 . 007Figure 3—figure supplement 1 . HaloTagged Cas9 activity is indistinguishable from untagged Cas9 . ( A ) Diagram of the CDS region in the Cas9 expression plasmid used in this study . ( B ) Cleavage assay comparing the HaloTagged Cas9 construct used in this study with an untagged Cas9 commercially purchased from New England Biolabs ( NEB , Ipswich , MA ) . Both forms of Cas9 were incubated with either naked DNA or the same DNA assembled into a nucleosome ( see Figure 3A ) . A positive control used the restriction enzyme , StyI-HF ( from NEB ) , to target a sequence at a location within the DNA known to be fully protected upon assembly into a nucleosome . Unless explicitly labaled as NEB , all constructs of Cas9 ( and dCas9 ) used were histidine tagged and HaloTagged . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 007 Cleavage of DNA by Cas9 has been described as a stepwise process ( Wu et al . , 2014; Knight et al . , 2015; Sternberg et al . , 2014 ) in which Cas9 must first scan for PAMs , unzip the DNA duplex , and fully pair the guide RNA and target DNA prior to cleavage . While our in vitro data show that full pairing and cleavage is prevented by the presence of a nucleosome , we wondered if binding without cleavage , especially at the more accessible ends of the nucleosome , might still occur . Additionally , the first step in binding , PAM recognition , may be governed by helical location within the nucleosome as well as its proximity to the more dynamic ends . Histones make contacts with the DNA at every helical turn , thus a PAM may fall on the outside of the nucleosome , exposing it to solvent , or on the inside at the DNA-histone interface . To test the influence of target location within the nucleosome on Cas9 binding , we used an electrophoretic mobility shift assay to monitor binding of dCas9 to Cy3 end-labeled 601 DNA ( Figures 4A , B ) , either free or assembled into a nucleosome containing Alexa Fluor 647-labeled histone H2B ( Figure 4—figure supplement 1A ) . Consistent with our cleavage results , binding by dCas9 was abolished by the presence of the nucleosome , regardless of the targeted dCas9 binding site ( Figure 4C–E and Figure 4—figure supplement 2A , B ) . 10 . 7554/eLife . 12677 . 008Figure 4 . dCas9 is unable to bind nucleosomal DNA in vitro . ( A ) Schematic of the experimental setup for in vitro binding assays . Either naked DNA or assembled nucleosomes were incubated with catalytically dead Cas9 ( dCas9 , histidine tagged and HaloTagged ) , and binding was assessed by an electrophoretic mobility shift assay ( EMSA ) . ( B ) Experimental conditions and timeline for binding assays . ( C ) A native PAGE gel showing the results of an EMSA in which dCas9 was targeted to the indicated PAMs on either naked or nucleosomal DNA . Gels were scanned for fluorescence from Cy3 on the DNA ( green ) and Alexa Fluor 647 on histone H2B ( magenta ) . The two color channels were merged to identify the location of intact nucleosomes ( white ) . See also Figure 4—figure supplement 1 and 3 for reagent preparation and experimental conditions , and Figure 4—figure supplement 2 for comparison with wtCas9 binding . ( D ) Quantification of the gel in ( C ) . Percent bound was determined by calculating the percent of the total band signal in each lane corresponding to Cas9-bound target as determined by a shift in mobility within the gel . ( E ) Summary of binding and cleavage results for each PAM tested . The ability of Cas9 to bind or cleave nucleosomal DNA at a targeted PAM is displayed as percent protection by the nucleosome , and was calculated by taking the ratio of binding or cleavage on nucleosomal DNA to that on naked DNA . While only the largest error bars are visible , replicates were performed for 15 of the 30 data points and are displayed with error bars showing standard deviation from the mean . The DNA positions plotted correspond to the three nucleotides of the targeted PAM . In order to compare percent protection from binding and cleavage with solvent accessibility , the PAMs are overlaid with a plot of the solvent accessible surface area for each strand ( Watson or Crick ) of DNA in the 601 nucleosome structure . The percent protection at each PAM , as well as the solvent accessibility were plotted so that the 5’ end of each DNA strand begins at the left of the graph , where position 0 indicates the dyad . ( F ) Structural assessment of the ability of Cas9 to bind nucleosomal DNA . Superposition of the Cas9-guideRNA-DNA crystal structure ( PDBID 4UN3 , Anders et al . , 2014 ) onto the 601 nucleosome crystal structure ( PDBID 3LZ0 , Vasudevan et al . , 2010 ) was achieved by alignment of the DNA path in both structures . ( Left ) To better view the alignment of the Cas9 target DNA with the nucleosomal DNA , the Cas9 protein density has been removed . ( Right ) After alignment of the DNA , inclusion of the Cas9 protein density reveals extensive steric clashes with the histones . Histone surface area was made partially transparent to better reveal the overlapping densities with Cas9 . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 00810 . 7554/eLife . 12677 . 009Figure 4—figure supplement 1 . Quality controls . ( A ) H2B labeling is near complete . An SDS-PAGE gel after refolding and purifying the histone octamer containing fully labeled versus poorly labeled H2B . The arrow indicates a mobility shift of H2B corresponding to a fully labeled band . Each histone is present at equimolar ratios as indicated by PageBlue protein stain . ( B ) Nucleosome stability in Cas9 binding and cleavage assays is ensured by including a nonspecific protein in solution . On the left , a native PAGE gel showing naked DNA and nucleosomes under Cas9 binding and cleavage reaction conditions without a nonspecific protein in solution . On the right , the same conditions plus inclusion of a nonspecific competitor protein , insulin . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 00910 . 7554/eLife . 12677 . 010Figure 4—figure supplement 2 . DNA binding by dCas9 is also representative of wtCas9 binding . ( A ) Two native PAGE gels showing the results of an EMSA binding assay comparing dCas9 and wtCas9 . Binding to naked DNA is shown in the gel on the left , while binding to nucleosomes is shown in the gel on the right . Both gels were imaged in the Cy3-DNA channel , while the gel on the right was also imaged in the Alexa Fluor 647 – H2B channel ( inset image ) . Binding conditions for both dCas9 and wtCas9 were identical , and were as described in the methods . ( B ) Quantification of the Native PAGE gel in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 01010 . 7554/eLife . 12677 . 011Figure 4—figure supplement 3 . ( wt/d ) Cas9 purification strategy . ( A ) Schematic showing the ( wt/d ) Cas9 purification strategy used in this study ( left ) . SDS-PAGE gel stained with PageBlue ( Life Technologies , Carlsbad , CA ) showing the elution fractions from the HiTrap SP-HP column during purification of 6His-dCas9-HaloTag ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 011 To better understand how a nucleosome might impede Cas9 binding , we aligned the available crystal structures of DNA-bound Cas9 and the structure of the 601 nucleosome ( Vasudevan et al . , 2010 , PDB ID 3LZ0; Anders et al . , 2014 , PDB ID 4UN3 ) . We superimposed the target DNA in the Cas9 crystal structure with the DNA in the nucleosome structure at a site where the two DNA paths gave the best fit . The resulting combined structure reveals that the Cas9 protein poses significant steric clashes with the histones ( Figure 4F ) . Given the extent of overlapping densities in the two structures , it seems unlikely that the histones and Cas9 could co-occupy the same piece of DNA . Additionally , it may be important to note that unlike other DNA binders such as transcription factors , binding by Cas9 constitutes melting of target DNA , which may pose an additional barrier to binding on a nucleosome . This hypothesis leaves two possible outcomes of targeting Cas9 to nucleosomal DNA: either Cas9 is capable of displacing histones in order to engage nucleosomal DNA , or it is excluded altogether . Our data support the latter conclusion . The nucleosome landscape in eukaryotic chromatin is dictated by both intrinsic DNA sequence preferences as well as extrinsic factors such as chromatin remodeling enzymes ( Jiang and Pugh , 2009; Flaus and Owen-Hughes , 2011; Hargreaves and Crabtree , 2011 ) . In order to model how these dynamics affect Cas9 access to nucleosomal DNA in vitro , we turned to a chromatinized plasmid system . We dialyzed plasmid DNA containing a single 601 nucleosome positioning sequence with a sub-saturating quantity of purified histone octamers , and confirmed the quality of the resulting chromatin assemblies by MNase digestion ( Figures 5A , B ) . We tested whether a nucleosome was positioned at the 601 sequence using restriction enzyme accessibility mapping , and found that sites within the 601 sequence were protected from digestion while sites immediately adjacent were not , suggesting precise positioning of a high occupancy nucleosome ( Figure 5C ) . To recapitulate the effects of chromatin remodeling in vitro , we used a purified , truncated form of the Snf2-like chromatin remodeling enzyme Chd1 from Saccharomyces cerevisiae ( yChd1 ) , which had previously been shown to mediate nucleosome sliding in an ATP-dependent manner without additional co-factors ( Patel et al . , 2011; 2013 ) . To confirm yChd1 activity on our chromatinized plasmid , we used the frequent cutter , HaeIII , to digest the chromatin in the presence or absence of the remodeler . Upon addition of yChd1 , we observed a shift toward lower molecular weight bands , indicative of dimished protection at HaeIII sites while still maintaining an overall chromatinized state ( Figure 5D ) . 10 . 7554/eLife . 12677 . 012Figure 5 . Nucleosomes within chromatinized DNA can block cleavage by Cas9 , but a chromatin remodeling factor can restore Cas9 access . ( A ) Schematic of the experimental setup . Supercoiled plasmid containing the 601 sequence inserted into a pBlueScript II SK ( + ) backbone ( pBSIISK+601 ) was chromatinized by step gradient salt dialysis in the presence of histone octamer . Purified yeast Chd1 ( yChd1 ) remodeling factor was used to test the effect of ATP-dependent remodeling factors on Cas9 access to nucleosomal DNA . ( B ) Quality assessment of the chromatinized plasmid used in this study . Titrated amounts of Micrococcal Nuclease ( MNase ) were incubated with the chromatinized plasmid , and the resulting pattern of protection by assembled nucleosomes was visualized on a 1 . 3% agarose gel post-stained with ethidium bromide ( EtBr ) . As a control , the supercoiled plasmid was also incubated with the lowest concentration of MNase . ( C ) A restriction enzyme accessibility assay ( REAA ) was used to assess the occupancy and position of the nucleosome assembled at the 601 sequence within the chromatinized plasmid . A panel of unique restriction enzyme sites spanning the 601 sequence were incubated with either the supercoiled plasmid , or the chromatinized plasmid . Cleavage was stopped , and protein was removed by incubation with proteinase K followed by Phenol:Chloroform:Isoamyl alcohol extraction and ethanol precipitation . ( Top ) The resulting DNA was then linearized using DraIII , and the level of cleavage by the restriction enzyme panel was visualized on a 1% agarose gel post-stained with EtBr . The label 'P' represents supercoiled plasmid , while 'C' represents chromatinized plasmid . ( Bottom right ) The location of the restriction sites used are indicated on a diagram of the plasmid . ( Bottom left ) After quantification of the gel , the percent protection from cleavage experienced in the chromatinized plasmid was plotted versus the location of the cleavage sites on the top strand of the 601 sequence . Experiment 1 refers to the REAA experiment shown in the gel above , while experiment 2 refers to the REAA experiment without remodeler shown in Figure 5F . The grey shading indicates the borders of the 601 sequence , and the grey oval represents the corresponding nucleosome . ( D ) REAA experiment using the frequent cutter , HaeIII , to assess the remodeling activity around the chromatinized plasmid by the purified yChd1 chromatin remodeler . The resulting banding patterns were visualized on a 1 . 5% agarose gel post-stained with EtBr . Low molecular weight fragments indicate a high degree of HaeIII accessibility , while higher weight bands indicate protection from digestion . ( E ) Diagram showing the location of the restriction enzyme cleavage sites and the PAMs targeted by Cas9/sgRNA in the experiment shown in ( F ) and ( G ) . ( F ) An accessibility assay was performed essentially as in C using either restriction enzymes or Cas9/sgRNAs in the presence or absence of the remodeler yChd1 . The level of cleavage by the restriction enzyme panel ( left ) or Cas9/sgRNAs ( right ) was visualized on a 1 . 3% agarose gel post-stained with EtBr . A negative control was conducted with an sgRNA that had no sequence complementarity to the plasmid used ( non-sense guide ) . The concentration of yChd1 used was the same as in panel ( D ) ( G ) Quantification of the gels shown in F . Percent protection from cleavage of the chromatinized plasmid in the presence or absence of the chromatin remodeler was calculated relative to the percent cleavage in the corresponding supercoiled plasmid control , and plotted at the location of the restriction enzyme cleavage sites or the center of the PAMs with respect to the 601 dyad . DOI: http://dx . doi . org/10 . 7554/eLife . 12677 . 012 We next sought to test whether yChd1 could affect Cas9’s ability to access nucleosomal DNA . Before addition of the remodeler to our chromatinized plasmid , we found that sites within the 601 nucleosome were strongly protected from cleavage by Cas9 , consistent with our mononucleosome results ( Figure 5E-G , PAM sites without remodeler ) . However , when the 601 nucleosome was remodeled by yChd1 , as indicated by a loss of protection from restriction enzyme cleavage ( approaching protection levels similar to those in the linker region ) , Cas9 cleavage efficiency was restored to around 80% of the corresponding naked plasmid control ( Figures 5E–G ) . Notably , the percent protection at the EcoRI site adjacent to the positioned nucleosome did not decrease upon addition of yChd1 , demonstrating that the decrease in protection along the 601 sequence was mediated by the nucleosome displacement activity of yChd1 rather than by a non-specific effect on cleavage efficiency . While our data with the chromatinized plasmid system confirm our findings that a well-positioned nucleosome provides a profound block to Cas9 cleavage , our further finding that chromatin remodeling restores access to nucleosomal DNA provides one potential mechanism by which Cas9 may efficiently modify broad portions of eukaryotic genomes . This plasmid model could be further exploited to assay the activity of Cas9 at nucleosome-free and boundary sites , and thus derive biophysical parameters governing Cas9-chromatin interactions .
Despite its swift success as a repurposed tool for gene editing , imaging , and transcription modulation , the ability of the prokaryotic CRISPR/Cas9 system to effectively navigate eukaryotic chromatin has remained poorly understood . Here , we show that the nucleosome , the basic unit of chromatin , poses a strong barrier to Cas9 , both in vitro and in vivo . By masking ~147 bp of DNA , the nucleosome effectively reduces the size of the eukaryotic genome available to Cas9 . Previous studies using ChIP-seq to assay Cas9 binding have shown that off-target binding at PAM plus seed sequences more frequently occurs in regions of open chromatin ( Wu et al . , 2014; O'Geen et al . , 2015 ) . Our data expand upon these findings to show that the discrete pattern of nucleosome organization is able to modulate the efficiency of Cas9 binding and cleavage at on-target sites . The practical implications of these observations are underscored by our finding that accounting for nucleosome occupancy offers a significant improvement in predictive power for sgRNA design . While our data show that nucleosomes strongly protect their DNA from Cas9 binding and cleavage in vitro , their organization in cells is not static . Transient displacement of nucleosomes occurs during replication , remodeling , and transcription . By adding the chromatin remodeling enzyme yChd1 to nucleosomes in vitro , we demonstrate that this displacement can in fact restore Cas9 access to DNA . However , despite brief exposure of nucleosomal DNA during remodeling and various other cellular processes , we still observe a clear anti-correlation between Cas9 activity and nucleosome occupancy in vivo , suggesting that the barrier to Cas9 target recognition exists even in a cellular environment . Indeed , the balance between nucleosome disruption , turnover , and repositioning in the cell leads to the average level of occupancy and positioning at each site observed by MNase-seq ( Valouev et al . , 2011; Jiang and Pugh , 2009 ) . Thus , our data suggest that it is likely the overall effect of this average nucleosome positioning that leads to the observed anti-correlation with Cas9/sgRNA activity . Additionally , it is important to note that there is likely a fundamental difference between applications that use dCas9 versus nuclease-active Cas9 . Knock-down of transcription by CRISPRi likely requires persistent binding by dCas9 to continually block transcription , and would be largely ineffective during S-phase when transcription is globally shut down . In contrast , to make a genomic edit , Cas9 must succeed in cleaving DNA only once , and could potentially take advantage of nucleosome turnover during replication . The role these differences play in the dependence of Cas9 on nucleosome position is still not clear . We expect , however , that nucleosome position and occupancy will be of particular concern to applications that use the nuclease-dead Cas9 and require sustained binding . Future investigations into the role of cell cycle and nucleosome disruption may provide an additional piece to our understanding of the mechanism of Cas9 in eukaryotic cells . Furthermore , nucleosome organization represents only one aspect of eukaryotic chromatin , and thus , our results contribute a first step in understanding and exploiting how chromatin affects Cas9 activity to enable more sophisticated and precise rules for targeting Cas9 .
The K562 dCas9-KRAB-BFP cell line was obtained from ( Gilbert et al . , 2014 ) and had been constructed from K562 cells obtained from ATCC . The resulting cell line tested negative for mycoplasma ( MycoAlert Kit , Lonza , Basel , Switzerland ) in regular screenings , and cytogenetic profiling by array comparative genomic hybridization ( not shown ) closely matched previous characterizations of the K562 cell line ( Naumann et al . , 2001 ) . Data from 30 published ( Gilbert et al . , 2014 ) and unpublished screens ( ML , BA , BB , CYP , MK , YC , JF , and JN , personal communication ) , conducted using the CRISPRi sgRNA library described in Gilbert et al . , ( 2014 ) in K562 cells constitutively expressing dCas9-KRAB-BFP , were processed through a standardized pipeline adapted from Bassik et al . ( 2013 ) , and Kampmann et al . ( 2013 ) . Briefly , sgRNA phenotypes were calculated as the log2 enrichment of sequencing read counts between two conditions ( e . g . initial and final timepoints for growth screens , untreated and treated for drug/toxin screens ) and normalized to cell doubling differences where appropriate . Most screens were conducted in duplicate , and sgRNA phenotypes from the duplicates were averaged . To determine hit genes , each gene was given an effect size ( average of strongest 3 sgRNA phenotypes by absolute value ) and a confidence value ( Mann-Whitney p-value of all ~10 sgRNAs compared to negative controls ) , and hits were selected using a score that integrates effect size and statistical confidence ( | effect Z score * log10 p-value | ≧ 20 in any screen ) . For genes with multiple TSS , each TSS was analyzed separately and the gene was assigned the highest score . Finally , sgRNA phenotypes were extracted for hit genes from the screen in which the gene scored as a hit and normalized to the average of the strongest 3 phenotypes to generate the 'sgRNA activity score' . sgRNA positions were defined as the genomic coordinate of the 3’ G of the NGG PAM ( all genomic coordinates referenced in this text are from hg19 ) . TSS positions were determined from the FANTOM5 project annotation ( Riken ) ( http://fantom . gsc . riken . jp/5/datafiles/phase1 . 3/extra/TSS_classifier/TSS_human . bed . gz; accessed March 2 , 2015 ) , using the downstream genomic coordinate of the corresponding 'p1@gene' BED file entry . All local averages were calculated in 50 bp windows centered around the indicated point . As sgRNA lengths including the PAM were ~24 bp and position was calculated relative to PAM , a window size of 50 bp captures all sgRNAs that directly neighbor the center point at either the 3’ or 5’ end . In order to quantify amplitude , the local averages were first smoothed using a low-pass Butterworth filter ( N = 4 , Wn = 0 . 03; SciPy signal processing module ) and then peaks and troughs were calculated by determining the local maxima and minima , respectively . As described for Figure 1D , Peak 0 was defined as the local maximum closest to the TSS , peak 1 was defined as the next maximum downstream of peak 0 , and troughs were defined as the minima immediately downstream of the respective peaks . K562 MNase-seq data was obtained from the ENCODE consortium as processed continuous signal data ( BigWig file format; accession number ENCFF000VNN , Michael Snyder lab , Stanford University ) . sgRNA target site signal was calculated as the average signal at all positions between the 5’ end of the sgRNA and the 3’ end of the PAM . K562 RNA-seq data were obtained from the ENCODE consortium as transcript quantifications ( accession number ENCFF485YKK , Thomas Gingeras lab , Cold Spring Harbor Laboratories ) . Genes were assigned expression levels in units of FPKM according to their highest-expressed transcript . Sequence score was calculated by passing the specified 30 bp target site for each sgRNA to the on_target_score_calculator . py script provided by Doench and colleagues ( accessed October 9 , 2015 ) ( Doench et al . , 2014 ) . This sequence score , MNase signal as calculated above , and sgRNA length ( base pairs in protospacer and PAM ) were each compared to sgRNA activity scores by Pearson correlation . Linear fits of the specified parameters were computed using multidimensional linear regression ( Sci-kit learn linear_model package ) , and correction of the activity scores for sequence and length features was performed by subtracting the predicted scores based on the combined fit . Screens for ricin susceptibility were performed essentially as previously described ( Gilbert et al . , 2014 ) . Briefly , K562 cells constitutively expressing Cas9-BFP from an SFFV ( spleen focus-forming virus ) promoter were transduced with our previously described pooled sgRNA tiling library packaged into lentivirus for a multiplicity of infection below 1 . Duplicate screens were infected and subsequently treated independently . Infected cells were allowed to recover for 2 days , then selected with 0 . 75 μg/mL puromycin ( Tocris ) for 2 days , and finally allowed to recover from puromycin treatment for 2 days . Cells were then cultured for 19 days and were either treated with three pulses of 0 . 5 ng/mL ricin administered for 24 hr and followed by re-suspension in fresh media , or passaged untreated . Genomic DNA was harvested from the endpoint untreated and treated samples and processed for high-throughput Illumina sequencing as previously described ( Gilbert et al . , 2014 ) . Screens were conducted at a minimum library coverage of 1000 cells per sgRNA , and sequenced to a median depth of ~500 reads per sgRNA . Phenotypes were calculated as log2 enrichments of read counts between untreated and treated conditions , normalized to cell doubling differences , and averaged between duplicates . Phenotype-signed Z scores were calculated by dividing all scores by the standard deviation of negative control phenotypes and then multiplying phenotypes by -1 for sgRNAs targeting genes shown to produce sensitizing phenotypes upon knockdown ( Bassik et al . , 2013 ) such that positive values represent 'expected' phenotypes . Mouse histones H2B ( T115C ) and H4 were recombinantly expressed in BL21 ( DE3 ) pLysS cells from expression plasmids gifted by Dr . Karolin Luger . Expression and purification of mH4 was conducted as previously described by the Luger lab ( Dyer et al . , 2004; Luger et al . , 1999 ) , while mH2B ( T115C ) was expressed and purified as described by the Cairns lab ( Wittmeyer et al . , 2004 ) with the following exception: after purification , histones were dialyzed against multiple changes of double distilled water and 1 mM β-mercaptoethanol ( BME ) before lyophilizing for storage . Purified recombinant mouse ( Mus Musculus ) histones H2A and H3 were gifts from Dr . Karolin Luger . The labeling mutant , mH2B ( T115C ) was fluorescently labeled with ~five-fold molar excess Alexa Fluor 647 C2 maleimide dye ( Thermo Fisher Scientific , Waltham , MA , USA ) as follows . 2 mg of lyophilized mH2B ( T115C ) was dissolved at 2 mg/mL in labeling buffer ( 7 M GuHCl , 50 mM Tris-HCl pH 7 . 5 , 1 mM TCEP ) and nutated at room temperature for 30 min . 1 mg of the dye was then dissolved in 50 µl of anhydrous DMF under Argon gas , and approximately half of the dye solution was slowly mixed with the dissolved mH2B ( T115C ) in the dark at room temperature to begin the labeling reaction . After nutating the reaction for 1 hr , the rest of the dissolved dye was slowly added , and the reaction was moved to 4°C overnight in the dark . In the morning , the reaction was quenched by adding over hundred-fold molar excess of BME . Histone octamers were refolded and purified as previously described ( Dyer et al . , 2004 ) . Specifically , 110 nmoles each mH3 and mH4 were refolded with 130 nmoles each mH2A and mH2B ( T115C ) . The resulting octamer was concentrated using a 10 , 000 MWCO Spin-X UF concentrator ( Corning , Tewksbury , MA ) , then purified on a Superdex 200 HR ( 10/30 ) column ( GE Life Sciences , Pittsburgh , PA ) using an Akta Explorer FPLC ( GE Life Sciences , Pittsburgh , PA ) at 0 . 2 mL/min . Selected fractions were concentrated , flash frozen , and stored at -80°C until use . A new purification scheme was conceived to achieve exceptionally high purity ( d/wt ) Cas9 ( Figure 4—figure supplement 3 ) . Nuclease active S . pyogenes 6His-Cas9-HaloTag-NLS ( HaloTag is a registered trademark of Promega , Madison , WI ) was recombinantly expressed and purified from BL21 ( DE3 ) pLysS-Rosetta cells ( Novagen/EMD Millipore , Darmstadt , Germany ) using the expression plasmid pET302-6His-wtCas9-Halo-NLS , while the nuclease dead S . pyogenes 6His-dCas9 ( D10A , H840A ) -HaloTag was recombinantly expressed and purified from BL21 ( DE3 ) pLysS cells using the expression plasmid pET302-6His-dCas9-Halo ( Knight et al . , 2015 ) . Bacterial cultures were grown in Terrific Broth II ( MP Biomedicals , Santa Ana , CA ) at 37°C until an OD600 reached 0 . 4 . Cultures were then transferred to an ice bath for ~15 min until an OD600 reached 0 . 5 , at which point expression was induced with 0 . 2mM IPTG , and the cultures were moved to an 18°C shaker for 16 hr . Cells were harvested at 3000xg for 20 min , then resuspended in lysis buffer ( 500 mM NaCl , 50 mM Hepes pH 7 . 6 , 5% glycerol , 1% Triton X-100 , 10 mM imidazole , 1 mM benzamadine , 2 . 3 µg/mL aprotinin , 0 . 5 mM PMSF , and 1 tablet per 50 mL of Protease Inhibitor Cocktail ( Roche , Basel , Switzerland ) . Cells were lysed using sonication on ice at 50% duty cycle , power 8 , 30 seconds on , 1 minute off ( Misonix/ Qsonica , LLC , Newtown , CT ) . The lysed cells were then ultracentrifuged at 4° , 40K rpms , for 40 min to remove cell debris . The supernatant was then allowed to bind to Ni-NTA agarose resin ( Qiagen , Hilden , Germany ) by nutating at 4°C for 30 min . The resin containing bound ( d/wt ) Cas9 was poured into a mini column ( Bio-Rad , Hercules , CA ) and washed with 10 column volumes ( CV ) of lysis buffer , and 5 CVs of 20 mM Imidazole buffer ( same as lysis buffer but with 20 mM imidazole ) . Elution of the ( d/wt ) Cas9 was achieved using 250 mM Imidazole buffer ( same as lysis buffer but with 250 mM imidazole ) , and fractions were checked on an SDS-PAGE gel . Chosen fractions were pooled and diluted to a starting NaCl concentration of 200 mM using Buffer A ( 0 M NaCl , 50 mM Hepes pH 7 . 6 , 5% glycerol , 1 mM DTT , 0 . 5 mM PMSF ) . A 5 mL HiTrap Q-HP ( GE Life Sciences , Pittsburgh , PA ) and a 5 mL HiTrap SP-HP ( GE Life Sciences , Pittsburgh , PA ) column were attached in tandem ( Q first in line ) and equilibrated on an Akta FPLC at 10% Buffer B ( same as Buffer A except with 2 M NaCl ) . The pooled ( d/wt ) Cas9 was filtered , then loaded onto the tandem columns at 2 mL/min . The columns were washed with 10% Buffer B until A280 and A260 returned to baseline , at which point the Q column was removed and ( d/wt ) Cas9 was eluted from the SP column using a gradient from 10% to 50% Buffer B over 10 CVs . Fractions were chosen using SDS-PAGE , pooled , and dialyzed into storage buffer ( 200 mM NaCl , 50 mM Hepes pH 7 . 6 , 5% glycerol , 1 mM DTT ) ( Figure 4—figure supplement 3 ) . Aliquots were flash frozen in liquid nitrogen and stored at -80°C . To make the fluorescent DNA used in this study , the 601 DNA sequence was amplified by PCR from plasmid pBSIISK+601 ( parent 601 sequence plasmid gifted by K Luger ) with the primers listed below . Two different PCR products were produced; one labeling the Watson strand of the 601 sequence with a 5’ Cy3 dye ( IDT , Coralville , IA ) , and the other labeling the Crick strand . Large-scale ( ~2 mL ) PCR reactions using in-house produced Pfu DNA polymerase were ethanol precipitated before loading onto a 20cm x 20cm 12% native TBE-PAGE gel for DNA purification . The PCR product was cut out of the gel , and the gel slice was crushed and soaked in 0 . 3 M sodium acetate pH 5 . 2 with multiple buffer changes . The pooled extract was ethanol precipitated , resuspended in 10 mM Tris-HCl pH 7 . 5 , 10 mM NaCl , and stored at -20º until use . sgRNA was produced by T7 transcription of a short DNA oligo template . To create this template , two oligos , one containing the T7 promoter and DNA target sequence , and the other encoding the invariable scaffolding of the sgRNA , were annealed and filled in using a single PCR cycle . The DNA template was ethanol precipitated , then transcribed using the T7 Quick High Yield RNA Synthesis Kit ( New England Biolabs , Ipswich , MA ) according to vendor instructions . The resulting sgRNAs were purified via 6% Urea-PAGE gel . The correct band was cut out and crushed and soaked in 0 . 3M sodium acetate pH 5 . 2 . After ethanol precipitation , and resuspension in RNase-free water , aliquots were stored at -80°C until use . SgRNAs targeting the 601 sequence included 20 bp of complimentarity 5’ to the targeted PAM , with the exception of PAM 4 , which has room for only 19 bp of complimentarity . SgRNAs were used against PAMs 1 , 3–15 , and 17–20 . PAM 2 did not have a long enough target sequence available within the 601 . The non-sense guides used in this study contained the target sequences 5'-ACATGTTGATTTCCTGAAA-3' or 5'-GATTTCACCTCTCAGCGCAT-3' . Nucleosomes were assembled by salt gradient dialysis as previously described ( Dyer et al . , 2004 ) using 50 µL home-made dialysis buttons and ~1 µM DNA . The Watson and the Crick labeled 601 DNA were assembled into separate nucleosomes . The best DNA:octamer molar ratio for optimal nucleosome assembly was selected by titrating octamer . The most homogenous assembly as judged by 6% Native TBE-PAGE gel was chosen for future assays . Cleavage and binding assays in Figures 3 , 4 and their corresponding supplements were both conducted in the same manner , exchanging dCas9 for wtCas9 during binding reactions . First , the complete ribonucleoprotein complex was formed by incubating five-fold molar excess of the chosen sgRNA with ( d/wt ) Cas9 at 37°C for 10 min . Next , the DNA substrate ( either naked DNA or nucleosome ) was added to 40 nM ( five-fold less than Cas9 ) , and the reactions were returned to 37º for an hour . Reaction buffer contained 20 mM Hepes pH 7 . 5 , 100 mM NaCl , 5 mM MgCl2 , 1 mM EDTA , and 2 . 5 mg/mL insulin ( Roche , Basel , Switzerland ) . Importantly , we found that including a nonspecific protein such as insulin prevents the nucleosomes from falling apart and getting lost to surfaces during binding and cleavage assays ( Figure 4—figure supplement 1B ) . Additionally , inclusion of insulin reduces nonspecific protein-protein interactions and aggregation . Binding reactions also contained 14 . 6% sucrose to allow direct loading onto a native gel . At 1 hr , binding reactions were loaded onto a pre-run 6% native TBE-PAGE gel at 4°C , while cleavage reactions were stopped using a 5X stop buffer ( 250 mM EDTA , 2% SDS ) , and prepared to load onto a 10% Urea-PAGE gel by adding 2X loading buffer ( 95% formamide , 20% DMSO , 5 mM EDTA , 0 . 025% Orange G ) and heating to 95°C for 5 min before snap cooling on ice . Binding and cleavage reactions were repeated at least twice each for most of the PAMs targeted . Gels were imaged using a PharosFX Plus ( Bio-Rad , Hercules , CA ) and quantified using Image Lab ( Bio-Rad , Hercules , CA ) . For binding and cleavage reactions , gels were scanned in the Cy3 fluorescence channel , while Alexa Fluor 647 fluorescence was also imaged for binding reactions . For cleavage reactions , labeling of either the Watson or Crick strand was chosen such that the fluorophore was always attached to the strand complimentary to the sgRNA used . Molecular graphics and analyses were produced using the UCSF Chimera package ( supported by NIGMS P41-GM103311 ) ( Pettersen et al . , 2004 ) . The crystal structures for the 601 nucleosome ( Vasudevan et al . , 2010 , PDBID 3LZ0 ) and the Cas9-sgRNA-DNA ternary complex ( Anders et al . , 2014 , PDBID 4UN3 ) were superimposed by aligning the DNA path in both structures using MatchMaker and manual manipulation . The solvent accessible surface area of the DNA in the 601 nucleosome structure was computed using residue areaSAS . Supercoiled plasmid was produced using a Qiagen Maxi prep kit ( Hilden , Germany ) . Histone octamers were prepared from purified histones as described above . DNA and histone octamer were mixed at a weight ratio of 1:0 . 8 ( DNA to octamer ) at a concentration of 0 . 4 µg/µL of DNA in 1X TE with 2 M NaCl . 50 µL homemade dialysis buttons were used to dialyse the solution by a step-wise gradient in 500 mL at room temperature . The gradient was as follows; 1 . 5 hr in 1 M NaCl in TE , 2 hr in 0 . 8 M NaCl in TE , 1 . 5 hr in 0 . 6 M NaCl in TE , and 2 hr in 0 . 05 M NaCl in TE . The resulting chromatin was stored at 4°C until use . MNase assays were performed essentially as in Torigoe et al . ( 2013 ) and Alexiadis ( 2002 ) . MNase ( Sigma-Aldrich , St . Louis ) was resuspended at 200 units/mL in water and then disluted at 1:100 , 1:500 , or 1:1000 in a solution of 1X MNase reaction buffer ( 50 mM Tris-HCl pH7 . 9 and 5 mM CaCl2 dihydrate ) , 0 . 1 mg/mL insulin , and 10% glycerol . Chromatinized plasmid assemblies in 1X MNase buffer were added to each MNase dilution . As a control , an equal amount of unchromatinized super-coiled plasmid was added to the lowest MNase dilution . Each reaction was incubated for 11 min at room temperature , then stopped in a solution with final concentrations of 20 mM EGTA , 200 mM NaCl , 1% SDS , 20 mg/mL GlycoBlue ( Thermo Fisher Scientific , Waltham , MA ) , and 13 . 4 mg/mL Proteinase K ( Roche , Basel , Switzerland ) and incubated at 37°C for 30 min . Reactions wer Phenol/Chloroform/Isoamyl Alcohol extracted , ethanol precipitated , and run on a 1 . 3% agarose gel in 0 . 5X TBE . 3 μg of the 123 bp DNA ladder ( Thermo Fisher Scientific , Waltham , MA ) was used as a size standard . Either chromatin assemblies or super-coiled plasmid with equivalent amounts of DNA were added to restriction enzyme master mixes for final concentrations of 1X CutSmart buffer ( New England Biolabs , Ipswich , MA ) , 6 ng/μL DNA , and 0 . 4 U/μL of the indicated restriction enzyme ( New England Biolabs ) . Reactions were incubated for 1 hr at NEB recommended temperatures . Reactions were stopped in final concentrations of 15 mM EDTA , 0 . 75% SDS , 150 mM NaCl , and 15 mg/mL Proteinase K and incubated at 37°C for 30 min . DNA was extracted with Phenol:Chloroform:Isoamyl Alcohol and ethanol precipitated , then resuspended in 1X CutSmart buffer and linearized by digesting with 1 U/μL DraIII-HF ( New England Biolabs ) at 37°C for 1 hr . The full reactions were run on an agarose gel and quantified as above . Chromatin remodeling assays were performed simmilar to Torigoe et al . ( 2013 ) and Alexiadis ( 2002 ) . Reactions contained 40 mM NaCl , 0 . 1 mg/mL BSA , 25 mM Tris-Acetate pH 7 . 5 , 10 mM Mg-Acetate , 1 mM DTT , 1 . 2 mM ATP , and 6 ng/μL chromatinized or supercoiled plasmid . For restriction enzyme accessibility , the indicated restriction enzyme was added at 0 . 5 U/µL . For Cas9 accessiblity , Cas9/sgRNA ribonucleoproteins assembled as described above were added at 15 . 62 nM . A truncated version of the chromatin remodeling enzyme yChd1 was used spanning amino acids 118–1274 , also referred to as yChd1ΔNC ( gift of Dr . Ashok Patel and Dr . Gregory Bowman ) , and was added to the indicated reactions at a 0 . 2 μM final concentration . Reactions were incubated at 27°C for 1 hr , then processed as with the restriction enzyme accessibility assay above . | Many bacteria have a type of immune system known as CRISPR that can target and cut foreign DNA to protect it against viruses . Recently , the CRISPR system was adapted to allow scientists to easily manipulate the genome of humans and many other organisms . However , unlike the loosely organized DNA found in bacteria , the DNA that makes up the human genome is tightly packed and wrapped around complexes of proteins to form structures called nucleosomes . It was not clear whether the CRISPR system was able to effectively target the stretches of DNA in a nucleosome . In 2013 , researchers developed a modified version of CRISPR , known as CRISPR interference , to block gene activity and in 2014 used it to systematically repress many of the genes in the human genome . Now , Horlbeck , Witkowsky et al . – who include several of the researchers from the 2014 work – have analyzed existing data for a specific type of human cell grown in the laboratory and found that CRISPR interference activity was strongest in certain areas around the start of each gene . However , CRISPR interference was much weaker in other areas of genes that coincided well with stretches of DNA that are known to often be bound by nucleosomes . Nucleosomes also appeared to block CRISPR editing , although the effects were less pronounced . Horlbeck , Witkowsky et al . then directly tested whether nucleosomes could prevent the CRISPR system from binding or modifying the DNA . When the individual components were mixed in test tubes , the CRISPR system could readily target “naked” DNA . However , it could not access nucleosome-bound DNA , unless an enzyme that can move nucleosomes along the DNA in the human genome was also added to the mix . These findings suggest one way that CRISPR can manipulate much of the human genome despite the widespread presence of nucleosomes . Future work will now aim to develop computational methods that take the positions of nucleosomes into account when picking DNA sites to target with CRISPR . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"short",
"report"
] | 2016 | Nucleosomes impede Cas9 access to DNA in vivo and in vitro |
The vertebrate eye primordium consists of a pseudostratified neuroepithelium , the optic vesicle ( OV ) , in which cells acquire neural retina or retinal pigment epithelium ( RPE ) fates . As these fates arise , the OV assumes a cup shape , influenced by mechanical forces generated within the neural retina . Whether the RPE passively adapts to retinal changes or actively contributes to OV morphogenesis remains unexplored . We generated a zebrafish Tg ( E1-bhlhe40:GFP ) line to track RPE morphogenesis and interrogate its participation in OV folding . We show that , in virtual absence of proliferation , RPE cells stretch and flatten , thereby matching the retinal curvature and promoting OV folding . Localized interference with the RPE cytoskeleton disrupts tissue stretching and OV folding . Thus , extreme RPE flattening and accelerated differentiation are efficient solutions adopted by fast-developing species to enable timely optic cup formation . This mechanism differs in amniotes , in which proliferation drives RPE expansion with a much-reduced need of cell flattening .
The retinal pigment epithelium ( RPE ) is an essential component of the vertebrate eye , composed of a monolayer of pigment-enriched epithelial cells abutting the neural retina ( NR ) with a primary role in photoreception ( Letelier et al . , 2017 ) . Despite the acquisition of specialized epithelial properties , RPE cells have a neural origin and share progenitors with the NR . These progenitors are organized in a pseudostratified neuroepithelium , known as optic vesicle ( OV ) or eye primordium . In amniotes , the OVs appear as balloon-like structures positioned at the sides of the anterior neural tube ( Moreno-Marmol et al . , 2018 ) . In zebrafish instead , these primordia are flat and form two bi-layered structures with the outer and inner layers distally connected by a rim or hinge ( Li et al . , 2000 ) . Under the influence of inductive signals ( Gallardo and Bovolenta , 2018; Cardozo et al . , 2020 ) , the two layers activate different genetic programs that specify the cells of the inner layer and ventral outer layer as NR and those of the dorsal outer layer as RPE ( Beccari et al . , 2013; Buono and Martinez-Morales , 2020; Buono et al . , 2021 ) . Whilst this specification occurs , the OV bends assuming a cup-like shape ( Martinez-Morales et al . , 2017 ) . The discovery of the ojoplano medaka fish mutant – affecting a transmembrane protein localized at the basal end feet of NR cells ( Martinez-Morales et al . , 2009 ) – in which the OV remains unfolded , was instrumental to propose that basal constriction of NR progenitors is at the basis of OV bending ( Martinez-Morales et al . , 2009 ) . This basal constriction is mediated by the redistribution of the actomyosin cytoskeleton ( Martinez-Morales et al . , 2009; Nicolas-Perez et al . , 2016; Bryan et al . , 2016 ) , which also enables the apical relaxation of retinal cells ( Sidhaye and Norden , 2017 ) , enhanced by focal adhesions of the apical surface with the extracellular matrix molecules ( ECM ) such as laminin ( Bryan et al . , 2016 ) . The importance of concomitant apical relaxation , especially of the cells positioned at the hinge , has also been supported in studies of mammalian retinal organoids ( Eiraku et al . , 2011; Okuda et al . , 2018 ) . Nevertheless and independently of their relative contribution , the acquisition of apical convexity and basal concavity in the NR epithelium are accepted drivers of the biomechanical forces that induce OV folding ( Okuda et al . , 2018 ) . In zebrafish , this mechanism is reinforced by rim involution or epithelial flow , a process whereby progenitors at the hinge emit dynamic lamellipodia at the basal side and actively translocate from the ventral outer layer of the OV into the inner/retinal layer ( Li et al . , 2000; Sidhaye and Norden , 2017; Zheng et al . , 2000; Heermann et al . , 2015; Kwan et al . , 2012; Picker et al . , 2009 ) . Periocular neural crest cells appear to facilitate this flow , in part by the deposition of the ECM ( Bryan et al . , 2020 ) to which the lamellipodia attach ( Sidhaye and Norden , 2017; Heermann et al . , 2015; Kwan et al . , 2012 ) . The result of this flow is an unbalanced cell number between the two layers , which should favour NR bending ( Sidhaye and Norden , 2017; Heermann et al . , 2015; Kwan et al . , 2012 ) . Whether this flow may also contribute to the concomitant cell shape modifications that the remaining outer layer cells undergo as they become specified into RPE , or conversely whether RPE specification favours the flow ( Heermann et al . , 2015 ) , remain open questions . Indeed as the OV folds , the pseudostratified neuroepithelial cells of the OV dorsal outer layer progressively align their nuclei becoming a cuboidal monolayer in amniotes species ( Moreno-Marmol et al . , 2018; Martinez-Morales et al . , 2004 ) . In zebrafish , cuboidal cells further differentiate to a flat/squamous epithelium ( Zheng et al . , 2000; Kwan et al . , 2012 ) that spreads to cover the whole apical surface of the NR ( Zheng et al . , 2000; Cechmanek and McFarlane , 2017 ) . In mice , failure of RPE specification , as observed after genetic inactivation of key specifier genes ( i . e . Otx1/Otx2 , Mitf , Yap/Taz ) , enables RPE progenitors to acquire an NR fate ( Martinez-Morales et al . , 2001; Bharti et al . , 2006; Kim et al . , 2016 ) . The resulting optic cups ( OCs ) present evident folding defects ( Martinez-Morales et al . , 2001 ) , raising the possibility that specific RPE features are needed for OC formation . In line with this idea , a differential stiffness of the RPE vs . the NR layer has been proposed to drive the self-organization of mammalian organoids into an OC ( Eiraku et al . , 2011; Okuda et al . , 2018; Nakano et al . , 2012 ) . Furthermore , generation of proper RPE cell numbers seems a requirement for correct OC folding in mice ( Carpenter et al . , 2015 ) . However , studies addressing the specific contribution of the RPE to OV folding are currently lacking . Here , we report the generation of a Tg ( E1-bhlhe40:GFP ) zebrafish transgenic line with which we followed the beginning of RPE morphogenesis under both normal and interfered conditions . We show that , whereas in amniotes , including humans , the developing RPE undergo proliferation to increase its surface with a less evident cell flattening , zebrafish RPE cells rapidly cease proliferation and expand their surface by reducing their length along the apico-basal axis and extending in the medio-lateral direction with a tissue autonomous process that depends on cytoskeletal reorganization . Localized interference with either the retinal or the RPE actomyosin and microtubule cytoskeleton shows that RPE flattening generates a mechanical force that actively contributes to OV folding , complementing the force generated by the basal constriction of the NR . This mechanism represents an efficient solution to match the increased apical surface of the NR layer in a fast-developing vertebrate species such as zebrafish .
Detailed analysis of zebrafish RPE morphogenesis has been hampered by the lack of a suitable transgenic line , in which RPE cells could be followed from their initial commitment . The E40 ( bhlhe40 ) gene , a basic helix-loop-helix family member , encodes a light and hypoxia-induced transcription factor ( also known as Dec1 , Stra13 , Sharp2 , or Bhlhb2 ) involved in cell proliferation and differentiation as well as in the control of circadian rhythms ( Yamada and Miyamoto , 2005 ) . In neurulating zebrafish embryos , its expression is limited to cells of the prospective RPE ( Figure 1A; Cechmanek and McFarlane , 2017; Yao et al . , 2006 ) , representing a potentially suitable tissue marker . We used predictive enhancer and promoter epigenetic marks at different zebrafish developmental stages ( Bogdanovic et al . , 2012 ) to scan the bhlhe40 locus for the presence of conserved and active regulatory regions . The promoter and four potential enhancers ( E1–4; Figure 1B ) appeared to be active between 80% epiboly and 24 hpf , encompassing the early stages of zebrafish eye development ( Bogdanovic et al . , 2012 ) . These enhancers were selected , amplified , and tested using the ZED vector ( Bessa et al . , 2009 ) as potential drivers of gene expression in the prospective RPE . The resulting F0 embryos were raised to adulthood and screened . Only the E1 enhancer drove specific and restricted GFP reporter expression into the prospective RPE . The corresponding fishes were further crossed to establish the stable transgenic line Tg ( E1-bhlhe40:GFP ) used in this study . Time-lapse studies of the Tg ( E1-bhlhe40:GFP ) progeny confirmed that the transgenic line faithfully recapitulated the bhlhe40 mRNA expression profile detected with ISH ( Figure 1A and C ) . GFP reporter expression appeared in a discrete group of neuroepithelial cells in the dorso-medial region of the OV ( 16–17 hpf ) and expanded both posteriorly and ventrally ( Figure 1C; Figure 1—video 1 and Figure 1—video 2 ) , so that , by 24 hpf , GFP-positive cells appeared to wrap around the entire inner NR layer . 3D reconstructions of selected embryos further confirmed the fast ( about 7 hr ) expansion of the GFP-positive domain forming an outer shell for the eye ( Figure 1D ) . Apart from a faint and very transient signal in some early NR progenitors — likely due to the existence of negative regulatory elements not included in the construct — no GFP expression was observed in regions other than the RPE during this process . However , after the formation of the OC , reporter expression appeared also in the ciliary marginal zone ( CMZ ) , the pineal gland , and few neural crest cells surrounding the eye ( Figure 1C; Figure 1—videos 1–3 ) . These additional domains of expression coincided with the reported bhlhe40 mRNA distribution ( Yao et al . , 2006 ) and represented no obstacle for using the transgenic line as a tool to follow the early phases of RPE generation . Indeed , very early activation represents an important advantage of the Tg ( E1-bhlhe40:GFP ) line over other presently available transgenic lines that allow visualizing the RPE ( Zou et al . , 2006; Miesfeld and Link , 2014 ) . The suitability of the Tg ( E1-bhlhe40:GFP ) line for the identification of the very first RPE cells is supported by the onset of the reporter expression in the dorso-medial OV region , coinciding with previous fate map predictions ( Zheng et al . , 2000; Kwan et al . , 2012 ) . To further verify this notion , we took advantage of the characteristic of the fluorescent Kaede protein ( Ando et al . , 2002 ) that switches from green to red emission upon UV illumination . Embryos were injected with Kaede mRNA and neuroepithelial cells located at the most dorso-medial region of the OV were UV illuminated at the 15 hpf stage to ensure that no differentiation had yet occurred ( Figure 1E ) . Embryos were let develop until 30 hpf . Photoconverted cells were found throughout the thin outer layer of the OC ( Figure 1E ) , confirming that the entire RPE derives from the dorso-medial OV region . Tg ( E1-bhlhe40:GFP ) embryos were thereafter used to dissect the extensive changes in cell shape that are associated with the acquisition of RPE identity ( Zheng et al . , 2000; Cechmanek and McFarlane , 2017 ) . At OV stage all retinal progenitors present a columnar-like morphology characteristic of embryonic neuroepithelia ( Figure 2A and A’ ) . As soon as RPE progenitors begin to express the transgenic GFP reporter , their apico-basal length rapidly and progressively reduces ( Figure 2A–C’ ) , so that the cells first assume a cuboidal shape ( Figure 2B and B’ ) and then become flat , forming a squamous epithelial monolayer overlaying the apical surface of the NR ( Figure 2C and C’ ) . At 30 hpf , RPE cells presented a polygonal , frequently hexagonal , morphology ( Figure 2D and D’ ) , with an apical surface area that , on average , became about eightfold larger than that observed in progenitor ( PN ) cells ( Figure 2F; RPE‾a: 354 . 8 ± 100 . 3 μm2 vs . PN‾a: 43 . 7 ± 7 . 8 μm2 ) . In contrast , the abutting apical surface of NR cells slightly shrank as compared to that of PN cells ( Figure 2E , E’ and F; NR‾a: 22 . 5 ± 2 . 9 μm2 vs . PN‾a: 43 . 7 ± 7 . 8 μm2 ) while maintaining a constant apico-basal length . The latter observation agrees with previous reports showing that the cone-like morphology of NR progenitors represents only a slight modification of the progenitor columnar shape ( Nicolas-Perez et al . , 2016; Sidhaye and Norden , 2017 ) . To obtain a quantitative analysis of the dynamic changes that RPE tissue , as whole , underwent during OV folding , we performed a morphometric characterization of the images from Figure 1—videos 1–3 . To this end , the fluorescent information from the Tg ( E1-bhlhe40:GFP ) reporter was discretized into seven different segments that were individually analysed along the recording time ( Figure 3—figure supplement 1; Materials and methods ) . The combined quantification of the different segments ( Figure 3A; Figure 3—figure supplement 1 ) showed that , between stages 17 and 21 hpf , the overall thickness of the RPE tissue underwent , on average , a flattening of more than threefold ( from a mean of about 24–8 µm; Figure 3B ) . Flattening occurred with a central to peripheral direction , so that RPE cells closer to the hinges were the last ones to flatten ( Figure 2C and C’ ) . In parallel , the overall RPE surface underwent an approximately twofold expansion between 17 and 22 hpf ( from approximately 1 . 1–2 . 2 × 103 μm2; Figure 3C; Figure 1—video 1 and Figure 1—video 2 ) , reflecting the large increase in the apical area observed in each individual cell at later stages ( Figure 2 ) . In line with the idea that cell flattening is per se sufficient to account for whole tissue enlargement , the RPE volume only slightly changed between 17 and 20 hpf with a slope increase of 0 . 47 × 103 μm3/h ( Figure 3D ) . To provide further support to this idea , we analysed the RPE volume variation in comparison with the growth of the entire OC in two time windows: from 17 to 22 hpf ( Figure 1—videos 1–3 ) and from 24 to 37 hpf ( Figure 3—video 1 ) , using GFP ( RPE ) and RFP ( eye ) reporter signals from the double Tg ( E1-bhlhe40:GFP; rx3:GAL4;UAS;RFP ) line or from the Tg ( E1-bhlhe40:GFP ) line injected with the pCS2:H2B-RFP mRNA ( Figure 3E and F ) . Signal quantification showed that the eye underwent a marked and linear volume increase ( slope: 5 . 54 × 104 μm3/hr from 17 to 22 hpf and 3 . 6 × 104 μm3/hr from 24 to 37 hpf ) as compared to that of the RPE ( Figure 3D and G ) . Between 20 and 22 hpf the reporter starts being expressed in the posterior and , to a lesser extent , in the anterior CMZ ( GFP-CMZ domain , Figure 1—videos 1 and 2 ) . Consistently with the onset of GFP-CMZ expression , RPE reporter volume suddenly expanded between 20 and 22 hpf ( slope: 1 . 25 × 104 μm3/hr; Figure 3G ) to then slow back between 24 and 37 hpf ( slope: 1 . 2 × 103 μm3/hr; Figure 3G ) . Confirming this association , only the tissue segments very close to the posterior CMZ had a volume larger than that of the RPE at 17–20 hpf ( Figure 3—figure supplement 1 ) , whereas the GFP-positive RPE domain located in the most central regions presented a volume undistinguishable from that detected at previous stages . In sum , a comparison of the dynamics slopes from the GFP-RPE domain and OV regions suggests that the volume of the RPE grows at very low pace ( 0 . 47 × 103 μm3/hr ) – despite the rather drastic morphological changes of its cells – whereas the whole OV expands at a pace ~25 times faster ( 1 . 25 × 104 μm3/hr; Figure 3D ) . Taken all together , this morphometric analysis indicates that the expansion of the RPE in zebrafish occurs by recruiting a limited number of cells that undergo profound cell shape changes: from a neuroepithelial to squamous morphology . Both external interactions and intracellular processes determine the shape of a cell and define its mechanical properties ( Totaro et al . , 2018 ) . Thus , in principle , RPE flattening might occur as a ‘passive’ process , triggered by the forces that the NR and hinge cells exert on the RPE ( Moreno-Marmol et al . , 2018; Heermann et al . , 2015 ) . Alternatively , it might depend on cell or tissue autonomous cytoskeletal rearrangements , involving , for example , myosin II activity , which controls the acquisition of a flat epithelial morphology in other contexts ( Tee et al . , 2011; Vishavkarma et al . , 2014 ) . Discriminating between these two possibilities has been technically difficult . Experiments directed to assess the mechanisms of OV folding have used whole embryo bathing in drugs such as blebbistatin ( Nicolas-Perez et al . , 2016; Sidhaye and Norden , 2017 ) , a specific myosin II inhibitor ( Rauscher et al . , 2018 ) . Such an approach hampers the assessment of the potential influence of NR over RPE morphogenesis ( and vice versa ) as well as the relative contribution of the two tissues to OV folding . We sought to overcome this limitation by spatially localized interference with the cytoskeletal organization of either the RPE or NR and by recording the tissue autonomous and non-autonomous consequences . Nevertheless , to begin with , we reproduced the whole embryo bathing approach used by others ( Nicolas-Perez et al . , 2016; Sidhaye and Norden , 2017 ) , focusing on the yet unreported effect that blebbistatin had on the RPE . Tg ( E1-bhlhe40:GFP ) embryos were bathed either in blebbistatin or its diluent ( DMSO ) at 17 hpf ( the onset of RPE specification; Figure 4A–C ) and then let develop up to 19 . 5 hpf , when embryos were analysed . DMSO-treated ( control ) embryos developed normally forming an OC surrounded by a squamous RPE ( Figure 4B ) . In blebbistatin-treated embryos , NR cells did not undergo basal constriction and the OV remained unfolded ( Figure 4C ) , as previously described ( Nicolas-Perez et al . , 2016; Sidhaye and Norden , 2017 ) . Notably , in almost all the embryos analysed ( n = 44/49 ) , RPE cells did not flatten but remained cuboidal in shape ( Figure 4C ) . A similar phenotype was observed after treatment with paranitroblebbistatin , a non-cytotoxic and photostable version of blebbistatin ( Figure 4D ) . These observations support that lack of OV folding is associated with alterations in both the retina and RPE . To uncouple the two events , we turned to the photoactivable compound azidoblebbistatin ( Ableb ) , which binds covalently to myosin II upon two-photon irradiation , thus permanently interfering with myosin II activity in a spatially restricted manner , as already proven ( Kepiro et al . , 2012; Kepiro et al . , 2015 ) . Tg ( E1-bhlhe40:GFP ) 17 hpf embryos were bathed in Ableb or in DMSO and irradiated in a small region of either the dorsal outer layer ( RPE ) or the inner retinal layer ( retina ) of the OV ( see Materials and methods ) . Embryos were then let develop until 24 hpf . During this period , the irradiated RPE cells underwent anterior and medio-lateral spreading – likely coinciding with the reported pinwheel ‘movement’ ( Kwan et al . , 2012 ) – and were mostly found in the ventral half of the OV . Notably , Ableb photoactivation in the prospective RPE cells reproduced , although slightly less efficiently ( n = 30/44 embryos ) , the phenotype observed upon whole embryo bathing in blebbistatin , in which RPE cells acquired a cuboidal morphology ( Figure 4C , D , G , G' ) . No detectable alterations were found in the OV of irradiated/DMSO-treated embryos or in the contralateral non-irradiated OV of embryos incubated in Ableb regardless of the irradiated region ( Figure 4E , F' , H , I' ) . Cell shape quantifications showed a significantly longer apico-basal axis ( Figure 4E’–G’ ) in irradiated Ableb RPE cells , normalized to that of control ( DMSO- and Ableb-treated non-irradiated ) OVs ( Figure 4K; Mann-Whitney U test , z = −5 . 088 , p < 0 . 001 , control mean length 15 . 96 vs . Ableb-treated 38 . 03 ) . Failure of cell flattening in the irradiated region of the RPE was consistently associated with a significant reduction of OV folding ( Figure 4E–G ) , as assessed by measuring the invagination angle ( Sidhaye and Norden , 2017 ) , which was normalized to that of control embryos ( Figure 4L; Mann-Whitney U test: z = −2 . 704 , p < 0 . 01 , mean rank for control 21 . 60 vs . Ableb-treated 33 . 33 ) . Photoactivation of Ableb in similar areas of the prospective NR basal region resulted in an elongated NR and a significantly impaired OV folding ( Figure 4J ) , as determined by the invagination angles normalized to those of control OV ( Figure 4N; Mann-Whitney U test: z = −3 . 035 , p < 0 . 01 , mean rank control 10 . 29 vs . Ableb 20 . 06 ) . Notably , disruption of NR morphogenesis had no consequences on RPE development in all the analysed embryos ( n = 16/22 ) : cells underwent normal flattening with apico-basal lengths comparable to those of controls ( Figure 4H’–J’ and M; Mann-Whitney U test: z = 0 . 582 , p > 0 . 05 , mean rank control 14 . 50 vs . Ableb 16 . 38 ) . These data strongly support that RPE flattening is not secondary to NR folding but rather a tissue autonomous event . They also indicate that OV folding requires forces independently generated in both the NR and RPE . Notably , blebbistatin or Ableb treatments did not compromise the expression of the Tg ( E1-bhlhe40:GFP ) transgene in any experimental condition , indicating that cellular tension and morphology did not affect RPE specification . Microtubule dynamics has an important role in determining the shape of a cell ( Yevick and Martin , 2018 ) . For example , reorientation of the microtubule cytoskeleton from the apico-basal to the medio-lateral cell axis together with actin filaments redistribution seems to drive the conversion of the Drosophila amnioserosa cells from a columnar to squamous epithelium ( Pope and Harris , 2008 ) . To determine if a similar reorientation occurs in the RPE , we used time-lapse analysis of Tg ( E1-bhlhe40:GFP ) embryos injected with the mRNA of EB3:GFP , a protein that binds to the plus end of growing microtubules ( Stepanova et al . , 2003 ) . In neuroepithelial RPE progenitors , microtubules grew in the apico-basal direction , whereas growth turned to the medio-lateral plane , as the RPE cells became squamous ( Figure 5—figure supplement 1; Video 1 ) . To determine if this reorientation is important for cell flattening , we bathed Tg ( E1-bhlhe40:GFP ) embryos in nocodazole , a drug that interferes with microtubule polymerization , or its vehicle ( DMSO ) at either 16 or 17 hpf ( Figure 5A and D ) and then analysed them at 18 . 5 or 19 . 5 hpf , respectively . The eye of DMSO-treated embryos developed normally ( Figure 5B and E ) , whereas in the presence of nocodazole RPE cells retained a columnar-like morphology with a stronger phenotype in embryos exposed to the drug at an earlier stage ( Figure 5C ) . Nocodazole treatment did not prevent the activation of the GFP reporter expression ( Figure 5C ) or the acquisition/distribution of expected specification ( otx1 and mitf ) and apico-basal polarity ( zo-1 and laminin ) markers ( Figure 5—figure supplement 2 ) . Notably , although the NR layer appeared to bend inward , the RPE layer remained unfolded ( Figure 5F ) and outer layer cells accumulated at the hinge , suggesting a defect in rim involution . This defect may be due to the alteration of microtubule polymerization in rim cells . Alternatively , the lack of RPE stretching may prevent the translocation of rim cells to the NR layer . The whole embryo treatments described above did not allow us to determine the differential requirement of microtubule dynamics in the RPE and the adjacent NR layer . However , we were unable to uncouple the effect of microtubule alterations in the two OV layers with localized drug interference . We thus resorted to use stathmin 1 ( STMN1 ) , a key regulator of microtubule depolymerization ( Belmont and Mitchison , 1996 ) . We generated a bidirectional UAS construct ( UAS:STMN1 ) driving the simultaneous production of GFP and STMN1 under the same regulatory sequences ( Paquet et al . , 2009; Distel et al . , 2010 ) , which we injected in Tg ( rx3:GAL4 ) embryos . We reasoned that , although rx3 drives transgene expression in both NR and RPE progenitors , the random and sparse expression that occurs in F0 would be sufficient to separate the effect in the two tissues . RPE cells expressing STMN1 – and notably also those nearby – retained a cuboidal-like shape with an abnormally increased apico-basal axis as compared to GFP-positive cells in control UAS:GFP-injected embryos . Even though cells in the inner OV layer appeared to still undergo basal constriction ( Figure 5G–I ) , the OV as a whole underwent poor invagination ( Figure 5J ) . All in all , the data derived from the manipulation of the actomyosin and microtubule cytoskeleton suggest that the RPE actively participates in OV folding by undergoing a tissue autonomous stretching driven by cell cytoskeletal rearrangements . Our finding that the zebrafish RPE largely grows through autonomous cell flattening agrees with the observation that zebrafish RPE cells barely proliferate during OV folding ( Cechmanek and McFarlane , 2017 ) . Furthermore , pharmacological treatment of embryos to block cell division during OV folding has little or no consequences on RPE expansion ( Cechmanek and McFarlane , 2017 ) . These observations however differ from reports in mouse embryos , in which RPE proliferation seems a requirement for OV folding ( Carpenter et al . , 2015 ) and suggest the existence of species-specific modes of early RPE growth . We hypothesized that these modes may be related to the speed of embryonic development with final consequences on the epithelial characteristic of the RPE . To test this possibility , we compared proliferation rate and apico-basal length of the zebrafish RPE ( Figure 6 ) with those of the medaka , chick , mouse , and human embryos at equivalent OV/OC stages ( Figure 7 ) . In these species , Otx2 and N-cadherin immunostaining was used to identify the RPE domain ( Martinez-Morales et al . , 2001; Bovolenta et al . , 1997 ) and the cell shape ( Figure 7—figure supplement 1 ) , respectively . 5-Bromo-2′-deoxyuridine ( BrdU ) incorporation in Tg ( E1-bhlhe40:GFP ) embryos from early OV ( 17 hpf ) to late OC stages ( 48 hpf ) showed a marked reduction of cell proliferation in the OV outer layer ( Figure 6A–B ) , very much in line with the report that only 2% of the outer layer cells undergo mitosis during this period ( Cechmanek and McFarlane , 2017 ) . At the earlier stage ( 17 hpf ) , BrdU-positive cells were scattered across the RPE with no easily identifiable geometry and accounted for 49% of the total RPE cells . This fraction dropped to about 20% at 19 hpf , when cells are flat , and then to 12% at 48 hpf ( Figure 6B ) when the epithelium is maturing . Statistical analysis showed significant differences between 17 and 20 hpf ( Mann-Whitney U test: z = −2 . 619 , p < 0 . 01 , mean rank for 17 hpf is 8 and for 19 hpf is 3 ) and a clear correlation between proliferation rate and developmental stage ( Kruskal-Wallis test: χ2 ( df = 7 , n = 40 ) = 32 , 023; p < 0 . 001 ) . During this period , the apico-basal axis of individual RPE cells flattened reaching a length of 3 μm at 22–23 hpf . Thus , acquisition of RPE identity , cell shape changes , and OV folding are associated with a progressive reduction of cell proliferation in the OV outer layer . OC morphogenesis in the teleost medaka fish occurs with a choreography comparable to that of the zebrafish ( Heermann et al . , 2015; Kwan et al . , 2012 ) but the medaka fish RPE does not adopt an extreme squamous morphology ( Figure 7A ) . Notably , medaka fish develop slower than zebrafish embryos , so that , from first appearance , their OVs take about 8 hr more to reach a fully developed OC ( 26 vs . 18 hr ) ( Furutani-Seiki and Wittbrodt , 2004 ) , a time compatible with an additional round of cell division . Consistent with this idea , BrdU incorporation in st18 to 22–23 medaka embryos showed that about 70% of the cells in the OV outer layer were actively cycling and this proportion dropped to about 48% at OC stage ( Figure 7A and E ) with a slightly less evident decrease of the average apico-basal axis ( st18: 21 . 3 µm vs . st23 3 . 5 µm; Figure 7A ) as compared to the changes observed in zebrafish . An equivalent analysis in chick and mouse embryos showed similar results . In these species OV conversion into an OC takes about 27 and 48 hr , respectively . During this period a similar and almost constant proportion of RPE progenitors incorporated BrdU ( Figure 7B , C and E ) , including when cells acquired the expression of the RPE differentiation marker Otx2 ( Figure 7—figure supplement 1 ) . Furthermore , RPE cells only roughly halved their apico-basal axis ( chick: HH12: 30 . 1 µm vs . HH18: 15 . 8 µm; mouse: E9 . 5: 23 . 7 µm vs . E11 . 5 13 µm Figure 7B and C ) , suggesting that in slower developing species , proliferation but not stretching accounts for RPE surface increase . To corroborate this idea , we next analysed human embryos . The human eye primordium is first visible at about 4–5 weeks of gestation corresponding to Carnegie stage ( CS ) 13 ( O’Rahilly , 1983 ) . A fully formed OC is reached only roughly 10 days after , at CS16 ( O’Rahilly , 1983 ) . Immunostaining of paraffin sections from CS13 to CS16 embryos with antibodies against Ki67 , a marker of the active phases of the cell cycle , demonstrated that the large majority of prospective RPE cells undergo a marked proliferation during the transition from OV to OC ( Figure 7D ) . Owing to the difficulties in obtaining early human embryonic samples , the percentage of proliferating cells could only be estimated , showing that in the OTX2-positive domain ( Figure 4 , Figure 7—figure supplement 1B ) , Ki67-positive RPE cells represented about 85–75% of the total between CS13 and CS16 . During this period , the prospective RPE layer always appeared as a rather thick pseudostratified epithelium with an organization resembling that of the NR composed of densely packed and elongated neuroepithelial cells ( Figure 7D; Figure 7—figure supplement 1 ) . During the formation of the OC , the RPE neuroepithelium only slightly flattened ( apico-basal thickness: CS13: 45 µm vs . CS16: 33 . 6 µm ) , far from reaching the cuboidal appearance seen at postnatal ages ( Figure 7—figure supplement 1 ) . Collectively , these data indicate that , in the absence of sufficient time for cell proliferation , flattening is an efficient solution adopted by zebrafish RPE cells to enlarge the whole tissue to the extent needed for OV folding . In other vertebrates , in which slower development allows for more rounds of cell division , the RPE grows in a conventional proliferation-based mode that correlates with a less evident flattening of RPE cells ( Figure 7F ) .
The cup shape of the vertebrate eye is thought to optimize vision ( Goldsmith , 1990 ) . This shape is acquired very early in development as the result of specification and morphogenetic events , during which the NR and the RPE arise . Studies in teleosts ( zebrafish and medaka ) together with mammalian organoid cultures have recently demonstrated a fundamental contribution of NR progenitors in driving the acquisition of this cup shape ( Moreno-Marmol et al . , 2018; Martinez-Morales et al . , 2017 ) . The role of the RPE progenitors in this process has instead not been properly clarified . In this study , we have filled this gap and analysed the folding of the zebrafish OV from the RPE perspective . This analysis has been possible thanks to the generation of a new RPE reporter line Tg ( E1-bhlhe40:GFP ) , in which GFP expression appears in the domain fated to originate the RPE . Following the cells arising from this domain , we show that RPE surface expansion is an active and tissue autonomous process required for OV folding . This expansion largely occurs by extreme cell flattening with little contribution of cell proliferation , a mechanism that sets zebrafish RPE morphogenesis apart from that of other analysed vertebrate species , in which proliferation accounts for RPE growth . Our analysis together with a previous report ( Cechmanek and McFarlane , 2017 ) shows that the onset bhlhe40 expression coincides spatially and temporally with that of zebrafish RPE specification . Thus , the Tg ( E1-bhlhe40:GFP ) line serves as an early tissue-specific marker that even precedes the appearance of previously accepted Otx or Mitf tissue specifiers , as confirmed in a parallel transcriptomic analysis ( Buono et al . , 2021 ) . Bhlhe40 expression in the RPE is conserved at least in mouse and humans ( Buono et al . , 2021; Cohen-Tayar et al . , 2018; Hu et al . , 2019 ) , suggesting a possible relevant function in this tissue . However , its CRISP/Cas9 inactivation , alone or in conjunction with that of the related bhlhe41 , mitfa , and mitfb , had no evident consequences on zebrafish RPE development , at least in our hands ( data not shown ) . One possible reason for the absence of an evident RPE phenotype is functional redundancy with other untested members of the large family of the BHLH transcription factors or that the gene has only later functions as reported ( Abe et al . , 2006 ) . However , we favour the alternative possibility that zebrafish RPE specification does not occur stepwise as in other species ( Martinez-Morales et al . , 2004; Fuhrmann et al . , 2014 ) but ‘en bloc’ with an almost simultaneous activation of all differentiation genes . This would make the inactivation of one or two genes insufficient to perturb fate acquisition . Such a mechanism is expected to provide robustness to a process that takes place in just few hours and finds support in present and past findings ( Buono et al . , 2021; Cechmanek and McFarlane , 2017 ) . Indeed , we and others Cechmanek and McFarlane , 2017 have shown that , by the time the OV starts to bend , the large majority of RPE cells have already left the cell cycle and have acquired a differentiated squamous morphology by undergoing a marked surface enlargement in the medio-lateral direction and a reduction of the apico-basal axis . The net result is an overall modest volume increase . Furthermore , blocking cell division as the OC forms does not interfere with RPE expansion ( Cechmanek and McFarlane , 2017 ) , strongly supporting a primary role of cell stretching in RPE expansion . Consistently , transcriptomic analysis shows that during this same lag of time , RPE cells repress genes characteristic of 16 hpf OV progenitors , such as vsx1 , and acquire the expression of RPE-specific genes . These include blocks of transcription factors , such as known RPE specifiers ( i . e . otx , mitf ) and regulators of epidermal specification ( i . e . tfap family members , known regulator of keratin gene expression; Leask et al . , 1991 ) as well as several cytoskeletal components , most prominently a large number of keratins and other desmosomal components found in squamous epithelia ( Buono et al . , 2021 ) . Thus , in just few hours ( from 16 to 18 hpf ) RPE cells acquire the molecular machinery required for their conversion from a neuroepithelial to a squamous and likely highly coupled epithelium . Our study shows that this conversion relays on a tissue autonomous cytoskeletal reorganization without the influence of the morphogenetic events occurring in the nearby NR . Indeed , local interference with actomyosin or microtubule dynamics is sufficient to retain RPE cells into a cuboidal or neuroepithelial configuration , respectively , without affecting their specification . In contrast , localized interference with NR bending has no effect on RPE flattening . Notably , our studies also suggest that the RPE acts in a ‘syncytial-like’ manner , as mosaic interference with microtubule polymerization seems to impact in the shape of the adjacent cells if not on the entire tissue . This is perhaps not surprising given that mature RPE cells have been reported to be chemically coupled ( Pearson et al . , 2004; Bao et al . , 2019 ) . Furthermore , the presumptive RPE of the chick ( unpublished observations ) and zebrafish ( Buono et al . , 2021 ) expresses high levels of connexion proteins ( i . e . Gap-43 ) , which are responsible for the ‘syncytial-like’ behaviour observed in brain astrocytes ( Buskila et al . , 2019 ) . This together with the additional observation that st18 RPE cells express many desmosomal proteins ( Buono et al . , 2021 ) indicate that the tissue becomes tightly connected very soon , perhaps behaving as a community ( Gurdon , 1988 ) . The extreme flattening of the zebrafish RPE cells makes the resolution of their cytoskeletal components difficult with in vivo confocal microscopy , hampering the complete understanding of how the actomyosin cytoskeleton promotes the acquisition of a squamous configuration . In other contexts , a flat morphology is associated with the presence of acto-myosin stress fibres that compress the nucleus ( Tee et al . , 2011; Vishavkarma et al . , 2014 ) . Myosin II is essential for this compressive role and its inhibition with blebbistatin causes the loss of the flat morphology ( Tee et al . , 2011; Vishavkarma et al . , 2014 ) , as we have observed in blebbistatin- and Ableb-treated embryos . It is thus possible that a similar nuclear compression may occur in the RPE cells as they flatten , although we were unable to detect stress fibres around the nucleus , likely due to plasma membrane proximity . Remodelling of the microtubular cytoskeleton seems to aid further RPE cell flattening . Microtubules change their orientation during RPE morphogenesis , from being aligned along the apico-basal axis of the cells at the onset of RPE morphogenesis , to becoming aligned with the planar axis in squamous RPE cells . A similar process has been described during the morphogenesis of the Drosophila amnioserosa ( Pope and Harris , 2008 ) , in which cells also change from a columnar to a squamous morphology . In these cells , actin accumulation at the apical edge seems to provide resistance to the elongation of microtubules , which thus bend , leading to a 90° rotation of all subcellular components . This rotation is accompanied by a myosin-dependent remodelling of the adherens junctions ( Pope and Harris , 2008 ) , a process that may also take place during RPE flattening . Although additional studies are needed to clarify the precise dynamics of the cytoskeletal reorganization underlying RPE differentiation , our study demonstrates that cytoskeletal dynamics occurs in a tissue autonomous manner . In contrast to other studies ( Nicolas-Perez et al . , 2016; Sidhaye and Norden , 2017 ) , we have used a photoactivable version of blebbistatin that has allowed us to determine the individual contribution of the NR and RPE to OV folding . As a drawback , this approach allows to activate the drug only in relatively small patches of tissue . It was thus rather remarkable to observe that failure of RPE flattening in small regions was sufficient to decrease OV folding . This suggests that RPE stretching represents an additional and relevant mechanical force that , together with retinal basal constriction and rim involution , contributes to zebrafish eye morphogenesis ( Figure 8A ) . This flattening and stretching together with a substantial expression of keratins ( Buono et al . , 2021 ) may confer a particular mechanical strength to the zebrafish RPE , which , in turn , may constrain the NR at the same time favouring rim involution ( Heermann et al . , 2015 ) . The latter possibility is supported by the observation that inner layer cells seem to accumulate at the hinge in the absence of RPE flattening . Alternatively , this accumulation may simply reflect that rim cell involution depends on intrinsic microtubule polymerization , although previous studies have discarded this possibility ( Sidhaye and Norden , 2017 ) . These marked morphogenetic rearrangements can thus be seen as an efficient solution adopted in fast-developing species to make eye morphogenesis feasible in a period that does not allow for proliferation-based tissue growth . The perhaps obvious question is whether similar morphogenetic rearrangements are needed in other vertebrates to form the remarkably conserved cup shape of the eye . So far , rim involution has been reported only in teleost species where it may represent a fast mode of increasing the surface of the inner layer of the OV , thus favouring its bending ( Sidhaye and Norden , 2017; Heermann et al . , 2015; Kwan et al . , 2012 ) . This idea is well in agreement with previous data showing that between 16 and 27 hpf the number of cells in the outer layer of the OV decreases from about 587 to 432 , whereas that of the inner layer increases in a way that cannot be explained solely by proliferation ( Zheng et al . , 2000 ) . In other species , this cell displacement may not be needed as the layer can grow by cell division . In a similar way , we have shown here that in slower developing species , RPE cells maintain a higher proliferation rate that contributes substantially to the increase of RPE surface while undergoing less marked changes in cell shape ( Figure 8B ) . This correlation is visible in medaka , despite its relative evolutionary proximity to zebrafish ( Furutani-Seiki and Wittbrodt , 2004 ) , and is maximal in human embryos . Indeed , in humans , the RPE layer is composed of cells with a neuroepithelial appearance and a high proliferation rate , despite the expression of OTX2 , considered a tissue specifier . Thus , in mammals , full commitment of the OV outer layer to an RPE identity may occur over a prolonged period of time and not ‘en bloc’ as in zebrafish , as suggested by comparing RNA-seq data of RPE cells from human CS13–16 embryos ( Hu et al . , 2019 ) with those from equivalent stages in zebrafish ( Buono et al . , 2021 ) . Human RPE cells from CS13 to CS16 embryos are still enriched in the expression of proliferation associated genes ( Hu et al . , 2019 ) but not of those typical of squamous epithelia as in zebrafish ( Buono et al . , 2021 ) . A slow acquisition of RPE identity may also explain why , in mice , inactivation of genes such as Otx2 , Mitf , or Yap causes the RPE layer to adopt NR characteristic ( Martinez-Morales et al . , 2001; Kim et al . , 2016; Nguyen and Arnheiter , 2000 ) , whereas this feature that has never been reported after equivalent manipulations in zebrafish ( Lane and Lister , 2012; Miesfeld et al . , 2015 ) , or why FGF8 can push the amniote but not the zebrafish RPE layer to acquire an NR identity ( Martinez-Morales et al . , 2005 ) . As a reflection of this slower differentiation in amniotes , RPE cells can largely retain their neuroepithelial morphology and adopt a final cuboidal – but not squamous – appearance at a slower and species-specific pace . We thus propose that RPE cell stretching vs . cell addition are different solutions adopted by species with different rates of development to reach a common goal: an appropriate equilibrium between the surface of the RPE and that of the NR . Indeed , the present study together with previous observations ( Carpenter et al . , 2015 ) and in silico models ( Okuda et al . , 2018; Eiraku et al . , 2012 ) support that this equilibrium is a prerequisite for proper OV folding .
Adult zebrafish ( Danio rerio ) were maintained under standard conditions at 28°C on 14/10 hr light/dark cycles . AB/Tübingen strain was used to generate the transgenic lines and as control wild type . Embryos and larvae were kept in E3 medium ( 5 mM NaCl , 0 . 17 mM KCl , 0 . 33 mM CaCl2 , 0 . 33 mM MgSO4 ) supplemented with Methylene Blue ( Sigma ) at 28°C and staged according to somite number and morphology ( Kimmel et al . , 1995 ) . The Tg ( E1-bhlhe40:GFP ) and Tg ( rx3:Gal4;UAS:RFP ) ( Weiss et al . , 2012 ) lines were maintained in the same conditions and crossed to generate the Tg ( E1-bhlhe40:GFP;rx3:GAL4;UAS;RFP ) line . Wild-type medaka fish ( Oryzias latipes ) of the cab strain were maintained at 28°C on a 14/10 hr light/dark cycle . Embryos were staged as described ( Iwamatsu , 2004 ) . Fertilized chick embryos ( Santa Isabel Farm , Cordoba , Spain ) were incubated at 38°C in a humidified rotating incubator until the desired stage . Embryos were inspected for normal development and staged according to Hamburger and Hamilton , 1992 . Wild-type BALB/c mice were in pathogen-free conditions at the CBMSO animal facilities , following current national and European guidelines ( Directive 2010/63/EU ) . The day of the appearance of the vaginal plug was considered as embryonic day ( E ) 0 . 5 . All experimental procedures were approved by the CBMSO and Comunidad Autónoma de Madrid ethical committees . Paraffin sections of human embryonic eye primordia were provided by the Joint MRC/Wellcome Trust ( grant# MR/R006237/1 ) Human Developmental Biology Resource ( http://hdbr . org ) . Sections corresponded to samples CS13 , -14 , -15 , and -16 . CS staging allowed to determine the age of embryo as days post ovulation based on morphological landmarks ( O’Rahilly and Müller , 2010 ) . Predictive enhancer and promoter epigenetic marks ( Bogdanovic et al . , 2012 ) were used to identify different potential regulatory elements of the bhlhe40 gene ( Figure 1B ) . Each region was amplified by PCR with specific primers ( Supplementary file 1 ) and cloned using the pCR8/GW/TOPO TA Cloning Kit ( Invitrogen ) . Plasmids were checked for enhancer insertion and the Gateway LR Clonase Enzyme Mix ( Invitrogen ) was used for recombination with the ZED vector ( Bessa et al . , 2009 ) . The resulting constructs were injected together with Tol2 mRNA to generate the corresponding transgenic embryos , which were screened using a transgenesis efficiency marker present in the ZED vector ( cardiac actin promoter:RFP ) . Positive larvae were grown to adulthood ( F0 ) and then individually outcrossed with wild-type partners to identify founders . Founders were analysed using confocal microscopy . One of the lines corresponding to the enhancer E1 was finally selected and used for subsequent studies . The UAS:STMN1 construct was generated from the bidirectional UAS:GFP vector , which allows simultaneous and comparable production of GFP and the gene product of interest under the same regulatory sequences ( Paquet et al . , 2009; Distel et al . , 2010 ) . The gene was amplified by PCR using specific primers ( Supplementary file 1 ) flanked by StuI restriction sites and the Expand High Fidelity PCR System , using the pQTEV-STMN1 ( Addgene# 31326 ) construct as a mould . The PCR product was digested with StuI ( Takara ) and cloned into the pCS2 vector and thereafter isolated together with the polyA sequence of the vector by digestion with HindIII and SacII ( Takara ) and sub-cloned into the UAS:GFP plasmid . The generated plasmid ( 30 pg ) was injected into the Tg ( rx3:Gal4;UAS:RFP ) ( Weiss et al . , 2012 ) line , together with Tol2 mRNA ( 50 pg ) to increase efficiency . Embryos at one cell stage were injected using a Narishige micro-injector and glass needles prepared by horizontally pulling standard capillaries ( filament , 1 . 0 mm , World Precision Instruments ) with aP-97 Flaming/Brown Micropipette Puller ( Sutter Instrument Company ) . A total of 30 pg for DNA and between 50 and 100 pg for mRNA in 1 nl volume were injected in the embryos in the cell or the yolk , respectively . Drug treatments were performed on manually dechorionated embryos at the desired developmental stage in E3 medium . The following compounds were used: blebbistatin ( 100 μM for 2 . 5 hr; Calbiochem ) ; paranitroblebbistatin ( 20 μM; Optopharma ) , Ableb ( 5 μM for 15 min before photoactivation; Optopharma ) , and nocodazole ( 10 ng/μl for 2 . 5 hr; Sigma ) . The pCS2:Kaede , pCS2:EB3-GFP , and pCS2:H2B-RFP constructs were linearized and transcribed using the mMessage mMachine SP6 transcription kit ( Invitrogen ) , following manufacturer’s instructions . After transcription mRNAs were purified using the NucleoSpin RNA Clean-up kit ( Machery Nagel ) . In situ hybridization ( ISH ) otx1 ( previously known as otx1b ) and mitfa probes were gifts from Prof . Steve Wilson ( UCL , London , UK ) . The bhlhe40 probe was generated by PCR from 24 hpf cDNA with specific primers Supplementary file 1 using the Expand High Fidelity PCR System . Reverse primers included the T3 promoter sequence to in vitro transcribe the PCR product . In vitro transcription was performed using T3 RNA polymerase and DIG RNA labelling Mix ( Roche ) following manufacturer’s instructions . Transcription products were precipitated with LiCl 0 . 4 M and 3 volumes of ethanol 100% overnight at –20°C . Samples were centrifuged at 4°C and 12 , 000 g for 30 min , washed with ethanol 70% , and re-suspended in 15 µl of RNAse-free water and 15 µl Ultra-Pure Formamide ( Panreac ) . ISH were performed as described ( Cardozo et al . , 2014 ) . BrdU ( Roche ) was re-suspended in DMSO ( Sigma ) to generate stocks of 50 mg/ml that were kept at –20°C . For Tg ( E1-bhlhe40:GFP ) zebra- and wild-type medaka fish groups of 15 embryos of stages comprised between 16 ss and 48 hpf were dechorionated and placed in BrdU solution ( 5 mg/ml in E3 medium ) for 30 min on ice and then washed with fresh E3 medium . Embryos were let recover at 28°C for 10 min before fixation in paraformaldehyde ( PFA ) 4% overnight at 4°C . For analysis in chick , BrdU ( 50 mg/egg ) was added to each embryo 30 min before fixation . For analysis in mouse , pregnant dams were injected intraperitoneally with BrdU ( 50 μg/g ) , sacrificed 1 hr later and fixed . Chick and mouse embryos were immersion fixed in 4% PFA in 0 . 1 M phosphate buffer , pH 7 at 4°C for 4 hr and then washed in PBS and cryoprotected in 15% and 30% saccharose in 0 . 1 M phosphate buffer . All embryos were cryo-sectioned and the sections hydrated with PBS 1X during 5 min and incubated in HCl during 40 min at 37°C . After HCl treatment , sections were rinsed with PBS 1X 10 times , and then processed for immunofluorescence as described below . The percentage of RPE proliferating progenitors was determined as the proportion of BrdU-positive cells over the total number of GFP ( for E1-bhlhe40:GFP ) or Otx2/Hoechst ( medaka fish , chick , mouse embryos ) positive cells in the RPE layer in each section . A minimum of three embryos and sections per embryo were counted ( both eyes ) . Zebrafish embryos at the corresponding stage for each experiment were fixed with 4% ( wt/vol ) PFA ( Merck ) in 0 . 1 M phosphate buffer overnight at 4°C . Whole-mount immunofluorescence was performed as described ( Cardozo et al . , 2014 ) . Alternatively , embryos were incubated in 15% sucrose – PBS overnight at 4°C , embedded in 7 . 5% gelatine ( Sigma ) 15% sucrose ( Merck ) , frozen in isopentane ( PanReac ) between –30°C and –40°C and kept at –80°C . Cryo-sectioning was performed with a cryostat ( Leica CM 1950 ) at 20 µm thickness and dried overnight at room temperature . Chick and mouse embryos were collected , fixed 4% PFA , equilibrated in sucrose , and cryo-sectioned as above . Paraffin sections of human embryonic tissue were de-paraffinized , washed in PBS , processed for antigen retrieval ( 10 mM citrate buffer , pH6 , for 5 min at 110°C in a boiling chamber , Biocaremedical ) , and subsequently processed together with all other samples for immunofluorescence . Immunostaining was performed as described ( Cardozo et al . , 2014 ) using the following primary antibodies: mouse anti-BrdU ( 1:200; Becton-Dickinson ) ; chick anti-GFP ( 1:2000; Abcam ) ; mouse anti-βcatenin ( 1:400 , BD Transduction Laboratories ) ; mouse anti-ZO-1 ( 1:400 , Invitrogen ) ; rabbit anti-laminin ( 1:200 , Sigma ) ; rabbit anti-Otx2 antibodies ( 1:1000; Abcam ) ; rabbit anti-Ki67 ( 1:500 , Abcam ) . The used secondary antibodies were conjugated with Alexa-488 , Alexa-594 , or Alexa-647 ( 1:500; Thermo Fisher ) . Sections were counterstained with Hoechst ( Invitrogen ) , mounted in Mowiol , and analysed by conventional and confocal microscopy . Wild-type embryos were injected with Kaede mRNA . Embryos at 15 hpf with homogeneous green fluorescence were selected , mounted , and visualized under the Nikon AR1+ Confocal Microscope using a 20×/0 . 75 Plan-Apochromat objective . A region of interest ( ROI ) was drawn in the outer layer , corresponding to the putative position of the RPE progenitors , at a specific z-position and irradiated with the 405 nm laser at 21% of power for 10 loops to switch Kaede emission from green to red fluorescence . Due to confocality , photoconversion occasionally extended further than the selected plane , so that the tissues present above or below ( i . e . ectoderm ) also underwent photoconversion . After photoconversion embryos were let develop up to approximately 30 hpf stage , fixed and analysed by confocal microscopy for red fluorescence distribution . Ableb ( Kepiro et al . , 2012 ) was photoactivated with a Zeiss LSM 780 Upright multiphoton FLIM system with a W Plan-Apochromat 20×/1 . 0 DIC M27 75 mm WD 1 . 8 mm dipping objective . For each eye a specific ROI was drawn including RPE cells identified by GFP fluorescence . Ableb was activated in the ROIs using 860 nm wavelength and 20 mW laser power ( this corresponds to 9–14 µW/µm2 inside the ROI ) . Embryos were mounted with the appropriate orientation in 1 . 5% low melting point agarose ( Conda ) diluted in E3 medium ( for in vivo recording ) or PBS ( for fixed samples ) . Images were acquired either with a Nikon A1R + High Definition Resonant Scanning Confocal Microscope connected to an Inverted Eclipse Ti-E Microscope ( 20×/0 . 75 Plan-Apochromat , 40×/1 . 3 oil Plan-Fluor and 60×/1 . 4 oil Plan-Apocromat objectives ) or with a Zeiss LSM710 Confocal Laser Scanning Microscope connected to a Vertical AxioImager M2 Microscope ( 40×/1 . 3 oil Plan-Apochromat , W N-Achroplan 20×/0 . 5 , W Plan-Apochromat 40×/1 . 0 DIC VIS-IR ) . 3D videos ( i . e . Figure 1—videos 1–3 ) were generated from full stacks using the 3D project option in Fiji ( Schindelin et al . , 2012 ) . RPE surface renderings were generated using Imaris ( Bitplane ) , with a value of 6 in Surface Area Detail and 7 in Background Subtraction . Unless otherwise specified , morphometric analysis of cells and tissues was performed using Matlab ( The Mathworks , Natick , MA ) using the XYZ coordinates of the processed images or Fiji ( Schindelin et al . , 2012 ) . This analysis was performed using previously processed fluorescent images from videos of Tg ( E1-bhlhe40:GFP; rx3:GAL4;UAS:RFP ) or Tg ( E1-bhlhe40:GFP ) and H2B-RFP-injected embryos ( Figure 1—video 2 and Figure 3—video 1 ) , from which the signal corresponding to the RPE or the whole OV/OC were isolated semi-manually with the help of Fiji macros and tools designed to select 3D structures . The RPE-specific GFP signal was processed with a median filter . In the case of Figure 3—video 1 , the background ramp for the GFP signal was neutralized in each frame via subtraction of a copy of itself after a grey-scale morphological operation ( Hassanpour et al . , 2015; Arce , 2005 ) . For all videos , the median intensity was thereafter established as the cutoff value for differentiating background and signal ( i . e . pixel with an intensity lower than the cutoff were set to zero ) for all images that were in both videos . The signal derived from H2B was localized in cell nuclei , and therefore it was post-processed with a grey-scale closing operation to fill empty spaces between nuclei . Morphometric analysis was performed in the resulting processed images . All values were calculated in microns by scaling the x , y , z coordinates according to the following: ( 0 . 62 μm × 0 . 62 μm × 1 . 37 μm ) for Figure 1—video 2 and ( 0 . 62 μm × 0 . 62 μm × 1 . 07 μm ) for Figure 3—video 1 . Volumes ( μm3 ) were calculated as the number of voxels with a value higher than 0 . RPE surface ( μm2 ) was calculated applying a second-order linear adjustment on the plane YZ corresponding to the plane of the OV/OC hinges with the fit function available in Matlab ( The Mathworks , Natick , MA ) . RPE thickness ( μm ) was determined as the result of volume ( μm3 ) /surface ( μm2 ) . Unfortunately , semi-manual RPE image extraction was not perfect , when GFP signal associated to CMZ development arises . To account for this problem , the GFP signal for each frame was divided into seven equivalent blocks using the x , y coordinates from the z-projection of each frame . In this case , RPE volume and surface were calculated independently in each one of the regions up to 20 hpf , when the most anterior block ( now corresponding to the arising CMZ ) was discarded from the analysis . For the subsequent frames the two anterior most blocks were discarded ( Figure 3—figure supplement 1 ) . The total OV/OC volume ( μm3 ) was determined using the red fluorescence from the Tg ( rx3:GAL4;UAS:RFP ) embryos at 17–22hpf . H2B expression was used to determine the volume of the OC ( H2B volume in Figure 3 ) as follows for each frame of Figure 3—video 1: the maximum , Gaussian blur and minimum filters were applied to the image; subsequently , the convex hull ( Hamburger and Hamilton , 1992 ) was calculated for the image to obtain the geometrical shape that covers all pixels with an intensity higher than 0 , including the lens; finally , only the regions present in the image and the convex hull are used to define the H2B volume . Individual cell area was determined in cells located at a medial position of the OV for each cell type ( progenitor , RPE , and NR ) ; cell contour was drawn using the segmented line tool in Fiji ( Schindelin et al . , 2012 ) . Apico-basal ( A-B ) length ( µm ) of individual cells was estimated by manually tracing a line from the basal to the apical membrane in the z-position in which the nucleus had its larger surface using the straight-line tool in Fiji ( Schindelin et al . , 2012 ) . To account for possible developmental asynchrony when eyes from the same embryo were differentially treated ( irradiated vs . non-irradiated ) , the A-B length of the irradiated eye was normalized with that of the non-irradiated eye . Values above 1 indicated less RPE cell flattening in experimental eyes . The invagination angle was determined as previously described ( Sidhaye and Norden , 2017 ) using manual drawing with the Fiji angle tool ( Schindelin et al . , 2012 ) . The vertex of the angle was placed approximately in the centre of the basal surface of the NR and the vectors were drawn up to the edges of the CMZ . Angles were measured in the z-positions in which the irradiated RPE was maximally affected and compared to equivalent positions of control non-irradiated eyes . Values were normalized with those of the contralateral non-treated eye , to account for possible asynchronies . All statistical analysis was performed with IBM SPSS Statistics version 20 . 0 . The method used is indicated in each case together with the sample size . | Rounded eyeballs help to optimize vision – but how do they acquire their distinctive shape ? In animals with backbones , including humans , the eye begins to form early in development . A single layer of embryonic tissue called the optic vesicle reorganizes itself into a two-layered structure: a thin outer layer of cells , known as the retinal pigmented epithelium ( RPE for short ) , and a thicker inner layer called the neural retina . If this process fails , the animal may be born blind or visually impaired . How this flat two-layered structure becomes round is still being investigated . In fish , studies have shown that the inner cell layer – the neural retina – generates mechanical forces that cause the developing tissue to curve inwards to form a cup-like shape . But it was unclear whether the outer layer of cells ( the RPE ) also contributed to this process . Moreno-Marmol et al . were able to investigate this question by genetically modifying zebrafish to make all new RPE cells fluoresce . Following the early development of the zebrafish eye under a microscope revealed that RPE cells flattened themselves into long thin structures that stretched to cover the entire neural retina . This change was made possible by the cell’s internal skeleton reorganizing . In fact , preventing this reorganization stopped the RPE cells from flattening , and precluded the optic cup from acquiring its curved shape . The results thus confirmed a direct role for the RPE in generating curvature . The entire process did not require the RPE to produce new cells , allowing the curved shape to emerge in just a few hours . This is a major advantage for fast-developing species such as zebrafish . In species whose embryos develop more slowly , such as mice and humans , the RPE instead grows by producing additional cells – a process that takes many days . The development of the eye thus shows how various species use different evolutionary approaches to achieve a common goal . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2021 | Stretching of the retinal pigment epithelium contributes to zebrafish optic cup morphogenesis |
Within a single generation time a growing yeast cell imports ∼14 million ribosomal proteins ( r-proteins ) into the nucleus for ribosome production . After import , it is unclear how these intrinsically unstable and aggregation-prone proteins are targeted to the ribosome assembly site in the nucleolus . Here , we report the discovery of a conserved nuclear carrier Tsr2 that coordinates transfer of the r-protein eS26 to the earliest assembling pre-ribosome , the 90S . In vitro studies revealed that Tsr2 efficiently dissociates importin:eS26 complexes via an atypical RanGTP-independent mechanism that terminates the import process . Subsequently , Tsr2 binds the released eS26 , shields it from proteolysis , and ensures its safe delivery to the 90S pre-ribosome . We anticipate similar carriers—termed here escortins—to securely connect the nuclear import machinery with pathways that deposit r-proteins onto developing pre-ribosomal particles .
Ribosome assembly is an essential process that is tightly connected to cellular growth and proliferation ( Warner , 1999 ) . In the eukaryotic model organism budding yeast , this universal translating machine is built of two subunits: a large subunit ( 60S ) consisting of three different rRNAs ( 25S , 5 . 8S , 5S ) and 46 ribosomal proteins ( r-proteins ) and a small subunit ( 40S ) that contains a single rRNA ( 18S ) and 33 r-proteins ( Ben-Shem et al . , 2011; Klinge et al . , 2011; Rabl et al . , 2011 ) . Assembly of the eukaryotic ribosome takes place in three distinct cellular territories: the nucleolus , the nucleoplasm and the cytoplasm ( Woolford and Baserga , 2013; Gerhardy et al . , 2014 ) . RNA polymerase I drives production of the 35S pre-rRNA transcript in the nucleolus , which initiates the assembly process . The emerging 35S pre-rRNA transcript undergoes co-transcriptional modification and processing ( Osheim et al . , 2004; Kos and Tollervey , 2010 ) , and associates primarily with 40S subunit r-proteins and ∼50 assembly factors to form the earliest pre-ribosome , the 90S ( Dragon et al . , 2002; Grandi et al . , 2002; Schäfer et al . , 2003 ) . Cleavage of 35S pre-rRNA releases the pre-40S particle , permitting the remaining pre-rRNA to associate with r-proteins of the 60S subunit and ∼200 additional assembly factors to undergo further maturation and pre-rRNA processing ( Fatica et al . , 2002; Grandi et al . , 2002; Nissan et al . , 2002 ) . Nuclear maturation of pre-ribosomal particles also requires the release of assembly factors , a process thought to require >50 energy consuming enzymes ( Strunk and Karbstein , 2009; Kressler et al . , 2010 ) . Export competent pre-ribosomal particles are separately transported through nuclear pore complexes ( NPCs ) into the cytoplasm by multiple export factors . In yeast , export factors include the exportin Xpo1 , which recognizes nuclear export sequences ( NESs ) in a RanGTP-dependent manner , and additional factors ( Tschochner and Hurt , 2003 ) . Export factors bind pre-ribosomal particles and interact simultaneously with FG-repeat nucleoporins lining the NPC channel ( Gadal et al . , 2001; Johnson et al . , 2001; Oeffinger et al . , 2004; Bradatsch et al . , 2007; Yao et al . , 2008 , 2010; Hackmann et al . , 2011; Altvater et al . , 2012; Bassler et al . , 2012; Faza et al . , 2012; Occhipinti et al . , 2013 ) . Following export , pre-ribosomal particles undergo final maturation prior to initiating translation . This involves the release of shuttling assembly factors , transport factors , incorporation of the remaining r-proteins and final pre-rRNA processing ( Panse and Johnson , 2010; Panse , 2011 ) . Within the pre-40S particle , immature 20S pre-rRNA is endonucleolytically cleaved into mature 18S rRNA by the nuclease Nob1 rendering the subunit translation competent ( Fatica et al . , 2004; Lamanna and Karbstein , 2009; Pertschy et al . , 2009 ) . Although , Nob1 is recruited to 40S pre-ribosomes in the nucleus , it is activated in the cytoplasm within an 80S-like pre-ribosomal particle formed upon interaction with a mature 60S subunit ( Lebaron et al . , 2012; Strunk et al . , 2012 ) . Additionally , multiple conserved ATPases Prp43 , Rio2 , Rli1 and Fap7 , the Prp43-activator Pfa1 , the kinase Rio1 , the assembly factor Ltv1 and the r-protein uS11 ( yeast Rps14 ) are implicated in this cleavage step ( Geerlings et al . , 2003; Vanrobays et al . , 2003; Jakovljevic et al . , 2004; Granneman et al . , 2005; Pertschy et al . , 2009; Strunk et al . , 2012; Hellmich et al . , 2013 ) . Despite the identification of a plethora of factors and their general order of action , how nuclear and cytoplasmic assembly steps are coordinated remains largely unknown . In addition to the tremendous energy required to assemble ribosomes , this process also accounts for the major proportion of the nucleocytoplasmic transport in a growing yeast cell ( Rout et al . , 1997; Sydorskyy et al . , 2003 ) . All mRNAs encoding r-proteins must be exported into the cytoplasm , where translation occurs . Nearly all newly synthesized r-proteins are then imported into the nucleus . In yeast , the importin Kap123 has been shown to be an important mediator of r-protein import , but the related importin Pse1 can functionally substitute Kap123 in vivo ( Rout et al . , 1997; Schlenstedt et al . , 1997 ) . Unlike other cargos , r-proteins contain large unstructured regions that form intricate interactions with rRNA within the mature ribosome and are prone to non-specific interactions with nucleic acids , aggregation and proteolytic degradation in their non-assembled state ( Jäkel and Görlich , 1998; Jäkel et al . , 2002; Klinge et al . , 2011; Rabl et al . , 2011 ) . In contrast to typical protein transport events , nuclear import of r-proteins and subsequent transfer to the ribosome production site pose logistical challenges . In addition to their transport role , importins have been implicated to chaperone basic r-proteins during their transport to the nucleus ( Jäkel et al . , 2002 ) . How these intrinsically unstable and aggregation-prone proteins are targeted to assembling pre-ribosomal particles after dissociating from importins remains unclear . Here , we report the discovery of a carrier Tsr2 that coordinates transfer of the eukaryote specific r-protein eS26 ( yeast Rps26; Ban et al . , 2014 ) after nuclear import to the assembling 90S pre-ribosome . Tsr2 extracts eS26 from its importins to terminate its import process . Hereby , we reveal an atypical RanGTP-independent mechanism to dissociate an importin:cargo complex . Tsr2 binds and protects the released eS26 from aggregation and proteolysis thereby ensuring its safe transfer to the 90S pre-ribosome . Our data raise the possibility of a yet unidentified fleet of carriers that securely link the nuclear import machinery with the ribosome assembly pathway .
Previous genome-wide studies revealed a strong accumulation of immature 20S pre-rRNA in a TSR2 ( 20 S rRNA accumulation 2 ) deficient yeast strain ( tsr2Δ ) ( Peng et al . , 2003 ) . Tsr2 is a conserved 23 . 7 kDa protein ( Figure 1—figure supplement 1A ) with no identified structural homologues that could provide clues into its role in 20S pre-rRNA processing . To dissect the function of Tsr2 , we generated a conditional mutant in which the endogenous TSR2 was placed under the control of the GAL1 promoter ( PGAL1-TSR2 ) . On repressive glucose media , Tsr2 protein levels were undetectable and the PGAL1-TSR2 strain was severely impaired in growth compared to a wild-type ( WT ) strain between 20–37°C ( Figure 1A ) . 10 . 7554/eLife . 03473 . 003Figure 1 . Tsr2 is required for cytoplasmic processing of 20S pre-rRNA to mature 18S rRNA , and directly binds eS26 . ( A ) Tsr2-TAP , Tsr2-GFP and Tsr2-3xGFP cells are not impaired in growth . Left panel: indicated strains were spotted on glucose containing media in 10-fold dilutions and grown at indicated temperatures for 3–7 days . Right panel: Tsr2 protein levels in whole cell extracts derived from the indicated strains were determined by Western analyses using α-Tsr2 antibodies . Protein levels of Arc1 served as loading control . ( B ) Tsr2 localizes predominantly to the nucleus . The Tsr2-TAP and the Tsr2-GFP strain and the PGAL1-TSR2 strain containing a centrometric plasmids encoding Tsr2-3xGFP were grown at 30°C to mid-log phase . Localization of Tsr2-TAP was visualized by indirect immunofluorescence microscopy using polyclonal α-TAP antibody ( red ) . Nuclear and mitochondrial DNA was stained with DAPI ( blue ) . Localization of Tsr2-GFP and Tsr2-3xGFP was analyzed by fluorescence microscopy . Scale bar = 5 µm . ( C ) Tsr2-deficient cells accumulate immature 20S pre-rRNA in the cytoplasm . WT and PGAL1-TSR2 cells were grown at 30°C in glucose containing media to mid-log phase . Localization of 20S pre-rRNA was analyzed by FISH using a Cy3-labeled oligonucleotide complementary to the 5′ portion of ITS1 ( red ) . Nuclear and mitochondrial DNA was stained with DAPI ( blue ) . Scale bar = 5 µm . ( D ) Tsr2-depleted cells accumulate 80S-like particles . WT and PGAL1-TSR2 cells were grown at 30°C in glucose containing media to mid-log phase . Cell extracts were prepared after cycloheximde treatment to preserve polysomes and subjected to sedimentation centrifugation on 7–50% sucrose gradients . Polysome profiles at OD254nm were recorded and the peaks for 40S and 60S subunits , 80S ribosomes and polysomes are indicated ( top panels ) . The gradients were fractionated and the RNA was extracted , separated on a 2% Agarose gel , stained with GelRed ( Biotium , middle panels ) and subsequently analyzed by Northern Blotting using probes against indicated rRNAs ( bottom panels ) . Exposure times for phosphoimager screens were 20 min for 25S and 18S rRNA , and 3–4 hr for 20S pre-rRNAs . ( E ) Tsr2 does not co-sediment with 40S subunits . WT cells were grown at 30°C to mid-log phase , extracts were prepared and fractionated as described in ( D ) . The polysome profile at OD254nm is shown in the upper panel . The peaks for 40S and 60S subunits , 80S ribosomes and polysomes are indicated . The gradient was fractionated , TCA precipitated and the protein content was assessed by Western analyses using the indicated antibodies . ( F ) Tsr2-TAP co-enriches the r-protein eS26 . Tsr2-TAP was isolated by tandem affinity purification and the Calmodulin-eluate was separated by 4–12% gradient SDS-PAGE and analyzed by Silver staining . The indicated proteins were identified by mass spectrometry . ( G ) Tsr2 interacts with eS26 in a yeast two-hybrid assay . Plasmids encoding the indicated GAL4 DNA-binding domain ( BD ) and GAL4 activation domain ( AD ) fusion proteins were transformed into the yeast reporter strain NMY32 . Transformants were spotted in 10-fold serial dilutions onto SDC-Leu-Trp ( -Leu-Trp ) or SDC-Ade ( −Ade ) and incubated at 30°C for 4 days . Growth on SDC-Ade indicates a strong two-hybrid interaction . The SV40 Large T antigen served as negative control for these analyses . ( H ) Tsr2 directly binds eS26 in vitro . GST-Tsr2 was immobilized on Glutathione Sepharose before incubation with an E . coli lysate containing recombinant eS26 . After incubation , bound proteins were eluted by SDS sample buffer , separated by SDS-PAGE and visualized by Coomassie Blue staining . L = input . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 00310 . 7554/eLife . 03473 . 004Figure 1—figure supplement 1 . Tsr2 and eS26 depletion does not impair pre-40S nuclear export . ( A ) Sequence alignment of Tsr2 from the indicated organisms done by ClustalO ( Sievers and Higgins , 2014; Sievers et al . , 2011 ) . Conservation at each position is depicted as a gradient from light blue ( 50% identity ) to dark blue ( 100% identity ) . ( B ) Tsr2- and eS26-depletion does not impair pre-40S subunit nuclear export . The indicated strains expressing uS5-GFP were grown in repressive glucose containing liquid media to mid-log phase at 30°C . Localization of uS5-GFP was monitored by fluorescence microscopy . Scale bar = 5 µm . ( C ) Human Tsr2 rescues the slow growth of Tsr2-depleted cells . The PGAL1-TSR2 cells transformed with indicated plasmids were spotted in 10-fold dilutions on selective glucose containing plates and grown at indicated temperatures for 3–7 days . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 004 Next , we localized Tsr2 using an integrated C-terminal -GFP and -TAP tag at the genomic locus . These cell-biological studies revealed that both fusion proteins predominantly localize to the nucleus ( Figure 1B ) . A similar location for the Tsr2-3xGFP fusion protein ( expressed from a CEN plasmid under its natural promoter and terminator regions ) was observed in a Tsr2-depleted strain . The strains expressing the various fusion proteins were not impaired in growth ( Figure 1A ) suggesting that addition of the -GFP , -TAP , and -3xGFP tags did not affect Tsr2 function . We conclude that Tsr2 mainly localizes to the nucleus . The location of Tsr2 led us to test whether the accumulation of 20S pre-rRNA in tsr2Δ cells Peng et al . ( 2003 ) is due to impaired nuclear export of pre-40S subunits . To this end , we monitored localization of 40S subunits in Tsr2-depleted cells using the established reporter uS5-GFP ( yeast Rps2-GFP; Milkereit et al . , 2001 ) . We used the yrb2Δ mutant , which is specifically impaired in pre-40S subunit export , as a control ( Moy and Silver , 2002 ) . As expected , the yrb2Δ mutant showed a nuclear accumulation of uS5-GFP , in contrast to WT , which displayed cytoplasmic localization of this reporter ( Figure 1—figure supplement 1B ) . Surprisingly , PGAL1-TSR2 cells grown on glucose also showed cytoplasmic uS5-GFP localization ( Figure 1—figure supplement 1B ) , indicating no apparent impairment in nuclear export of pre-40S subunits . The data above raised the possibility that cytoplasmic processing of 20S pre-rRNA is impaired in Tsr2-depleted cells . To this end , we monitored the localization of the 5′ portion of the internal transcribed spacer 1 ( ITS1 ) that is present within immature 20S pre-rRNA , but not in mature 18S rRNA , by fluorescence in situ hybridization ( FISH ) . In a WT strain , due to efficient nuclear export of pre-40S subunits , Cy3-ITS1 ( red ) is detectable only in the nucleolus ( Figure 1C ) . After nuclear export , ITS1 is cleaved from 20S pre-rRNA by the endonuclease Nob1 and degraded by the exonuclease Xrn1 ( Stevens et al . , 1991; Moy and Silver , 2002 ) . Tsr2-depleted cells exhibited strong cytoplasmic accumulation of Cy3-ITS1 ( Figure 1C ) , indicating that cytoplasmic processing is impaired in these cells . Two studies proposed that 20S pre-rRNA processing occurs within an 80S-like particle formed via interaction between a mature 60S subunit and a pre-40S subunit in the cytoplasm ( Lebaron et al . , 2012; Strunk et al . , 2012 ) . One possibility is that formation of this particle is impaired in Tsr2-depleted cells , thereby indirectly interfering with 20S pre-rRNA processing . To test this , we performed polysome analyses . Cell extracts from WT and Tsr2-depleted cells prepared under polysome preserving conditions were analyzed by sucrose gradient centrifugation . In agreement with a role in the 40S biogenesis pathway , the polysome profile of Tsr2-depleted cell extracts revealed strongly reduced levels of free 40S subunits and polysomes ( Figure 1D , top panel ) . Northern analyses revealed that mature 25S rRNA and immature 20S pre-rRNA co-peak ( Figure 1D , bottom panel ) , indicating accumulation of 80S-like particles , similar to the one seen upon Fap7-depletion ( Granneman et al . , 2005; Strunk et al . , 2012 ) . Thus , pre-40S subunits that are exported into the cytoplasm in Tsr2-depleted cells interact with mature 60S subunits , but fail to undergo 20S pre-rRNA processing . We conclude that Tsr2 is required for cytoplasmic maturation of pre-40S subunits . Next , we analyzed the sedimentation behavior of Tsr2 on sucrose density gradients . Cell extracts from WT cells were subjected to polysome analyses . The gradient was fractionated and analyzed by Western analyses . Unexpectedly , Tsr2 did not co-sediment with the 40S peak or with heavier fractions , but was found exclusively in lighter fractions at the top of the gradient ( Figure 1E ) . These data indicate that Tsr2 does not stably associate with pre-ribosomal particles in the 40S biogenesis pathway . To identify interaction partners of Tsr2 , we isolated Tsr2-TAP . In agreement with the sedimentation studies above , Tsr2-TAP did not isolate a pre-40S particle . Instead , Tsr2-TAP co-enriched stoichiometric amounts of the eukaryotic specific r-protein eS26 ( Figure 1F; Peng et al . , 2003 ) . Further , yeast two-hybrid analysis revealed a strong interaction between Tsr2 and eS26 , as determined by growth on stringent adenine deficient media ( Figure 1G ) . In vitro binding studies using recombinant proteins showed that eS26 and Tsr2 formed a robust complex ( Figure 1H ) . We conclude that eS26 directly binds Tsr2 . In budding yeast , two non-essential genes , RPS26A and RPS26B , encode the r-protein eS26 . To investigate the phenotypes of RPS26 deficiency , we created a conditional double mutant in which the endogenous promoter of RPS26A in the rps26bΔ strain was replaced with the GAL1 promoter ( PGAL1-RPS26A ) . Consistent with an essential function of eS26 in yeast , the PGAL1-RPS26Arps26bΔ strain did not grow on repressive glucose containing medium ( Figure 2A ) . Using this strain , we investigated whether eS26 is required for nuclear export of pre-40S subunits and/or cytoplasmic 20S pre-rRNA processing by monitoring the localization of uS5-GFP and Cy3-ITS1 . eS26-depletion did not induce nuclear accumulation of uS5-GFP ( Figure 1—figure supplement 1B ) , indicating no apparent impairment in pre-40S subunit nuclear export . However , these cells showed a strong cytoplasmic accumulation of Cy3-ITS1 ( Figure 2B ) , indicating impairment in final 20S pre-rRNA processing . Further , polysome analyses of eS26-depleted cell extracts revealed strongly reduced levels of free 40S subunits ( Figure 2C , top panel ) . Northern analyses revealed that mature 25S rRNA and immature 20S pre-rRNA co-peaked ( Figure 2C , bottom panel ) , indicating an accumulation of 80S-like particles . Thus , as observed in Tsr2-depleted cells , eS26-depleted cells contain pre-40S subunits that fail to process 20S pre-rRNA in the cytoplasm . Based on these data we conclude that eS26 is required for cytoplasmic maturation of pre-40S subunits . 10 . 7554/eLife . 03473 . 005Figure 2 . eS26 is required for cytoplasmic processing of immature 20S pre-rRNA to mature 18S rRNA . ( A ) eS26 is essential for viability in yeast . Left panel: WT , rps26aΔ , rps26bΔ and the conditional mutant PGAL1-RPS26Arps26bΔ were spotted in 10-fold dilutions on galactose and repressive glucose containing media and grown at 30°C for 2–4 days . Right panel: protein levels of eS26 in whole cell extracts of indicated strains were determined by Western analyses using α-eS26 antibodies . Arc1 protein levels served as loading control . ( B ) eS26-depleted cells accumulate immature 20S pre-rRNA in the cytoplasm . PGAL1-RPS26Arps26bΔ cells transformed with indicated plasmids were grown in glucose containing liquid media at 37°C to mid-log phase . Localization of 20S pre-rRNA was analyzed by FISH using a Cy3-labeled oligonucleotide complementary to the 5′ portion of ITS1 ( red ) . Nuclear and mitochondrial DNA was stained with DAPI ( blue ) . Scale bar = 5 µm . ( C ) eS26-depleted cells accumulate 80S-like particles . The indicated strains were grown in glucose containing liquid media at 30°C to mid-log phase . Cell extracts were prepared after cycloheximide treatment and subjected to sedimentation centrifugation on 7–50% sucrose density gradients . Polysome profiles were recorded at OD254nm ( top panels ) . The peaks for 40S and 60S subunits , 80S ribosomes and polysomes are indicated . Sucrose gradients were fractionated , the RNA was extracted , separated on a 2% Agarose gel , stained with GelRed ( Biotium , middle panels ) and subsequently analyzed by Northern blotting using probes against the indicated rRNAs ( bottom panels ) . Exposure times for phosphoimager screens were 20 min for 25S and 18S rRNA , and 3–4 hr for 20S pre-rRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 005 The robust interaction between the predominantly nuclear localized Tsr2 and eS26 prompted us to investigate at which stage eS26 is recruited to pre-40S subunits . To address this , we isolated pre-ribosomal particles at different maturation stages along the 40S biogenesis pathway ( Grandi et al . , 2002; Schäfer et al . , 2003 ) . Noc4-TAP purifies the earliest precursor of the pre-40S subunit , the 90S pre-ribosome; Enp1-TAP purifies both the 90S and early pre-40S subunits; Rio2-TAP purifies a late pre-40S subunit containing immature 20S pre-rRNA; and Asc1-TAP purifies a 40S subunit containing mature 18S rRNA and devoid of late assembly factors ( Figure 3A , Figure 3—figure supplement 1A ) . Co-enrichment of eS26 with pre-ribosomal particles was assessed by ( 1 ) Western analyses using antibodies that recognize eS26 and ( 2 ) selected reaction monitoring mass spectrometry ( SRM-MS ) . SRM-MS is a reliable tool that overcomes stochastic under sampling of peptides , a critical deficit in shotgun mass spectrometry which complicates the reproducible detection and precise quantitation of proteins in a complex mixture ( Picotti and Aebersold , 2012 ) . SRM relies on the development of specific mass spectrometric-based assays for every target protein and their subsequent application to the relative or absolute quantification within multiple biological samples . We developed a set of SRM assays that enabled us to simultaneously monitor the co-enrichment of eS26 and different r-proteins: uS7 ( Rps5 ) , eS28 ( Rps28 ) , eS1 ( Rps1 ) and uS11 ( Rps14 ) ( Figure 3B ) with multiple pre-ribosomal particles . Both Western and SRM analyses revealed that eS26 co-enriches efficiently with the earliest ribosomal precursor , the 90S , and different pre-ribosomes along the 40S maturation pathway ( Figure 3A , B ) . The Western signal for eS26 on the 90S pre-ribosome ( Noc4-TAP ) is specific since no association was detected with the earliest 60S pre-ribosome ( Ssf1-TAP ) ( Figure 3—figure supplement 1B ) . 10 . 7554/eLife . 03473 . 006Figure 3 . eS26 is incorporated into the earliest pre-ribosome , the 90S . ( A ) eS26 co-enriches with pre-ribosomal particles along the 40S maturation pathway . Pre-ribosomal particles in the 40S maturation pathway were purified using the indicated TAP-tagged baits . Calmodulin-eluates were analyzed by Silver staining and Western analyses using the indicated antibodies . The r-protein uS7 served as loading controls for the TAPs . ( B ) SRM-MS reveals co-enrichment of eS26 with pre-ribosomal particles . Upper panel: the relative abundance of different r-proteins was normalized to uS7 levels in the indicated TAP purifications ( three independent biological replicates ) . The error bars show the standard deviation . Lower panel: the intensity of different transitions ( listed in the box ) of two specific peptides of eS26 was determined by SRM mass spectrometry in the indicated TAP purifications . ( C ) eS26-GFP accumulates in the nucleus in a yrb2Δ strain . Left panel: WT , rps26aΔ and RPS26A-GFP cells were spotted in 10-fold dilutions and grown at indicated temperatures for 3–7 days . Right panel: WT and yrb2Δ cells expressing eS26-GFP were grown in glucose containing liquid media to mid-log phase at 20°C . Localization of eS26-GFP was monitored by fluorescence microscopy . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 00610 . 7554/eLife . 03473 . 007Figure 3—figure supplement 1 . Tsr2 and eS26 protein levels in the indicated TAP strains and levels of 20S pre-rRNA and 18S rRNA in the indicated TAP purified particles . ( A ) Noc4- , Enp1- and Rio2-TAP purify pre-40S subunits containing immature 20S pre-rRNA whereas Asc1-TAP purifies a 40S subunit containing mature 18S rRNA . 1 µg of RNA isolated from the indicated pre-40S TAP-eluates was separated on a 2% Agarose gel and probed against indicated rRNAs by Northern blotting . 1 µg of total RNA extracted from WT cells was used as a control . ( B ) eS26 does not co-enrich with the earliest 60S pre-ribosome . Noc4-TAP , the earliest pre-ribosomal particle and Ssf1-TAP , the earliest pre-ribosome in the 60S maturation pathway were isolated . The Calmodulin eluates were visualized by Silver staining and by Western analyses using the indicated antibodies . The CBP signal served as loading controls for the TAPs . ( C ) Tsr2 and eS26 protein levels in indicated TAP strains ( also used in Figure 3A ) are equal to levels in WT cells . Whole cell extracts ( WCE ) were prepared from the indicated strains and analyzed by Western analyses using antibodies against Tsr2 and eS26 . The protein Arc1 served as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 007 To support these biochemical data , we performed a complementary cell-biological experiment . If eS26 were targeted to the 90S pre-ribosome , then impairment in pre-40S subunit export should result in its accumulation in the nucleus . To monitor eS26 localization in vivo , we tagged RPS26A with GFP at the C-terminus ( eS26-GFP ) in WT and yrb2Δ cells at the genomic locus . Unlike the rps26aΔ mutant , the RPS26A-GFP strain was not impaired in growth at 20°C and 37°C indicating that addition of GFP does not impair its function on the 40S subunit ( Figure 3C , left panel ) . As expected , WT cells displayed a strong cytoplasmic localization of eS26-GFP ( Figure 3C , right panel ) . In contrast , in yrb2Δ cells eS26-GFP accumulated in the nucleus ( Figure 3C , right panel ) . Together all these data suggest that eS26 is transported to the nucleus for loading on the 90S pre-ribosome . Consistent with the sedimentation studies and direct binding to only eS26 ( Figure 1 ) , Tsr2 did not detectably co-enrich with affinity purified pre-ribosomal particles in the 40S maturation pathway ( Figure 3A ) . This lack of co-enrichment was not due to altered protein levels in the different TAP strains , since Western analyses of whole cell extracts revealed that Tsr2 was expressed at WT levels ( Figure 3—figure supplement 1C ) . Altogether , these results suggest that there are at least two populations of eS26 in vivo , one bound to ribosomes and another bound to Tsr2 . We next investigated how eS26 is imported into the nucleus prior to its incorporation into the 90S pre-ribosome . In yeast , the most abundant importin Kap123 transports various r-proteins into the nucleus ( Rout et al . , 1997; Schlenstedt et al . , 1997 ) . However , r-proteins also utilize additional importins , including Pse1 , Kap104 , Sxm1 and Nmd5 ( Rout et al . , 1997; Sydorskyy et al . , 2003 ) . We investigated the interaction between eS26 and all yeast importins in vitro . These studies revealed that the importins Kap123 , Kap104 and Pse1 efficiently bound eS26 ( Figure 4A ) . A very weak interaction was observed between Sxm1 , Kap95 and Nmd5 and eS26 , and no binding was observed with the remaining importins ( Figure 4—figure supplement 1 ) . In contrast , none of the importins bound to either Tsr2 or the Tsr2:eS26 complex ( Figure 4A , Figure 4—figure supplement 1 ) , indicating that eS26 alone specifically interacts with importins . 10 . 7554/eLife . 03473 . 008Figure 4 . Kap123 , Kap104 and Pse1 transport eS26 to the nucleus . ( A ) eS26 , but not Tsr2:eS26 or Tsr2 , interacts with Kap123 , Kap104 and Pse1 . Recombinant , GST-Kap123 , GST-Kap104 , GST-Pse1 and GST alone were immobilized on Glutathione Sepharose and incubated with purified 3 . 4 µM Tsr2 , 4 µM Tsr2:eS26 , or E . coli lysate containing ∼4 µM eS26FLAG in PBSKMT combined with competing E . coli lysates for 1 hr at 4°C . After washing with PBSKMT , bound proteins were eluted in SDS sample buffer and separated by SDS-PAGE . Proteins were visualized by Coomassie Blue staining or Western analyses using indicated antibodies . L = input . GST-tagged importins are indicated with asterisks . ( B ) Nuclear uptake of GFP-eS26 is impaired in kap123Δ and kap104Δ mutants . Strains expressing GFP-eS26 were grown in synthetic media at 25°C ( ts-mutants: pse1-1 and kap104Δ ) or 30°C to mid-log phase . Ts-mutant strains were then shifted to 37°C for 4 hr and localization of GFP-eS26 was analyzed by fluorescence microscopy . Percentage of cells displaying cytoplasmic mislocalization of the GFP-eS26 fusion is indicated . Scale bar = 5 µm . ( C ) Tsr2-3xGFP is targeted to the nucleus by Kap123 . Importin mutant strains expressing Tsr2-3xGFP were grown in synthetic media at 25°C ( ts-mutants: pse1-1 and kap104Δ ) or 30°C to mid-log phase . Pse1-1 and kap104Δ cells were then shifted to 37°C for 4 hr . PGAL1-RPS26Arps26bΔ cells containing Tsr2-3xGFP were grown for 15 hr in glucose containing media . Localization of Tsr2-3xGFP was analyzed by fluorescence microscopy . Scale bar = 5 µm . ( D ) RanGTP ( His6-Gsp1Q71L-GTP ) does not efficiently release eS26 from Kap123 and Pse1 . GST-importin:eS26FLAG complexes immobilized on Glutathione Sepharose were incubated with either buffer alone or with 1 . 5 µM RanGTP or 3 nM 3′-end of 18S rRNA for 1 hr at 4°C . Washing , elution , and visualization were performed as in ( A ) . GST-tagged importins are indicated with asterisks . ( E ) Tsr2 efficiently dissociates the Kap123:eS26FLAG complex . The GST-Kap123: eS26FLAG complex immobilized on Glutathione Sepharose was incubated with either buffer alone or with 1 . 5 µM or 375 nM RanGTP or 1 . 5 µM or 375 nM Tsr2 . Samples were withdrawn at the indicated time points . Washing , elution , and visualization were performed as in ( A ) . GST-tagged Kap123 is indicated with an asterisk . ( F ) eS26 stably associates with Tsr2 after its release from Kap123 . Left panel indicates the experimental setup as flowchart . Immobilized GST-Kap123:eS26FLAG complex was incubated with 1 . 5 µM His6-Tsr2 or buffer alone . As shown in the flowchart , the supernatant was incubated with Ni-NTA Agarose for 1 hr at 4°C ( IP-Sup ) . Washing , elution , and visualization were performed as in ( A ) . GST-tagged Kap123 is indicated with an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 00810 . 7554/eLife . 03473 . 009Figure 4—figure supplement 1 . eS26 , but not Tsr2:eS26 or Tsr2 , interacts with importins . Recombinant GST tagged importins , immobilized on Glutathione Sepharose , were incubated with purified 3 . 4 µM Tsr2 , 4 µM Tsr2:eS26 or E . coli lysate containing ∼4 µM eS26FLAG in PBSKMT and competing E . coli lysates for 1 hr at 4°C . After washing , bound proteins were eluted in SDS sample buffer , separated by SDS-PAGE , and visualized by either Coomassie Blue staining or Western analyses using indicated antibodies . L = input . GST-tagged importins are indicated with asterisk , His6-Srp1 is indicated with a rectangle . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 00910 . 7554/eLife . 03473 . 010Figure 4—figure supplement 2 . GFP-eS26 binds to importins and Tsr2 but is not incorporated into pre-ribosomes . ( A ) Location of N- and C-terminus of eS26 within the mature 40S subunit ( Rabl et al . , 2011 ) . eS26 N-terminus ( green ) is embedded deeply within the 40S subunit whereas the C-terminus ( red ) projects away from the body of the 40S subunit . Red letters indicate the 20 C-terminal residues that are not visualized in the structure ( B ) GFP-eS26 is not found in heavier fractions on sucrose gradients . WT lysates and lysates containing GFP-eS26 were subjected to sucrose gradient sedimentation as described in Figure 1D . The peaks for 40S and 60S subunits , 80S ribosomes and polysomes are indicated . The proteins in the gradient were detected by Western analyses using the indicated antibodies . ( C ) GFP-eS26 binds to Kap123 , Kap104 and Pse1 . Recombinant GST-Kap123 , -Kap104 , -Pse1 and GST alone were immobilized on Glutathione Sepharose and then incubated with E . coli lysate containing GFP-eS26 for 1 hr at 4°C . Bound proteins were eluted in SDS sample buffer , separated by SDS-PAGE and visualized by Coomassie Blue staining and Western analyses using α-GFP antibody . L = input . ( D ) GFP-eS26 is unable to rescue the lethality of the eS26 deficient strain . The PGAL1-RPS26Arps26bΔ strain transformed with different plasmids encoding eS26 or GFP-eS26 were spotted in 10-fold dilutions on selective glucose containing plates and grown at indicated temperatures for 3–7 days . ( E ) GFP-eS26 and GFP-eS26D33N levels are strongly reduced in Tsr2-depleted cells . Whole cell extracts ( WCE ) prepared from WT and Tsr2-depleted cells were assessed by Western analyses using antibodies against the indicated proteins . Arc1 protein levels served as loading control . ( F ) Upper panel: the Zn2+-binding domain of eS26 is required for efficient nuclear uptake . WT cells expressing GFP-eS26 truncations were grown in synthetic media at 30°C to mid-log phase and the localization of GFP-eS26 truncations was analyzed by fluorescence microscopy . Scale bar = 5 µm . Lower panel: Schematic for the eS26 truncations used for fluorescence microscopy . ( G ) GFP-eS26C77W protein levels are strongly reduced in ( WCE ) extracts . Whole cell extracts were prepared from PGAL1-RPS26Arps26bΔ cells transformed with plasmids encoding for GFP-eS26 WT and mutant proteins . eS26 protein levels were assessed by Western analyses using α-GFP antibodies . Arc1 served as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 01010 . 7554/eLife . 03473 . 011Figure 4—figure supplement 3 . Tsr2 efficiently releases the conserved eS26 from importins . ( A ) Left panel: sequence alignment of eS26 from the indicated organisms done by ClustalO ( Sievers and Higgins , 2014; Sievers et al . , 2011 ) . Conservation at each position is depicted as a gradient from light blue ( 50% identity ) to dark blue ( 100% identity ) . Mutated residues linked to DBA are depicted with orange ( Asp33 ) and green ( Cys77 ) dots . Right panel: location of eS26 within the mature 40S subunit ( Rabl et al . , 2011 ) . eS26 clamps the 3′-end of the mature 18S rRNA at the site where the endonuclease Nob1 cleaves the immature 20S pre-rRNA . Inset depicts the 3′-end portion of 18S rRNA ( red ) in contact with eS26 ( blue ) . The position of amino acids D33 ( orange ) and C77 ( green ) that are mutated in DBA or KFS and the coordinated Zn2+ ion ( black ) are depicted . ( B ) RanGTP and the 3′-end of 18S rRNA cannot dissociate the Kap123:eS26 complex . GST-Kap123:eS26aFLAG complexes , immobilized on Glutathione Sepharose , were incubated with buffer alone or with 1 . 5 µM RanGTP , 1 . 5 µM Tsr2 , 3 nM of the 3′-end of 18S rRNA or the combination of RanGTP and the 3′ end of 18S rRNA for 1 hr at 4°C . Bound proteins were eluted in SDS sample buffer , separated by SDS-PAGE and visualized by Coomassie Blue staining and Western analyses using α-eS26 antibodies . L = input . GST-tagged importins are indicated with asterisks . ( C ) eS26 stably associates with Tsr2 after its release from Pse1 . Immobilized GST-Pse1:eS26FLAG complex was treated with 1 . 5 µM His6-Tsr2 or buffer alone . The supernatant was incubated with Ni-NTA Agarose for 1 hr at 4°C ( IP-Sup ) . Washing , elution , and visualization were performed as in Figure 4E . GST-tagged Pse1 is indicated with an asterisk . ( D ) RanGTP , but not Tsr2 dissociated the Pse1:Slx9 complex in vitro . Pse1:Slx9 complexes were immobilized on Glutathione Sepharose and incubated with buffer alone or with 1 . 5 µM RanGTP , 1 . 5 µM Tsr2 or 3 nM 3′-end of 18S rRNA for 1 hr at 4°C and analyzed as in Figure 4C . GST-tagged importins are indicated with asterisks . ( E ) Tsr2 efficiently dissociates importin:eS26FLAG complexes . GST-Kap104: eS26FLAG and GST-Pse1:eS26FLAG complexes immobilized on Glutathione Sepharose were incubated with either buffer alone or with 1 . 5 µM or 375 nM RanGTP or 1 . 5 µM or 375 nM Tsr2 . Samples were withdrawn at the indicated time points ( 1 , 2 , 4 , 8 min ) . Washing , elution , and visualization were performed as in Figure 4A . GST-tagged importins are indicated with asterisks . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 01110 . 7554/eLife . 03473 . 012Figure 4—figure supplement 4 . RanGTP and Tsr2 do not release eS31 , eS8 and uS14 from Kap123 . GST-Kap123 and GST alone were immobilized on Glutathione Sepharose and incubated with E . coli lysate containing ∼4 µM eS14FLAG , eS31FLAG or eS8FLAG in PBSKMT combined with competing E . coli lysates for 1 hr at 4°C . GST-Kap123:eS14FLAG , GST-Kap123:eS31FLAG , GST-Kap123:eS8FLAG complexes were incubated with either buffer alone or with 1 . 5 µM RanGTP or 1 . 5 µM Tsr2 for 1 hr at 4°C . Bound proteins were eluted in SDS sample buffer and separated by SDS-PAGE . Proteins were visualized by Coomassie Blue staining or Western analyses using α-FLAG-antibodies . L = input . GST-Kap123 is indicated with asterisks . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 012 To verify our interaction data in vivo , we monitored nuclear uptake of eS26 in WT cells and in different importin mutants . The r-protein eS26 is assembled into the 90S pre-ribosome and is then rapidly transported to the cytoplasm as part of the 40S pre-ribosome . To investigate eS26 nuclear uptake in vivo we uncoupled its import from its export . Structural analyses of the 40S subunit showed that the N-terminus of eS26 is embedded within the rRNA framework ( Figure 4—figure supplement 2A; Rabl et al . , 2011 ) . We fused GFP to the N-terminus of eS26 with the aim to impair its incorporation into the 90S pre-ribosome . Sucrose gradient analyses showed that GFP-eS26 co-sediments only in lighter fractions at the top of the gradient suggesting that it is not incorporated into pre-ribosomes ( Figure 4—figure supplement 2B ) . In vitro binding studies showed that like eS26 , GFP-eS26 interacts with Kap123 , Kap104 and Pse1 ( Figure 4—figure supplement 2C ) . Thus , the GFP-eS26 fusion protein is functional to recruit the import machinery , although it does not complement the eS26-depleted strain ( Figure 4—figure supplement 2D ) . Further , GFP-eS26 directly binds Tsr2 ( Figure 4—figure supplement 2C ) and importantly , like eS26 , GFP-eS26 is degraded upon Tsr2-depletion ( Figure 4—figure supplement 2E ) . We exploited the GFP-eS26 fusion protein as a tool to monitor the nuclear uptake of eS26 in different importin mutants . Consistent with in vitro binding assays , nuclear uptake of GFP-eS26 was reduced in kap123Δ and kap104Δ cells ( Figure 4B ) , indicating that eS26 import requires these importins . Nuclear localization of GFP-eS26 in the pse1-1 ts mutant at restrictive temperature remained unaffected ( Figure 4B ) indicating that impairment of this importin alone does not inhibit the nuclear import of eS26 . The pse1-1 kap123Δ mutant showed only a slight increase in cytoplasmic staining of GFP-eS26 ( Figure 4B ) . Nuclear import of GFP-eS26 was unaffected in the kap114Δ sxm1Δ double mutant and sxm1Δ kap120Δ nmd5Δ triple mutant ( Figure 4B ) . Next , we investigated which region of eS26 contributes to its nuclear uptake . For this , we monitored the localization of different truncated versions of eS26 fused to -GFP at the N-terminus . These cell-biological analyses revealed that the Zn2+-binding domain is required for efficient nuclear uptake of eS26 ( Figure 4—figure supplement 2F ) . If eS26 were imported into the nucleus in complex with Tsr2 , then we reasoned that depletion of eS26 would induce Tsr2 mislocalization to the cytoplasm . However , localization of Tsr2-3xGFP was not affected upon eS26-depletion ( Figure 4C ) . These studies together with the observation that Tsr2:eS26 complex is unable to recruit importins argue against the idea that eS26 is transported to the nucleus in complex with Tsr2 . We conclude that Kap123 and Kap104 target eS26 to the nucleus and that Tsr2 is not a component of this import complex . Next , we investigated how Tsr2 is targeted to the nucleus in vivo . For this , we monitored the location of Tsr2-3xGFP in different importin mutants . We found that Tsr2-3xGFP mislocalizes to the cytoplasm in the kap123Δ mutant , but not in other importin mutants for e . g . kap104Δ and pse1-1 ( Figure 4C ) . Thus , Kap123 seems to be the major import receptor for Tsr2 . However , we did not observe a direct interaction between Tsr2 and Kap123 or any other importin in vitro ( Figure 4A , Figure 4—figure supplement 1 ) . One possibility could be that import of Tsr2 by Kap123 is regulated by post-translational modification . Alternatively , Tsr2 might be transported into the nucleus via a ‘piggy bag’ mechanism bound to another yet unknown Kap123 cargo . We can exclude the possibility that eS26 serves as an adaptor to import Tsr2 since ( 1 ) Tsr2 does not mislocalize to the cytoplasm in a eS26-depleted strain ( Figure 4C ) and ( 2 ) in vitro binding assays show that the Tsr2:eS26 complex does not interact with Kap123 ( Figure 4A ) . After transport of an importin:cargo complex into the nucleus , RanGTP binds to the N-terminal region of the importin , triggering cargo release and allowing recycling of the importin to participate in subsequent import cycles ( Lee et al . , 2005; Cook et al . , 2007; Kobayashi and Matsuura , 2013 ) . To test whether the release of eS26 from the importins is RanGTP-dependent , we performed in vitro dissociation assays . A pre-formed importin:eS26 complex was incubated with 1 . 5 µM Gsp1Q71L-GTP ( equivalent to the human RanQ69L mutant that cannot efficiently hydrolyze GTP , hereafter Gsp1Q71L-GTP is termed RanGTP; Bischoff et al . , 1994; Maurer et al . , 2001 ) . Although RanGTP was able to dissociate the Kap104:eS26 complex , and partially dissociate the Pse1:eS26 complex , we did not observe dissociation of the Kap123:eS26 complex even after 1 hr incubation ( Figure 4D ) . It was reported that both RNA and RanGTP are required to release of the mRNA binding proteins Nab2 and Nab4 from Kap104 and the mRNA export factor Npl3 from Mtr10 ( Senger et al . , 1998; Lee and Aitchison , 1999 ) . Because eS26 directly interacts with the 3′-end of the 18S rRNA ( Figure 4—figure supplement 3A , right panel ) , we tested if this region of the 18S rRNA is required to release eS26 from Kap123 . However , eS26 remained stably bound to Kap123 in the presence of this RNA , either alone or in combination with RanGTP ( Figure 4D , Figure 4—figure supplement 3B ) . Since the Tsr2:eS26 complex was unable to interact with importins , we tested whether Tsr2 stimulates the release of eS26 from importins . Surprisingly , Tsr2 alone efficiently removed eS26 from Kap123 , Pse1 and Kap104 ( Figure 4—figure supplement 3B , C and data not shown ) . This release was specific , since only RanGTP , but not Tsr2 , was able to remove the 40S assembly factor Slx9 ( Faza et al . , 2012 ) from the Pse1:Slx9 complex under the same conditions ( Figure 4—figure supplement 3D ) . Moreover , Tsr2 specifically releases eS26 from the importin:eS26 complex , since it did not dissociate other tested importin:r-protein complexes ( Kap123:uS14 , Kap123:eS31 and Kap123:eS8 ) ( Figure 4—figure supplement 4 ) . Since eS26 was inefficiently removed from the Pse1:eS26 complex after 1 hr incubation with RanGTP ( Figure 4D ) , we investigated the dissociation kinetics of importin:eS26 complexes in the presence of RanGTP or Tsr2 . For this , the importin:eS26 complex was incubated with 1 . 5 µM of either RanGTP or Tsr2 and the release of eS26 from the importin was monitored over time . We found that the amount of eS26 bound to Kap123 , Pse1 or Kap104 was only slightly reduced after 8 min , even though RanGTP was efficiently recruited to the different importins:eS26 complexes ( Figure 4E , left panel and Figure 4—figure supplement 3E , left panel ) . In contrast , Tsr2 completely removed eS26 from these importins within 1 min incubation ( Figure 4E , right panel and Figure 4—figure supplement 3E , right panel ) . Notably , even at lower concentrations ( 375 nM ) Tsr2 was able to release eS26 from the importin:eS26 complex ( Figure 4E , Figure 4—figure supplement 3E ) . Moreover , Tsr2 stably associated with the released eS26 ( Figure 4F , Figure 4—figure supplement 3C ) . The observation that Tsr2 is able to extract eS26 from importins , prompted us to investigate whether Tsr2 plays a role in the transfer of eS26 to the assembling pre-ribosome . To test this , we isolated Enp1-TAP , which purifies both the 90S pre-ribosome and an early pre-40S subunit , from WT and Tsr2-depleted cells and assessed co-enrichment of eS26 by Western analyses . Consistent with a role for Tsr2 in supplying eS26 to the 90S pre-ribosome , we found that eS26 does not efficiently co-enrich with Enp1-TAP in Tsr2-depleted cells ( Figure 5A ) . This was specific for eS26 , since the recruitment of uS7 and uS3 to Enp1-TAP particles was not affected in these cells ( Figure 5A ) . This lack of enrichment was due to decreased eS26 protein levels , since Western analyses of whole cell extracts derived from Tsr2-depleted cells revealed strongly reduced eS26 protein levels ( Figure 5B ) . These data led us to test whether eS26 becomes susceptible to proteolysis in Tsr2-depleted cells . To this end , we monitored eS26 protein levels over time in whole cell extracts after switching the PGAL1-TSR2 strain to repressive glucose containing media . These analyses revealed that eS26 protein levels decreased over time upon Tsr2-depletion ( Figure 5C ) . 10 . 7554/eLife . 03473 . 013Figure 5 . Tsr2 shields eS26 from proteolysis and aggregation , and promotes safe transfer to the 90S pre-ribosome . ( A ) Efficient recruitment of eS26 to Enp1-TAP requires Tsr2 . Enp1-TAP was isolated from WT and Tsr2-depleted cells . After tandem affinity purification , eluates were separated by 4–12% gradient SDS-PAGE and subjected to Western analyses using indicated antibodies . CBP ( α-TAP ) levels served as loading control . ( B ) eS26 levels are strongly reduced in Tsr2-depleted cells . Whole cell extracts ( WCE ) prepared from WT and Tsr2-depleted cells were assessed by Western analyses using antibodies against the indicated proteins . Arc1 protein levels served as loading control . ( C ) Tsr2 protects eS26 from proteolysis in vivo . The conditional mutant strain PGAL1-TSR2 growing on galactose medium was transferred to repressive glucose containing liquid media at 30°C . Cells were withdrawn at the indicated time points and whole cell extracts were prepared . Western analyses were performed to determine the levels of the indicated proteins . Arc1 served as loading control . ( D ) Tsr2 prevents aggregation of recombinant eS26 in vitro . The aggregation assay was performed in a 384-well plate . In each well 33 µM GST-eS26 and a given concentration of Tsr2 ( 0 up to 266 µM ) in PBSKMT was pre-incubated for 1 hr at 4°C ( final volume: 90 µl ) . 250 nM of PreScission protease was added to initiate aggregation . After 1 hr of incubation , the scattering signal of the aggregated eS26 was monitored by a 384-well plate reader by measuring the intensity at 450 nm ( Y-axes ) . Concentration of Tsr2 used in the assay ( X-axes ) are expressed as a molar ratio of eS26:Tsr2 . Four replicates for each well were measured . The error bars show the standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 013 We observed that purified recombinant eS26 was highly prone to aggregation . Expressing eS26 as a fusion protein with a highly soluble GST tag suppressed its tendency to aggregate . However , removal of the GST tag after cleavage by PreScission protease resulted in immediate aggregation of free eS26 , as determined by a massive increase in the light scattering intensity ( Figure 5D ) . We tested whether Tsr2 could suppress the aggregating ability of recombinant eS26 . We treated GST-eS26 with PreScission protease in absence and presence of Tsr2 . A concomitant decrease in the light scattering of the reaction mixture was observed ( Figure 5D ) , as the Tsr2 concentration in the cleavage buffer was increased , indicating aggregation of free eS26 was suppressed . Altogether , these data indicate that Tsr2 protects eS26 , and thereby ensures a safe transfer to the 90S pre-ribosome . Mutations in r-proteins have been linked to Diamond-Blackfan anemia ( DBA ) , a rare congenital red blood cell aplasia ( Ellis and Lipton , 2008; Ganapathi and Shimamura , 2008; Narla and Ebert , 2010; Ellis and Gleizes , 2011; McCann and Baserga , 2013; Ellis , 2014 ) . Several mutations in the start codon of RPS26 , including two mutations within eS26 , D33N and C77W have been linked to DBA ( Doherty et al . , 2010; Cmejla et al . , 2011 ) . Both residues are highly conserved from yeast to humans ( Figure 4—figure supplement 3A , left panel ) . The C77W mutation is additionally linked to Klippel-Feil syndrome ( KFS ) , a skeletal developmental disorder in DBA patients ( Cmejla et al . , 2011 ) . In order to analyze the phenotypes induced by the D33N and C77W mutations , we introduced the individual mutations into yeast RPS26A . First , plasmids encoding DBA-linked mutants were transformed into the PGAL1-RPS26Arps26bΔ strain and growth was analyzed on glucose containing media . Whereas the D33N mutant partially rescued the lethality of the eS26-conditional mutant , the C77W variant did not allow any growth ( Figure 6A ) . Further , as in the PGAL1-TSR2 strain under repressive conditions , both variants resulted in strongly reduced eS26 protein levels ( Figure 6B ) . Neither strain displayed defects in the nuclear export of pre-40S subunits ( Figure 1—figure supplement 1B ) . As expected , neither variant was able to rescue the 20S pre-rRNA processing defect of eS26-deficient cells , as determined by the strong cytoplasmic localization of Cy3-ITS1 ( Figure 6C ) . Thus , eS26 mutants linked to DBA are impaired in cytoplasmic processing of 20S pre-rRNA . 10 . 7554/eLife . 03473 . 014Figure 6 . The eS26C77W mutant associated with Klippel-Feil syndrome in Diamond-Blackfan anemia patients is impaired in binding importins . ( A ) The DBA linked eS26D33N and eS26C77W mutants are unable to fully rescue the growth defect of eS26-depleted cells . The PGAL1-RPS26Arps26bΔ strain transformed with different plasmids encoding eS26 mutants were spotted in 10-fold dilutions on selective glucose containing plates and grown at indicated temperatures for 3–7 days . Residues mutated in DBA are depicted in Figure 4—figure supplement 3A . ( B ) DBA linked mutations cause strongly reduced eS26 protein levels . Whole cell extracts were prepared from PGAL1-RPS26Arps26bΔ cells transformed with indicated plasmids encoding for eS26 WT and mutant proteins . eS26 protein levels were assessed by Western analyses using α-eS26 antibodies . Arc1 served as loading control . ( C ) eS26 mutants linked to DBA accumulate 20S pre-rRNA in the cytoplasm . PGAL1-RPS26Arps26bΔ cells transformed with plasmids encoding for eS26 WT and mutant proteins were grown at 37°C to mid-log phase in glucose containing medium . Localization of 20S pre-rRNA was analyzed by FISH using a Cy3-labeled oligonucleotide complementary to the 5′ portion of ITS1 ( red ) . Nuclear and mitochondrial DNA was stained with DAPI ( blue ) . Scale bar = 5 µm . ( D ) Tsr2 interacts with eS26 mutants linked to DBA . Recombinant GST-Tsr2 was immobilized on Glutathione Sepharose and then incubated with E . coli lysates containing eS26aFLAG , eS26D33NFLAG or eS26C77WFLAG lysates for 1 hr at 4°C . Bound proteins were eluted by SDS sample buffer , separated by SDS-PAGE and detected by Coomassie Blue staining . L = input . ( E ) eS26C77W is impaired in binding to Kap123 , Kap104 and Pse1 . Recombinant GST-Kap123 , -Kap104 , -Pse1 and GST alone were immobilized on Glutathione Sepharose and then incubated with E . coli lysate containing eS26FLAG , eS26D33NFLAG or eS26C77WFLAG for 1 hr at 4°C . Bound proteins were eluted in SDS sample buffer , separated by SDS-PAGE and visualized by Coomassie Blue staining and Western analyses using α-eS26 antibody . L = input . ( F ) The GFP-eS26D33N fusion protein is efficiently targeted to the nucleus . WT cells expressing GFP-eS26 and GFP-eS26D33N were grown in synthetic media at 30°C to mid-log phase and the localization of GFP-eS26 was analyzed by fluorescence microscopy . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 014 We tested whether the identified eS26 binders could interact with D33N and C77W variants in vitro . Pull-down assays demonstrated that both mutant proteins efficiently bound Tsr2 ( Figure 6D ) , suggesting these mutations do not contribute to the Tsr2:eS26 interaction surface . The eS26D33N mutant efficiently binds to Kap123 , Kap104 and Pse1 ( Figure 6E ) . In agreement with these interaction studies , nuclear uptake of GFP-eS26D33N was not affected ( Figure 6F ) and the levels of GFP-eS26D33N were strongly reduced upon Tsr2-depletion ( Figure 4—figure supplement 2E ) . In contrast , the eS26C77W mutant interacted weakly with these importins ( Figure 6E ) . We were unable to localize GFP-eS26C77W; whole cells extracts revealed that GFP-eS26C77W protein levels were strongly reduced ( Figure 4—figure supplement 2G ) .
Using Western analyses and targeted SRM assays , we found that untagged eS26 is recruited to Noc4-TAP and co-enriches with nuclear pre-40S subunits that contain 20S pre-rRNA ( Figure 3A , B , Figure 3—figure supplement 1A ) . Moreover , eS26-GFP accumulated in the nucleus of yrb2Δ cells that are specifically impaired in 40S pre-ribosome export ( Figure 3C ) . These data suggest that eS26 can be transported to the 90S pre-ribosome . Our findings contrast a previous report wherein a FLAG-tagged eS26 immunoprecipitated mainly 18S rRNA ( Ferreira-Cerca et al . , 2007 ) . In addition , eS26 was not identified in mass spectrometry studies of a pre-40S subunit , and was suggested to replace the assembly factor Pno1 ( Strunk et al . , 2011 , 2012 ) . However , we found that Pno1-TAP efficiently co-enriched eS26 ( Figure 3A ) . Further , Tsr2-depletion impaired recruitment of Pno1 to pre-40S subunits , suggesting that eS26 helps to recruit Pno1 ( Figure 5A ) . eS26-depletion impaired only 20S pre-rRNA processing in the cytoplasm , suggesting that eS26 does not apparently affect 90S assembly per se , but is specifically required for final maturation ( Figure 2B ) . Although the precise timing of eS26 recruitment remains unclear , based on our data , we propose that it is a late event during 90S assembly . eS26 clamps the 3′-end of mature 18S rRNA ( Figure 4—figure supplement 3A , right panel; Rabl et al . , 2011 ) , precisely at the site where the endonuclease Nob1 cleaves the 20S pre-rRNA . Pre-40S subunits that lack eS26 escape nuclear proofreading and are efficiently transported into the cytoplasm . These incompletely assembled pre-40S subunits recruit the endonuclease Nob1 ( Figure 5A ) and form 80S-like particles ( Figures 1D and 2C ) . However , they fail to process 20S pre-rRNA ( Figures 1C and 2B ) , an essential pre-requisite to form a mature 40S subunit . Thus , 20S pre-rRNA within an 80S-like particle becomes an optimal substrate for Nob1 only when the pre-40S subunit has satisfied a checklist that assesses its potential to translate , including the incorporation of eS26 . We propose that the cytoplasmic 20S pre-rRNA cleavage functions as one of the checkpoints that prevent incompletely assembled , pre-40S subunits from entering translation . eS26 is targeted to the 90S pre-ribosome and therefore must reach the nucleolus . Unlike the Kap104 adaptor Syo1 that co-imports uL18 ( yeast Rpl5 ) and uL5 ( yeast Rpl11 ) ( Kressler et al . , 2012 ) , Tsr2 does not mediate interactions between eS26 and importins . Instead , our data identified Kap123 and Kap104 as the major importins that directly bind and transport eS26 into the nucleus ( Figure 4 ) . Recruitment of RanGTP did not efficiently trigger the dissociation of importin:eS26 complexes ( Figure 4 ) . One possibility could be that eS26 engages in a novel interaction with the importins , thereby delaying its release . Such a delay may ensure the coordinated handover to the next binding factor , Tsr2 . Structural analyses of the importin:eS26 complex should provide clues into why eS26 is inefficiently released from importins by RanGTP . In contrast to RanGTP , Tsr2 efficiently removed eS26 from its importins ( Figure 4 , Figure 4—figure supplement 3 ) , identifying an atypical RanGTP-independent mechanism to terminate the import cycle . The observation that Tsr2 prevents proteolysis and aggregation of eS26 ( Figure 5 ) indicates an additional ‘private’ chaperone function . Thus our study adds Tsr2:eS26 to the growing list of known chaperones:r-proteins pairs ( Sqt1:uL16; Rrb1:uL3; Yar1:uS3 ) required for ribosome assembly ( Eisinger et al . , 1997; Iouk et al . , 2001; Schaper et al . , 2001; Koch et al . , 2012 ) . Tsr2 may prevent eS26 from undergoing non-specific interactions with nucleic acids during its journey towards the 90S pre-ribosome . How eS26 is transferred from Tsr2 to 90S pre-ribosomes remains unclear . It is tempting to speculate that posttranslational modifications and/or energy consuming enzymes couple the extraction of eS26 from Tsr2 and subsequent incorporation . Based on our data , we propose a model in which eS26 is transported to the nuclear compartment predominantly by importins Kap123 and Kap104 ( Figure 7 ) . Inside the nucleus , eS26 is removed from its importins in a RanGTP-independent mechanism mediated by Tsr2 . The released eS26 forms a stable complex with Tsr2 . After Tsr2:eS26 complex formation , Tsr2 guarantees a safe transfer of eS26 to the 90S pre-ribosome . Although RanGTP is able to inefficiently release eS26 from its importin , failure to immediately bind Tsr2 results in eS26 degradation . Therefore , in absence of Tsr2 , only a smaller fraction of eS26 may reach the 90S pre-ribosome , providing a possible explanation as to why Tsr2-deficient cells are severely impaired in growth but are still viable , although the r-protein eS26 is essential . Notably , human Tsr2 can rescue the severe growth defect of the Tsr2-depleted strain ( Figure 1—figure supplement 1C ) , strongly suggesting an evolutionarily conserved role of Tsr2 in 40S assembly . 10 . 7554/eLife . 03473 . 015Figure 7 . A model for the transport of eS26 to the 90S pre-ribosome . Newly synthesized eS26 is transported from the cytoplasm into the nucleus by importins . In the nucleus , Tsr2 alone removes eS26 from importins by a RanGTP-independent mechanism . Subsequently , Tsr2 binds the released eS26 , protects it from proteolysis and aggregation , and enables safe transfer to the 90S pre-ribosome . If eS26 is released from the importin by RanGTP it may not immediately encounter Tsr2 , resulting in a smaller fraction reaching the 90S pre-ribosome . See ‘Discussion’ for details of the proposed model . DOI: http://dx . doi . org/10 . 7554/eLife . 03473 . 015 Similar to the Tsr2-depletion , both DBA mutants ( eS26D33N and eS26C77W ) accumulate 20S pre-rRNA in the cytoplasm ( Figure 6C ) . The eS26C77W mutant interacted poorly with its import receptors , suggesting that the inability to interact with importins may cause its degradation . Cysteine 77 is one of four conserved cysteines within eS26 that coordinates a Zn2+ ion ( Figure 4—figure supplement 3A , right panel; Rabl et al . , 2011 ) . Our data raise an intriguing possibility that the NLS within eS26 becomes available to interact with importins only when the Zn2+ ion is correctly coordinated . In addition to their transport role , importins may select correctly folded eS26 . Notably , the eS26D33N mutant interacted with Kap123 , Kap104 , Pse1 and Tsr2 in vitro ( Figure 6D , E ) . We speculate that the in vivo instability of this variant might be due to a failure to incorporate eS26 into the 90S pre-ribosome . Several mutations in eS26 have been linked to DBA , the majority of which are in the start codon , thereby causing eS26 haploinsufficiency ( Doherty et al . , 2010 ) . Notably , eS26 levels are strongly reduced in Tsr2-depleted cells . Interestingly , about half of DBA cases are due to unidentified mutations . Based on these data , we speculate that the TSR2 gene may be a potential hotspot for DBA . More than 20 years ago , a system was envisioned to efficiently transfer r-proteins from the NPCs towards the nucleolus ( Russell and Tollervey , 1992 ) . Here , we identify Tsr2 as the first component of this transfer system that connects the nuclear import machinery with the ribosome assembly pathway . We propose the term ‘escortin’ to describe this ‘linker’ function . Aggregating r-proteins in the nucleolus aggravate the toxicity of a Caenorhabditis elegans Huntington disease model and decrease their lifespan David et al . ( 2010 ) , emphasizing the importance to safely transfer r-proteins to the assembling pre-ribosomes . Due to their unstable and aggregation-prone nature Koplin et al . ( 2010 ) we envisage an escortin network to securely connect the nuclear import machinery with the ribosome assembly pathway . Intriguingly , like in the case of Kap123:eS26 complex , RanGTP is unable to efficiently release uL14 ( human Rpl23a ) from importin 7 ( RanBP7 ) ( Jäkel and Görlich , 1998 ) . Moreover , we found that yeast r-proteins ( uS14 , eS31a and eS8a ) bound to Kap123 were not released upon RanGTP treatment ( Figure 4—figure supplement 4 ) suggesting that these r-proteins may require specific escortins for their release . Affinity purifications coupled to mass spectrometry have identified >200 non-ribosomal factors that are directly involved in ribosome assembly ( Bassler et al . , 2001; Harnpicharnchai et al . , 2001; Dragon et al . , 2002; Fatica et al . , 2002; Grandi et al . , 2002; Nissan et al . , 2002; Schäfer et al . , 2003 ) . However , escortins , which are not stably bound to pre-ribosomal particles , may have escaped identification . Individual subunits/sub-complexes of other macromolecular complexes involved in genome replication , genomic stability and gene expression must be imported into the nucleus prior to their assembly . The fate of these cargoes after being released from importins in the nucleus remains largely unexplored . Many of these components may rely on escortins that will ensure their transfer to their assembly site . Thus , we expect that the list of escortins for ribosome assembly and other biological pathways will expand in the near future .
The Saccharomyces cerevisiae strains used in this study are listed in Supplementary file 1A . Genomic disruptions , C-terminal tagging and promoter switches at genomic loci were performed as described previously ( Longtine et al . , 1998; Puig et al . , 2001; Janke et al . , 2004 ) . Preparation of media , yeast transformations and genetic manipulations were performed according to established procedures . Plasmids used in this study are listed in Supplementary file 1B . Details of plasmid construction will be provided upon request . All recombinant DNA techniques were performed according to established procedures using E . coli XL1 blue cells for cloning and plasmid propagation . Point mutations in RPS26A were generated using the QuikChange site-directed mutagenesis kit ( Agilent Technologies , Santa Clara , CA , USA ) . All cloned DNA fragments and mutagenized plasmids were verified by sequencing . Localization of 20S pre-rRNA was analyzed using a Cy3-labeled oligonucleotide probe ( 5′-Cy3-ATG CTC TTG CCA AAA CAA AAA AAT CCA TTT TCA AAA TTA TTA AAT TTC TT-3′ ) that is complementary to the 5′ portion of ITS1 as previously described ( Faza et al . , 2012 ) . Pre-40S subunit export , monitored by localization of uS5-GFP and localization of GFP-eS26 was performed as previously described ( Faza et al . , 2012; Altvater et al . , 2014 ) . Indirect immunofluorescence using affinity-purified polyclonal antibodies against the TAP-tag ( 1:1000; Thermo Scientific; Rockford , IL , USA ) and staining of the nuclear and mitochondrial DNA with DAPI was performed as described previously ( Schlenstedt et al . , 1997; Solsbacher et al . , 1998 ) . Cells were visualized using DM6000B microscope ( Leica , Germany ) equipped with HCX PL Fluotar 63 × /1 . 25 NA oil immersion objective ( Leica , Solms , Germany ) . Images were acquired with a fitted digital camera ( ORCA-ER; Hamamatsu Photonics , Hamamatsu , SZK , Japan ) and Openlab software ( Perkin–Elmer , Waltham , MA , USA ) . Sedimentation analysis of yeast lysates by sucrose gradient ultracentrifugation was performed as described previously ( Kemmler et al . , 2009; Altvater et al . , 2014 ) . For Western analyses , 500 μl fractions were precipitated by TCA ( trichloroacetic acid ) , washed in acetone , resuspended in 100 μl of onefold SDS sample buffer and separated by SDS-PAGE . Tsr2 , eS26 and uL3 were detected by Western analyses . For rRNA analysis , 500 µl fractions were collected and diluted with an equal volume of lysis buffer . RNA was extracted with Phenol-Chlorofom-Isoamylalcohol and precipitated in isopropanol . RNA pellets were washed with 80% ethanol and resuspended in 20 µl water . rRNAs were then separated on a 1 . 2% Agarose/formaldehyde gel for 1 . 5 hr at 200 V . For Northern analysis , rRNAs were blotted onto a Hybond-XL ( Amersham , Pittsburg , PA , USA ) membrane by capillary transfer and probed for 18S ( 5′-CATGCATGGCTTAATCTTTGAGAC ) , 20S ( 5′-GGTTTTAATTGTCCTATAACAAAAGC ) and 25S rRNA ( 5′-TGCCGCTTCACTCGCCGTTAC ) using radioactively labeled probes . rRNAs were detected using phosphoimaging screens ( GE Healthcare , Pittsburg , PA , USA ) . Whole cell extracts were prepared by alkaline lysis of yeast cells as previously described ( Kemmler et al . , 2009 ) . Tandem affinity purifications ( TAP ) of pre-ribosomal particles were carried out as previously described ( Faza et al . , 2012; Altvater et al . , 2014 ) . Calmodulin-eluates were separated on NuPAGE 4–12% Bis-Tris gradient gels ( Invitrogen , Carlsbad , CA , USA ) and visualized by either Silver staining or Western analyses using indicated antibodies . To analyze RNAs after TAP purification , RNA was extracted with Phenol-Chlorofom-Isoamylalcohol from Calmodulin-eluates and precipitated in isopropanol . RNA pellets were washed with 80% ethanol and finally resuspended in 20 µl water . 1 µg of total RNA was separated on a 1 . 2% Agarose/formaldehyde gel for 1 . 5 hr at 200 V . Western analyses were performed as previously described ( Kemmler et al . , 2009 ) . The following primary antibodies were used in this study: α-Tsr2/S26 ( 1:3000; this study ) , α-Arc1 ( 1:4000; E Hurt , University of Heidelberg , Heidelberg , Germany ) , α-uL3 ( yeast Rpl3 ) ( 1:5000; J Warner , Albert Einstein College of Medicine , Bronx , NY , USA ) , α-uS7 ( yeast Rps5 ) ( 1:4000; Proteintech Group Inc . , Chicago , IL , USA ) , α-uS3 ( yeast Rps3 ) ( 1:3000; M Seedorf , University of Heidelberg , Heidelberg , Germany ) ; α-TAP ( CBP ) ( 1:4000; Thermo Scientific , Rockford , IL , USA ) , α-Pno1 ( 1:10 , 000; K Karbstein , Scripps Research Institute , Jupiter , FL , USA ) , α-Dim1 ( 1:10 , 000; K Karbstein , Scripps Research Institute , Jupiter , FL , USA ) , α-Nob1 ( 1:500; Proteintech Group Inc . , Chicago , IL , USA ) , α-Tsr1 ( 1:10 , 000; K Karbstein , Scripps Research Institute , Jupiter , FL , USA ) , α-Ltv1 ( 1:5000; K Karbstein , Scripps Research Institute , Jupiter , FL , USA ) , α-Rio2 ( 1:1000; Proteintech Group Inc . , Chicago , IL , USA ) , α-FLAG ( 1:3000; Sigma-Aldrich , St . Louis , MO , USA ) . The secondary HRP-conjugated α-rabbit and α-mouse antibodies ( Sigma-Aldrich , USA ) were used at 1:1000-1:5000 dilutions . Protein signals were visualized using Immun-Star HRP chemiluminescence kit ( Bio-Rad Laboratories , Hercules , CA , USA ) and captured by Fuji Super RX X-ray films ( Fujifilm , Tokyo , Japan ) . All recombinant proteins were expressed in E . coli BL21 cells by IPTG induction . His6-tagged proteins were affinity purified in 50 mM Hepes pH 7 . 5 , 50 mM NaCl , 10% glycerol using Ni-NTA Agarose ( GE healthcare ) , GST fusion proteins were purified in PBSKMT ( 150 mM NaCl , 25 mM sodium phosphate , 3 mM KCl , 1 mM MgCl2 , 0 . 1% Tween , pH 7 . 3 ) using Glutathione Sepharose ( GE healthcare ) . GST-tagged importins , His6-taggged importins and RanGTP ( His6-Gsp1Q71L-GTP ) were expressed and purified as previously described ( Solsbacher et al . , 1998; Maurer et al . , 2001; Fries et al . , 2007 ) . Recombinant GST-Tsr2 was immobilized in PBSKMT on Glutathione Sepharose ( GE healthcare ) , and incubated with E . coli lysates containing recombinant eS26 , eS26FLAG , eS26D33NFLAG , eS26C77WFLAG for 1 hr at 4°C . After incubation , the immobilized GST-proteins were washed three times with PBSKMT 4°C . The bound proteins were eluted with LDS . The in vitro binding studies between recombinant eS26FLAG , eS26D33NFLAG , eS26C77WFLAG , Tsr2 , Tsr2:eS26 complex and yeast importins as GST-fusion proteins were performed as previously described ( Solsbacher et al . , 1998 ) . 1/5th of the bound proteins and input ( eS26 , eS26FLAG , eS26D33NFLAG , eS26C77WFLAG ) were analyzed on a Coomassie Blue stained gel . 1/10th of the bound proteins and 1/1000th of the input was used for Western analyses . To dissociate the GST-importin:eS26FLAG ( Kap123 , Pse1 and Kap104 ) complex or GST-Kap123:eS31FLAG , GST-Kap123:eS8FLAG , GST-Kap123:eS14FLAG complexes pre-immobilized GST-importin:ribosomal protein complexes were incubated with buffer alone or 3 nM of 3′-end of 18S rRNA ( only for eS26FLAG ) , 1 . 5 µM Tsr2 , 1 . 5 µM His6-Tsr2 ( only eS26FLAG ) and/or 1 . 5 µM RanGTP ( His6-Gsp1Q71L-GTP ) for 1 hr at 4°C ( protocol modified from Rothenbusch et al . , 2012 ) . To show that eS26 stably associated with Tsr2 after release from importins , the supernatant of the samples with buffer alone and His6-Tsr2 were incubated with Ni-NTA Agarose for 1 hr at 4°C . For dissociation kinetics , 1 . 5 µM RanGTP ( His6-Gsp1Q71L-GTP ) or Tsr2 were added to pre-immobilized importin:eS26FLAG complexes and samples were withdrawn at 1 , 2 , 4 and 8 min . Bound proteins were eluted in twofold LDS/SDS-sample buffer by incubating at 70–95°C and separated by SDS-PAGE . Proteins were visualized by Coomassie Blue staining or by Western analyses using antibodies against Tsr2 and eS26 . The aggregation assay was performed in a 384-well plate ( Polystyrene , clear bottom , low volume , Corning , USA ) . In each well 33 µM GST-eS26 and a given concentration of Tsr2 ( 0 up to 266 µM ) in PBSKMT was pre-incubated for 1 hr at 4°C ( final volume: 90 µl ) . 250 nM of PreScission protease was added to initiate aggregation . Aggregation of free eS26 was measured at 450 nm using a Multiskan GO plate reader ( Thermo Scientific , USA ) . As controls , scattering intensities of individual components used in the aggregation assay such as 33 µM of GST-eS26 alone , 266 µM of Tsr2 alone , PreScission protease and buffer were measured . Four replicates were performed for each sample measured . To develop SRM assays , peptide samples of the affinity purified pre-40S particles were analyzed on a nanoLC 1Dplus system ( Eksigent ) connected to a TripleTOF 5600+ mass spectrometer ( ABSciex ) . Peptides were separated by reversed-phase liquid chromatography on a 20-cm fused silica microcapillary ( 75 µm inner diameter , New Objective ) packed in-house with 3 µm C18 beads ( Magic C18 AQ , 200 Å pore size; Michrom BioResources , Auburn , CA , USA ) with a linear gradient from 98% solvent A ( 98% acetonitrile , 0 . 1% formic ) and 2% solvent B ( 98% acetonitrile , 0 . 1% formic acid ) to 35% solvent B over 120 min at a flow rate of 300 nl/min . The mass spectrometer was operated in information-dependent acquisition ( IDA ) mode . MS1 spectra were recorded in the range of 360–1460 m/z for 500 ms . Up to 20 precursor ions with charge state 2–5 were selected for fragmentation and MS2 spectra were recorded in the range of 50–2000 m/z for 150 ms in high sensitivity mode . Selected precursor ions were excluded for 20 s after one occurrence . Raw data files were centroided and converted to mzML format using the ABSciex Data Converter and then converted to mzXML format using ProteoWizard MSConvert ( Kessner et al . , 2008 ) . MS2 spectra were searched with Sorcerer-SEQUEST ( SageN Research ) against a S . cerevisiae protein database ( SGD , May 2013 ) to which the sequences of the 11 spiked-in iRT peptides and various common contaminants were added . Reversed sequences of all proteins were appended to the protein database to assess the number of false positive peptide-spectrum matches ( Elias and Gygi , 2007 ) . Tryptic cleavage was defined to occur after lysine and arginine , unless followed by a proline residue , and peptides were allowed to have up to one non-tryptic end and up to two missed cleavages . Cysteine carbamidomethylation was added as static modification and methionine oxidation as variable modification . Precursor mass tolerance was set to 50 ppm . Resulting peptide-spectrum matches were statistically assessed using PeptideProphet and iProphet as part of the TPP ( Keller et al . , 2002; Deutsch et al . , 2010; Shteynberg et al . , 2011 ) . The iProphet output was processed with MAYU ( Reiter et al . , 2009 ) , which has been modified to work with iProphet probabilities . Peptide-spectrum matches were selected at a false discovery rate ( FDR ) of 0 . 07% to obtain a protein FDR of 1% . An in-house written script was used to convert all retention times into iRT values ( Escher et al . , 2012 ) . SpectraST ( Lam et al . , 2008 ) was used to generate a consensus spectral library from which the six most intense fragment ions ( b- or y-ions ) per peptide precursor were selected in Skyline ( MacLean et al . , 2010 ) . The final SRM assays for target proteins and iRT peptides are given in Supplementary file 2 . | The production of a protein in a cell starts with a region of DNA being transcribed to produce a molecule of messenger RNA . A large molecular machine called ribosome then reads the information in the messenger RNA molecule to produce a protein . Ribosomes themselves are made of RNA and several different proteins called r-proteins . The construction of a ribosome starts with the assembly of a pre-ribosome inside the cell nucleus , and the ribosome is completed in the cytosol of the cell . A yeast cell will divide about 30 times during its lifetime , and before each division event a single yeast cell needs to import about 14 million r-proteins into its nucleus in order to make about 200 , 000 ribosomes . However , many details of this process are mysterious . In particular , many r-proteins are known to be unstable: meaning that , left to their own devices , r-proteins are highly likely to aggregate , which would prevent them becoming part of a ribosome . Now , Schütz et al . have figured out how a carrier protein called Tsr2 makes sure that an r-protein called eS26 does indeed become part of a ribosome . The human disorder known as Diamond-Blackfan anemia is caused by a mutation in the gene for eS26 . The eS26 proteins are ferried to the cell nucleus on specialized transport vehicles . Schütz et al . have now shown that the Tsr2 carrier protein unloads the r-protein from the transport vehicle in the nucleus , and then binds it . This means that the r-protein does not form an aggregate . Finally , the Tsr2 carrier protein transfers the r-protein to the pre-ribosome . This is the first time that a carrier protein that unloads an r-protein cargo from its transport vehicle , to ensure safe delivery to the pre-ribosome , has been identified . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2014 | A RanGTP-independent mechanism allows ribosomal protein nuclear import for ribosome assembly |
Bitter compounds elicit an aversive response . In Drosophila , bitter-sensitive taste neurons coexpress many members of the Gr family of taste receptors . However , the molecular logic of bitter signaling is unknown . We used an in vivo expression approach to analyze the logic of bitter taste signaling . Ectopic or overexpression of bitter Grs increased endogenous responses or conferred novel responses . Surprisingly , expression of Grs also suppressed many endogenous bitter responses . Conversely , deletion of an endogenous Gr led to novel responses . Expression of individual Grs conferred strikingly different effects in different neurons . The results support a model in which bitter Grs interact , exhibiting competition , inhibition , or activation . The results have broad implications for the problem of how taste systems evolve to detect new environmental dangers .
The sense of taste allows animals to identify food sources that are nutritious but not toxic ( Liman et al . , 2014 ) . Many toxins elicit a bitter taste and an aversive response that is widely conserved across phyla . Bitter-tasting compounds are of diverse structures and chemical classes . Many bitter compounds are synthesized by plants and deter insects from feeding on them ( Pontes et al . , 2014; Briscoe et al . , 2013; Salloum et al . , 2011; Sellier et al . , 2011; Freeman and Dahanukar , 2015 ) . Insects in turn have evolved gustatory systems that detect these compounds and signal the danger of toxicity ( Wada-Katsumata et al . , 2013 ) . The principal taste organ of the Drosophila head is the labellum ( Figure 1 ) . The labellum contains ~31 taste sensilla that fall into morphological classes based on length: short ( S; green and blue in Figure 1 ) , intermediate ( I; purple and red ) , and long ( L; gray in Figure 1 ) ( Stocker , 1994 ) . Each sensillum has a pore at its tip . When a sensillum makes contact with a potential food source , compounds from the food source diffuse through the pore and activate gustatory receptor neurons inside . 10 . 7554/eLife . 11181 . 003Figure 1 . Sensillum types on the labellum . Left: The four bitter-responsive types , S-a ( green ) , S-b ( blue ) , I-a ( purple ) and I-b ( red ) , are differently distributed on the labellar surface . L sensilla ( gray ) show little if any response to bitter compounds . Right: S-a and S-b have four gustatory receptor neurons ( GRNs ) , one of which is bitter-responsive , and I-a and I-b sensilla have two GRNs , one of which is bitter-responsive . The bitter-responsive neuron ( B ) of each sensillum type expresses a different combination of Grs . Five Grs , referred to as 'Commonly Expressed Receptors' ( CERs; in rectangles ) , are expressed in every bitter neuron on the labellum . Many or all sensilla also contain a sugar-sensitive neuron ( S ) . Mapping of Grs to neurons is based on GAL4 driver expression . Figure adapted from Weiss et al . ( 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 003 Electrophysiological analysis of all 31 sensilla with a panel of 16 bitter compounds revealed that bitter compounds elicit different responses from different sensilla ( Weiss et al . , 2011 ) . Four functional classes of bitter-sensitive sensilla were identified , two of short length , S-a and S-b , and two of intermediate length , I-a and I-b . Each class contains a neuron excited by bitter compounds , referred to as a 'bitter neuron , ' with a distinguishable response spectrum . Expression analysis of all 68 members of the Gustatory receptor ( Gr ) family identified four corresponding classes of bitter neurons , each expressing a distinct subset of receptors ( Figure 1 ) ( Weiss et al . , 2011 ) . The receptors expressed in bitter neurons are referred to for convenience as 'bitter receptors . ' They show extensive coexpression: 29 are coexpressed in the bitter neuron of S-a; 16 in S-b; 6 in I-a; and 10 in I-b . In addition to the labellum , the legs and pharynx of the fly also have gustatory function , as does the larval head ( Weiss et al . , 2011; Kwon et al . , 2011; LeDue et al . , 2015; Ling et al . , 2014; Meunier et al . , 2003; Oppliger et al . , 2000 ) . We note that expression analysis of Grs has been based primarily on Gr-GAL4 drivers , since success in analyzing Gr expression with in situ hybridization has been extremely limited . Loss-of-function studies have clearly shown that some Grs are required for the responses to certain bitter compounds . For example , Gr93a is required for the behavioral and physiological response to caffeine; Gr8a is required for response to L-canavanine; Gr66a , Gr33a and Gr32a are broadly required for the response to several bitter tastants ( Lee et al . , 2009; 2010; 2012; Moon et al . , 2006; Moon et al . , 2009 ) . However , the roles of individual Grs in bitter response have been difficult to discern in detail , in part because of the coexpression of bitter receptors and in part because of the difficulty of expressing bitter receptors in conventional expression systems such as cultured cells or oocytes ( Liman et al . , 2014 ) . In vivo expression studies provide a complementary approach for analysis of Gr function . Here , we analyze eight bitter Grs , most of which have not been functionally studied before . We express them in a natural laboratory: in bitter neurons of the labellum that either do or do not express them endogenously , which we refer to as overexpression or ectopic expression , respectively . Specifically , we express individual Grs in four different bitter neurons , three of wild type and one of a Gr mutant , and we measure the effects of their expression using electrophysiology and a panel of 21 bitter tastants . From an analysis of ~600 receptor-neuron-tastant combinations ( n≥27 , 000 total recordings ) we find several surprising results . While expression of Grs led to increases in many responses , expression of some Grs led to decreases in certain responses . Conversely , the deletion of one Gr from a neuron led to novel responses not observed in wild type . Overexpression of a Gr in some neurons led not only to increased responses but also to novel responses . A recurrent theme was that the expression of the same Gr in different neurons led to strikingly different results . Taken together , our findings provide support for a model in which bitter Grs interact , exhibiting competition , inhibition , or activation in different contexts . The results may have profound evolutionary implications: they suggest a rich source of means by which the taste system can evolve novel , increased , or decreased responses to new environmental opportunities and dangers , or modulate its response to accommodate changes in the internal state of the fly .
To investigate the function of bitter Grs , we ectopically expressed them in bitter neurons of the labellum via the GAL4-UAS system ( Brand and Perrimon , 1993 ) . We then measured the effects of this expression via electrophysiology ( Benton and Dahanukar , 2011; Delventhal et al . , 2014 ) . The aim of this approach was to study bitter Grs in an environment similar to their native environment—that is , in bitter taste neurons , in taste sensilla , in vivo . We initially expressed Grs in bitter neurons of I-a sensilla ( Figure 1 , purple sensilla ) . These neurons were selected because they have the narrowest response spectrum of the various classes of labellar bitter neurons ( Weiss et al . , 2011 ) . These sensilla also express the smallest number of bitter Gr genes , as determined in a systematic Gr-GAL4 expression analysis ( Figure 1 ) . As an initial Gr gene we examined Gr22b , which in wild-type flies is expressed in S-a sensilla , but not I-a sensilla . We used Gr89a-GAL4 , which drives expression in all labellar bitter neurons , and then measured responses of I-a sensilla to a broad panel of bitter tastants . We found that ectopic expression of Gr22b conferred an electrophysiological response to cucurbitacin ( CUC ) that is not observed in either of the parental control lines , GAL89a-GAL4; + or +/UAS-Gr22b ( Figure 2a , b ) . CUC is a plant-derived compound that has insecticidal activity and tastes bitter to humans ( Torkey et al . , 2009 ) . Expression of Gr22b also conferred strong responses to sucrose octaacetate ( SOA ) , another bitter-tasting compound ( Harwood et al . , 2012 ) , and to azadirachtin ( AZA ) , which is found in the seeds of the neem tree and which inhibits feeding of locusts and many other insects ( Table 1 ) ( Schmutterer , 1990; Sharma et al . , 1993 ) . 10 . 7554/eLife . 11181 . 004Figure 2 . Ectopic expression of Gr22b in I-a bitter neurons leads to novel responses to three bitter compounds . ( a ) Sample electrophysiological responses from I-a sensilla of parental controls and of flies ectopically expressing UAS-Gr22b . ( b ) Mean responses . Asterisks indicate responses that are different from both parental controls as measured by a two-way ANOVA , with Bonferroni multiple comparisons correction ( p ≤ 0 . 0001 , n ≥ 20 ) . Only compounds that do not elicit responses greater than that from a TCC control solution in wild type I-a sensilla , as measured by a one-way ANOVA , are shown . ( c ) Responses of I-a sensilla to SOA in Gr22b-expressing flies , across a range of concentrations , relative to both parental controls ( * indicates p ≤ 0 . 0001 , n ≥ 17 ) . Concentrations are graphed on a logarithmic scale . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 00410 . 7554/eLife . 11181 . 005Table 1 . Panel of 21 bitter taste compounds tested in electrophysiological recordings . a '−' indicates that insecticidal activity has not been described . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 005TastantAbbreviationConcentrationChemical classSourceInsecticidal activityaAristolochic acidARI1 mMphenanthreneAristolochia family of plants−AzadirachtinAZA1 mMterpenoidNeem tree+Berberine chlorideBER1 mMalkaloidGolden seal , bayberry , Oregon grape and goldthread−CaffeineCAF1 mMalkaloidCoffee , chocolate , tea , kola nut−CoumarinCOU10 mMbenzopyroneTonka bean , honey clover+Cucurbitacin I hydrateCUC1 mMglycosidePumpkins , gourds , cucumbers+N , N-Diethyl-m-toluamideDEET10 mMN , N-dialkylamidesynthetic+Denatonium benzoateDEN10 mMquaternary ammonium cationsynthetic−EscinESC1 mMterpenoidHorse chestnut tree−GossypolGOS1 mMterpenoidCotton+ ( - ) -lobeline HClLOB1 mMalkaloidIndian tobacco , Cardinal flower+MyricetinMYR1 mMflavonoidBerries , wine−QuinineQUI1mMalkaloidCinchona tree bark−RotenoneROT1 mMketoneJicama+SaponinSAP1%terpenoidSoapbark tree+D- ( + ) -sucrose octaacetateSOA1 mMacetylated sucrose derivativesynthetic+Sparteine sulfate saltSPS10 mMalkaloidScotch broom+Strychnine nitrate saltSTR10 mMalkaloidStrychnos seeds+TheobromineTHE1 mMalkaloidCacao , tea , kola nut , chocolate−TheophyllineTPH10 mMalkaloidTea leaves+UmbelliferoneUMB1 mMphenylpropanoidCarrot , coriander− To confirm and extend these results , we tested responses to SOA across a range of concentrations ( Figure 2c ) . Responses conferred by Gr22b increased above 0 . 1 mM and appeared to saturate at 20 spikes/s . Neither of the parental control lines responded at any concentration . Taken together , these results indicate that expression of Gr22b confers novel responses in a neuron that does not normally express it . We note that CUC , SOA , and AZA all elicit responses from wild type S-a sensilla , which express Gr22b endogenously . Next , we asked whether a number of other Grs might also confer responses to I-a , and if so , whether the responses differed . Six additional Grs were chosen to analyze in detail , including Grs expressed in the I-b , S-a , and S-b classes of bitter-sensitive labellar sensilla , in bitter-sensitive leg sensilla , in the pharynx , in the larva , and in the adult antenna ( Table 2 ) ( Weiss et al . , 2011; Kwon et al . , 2011; 2014; Scott et al . , 2001; Fishilevich and Vosshall , 2005 ) . We expressed each of these receptors individually using the Gr89a-GAL4 driver and measured the responses to an expanded panel of 21 chemically diverse bitter compounds in I-a sensilla . As part of this analysis we extended our analysis of Gr22b to include the entire tastant panel . 10 . 7554/eLife . 11181 . 006Table 2 . Endogenous expression patterns of Grs selected for analysis , as determined primarily by Gr-GAL4 analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 006GeneLabellumLegsPharynxLarvaAntennaGr2a−−++−Gr10a−−−++Gr22b+ ( I-b , S-a ) +++−Gr28a+ ( I-b , S-a , S-b ) +++−Gr28b . a+ ( I-b , S-a , S-b ) +++−Gr36a+ ( S-b ) +−−−Gr58c−+−−−Gr59c+ ( I-a , S-a ) −−+− Four of these seven Grs conferred responses to three or more compounds that elicited no response from the control line , referred to henceforth as novel responses ( 'N' in Figure 3a ) . Gr22b , in addition to conferring the novel responses shown in Figure 2b , also conferred greater responses to DEN , BER , and LOB than the control ( p<0 . 0001 , two-way ANOVA , Bonferroni test , n≥20 ) . 10 . 7554/eLife . 11181 . 007Figure 3 . Electrophysiological responses of sensilla in which seven individual Grs are expressed in I-a bitter neurons . ( a ) 'N' indicates a novel response , to which there was no significant response in the wild-type control . Each experimental genotype is Gr89a-GAL4; UAS-GrX . Asterisks indicate responses that are different from the Gr89a-GAL4;+ parental control ( two-way ANOVA , with Bonferroni multiple comparisons correction . p ≤ 0 . 0001 , except that p ≤ 0 . 001 for Gr58c/COU; p<0 . 01 for Gr2a/TPH and Gr10a/TPH; p ≤ 0 . 05 for Gr58c/MYR , Gr10a/CAF , and Gr28b . a/MYR . n ≥ 7 ) . ( b ) Responses to UMB from Gr2a-expressing flies , across a range of concentrations , and relative to both parental controls ( asterisks indicate p ≤ 0 . 0001 , n ≥ 17 ) . Concentrations are graphed on a logarithmic scale . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 007 Gr58c conferred novel responses to MYR , STR , and COU , and responses to QUI , LOB , and DEN that were greater than those of the control line ( p<0 . 0001 , except that for COU p<0 . 0005 , and for MYR p<0 . 05 , n≥16 ) . We note that Gr58c is expressed in leg sensilla that respond to five of these compounds; the sixth was not tested in legs ( Ling et al . , 2014 ) . Remarkably similar response profiles were conferred by expression of Gr2a and Gr10a . Both receptors conferred responses to caffeine ( CAF ) , umbelliferone ( UMB ) , and theophylline ( TPH ) , which elicit essentially no response from the control , and an increased response to LOB ( p<0 . 0001 , except p<0 . 01 for the TPH values and p<0 . 05 for CAF in the case of Gr10a , n≥8 ) . Despite this functional similarity , Gr2a and Gr10a are distantly related phylogenetically ( Robertson et al . , 2003 ) . CAF and TPH , which are present in coffee and tea , are closely related in structure; UMB is present in carrots and coriander and is structurally distinct ( Table 1 ) . In an independent dose-response experiment , the response to UMB increased above 0 . 3 mM , and appeared to saturate at a level between 10 spikes/s and 15 spikes/s ( Figure 3b; limited solubility of UMB precluded testing at higher concentrations ) . Neither parental control responded to any tested concentration of UMB . We note that saturation of the response to UMB conferred by Gr2a was lower than that for the response to SOA conferred by Gr22b ( Figure 2c ) ; the SOA response also appeared to have a lower threshold and to saturate at a lower concentration than the response to UMB . When Gr28b . a was expressed , we observed a modest response to MYR that was not observed in the control line ( p<0 . 05 , n≥26 ) , as well as an increased response to LOB ( p<0 . 0001 , n≥26 ) . Gr28a produced an increase in response to LOB ( p<0 . 0001 , n≥24 ) . Gr36a produced no changes in the response profile of I-a . Gr36a could require a co-factor not present in I-a , or could function in response to a tastant not included in the panel . We asked whether the three receptors that conferred the most modest effects , or no effects , to I-a sensilla , i . e . Gr28b . a , Gr28a , and Gr36a , behaved similarly in other sensilla . Again using the Gr89a-GAL4 driver , we measured the effect of expressing them on the responses of the bitter neurons of I-b and S-a sensilla . Expression of Gr28b . a in I-b sensilla , where it is expressed endogenously , conferred an increased response to aristolochic acid ( ARI ) , as well as to berberine ( BER ) ( Figure 4a ) . The increase to ARI was confirmed in an independent dose-response analysis: responses in I-b sensilla of the Gr89a-GAL4; UAS-Gr28b . a line were greater than in either parental control at the two higher concentrations tested ( Figure 4b; higher concentrations were not tested due to limited solubility ) . At the highest concentration , the response of Gr89a-GAL4; UAS-Gr28b . a approached 25 spikes/s . 10 . 7554/eLife . 11181 . 008Figure 4 . Electrophysiological responses of sensilla in which three individual Grs ( Gr28b . a , Gr28a , Gr36a ) are expressed in I-b ( a ) and S-a ( c ) bitter neurons . Tastant order and x-axis scales differ between panels a and c for clarity of presentation . The experimental genotypes were Gr89a-GAL4; UAS-GrX . ( a ) In I-b sensilla , Gr28b . a conferred an increased response to ARI ( p ≤ 0 . 0001 , n ≥ 10 ) and BER ( p ≤ 0 . 01 , n ≥ 10 ) relative to the GAL4 parental control line . Gr28a conferred response to SAP ( p ≤ 0 . 001 , n ≥ 13 ) and TPH ( p ≤ 0 . 05 , n ≥ 13 ) relative to the GAL4 parental control line . Gr36a conferred no increased responses ( n ≥ 6 ) . ( b ) A dose-response analysis using both parental controls revealed increases in ARI response in I-b sensilla ( * indicates p ≤ 0 . 05 , n ≥ 22 ) . Concentrations are graphed on a logarithmic scale . ( c ) In S-a sensilla , Gr28a conferred decreased responses ( ARI: p ≤ 0 . 0001 , UMB: p ≤ 0 . 001 , DEN and BER: p ≤ 0 . 02 . n ≥ 11 ) , while Gr28b . a and Gr36a did not ( n ≥ 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 008 Expression of Gr28a in I-b sensilla conferred a novel response to saponin ( SAP ) not observed in the GAL4 control line ( Figure 4a ) ; the response was not observed in the UAS-Gr28a parental control either ( 1 spike/s ± 1 , n=11 ) . The production of this novel response was surprising , since Gr28a is expressed endogenously in I-b sensilla . Gr36a conferred no response to I-b , as in I-a . Thus the effects of expression of Gr28b . a and Gr28a in I-b differ from those in I-a . In contrast to I-b , no responses to ARI or SAP were conferred by expression of Gr28b . a or Gr28a , respectively , in I-a sensilla . Conversely , the increases in response to MYR and LOB that were observed in I-a were not observed in I-b . The simplest interpretation of these results , taken together , is that some Grs have distinct functions in different neuronal contexts . We also expressed the three Grs in S-a sensilla . No increases in response were observed to any tastant , with any receptor . These results lend further support to the conclusion that the effects of Gr expression vary in different neuronal contexts . However , a striking effect was observed in the response to DEN , UMB , ARI , and BER upon expression of UAS-Gr28a in S-a: responses were lower than in the parental control line ( Figure 4c ) . A similar finding is presented in the next section , where it is considered in more detail . We asked whether the Grs that conferred responses to three or more compounds in I-a ( Gr22b , Gr58c , Gr2a and Gr10a ) produced similar effects in other sensilla . We first tested the effects of Gr22b expression in S-a sensilla . As in I-a , Gr22b expression conferred an increased response to SOA and CUC ( Figure 5a ) . The Gr89a-GAL4; UAS-Gr22b line also responded strongly to AZA , as in I-a , but the responses of the control lines were equally strong . 10 . 7554/eLife . 11181 . 009Figure 5 . Electrophysiological responses of sensilla in which four individual Grs are expressed in S-a and I-b bitter neurons . Each experimental genotype is Gr89a-GAL4; UAS-GrX . ( a ) Response profiles of both parental controls and flies expressing UAS-Gr22b in S-a bitter neurons . Asterisks indicate responses that are different from both parental controls ( two-way ANOVA , with Bonferroni multiple comparisons correction; p ≤ 0 . 0001 , except BER: p ≤ 0 . 0005 . n ≥ 7 ) . Response profiles generated by expression of Gr22b , Gr58c , Gr2a , Gr10a in S-a bitter neurons ( b ) and I-b bitter neurons ( d ) . Asterisks indicate responses that are different from the parental control ( two-way ANOVA , with Bonferroni multiple comparisons correction; p ≤ 0 . 0001 , except that for S-a sensilla , p≤0 . 001 for Gr2a/SPS , Gr2a/LOB , Gr10a/DEN , Gr10a/LOB; p ≤ 0 . 01 for Gr2a/DEN , Gr2a/AZA , Gr58c/ARI; p≤0 . 03 for Gr10a/SAP , Gr10a/SPS , and Gr58c/UMB . For I-b sensilla , p ≤ 0 . 001 for Gr22b/TPH; p≤0 . 01 for Gr22b/THE , Gr22b/CAF , Gr58c/LOB , and Gr58c/TPH; p ≤ 0 . 05 for Gr58c/CAF and Gr58c/UMB . n ≥ 6 ) . ( c ) Expression of Gr2a and Gr22b in S-a bitter neurons conferred suppression of the endogenous response to SPS . Asterisks indicate responses that are different from both parental controls ( p ≤ 0 . 002 , n ≥ 18 ) . Concentrations are graphed on a logarithmic scale . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 009 Dramatic reductions , however , were produced by Gr22b expression in the responses of S-a to a number of tastants , such as DEN ( Figure 5a ) . This decreased DEN response in S-a is in contrast to the increased DEN response in I-a produced by expression of the same receptor ( Figure 3a ) . This suppression is particularly interesting because Gr22b is expressed endogenously in S-a sensilla; thus suppression of response is observed in a cell that has been engineered to overexpress one of its endogenous receptors . Gr58c also suppressed the responses of S-a sensilla , in this case to UMB and ARI ( Figure 5b ) . Gr58c did not suppress the response to DEN; thus Gr58c and Gr22b both suppress responses to bitter compounds , but different compounds . Gr2a reduced responses in S-a to several tastants , including DEN , SPS , STR , and LOB . Gr10a also showed suppressive effects in S-a . Moreover , the profiles produced by Gr2a and Gr10a were very similar to each other ( Figure 5b ) , as in I-a ( Figure 3a ) . We confirmed the suppression of S-a responses via an independent dose-response analysis ( Figure 5c ) . At all concentrations tested , responses to SPS were lower in the line expressing Gr2a than in the parental control lines . Suppression of SPS response by Gr22b was observed at both 3 mM and 10 mM concentrations . Suppression of responses was observed not only in S-a sensilla but also in I-b sensilla ( Figure 5d ) . Although Gr22b expression conferred novel responses to SOA , CUC and AZA , as it did in I-a sensilla ( Figure 3a ) , it suppressed the responses to THE , CAF , TPH , and UMB . The ability of Gr22b to confer entirely novel responses and suppress endogenous ones in I-b sensilla is especially surprising given that Gr22b is endogenously expressed in I-b bitter neurons . Likewise , Gr58c showed reduced responses to CAF , TPH , and UMB , while conferring response to DEN and LOB , two compounds to which Gr58c expression also confers responses in I-a sensilla ( Figure 3a ) . Gr2a and Gr10a again behaved similarly in I-b sensilla , in that their expression had no effects . Their lack of effect in I-b sensilla is in contrast to their conferral of responses in I-a sensilla and their suppression of responses in S-a sensilla . In summary , suppression occurred broadly: in both S-a and I-b , by several different receptors , and with several different tastants . Expression of an individual receptor in a given sensillum increased the responses to some tastants and decreased the responses to others ( e . g . Gr22b; CUC and DEN in S-a ) . Expression of an individual receptor increased the response to a tastant in one sensillum type and decreased it in another ( e . g . Gr22b; DEN in I-a and S-a ) . Suppression was observed both in cases of overexpression and ectopic expression , i . e . expression of a receptor in a neuron that either does ( e . g . Gr22b in S-a ) or does not ( e . g . Gr2a in S-a ) express it endogenously in wild type . Finally , in some sensilla a receptor produced decreased responses but no increased responses ( e . g . Gr2a in S-a ) . The foregoing analysis concerned expression of Grs that are coexpressed in labellar bitter neurons with many other Grs , or that are not expressed in the labellum . We next considered Gr59c , which is the sole receptor in the I-a bitter neurons other than the commonly expressed receptors ( CERs ) ( Figure 1 ) ; it is also expressed in S-a but not I-b . We analyzed it through both in vivo expression analysis and loss-of-function analysis . First , we drove its expression in I-b , S-a , and I-a sensilla . A previous test with a limited number of tastants found that misexpression of Gr59c increased or decreased the responses to certain tastants ( Weiss et al . , 2011 ) . Using our complete panel of 21 tastants we confirmed and extended these results: we found that expression of Gr59c increased responses to LOB , DEN and BER in all three sensillum types , and suppressed many of the larger responses in I-b and S-a ( Figure 6 ) . Interestingly , the net result in each sensillum type was a profile in which LOB , DEN , and BER elicited the strongest responses . 10 . 7554/eLife . 11181 . 010Figure 6 . Electrophysiological response profiles generated by expression of Gr59c in I-b , S-a , and I-a , relative to the wild-type GAL4 parental control ( p ≤ 0 . 001 , except that p≤ 0 . 05 in the case of the response of I-b to THE and CAF . n ≥ 8 ) . The experimental genotype was Gr89a-GAL4; UAS-Gr59c . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 010 Second , we constructed a deletion of Gr59c ( △Gr59c ) ( Figure 7—figure supplement 1 ) . We hypothesized that since Gr59c expression confers a response to LOB , DEN , and BER , and since these three tastants elicit the greatest mean responses from I-a among all the tastants in the panel , that a △Gr59c I-a sensillum might display weak or no responses to all tastants of the panel . If so , moreover , it might provide a particularly useful in vivo system in which to express other Grs . As expected , deletion of Gr59c eliminated the responses to LOB , DEN , and BER ( Figure 7 ) . However , contrary to our expectations , △Gr59c I-a sensilla showed strong novel responses to THE , CAF , UMB , and TPH . Intriguingly , the response profile of the mutant I-a sensillum is similar to that of control I-b sensilla ( Figure 7 ) . Expression of a UAS-Gr59c construct in △Gr59c flies was sufficient to restore a profile similar to that of wild-type I-a sensilla , providing evidence that the mutant phenotype is in fact due to loss of Gr59c . We note that this rescue argues against the notion that the flanking Gr59d gene , which is also removed by the Gr59c deletion ( Figure 7—figure supplement 1 ) , contributes to the △Gr59c phenotype . Moreover , a GAL4-Gr59d driver shows expression in S-a bitter taste neurons but not in I-a sensilla , consistent with the interpretation that the △Gr59c phenotype observed in I-a sensilla arises from loss of Gr59c alone ( Weiss et al . , 2011 ) . 10 . 7554/eLife . 11181 . 011Figure 7 . Electrophysiological response profiles of w− Canton-S ( wCS ) control I-a sensilla , ΔGr59c mutant I-a sensilla , wCS control I-b sensilla , and ΔGr59c I-a sensilla that had been rescued with a UAS-Gr59c construct driven by Gr66a-GAL4 . Rescued ΔGr59c flies were tested with a reduced panel of 16 compounds; the other genotypes were tested with the full panel of 21 compounds . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 01110 . 7554/eLife . 11181 . 012Figure 7—figure supplement 1 . The ΔGr59c mutation was generated through FLP-FRT-mediated recombination between piggybac transposon lines f03881 and f04393 ( Parks et al . , 2004 ) . A ~17kb region of the genome was deleted; it encompassed Gr59c , as well as several other genes . This deletion was backcrossed to a wCS control background for 7 generations . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 01210 . 7554/eLife . 11181 . 013Figure 7—figure supplement 2 . Electrophysiological response profiles of wCS control S-a sensilla and ΔGr59c S-a sensilla . Response to DEN is reduced in ΔGr59c mutant S-a sensilla ( p ≤ 0 . 0001 , n ≥ 12 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 013 We asked whether △Gr59c has a similar effect on the S-a bitter neuron , which also expresses Gr59c . We found no effect , other than a decrease in the response to DEN ( Figure 7—figure supplement 2 , p<0 . 0001 , n≥12 ) . The lack of a strong mutant phenotype may reflect the much greater molecular complexity of S-a: apart from the CERs , the S-a bitter neuron expresses 24 Grs , whereas its I-a counterpart expresses only Gr59c . The difference between the △Gr59c phenotype in the two sensilla provides a further example of how the function of a bitter Gr may differ in different neuronal contexts . How might deletion of Gr59c cause an I-a bitter neuron to shift its response profile , particularly to one similar to that of an I-b neuron ? One possibility is that loss of Gr59c changes the developmental fate of I-a bitter neurons . We addressed this possibility in two ways , through a molecular analysis and through a developmental genetic analysis . We first considered the hypothesis that deletion of Gr59c induces expression of other Grs in the I-a sensillum . Our interest in this hypothesis was motivated by the finding that mutation of a green light-sensing receptor in the Drosophila visual system , Rh6 , leads to the ectopic expression of a blue-sensitive receptor , Rh5 ( Vasiliauskas et al . , 2011 ) . Likewise , deletion of a mammalian odor receptor gene leads to the expression of another odor receptor ( Lomvardas et al . , 2006; Magklara and Lomvardas , 2013 ) . We tested the expression of four Gr-GAL4 drivers , one of which is expressed in wild-type I-a sensilla but not I-b sensilla ( Gr59c-GAL4 ) , and three of which are not expressed in I-a but are expressed in I-b ( Gr28a- , Gr28b . a- and Gr22b-GAL4 ) ( Figures 1 , 8; arrowheads indicate the expected positions of cell bodies of representative I-a sensilla ) . All drivers were tested in both △Gr59c and control backgrounds . If the I-a bitter neurons underwent a complete change in cell fate to that of I-b bitter neurons , then one would expect these neurons to lose expression of Gr59c-GAL4 . This outcome was not observed ( Figure 8 , top two panels ) . In fact , expression of Gr59c-GAL4 appeared the same in the △Gr59c labellum as in the control . Likewise , none of the three drivers that are expressed in I-b were expressed in △Gr59c I-a sensilla ( Figure 8 , bottom six panels ) . These results argue against the hypothesis that I-a sensilla undergo a complete fate change switch to I-b sensilla . 10 . 7554/eLife . 11181 . 014Figure 8 . Fluorescent confocal microscopy of whole-mount labella reveals that the ΔGr59c mutation does not cause loss of Gr59c-GAL4 expression in I-a sensilla ( top two panels ) . The ΔGr59c mutation does not cause gain of Gr28b . a- , Gr28a- , or Gr22b-GAL4 expression in I-a sensilla ( bottom six panels ) . ( n ≥ 5 flies per genotype ) . White arrowheads indicate the positions of representative I-a neurons . Cell bodies of neurons that innervate I-a sensilla can be seen only in the top two panels , i . e . in Gr59c-GAL4 and ΔGr59c; Gr59c-GAL4 , as indicated by white arrowheads . Full genotypes tested: Sp/CyO; Gr-GAL4/UAS-GFP and ΔGr59c; Gr-GAL4/UAS-GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 014 We also tested the hypothesis that Gr59c plays a role during development in determining the fate of taste neurons . Using a △Gr59c mutant background , we restored Gr59c expression at different times , during and after development . Specifically , we introduced a UAS-Gr59c rescue construct under the control of the Gr89a-GAL4 driver , but we repressed the driver with a temperature-sensitive GAL80ts construct ( Lee and Luo , 1998; Zeidler et al . , 2004 ) . We increased the temperature at varying times , thereby inactivating GAL80ts , relieving the repression , and turning on Gr59c . We asked whether the I-a sensilla had a wild-type phenotype ( responses to BER , LOB , and DEN ) or a mutant phenotype ( responses to UMB , TPH , and CAF ) . The flies were examined at 7-9d , by testing the physiological responses of sensilla to a diagnostic panel of tastants . In the absence of a rescue construct , all flies showed the △Gr59c phenotype ( Figure 9a ) , as expected . In the presence of a UAS-Gr59c rescue construct with no GAL80ts construct to repress it , all flies showed a wild-type I-a phenotype , as expected of a rescued Gr59c mutant ( Figure 9b ) . In the presence of both a UAS-Gr59c rescue construct and a GAL80tsconstruct , the phenotypes of the sensilla fell into two categories depending on the temperature regime ( Figure 9c ) . When the temperature was not increased at all and repression of UAS-Gr59c was continuous , a △Gr59c phenotype was observed , as expected , demonstrating that GAL80ts successfully suppressed UAS-Gr59c expression at low temperature ( blue bar in Figure 9c ) . However , if the temperature was increased either during pupal development , at 1d after eclosion , or at 5d after eclosion , a wild-type I-a phenotype was restored in every case . Thus , rescue by the UAS-Gr59c transgene could occur even after eclosion , in a mature fly . These results argue against a developmental requirement for Gr59c . Rather , they are consistent with a physiological role for Gr59c in determining the response profile of the sensillum . 10 . 7554/eLife . 11181 . 015Figure 9 . UAS-Gr59c expression in adult flies is sufficient to restore wild-type responses to ΔGr59c mutant I-a sensilla . All genotypes have identical 2nd chromosomes: ΔGr59c , Gr89a-GAL4 . Flies were subjected to the four indicated temperature regimes . ( a ) GAL80ts parental control flies without UAS-Gr59c display mutant , elevated responses to UMB , TPH , and CAF in all regimes . ( b ) UAS-Gr59c parental control flies without GAL80ts in all cases display wild-type responses to BER , LOB , and DEN . ( c ) Experimental flies ( UAS-Gr59c/GAL80ts ) display mutant responses when kept continuously at GAL80ts permissive temperature ( 25°C ) , indicating that GAL80ts is suppressing UAS-Gr59c expression . However , when GAL80ts is inactivated at any time by shifting flies to higher temperature ( 29°C ) , thus activating UAS-Gr59c , flies display wild-type responses . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 015 What might be the nature of such a physiological role ? Expression of Gr59c increased response to LOB , DEN , and BER in all sensilla tested , and suppressed many other responses ( Figure 6 ) ; deletion of Gr59c decreased response to LOB , DEN , and BER and allowed an increase of other responses . One model consistent with these results is that Gr59c binds to one or more co-factors , perhaps co-receptors , to form a complex that responds to LOB , DEN , and BER . The binding to Gr59c would prevent these co-receptors from binding other Grs to form a complex that responds to other tastants . Thus Gr59c might compete strongly with other Grs for binding to essential co-receptors or other co-factors . This model of Gr59c as a strong competitor might explain why three Grs expressed in I-a ( Gr28b . a , Gr28a , and Gr36a ) conferred few if any responses . To test this model , we expressed these receptors in △Gr59c I-a sensilla , where the putative strong competitor is absent , to determine if they conferred stronger responses . Expression of Gr28a in △Gr59c I-a sensilla conferred response to eight tastants ( Figure 10 ) . By contrast , expression of Gr28a in wild type I-a sensilla increased responses to none of these tastants ( Figure 3a ) . Expression of Gr36a in △Gr59c I-a sensilla conferred responses to 6 tastants , while suppressing responses to two . By contrast , expression of Gr36a conferred no changes in response profiles of wild type I-a sensilla ( Figure 3a ) , or S-a or I-b sensilla ( Figure 4a , b ) . Expression of Gr28b . a conferred an increased response to ARI , which is consistent with the increased response to ARI found in I-b sensilla ( Figure 4a , b; MYR was not tested ) . 10 . 7554/eLife . 11181 . 016Figure 10 . Electrophysiological responses in sensilla ectopically expressing four individual Grs , in ΔGr59c I-a bitter neurons . Novel responses are indicated by 'N' . The experimental genotypes are: ΔGr59c , Gr89a-GAL4; UAS-GrX , except that in the case of Gr22b , the experimental genotype is ΔGr59c; UAS-Gr22b/Gr66a-GAL4 . Asterisks indicate significant changes relative to the ΔGr59c , Gr89a-GAL4; + parental control . p ≤ 0 . 0001 , except that p ≤ 0 . 0005 for Gr36a/SOA and Gr22b/STR; p ≤ 0 . 01 for Gr36a/CAF , Gr22b/TPH; p ≤ 0 . 05 for Gr28b . a/ARI . n ≥ 24 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 016 Gr22b had conferred strong , novel responses to CUC , SOA and AZA in I-a ( Figure 3a ) , and we wanted to determine if in the absence of Gr59c these responses were still conferred . We tested two of these responses , SOA and AZA , and both were conserved ( Figure 10 ) . Responses were also conferred to ESC and SPS , and several responses were suppressed . In summary , removal of Gr59c from I-a sensilla allowed Grs to confer novel responses that were not observed in the presence of Gr59c . These results provide further support for the hypothesis that the function of bitter Grs depends on the molecular context . The results are also consistent with the notion that Grs may interact with each other or compete for common co-factors .
We have carried out a systematic functional analysis of eight Gr genes , most of which have not been examined functionally before . While most previous analysis of bitter Gr genes has been accomplished via loss-of-function analysis , we have examined all eight Gr genes through in vivo expression analysis . We expressed them in bitter-sensitive neurons that either do or do not express them in wild type , i . e . overexpression or ectopic expression , respectively . These bitter-sensitive neurons also express different patterns of other bitter Grs . We have complemented some of this analysis via loss-of-function analysis as well . All of our studies have been carried out in vivo , in a natural laboratory: in taste neurons in gustatory sensilla of the labellum , which have evolved over the course of hundreds of millions of years to sample , detect , and signal taste information . The taste stimuli have been delivered by direct contact in a manner that simulates natural sampling of tastants , and the responses have been measured by recording action potentials , the signals transmitted to the brain . All expressed Grs had functional effects in at least one neuronal type . This uniformity is noteworthy in part because of the diversity of the tested Grs . Although most of them are expressed in the labellum of wild type , as determined by Gr-GAL4 analysis , some are expressed in the legs , the pharynx , the larval taste system , and even the antenna , where taste receptor function is enigmatic ( Weiss et al . , 2011; Kwon et al . , 2011; Kwon et al . , 2014; Scott et al . , 2001; Fishilevich and Vosshall , 2005; Menuz et al . , 2014; Fujii et al . , 2015 ) . Moreover , the tested receptors are dispersed widely not only across tissues and developmental time , but also across the phylogenetic tree of Grs ( Robertson et al . , 2003 ) . The effects of nearly all Grs , moreover , were easily distinguishable . There was one notable exception , however: the profiles of Gr2a and Gr10a were very similar in each of three sensillum types examined ( Figures 3 , 5b , d ) . Gr2a and Gr10a are not recently duplicated genes , but rather are on distant branches of the Gr phylogenetic tree ( Robertson et al . , 2003 ) . We have previously found that similar odor response spectra are conferred by the D . melanogaster and D . pseudoobscura orthologs of Or71a , which contain only 59% amino acid sequence identity ( Ray et al . , 2008 ) . Gr2a and Gr10a provide an even more extreme case: they exhibit only 32% identity . A key finding of this study is that expression of individual Grs produced different effects in different bitter taste neurons ( Figure 11 , Figure 11—figure supplement 1 ) . This finding supports an emerging theme in the field of taste: that bitter taste neurons are heterogeneous ( Weiss et al . , 2011; Meunier et al . , 2003; French et al . , 2015; Glendinning et al . , 2006; Glendinning et al . , 2002 ) . Historically , a long-held concept in the field has been that all bitter taste neurons are identical , and that they function identically to warn the animal indiscriminately of toxic compounds ( Freeman and Dahanukar , 2015 ) . An emerging concept is that bitter taste neurons are distinguishable , both in terms of response profile and receptor expression ( Weiss et al . , 2011; Ling et al . , 2014; Meunier et al . , 2003 ) . The differing effects of expressing an individual Gr in varying neuronal contexts adds another line of evidence to support the view that different bitter taste neurons are functionally distinct , thereby endowing the system with richer coding capacity . 10 . 7554/eLife . 11181 . 017Figure 11 . Gr ectopic expression produces receptor-specific and neuron-specific effects on the response profiles . Effects are delineated for each tastant in Figure 11—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 01710 . 7554/eLife . 11181 . 018Figure 11—figure supplement 1 . Summary of effects of UAS-Gr expression on responses to tastants in different sensillum types , based on comparison with tested control lines . '+' indicates an increase in the level of an endogenous response . '*' indicates a novel response not observed in parental control lines examined . '-' indicates suppression of an endogenous response . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 018 Expression of Grs produced three kinds of effects: a novel response to a particular tastant , not observed in either parental control; an increased response to a tastant , greater than that observed in controls; a decreased response to a tastant , i . e . suppression . All three effects were observed broadly , and most receptors exhibited all three kinds of effects ( Figure 11 , Figure 11—figure supplement 1 ) . However , there was specificity to the effects , in that different receptors conferred varying effects in different neurons or in response to different tastants . The degree of diversity was striking , in the case of both sensilla and receptors . For example , in the S-a sensillum , expression of Gr22b increased some responses and decreased others; Gr58c only decreased responses; Gr36a had no effects on any responses . In the case of the Gr10a receptor , which is expressed in the antenna as well as a larval taste organ , expression in S-a sensilla increased some responses and decreased others; expression in I-a produced increased responses and novel responses but no decreased responses; expression in I-b had no effects . According to the simplest model of Gr function , one might have expected that expression of a Gr would simply add to the response profile of the neuron . If the expressed Gr responded to a tastant that the endogenous Grs in the native neuron do not respond to , then a novel response would be added to the profile . If the expressed Gr responded to a tastant that the endogenous Grs respond to , then the responses to that tastant would then increase . The effects observed in this study are difficult to reconcile with this simple model of Gr function . A more complex model is required to explain , for example , the findings that: i ) expression of some Grs decreased endogenous responses; ii ) deletion of Gr59c increased the response to certain tastants; iii ) expression of some Grs in neurons in which they are endogenously expressed caused novel responses; iv ) the effects of expression of a Gr differed among neuron types ( Figures 11 , 12 , Figure 11—figure supplement 1 ) . Below we consider mechanisms that could explain these findings . 10 . 7554/eLife . 11181 . 019Figure 12 . Findings and models . Four findings of this study are indicated , along with one possible model to explain each . i ) Expression of a Gr , indicated by the blue sphere , decreases the response to a tastant , represented by the traces below . One possible model is that the expressed Gr ( blue sphere ) interacts with another , active Gr ( green ) and inhibits it ( represented by its conversion from a green , active Gr to a red , inactive Gr ) . ii ) Deletion of Gr59c ( blue sphere with X ) leads to an increased response . One possible model is that in wild type , the Gr inhibits an endogenous Gr ( red ) . Removal of the Gr allows the inactive Gr ( red ) to become active ( green ) . iii ) Overexpression of a Gr , in a neuron that contains the Gr endogenously , induces a response that is not observed in wild type . One model is that above a certain concentration threshold of the Gr , it is able to bind and convert another Gr from an inactive ( red ) to an active ( green ) form . iv ) Expression of a Gr in two different neurons , 'A' and 'B , ' produces different results . The neuron at left shows an increase in response to a tastant whereas the neuron at the right shows a decrease . One model is that in the A neuron , the expressed Gr ( blue ) binds to an inactive Gr ( red ) specific to neuron A , and activates it ( green ) . By contrast , in the B neuron , the expressed Gr ( blue ) binds to a different Gr ( green ) specific to neuron B , and inactivates it ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11181 . 019 The diverse effects we have observed from expressing Grs may have profound evolutionary implications . A major challenge for Drosophila and other organisms is to respond to changes in the chemical environment . For example , as new plants move into the habitat of a fly , or as a fly moves into a new habitat , there are novel opportunities for nutrition but also novel dangers of toxicity . How can the fly adapt so as to exploit the opportunities but avoid the dangers ? The ability of its taste system to encode and interpret new chemical signals is crucial . A classic means of detecting new signals is via changes in the amino acid sequence of taste receptors ( Ueno et al . , 2001 ) . A change in the binding site of a bitter receptor , for example , may allow the detection of a new toxin . Our results suggest a rich , alternative means of generating variation in chemosensory signaling . We have found that changes in the expression of a taste receptor can generate a wide variety of effects: increased responses , decreased responses , and novel responses . Accordingly , simple alterations in regulatory sequences that govern the cellular specificity of expression , or the level of expression , could change the response specificities of taste neurons in dramatic and beneficial ways . Thus , the organization of the taste system—featuring the coexpression of many Grs in individual neurons—may provide numerous degrees of freedom for altering the response patterns of the neurons , via either the gain or loss of receptor function . Such alterations could meet a specific environmental challenge , or could in a more general sense expand the potential of the system to encode and perhaps distinguish among bitter compounds of differing toxicity or significance . A network of coexpressed Grs , sensitive to relative levels of expression , could also provide a means of modulating taste sensitivity in response to short-term changes in either the environment or the internal state of the fly . We found that alterations in the expression of Gr59c in the mature fly induced changes in taste sensitivity ( Figure 9 ) . Thus , short-term changes in Gr transcription or post-transcriptional processing could in principle be a mechanism of altering taste response to meet the short-term needs of a hungry or satiated fly ( Prestia et al . , 2011; Zhou et al . , 2014 ) . To investigate whether alterations in Gr expression may provide a mechanism for the evolution or modulation of taste responses , it will be of interest to examine the behavioral consequences of such alterations . For example , does the expression of Gr22b in I-a , I-b , or S-a sensilla , which produces increased or novel electrophysiological responses to SOA , increase the fly’s behavioral sensitivity or degree of aversion to SOA ? Does expression of Gr58c in I-b and S-a , which suppresses electrophysiological responses to UMB , decrease the fly’s behavioral response to UMB ? Increased aversion to a bitter tastant in a food source could be adaptive in environments that contain a particularly rich selection of more beneficial food sources . Decreased aversion to a bitter compound in a food source might be adaptive following the starvation of a fly with few alternatives . In a more general sense , the functional organization of the taste system differs from that of the olfactory system in that the 60 Gr genes are more highly coexpressed in a smaller number of neuronal types than the 60 Or genes . This pattern of organization may in principle entail some loss of discriminatory power , but it could provide a degree of evolutionary and regulatory flexibility that may serve the needs of the animal well . Finally , we note that the present analysis of eight Gr genes lays a genetic foundation for a detailed biochemical analysis . It will be of interest , for example , to test the possibility of association between tagged Grs in vivo . The results also provide a framework and a focus for further testing of specific hypotheses about the functions of individual Gr genes , via the construction of a series of genotypes with CRISPR/Cas technology ( Bassett et al . , 2013; Gratz et al . , 2013 ) .
Flies were grown on standard cornmeal-agar medium . Flies used for physiological recordings and imaging were grown at 25°C in a humidity-controlled incubator . Only females , aged 6–8 days , were used for electrophysiological recording . Male and female flies , aged 7–14 days , were used for imaging . All transgenic genotypes tested were homozygous , except where indicated . Gr-GAL4 drivers ( Gr89a-GAL4 , Gr66a-GAL4 , Gr59c-GAL4 , Gr28a-GAL4 , Gr28b . aGAL4 , Gr22b-GAL4 ) were as described in Weiss et al . ( 2011 ) ; Gr28a-GAL4 was provided by H . Amrein . UAS-Gr lines were generated by amplification of Gr coding regions from Canton-S cDNA prepared from proboscis , legs , and larvae . UAS-Gr2a , -Gr22b , -Gr28a , -Gr36a , -Gr59c were cloned into the pUAST expression vector and inserted into the genome using random transposon integration . UAS-Gr10a , -Gr28b . a , and -Gr58c were cloned into the pBI-UASCG expression vector and inserted into the genome with phiC31 site-specific integration into the following strains: 9725 ( Gr10a ) , 9744 ( Gr28b . a ) , and attP2 ( Gr58c ) ( Groth et al . , 2004; Pfeiffer et al . , 2010 ) . Since UAS-Gr2a and UAS-Gr10a gave very similar results , we verified with PCR that the lines containing these constructs in fact contained the expected transgenes . The ΔGr59c line was generated using FLP-FRT mediated recombination ( Parks et al . , 2004 ) between two FRT-bearing piggybac transposons from the Exelixis collection ( pBacf04393 and pBacf03881 ) flanking a ~17 kb genomic region encompassing the Gr59c locus ( see Figure 7—figure supplement 1 ) . The deletion was confirmed through PCR and is homozygous viable . The ΔGr59c line was backcrossed into w- Canton-S flies ( wCS ) for 7 generations . We note that a mutant phenotype different from that shown in Figure 7 was observed in the original Exelixis w1118 isogenic background before the outcrossing to wCS; the phenotype shown in Figure 7 was also observed in all other genetic backgrounds and contexts tested ( n>10 ) . Tastants were obtained at the highest available purity from Sigma-Aldrich . All tastants were dissolved in 30 mM tricholine citrate ( TCC ) , an electrolyte that inhibits the water neuron ( Wieczorek and Wolff , 1989 ) . Tastant solution aliquots were stored at −20°C long-term and kept at 4°C while in use , for no more than a week . See Table 1 for concentrations used . Single-sensillum recordings were performed as described in Delventhal et al . ( 2014 ) . To quantify responses , the number of action potentials ( spikes ) was counted over a 500 ms period , starting 200 ms after contact . A high salt stimulus ( 400 mM NaCl ) , which activates the bitter neuron ( Hiroi et al . , 2003; Meunier et al . , 2003 ) was used as a positive control at the beginning and end of the recording session for each sensillum to ensure that the bitter GRN was functional . All recordings from sensilla that displayed an average NaCl response of less than 10 spikes/s at the beginning or end of a recording session were discarded . No more than eight tastants were tested on an individual sensillum of a given fly , with 2–3 minutes between presentations . Whole labella were dissected and their GFP fluorescence was imaged using a Zeiss LSM510 confocal microscope under 40X magnification . Images were processed using ImageJ . All error bars represent S . E . M . Experimental genotype response profiles were compared to the GrX-GAL4 ( and UAS-GrX , when available ) parental control response profiles with a two-way ANOVA and a Bonferroni multiple comparisons correction , using Prism software . A one-way ANOVA was performed within the parental control genotype to determine which responses were significantly different from the TCC diluent control . Responses that were determined to be significantly greater than TCC within the control genotype were then designated as 'increased' when elevated in experimental genotypes; responses that did not differ from TCC in the control response profile were designated as 'novel' when found to be elevated in experimental genotype profiles . | Insects and other animals use their sense of taste to tell if their food is safe to eat . Plant toxins , for example , often have a bitter flavor that animals can detect and avoid . Fruit flies have many bitter-sensitive nerve cells , but it is not known how the receptors on these nerve cells signal the detection of bitter-flavored compounds . Delventhal and Carlson have now used fruit flies to investigate how taste receptors of the so-called Gustatory receptor family detect bitter flavors . The experimental approach involved genetically modifying four different types of nerve cells that sense bitter compounds so that they produced higher levels of particular taste receptors than normal . Then , the flies were exposed to a range of bitter compounds while the electrical activity of each cell was measured . The analysis involved about 600 combinations of receptors , nerve cells and compounds . In some bitter-sensing nerve cells , increasing the number of taste receptors increased the cell’s responsiveness to bitter compounds . However , in other nerve cells , similar modifications suppressed an existing response or resulted in a new response . Delventhal and Carlson propose that these results suggest the specific response of a bitter-sensing nerve cell depends on the interactions between its different taste receptors . Furthermore , the ability of receptors to compete , inhibit or activate each other in different ways could have implications for evolution . For example , such flexible interactions might allow a taste system to evolve new , enhanced or diminished responses to new food sources and tastes in a changing environment . It now remains to be investigated how such receptor interactions take place at a molecular level . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Bitter taste receptors confer diverse functions to neurons |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.